Introduction
A common theory becomes informative when one choice of dynamical data restricts several observations at once. This requirement is familiar in physics: a model of matter should relate its motion, fluctuations and response through the same interactions. A related question arises in finance, where the prices of traded assets constrain the valuation of risks that cannot be replicated. In both settings, the central task is to identify what the stated dynamics and constraints determine, and what additional information an observation requires. The present programme uses this question to connect a geometric field construction with explicit models of relaxation and constrained probabilistic selection.
Research question and physical motivation
The four interactions conventionally described as gravitation, electromagnetism, the weak interaction, and the strong interaction are represented at presently accessible scales by theories with markedly different mathematical formulations. General relativity describes gravity through spacetime geometry, whereas the Standard Model describes the non-gravitational interactions through gauge fields and matter representations. The coexistence of these descriptions is experimentally successful over their respective domains, but their conceptual combination leaves a basic structural question: can the four sectors be regarded as different effective manifestations of one underlying dynamical principle rather than as theories placed side by side?
The question is addressed through a common variational and stochastic structure. Its guiding idea is that motion, fluctuation, relaxation and selection can be understood as connected operations on a specified space of possibilities. Geometry determines admissible changes; a reference law specifies their stochastic dynamics; constraints and observations determine how histories are selected or reweighted. The same questions also occur in communication, population evolution and decision-making, where the variables have different meanings. Their comparison can reveal common structures while making the additional physical, biological or economic conditions explicit.
For the physical realization, we fix a configuration space, geometric action, reference measure, mobility, noise law, objective and observation maps. The scientific question is which relationships these shared data impose between responses that would otherwise be specified separately. The following finite models explain this question constructively; the later geometric and stochastic developments supply its field realizations and effective limits.
To make the proposed connection testable rather than merely schematic, we also specify one finite four-dimensional interacting branch. Its link variables, regulated coframe coordinates, and order-parameter variables enter one positive norm-square functional; the same frozen registry determines the finite generator, Gibbs reference, and entropy-production identity. This branch is the smallest representative on which the unified idea can be checked without silently importing a second action. A finite Euclidean regulator is a controlled theorem about that branch, whereas the Lorentzian constraint algebra, regulator-independent limit, and interacting quantum theory require separate proofs.
The physical motivation is more specific than a search for formal similarity. A unified description must explain how four ingredients cooperate: a variational principle selects admissible field changes, a reference measure gives those changes a probabilistic meaning, a positive response operator describes relaxation, and an observation map determines which part of the dynamics is measurable. These roles are often separated when gravity, gauge fields, matter, and nonequilibrium thermodynamics are written in different languages. The proposed framework tests whether they can be generated from one declared set of geometric and probabilistic data while preserving the distinctions that each sector requires.
The thermodynamic sector provides the decisive test of this proposal. The conservative Lorentzian branch describes field evolution through a generally indefinite action, whereas entropy production uses a positive quadratic form on dissipative directions. Consequently, the second law cannot be obtained by relabelling a Lorentzian action as a free energy. It must emerge after the field variables, reference measure, bath coupling, and physical projection have been specified. This is why the manuscript treats the finite-regulator construction as an explicit bridge: it makes the positivity, the Gibbs measure, the noise covariance, and the heat convention visible before any continuum or quantum claim is made.
The smallest representative of this idea is the four-dimensional interacting branch introduced later in the paper. Link variables, regulated coframe coordinates, and an order parameter enter one shared functional. Its finite generator has a unique Gibbs law and an exact Dirichlet dissipation identity. The branch therefore supplies an explicit finite test of the proposed architecture, whose physical interpretation still requires matching. It also clarifies the logical boundary of the result: a finite Euclidean theorem does not by itself establish the full Lorentzian constraint algebra, a regulator-independent field limit, or a positive interacting quantum Hilbert space. Those requirements are stated as subsequent tests, so that the physical motivation and the mathematical claims remain aligned.
§ 14 develops the finite construction further. It gives a response-based inverse map, an explicit energy-scale ambiguity, and a constrained nonlinear example. An alternative symmetric temporal regulator then gives a positive interacting quantum model at fixed spatial resolution. The distinction between this model, physical gravitational constraints, and a spatial continuum limit determines the order of the remaining analysis.
The finite extensions also address how the common structure can be measured. A locally rotation-covariant logarithmic coframe model combines finite inverse moments with an explicit Gibbs generator (§ 14.10 ). Exact temporal endpoint normalization then distinguishes a removable row factor from the field-dependent physical inertia. At a fixed observation scale, determinant bounds quantify the probability of nearly singular averaged geometry. Finally, the retained gauge–scalar interaction gives an oscillatory backreaction and an explicit finite-time trajectory bound, illustrated with the same parameters in the coupled and frozen-scalar models (§ 14.14 ). These results connect symmetry, equilibrium, and response in calculable examples. Their model choices are stated so that finite constructions can be compared without identifying them with a common continuum theory by assumption.
1.2. Four organizing principles. The central proposal is to organize a theory by the possibilities for change, the law of its fluctuations, and the conditions that select or reweight its histories. Mechanical motion, relaxation, inference and selection then become different operations on specified dynamical data. This proposal is expressed by four principles.
Admissible change has a structure. Before assigning a cost, specify what can change and which relations must be preserved. In the field construction, a connection, its curvature, matter representations and constraints provide this structure. On a finite network, the corresponding data are states and allowed transition edges. An invariant pairing or a mobility determines how admissible changes are compared. These choices make symmetry and conservation part of the dynamics from the outset. They also explain why a displacement that is possible in one effective description may be forbidden in another.
Motion, fluctuations and relaxation use compatible data. A reference stochastic law assigns probabilities to histories. Its conservative drift can transport a state without lowering its energy; its dissipative drift and noise determine exchange with the environment. When a thermal realization is intended, the same reservoir scale and mobility must enter both damping and fluctuations. Equilibrium is then a stationary law, while a microscopic history continues to move. The conservative limit retains the transport law and removes damping and its matched noise together. This identifies the physical conditions under which reversible mechanics and irreversible relaxation belong to one description.
Selection compares histories relative to a reference. Boundary information, an observed signal, reproductive growth or an admissible control can change the weight assigned to a history. A relative-entropy cost expresses how far this new law departs from the reference law. With a declared dimensionless path score G G G , the elementary variational problem is
inf Q ≪ P { D ( Q ∥ P ) − E Q G } . (1) \inf_{Q \ll P}\{\mathcal{D}(Q\Vert P)-\mathbb{E}_{Q}G\}. \tag*{(1)} Q ≪ P inf { D ( Q ∥ P ) − E Q G } . (1)
Its minimizer is an exponentially reweighted path law. Section 2 derives this statement and shows how growth, Bayesian updating and a martingale constraint lead to different realizations. In the quadratic drift-control branch, the Girsanov identity supplies this relative-entropy geometry. The full joint residual objective retains its additional geometric and nonlinear terms. A mechanical action, in contrast, selects stationary trajectories through a first variation; its extremum can be a saddle. The common variational organization keeps these roles explicit.
A change of description must transport the law and its meaning. An observation or reduction maps the full state to a smaller description. The induced noise, memory, constraints and time scale must follow that map. A coarse observation can lose information and need not remain Markovian. Likewise, a distribution of present population types and a distribution along selected ancestral histories answer different questions even when one operator generates both. A common theory must specify these relations between descriptions, so that its parameters cannot be adjusted independently at each stage.
These principles shift the explanatory question. In addition to asking which equation fits each domain, we ask which parts of its motion, statistics, response and selection are fixed once the admissible changes and their reference law have been chosen. A shared mathematical structure can organize several disciplines without assigning identical physical meanings to their variables. A stronger physical unification arises when a specified microscopic realization also fixes the maps and common coefficients connecting its sectors.
Representative models and contributions
The argument begins with a finite model that can be solved completely. One transition generator gives a thermal relaxation law. Weighting its histories by reproductive growth gives a mutation–selection equation; weighting terminal states by an observation likelihood gives Bayesian updating. A financial example imposes a discounted-price martingale constraint and obtains a different probability law. An underdamped oscillator then exhibits conservative transport, matched noise and damping, and a fluctuation–response relation in one physical model. These examples establish the precise common operations before the field construction introduces its larger configuration space.
The contribution of this organization is a chain of explicit dependencies. For the finite example, an equilibrium fraction and one relaxation rate fix the transition generator; a separately specified growth difference then fixes the population equilibrium and a changed spectral rate. For the oscillator, the same kinetic and reservoir parameters fix both response and fluctuations. The proofs identify which relations follow from shared inputs, which require a further constraint, and which are lost when an observation is changed. The underlying entropy, selection and change-of-measure methods have established precedents [48], [32], [90], [96], and [91]. Their role here is to give a transparent finite realization of the proposed organization and a concrete guide to its field-theoretic use.
The field construction asks a more specific question: which gravitational and internal equations follow from a common geometric action, and how closely do their selected solutions agree with established physics? The Cartan curvature contains the coframe, so its squared norm generates an Einstein term through a cross contraction. The internal variation generates a constitutive gauge equation whose canonical form depends on its kinetic coefficients. § 7 puts these two mechanisms in one quantitative comparison. It derives the full coefficient-normalized gauge residual, separates equation consistency from solution stability, and demonstrates why resolving a mesh cannot remove a fixed physical correction. This turns the common variational origin into a sequence of explicit, testable requirements.
Quantitative consequences of shared inputs. Four calculations make the connection between a common structure and an observable more precise. Finite response measurements give an exact joint parameter set and a prediction bound that preserves their common calibration error (Section 10.2). An outgoing radial problem turns the prescribed source into a controlled surface response and radiated power (Section 22.5). An exactly solvable hidden-variable model measures the error caused by discarding memory and distinguishes two inequivalent short-memory scalings (Section V.§ 9 ). Finally, an incomplete market shows how a declared reference law selects one price from the no-arbitrage interval and which changes of that reference remain unidentifiable (§ 2.6 ). The common contribution is a calculable dependence between inputs and outputs. The entropy projection, Gaussian dynamics and coercive variational estimates use established methods; the formulas specify their role and limits in the chosen realizations.
The remainder develops three connected results. First, one regulated geometric action, reference measure and mobility give coupled field equations and a stochastic law. Under the stated matching and low-energy conditions, these recover Einstein–Yang–Mills and matter equations; the thermal and fluid projections have their corresponding balance laws. Second, coefficient identification and target-observable theorems determine what calibrated shared inputs actually fix. Third, finite quantum, scalar-source and laboratory models expose relations that can be tested beyond the observations used for calibration. The quantum and continuum completions retain their separate hypotheses.
A geometric control construction sharpens this separation. Appendix R proves the correspondence between a quadratic ergodic control problem, a positive ground state and its reversible diffusion on a fixed regulated configuration space. It also constructs dynamics tangent to regular constraints, gives a quantitative local bound for an independently matched Einstein branch, and identifies a positive crossing-kernel condition for Euclidean reconstruction. A curved product-space example shows why the control equation alone cannot select the Einstein field equation. The finite gauge calculation tests the control mechanism. Appendix S then verifies an exact Einstein branch of the independent-connection Cartan sector, computes the quadratic response error from its specified coefficients, and constructs the spatial ultraviolet limit of free physical gravitational waves. Between these constructions, an exact nonlinear Einstein initial-data family permits arbitrary smooth spatial rotation of its shear. Its stochastic sampling requires explicit curvature-drift terms; tangent noise alone has a computable nonzero constraint defect. It also proves that the resummed four-derivative metric kernel fails physical-time reflection positivity. This identifies a concrete obstruction that an interacting completion must overcome.
Time supplies a further test of this organization. A probability law can describe relaxation without specifying a physical light cone, and an equilibrium distribution can remain unchanged when all transport rates are rescaled. Section 26.3 separates these questions through an exact clock-scaling identity and a quadratic hyperbolicity criterion. Section 26.4 then derives the common characteristic set of the matched scalar, gauge, Dirac and Einstein perturbation blocks. The calculation uses their physical kinetic terms and gauge constraints; it does not assume that the ungauged field determinants are nonzero. A synthetic wave-packet comparison distinguishes characteristic propagation from group delay, and an independent-lag experiment illustrates how dynamical calibration supports a prediction beyond an equilibrium histogram.
This development has applications wherever admissible dynamics, uncertainty and selection must be considered together: physical transport, communication through fluctuating environments, biological type selection and constrained financial decisions. Extending the same approach to social evolution requires explicit actors, transitions, resources and institutional feedback; a historical trend alone does not specify those quantities. The finite models below provide a construction for stated assumptions, rather than a universal social objective.
§ 2 gives the representative models, their proofs and a synthetic comparison. § 3 introduces the master fields and objective. The subsequent sections develop equilibrium and effective field equations. The controlled comparison in § 7 then explains how their residuals and selected solutions are related. The matter sources lead into selection, physical matching, quantum structure and fluid transport. The conclusion returns to the common principles and the next physical questions. Appendix AH gives the detailed connections to the individual constructions, together with their assumption and notation tables.
The relation between equilibrium, geometry and time is sharpened by two explicit constructions below. A spatially correlated logarithmic coframe model has a nonlinear infinite-mode limit with uniform inverse moments (Section 27). A separate calibration result shows which additional transition measurement fixes the dynamical scale left undetermined by finite Gibbs populations (Section 29). Both constructions state their chosen reference laws and their relation to the full gravitational problem.
From shared principles to shared predictions. The eight companion studies follow different parts of one dependence chain. The particle construction fixes a representation and its matching rules; the fluid and second-law analyses identify the meaning of transport and heat; the quantum construction specifies the additional state and observable structure; and the cosmological and black-hole studies turn selected fields into source and detector responses. A coefficient can be transferred between these studies only after its field normalization, physical units, background and observation rule have been matched. This is where a common variational origin becomes more restrictive than a collection of separately calibrated models.
The scalar example makes that restriction concrete. Elementary coupling is only the first step: the material response and the exterior boundary determine the field produced by a body. Proposition 109 bounds this response even when the local restoring term vanishes or is weakly negative. It uses the energy required to continue the field into the exterior. The resulting residual bound separates an inaccurate numerical solution from a large physical response. This local advance supports the source calculation; the frequency-dependent response and the independently matched matter coefficients remain necessary for a binary-radiation prediction. Proposition 110 takes a further step by bounding an outgoing monopole response on a specified frequency band. Its scope is deliberately explicit: the binary dipole channel and the coupled matter response remain separate calculations.
The representative models now turn these principles into calculations. Their order follows the information needed for prediction: first a reference law, then a physical or economic constraint, and finally an observable that was not used to select the law. The financial example is expanded to an incomplete market so that the selection criterion has a visible role beyond the no-arbitrage equations themselves.
A common mechanism in representative finite models
A small model can reveal the organizing idea more clearly than a long list of applications. Consider finitely many states. The same transition law can describe a relaxing physical system, a changing population of types, or an unobserved environment in a communication channel. What changes between these interpretations is specified below: the observed quantity, the weight assigned to a history, or an additional constraint. The finite model realizes the stochastic and selection operations of the programme. Its states are not identified with a projection of the full gauge–gravity construction.
A reference law and the direction of relaxation
Let L L L be the row generator of an irreducible continuous-time Markov chain: L i j ≥ 0 L_{ij} \ge0 L ij ≥ 0 for i ≠ j i \ne j i = j and ∑ j L i j = 0 \sum_j L_{ij} = 0 ∑ j L ij = 0 . Column distributions obey p ˙ = L T p \dot{p} = L^{\mathsf{T}}p p ˙ = L T p . Assume detailed balance with a positive probability vector π \pi π :
a i j : = π i L i j = π j L j i ( i ≠ j ) . a_{ij} := \pi_i L_{ij} = \pi_j L_{ji} \qquad(i \ne j). a ij := π i L ij = π j L j i ( i = j ) .
Write u i = p i / π i u_i = p_i/\pi_i u i = p i / π i . Then p ˙ i = ∑ j ≠ i a i j ( u j − u i ) \dot{p}_i = \sum_{j \ne i} a_{ij}(u_j-u_i) p ˙ i = ∑ j = i a ij ( u j − u i ) . Consequently,
d d t D ( p ∥ π ) = − ∑ i < j a i j ( u i − u j ) ( log u i − log u j ) ≤ 0. (2) \frac{d}{dt}D(p\mathbin{\|}\pi) = -\sum_{i<j}a_{ij}(u_i-u_j)(\log u_i-\log u_j) \le0. \tag*{(2)} d t d D ( p ∥ π ) = − i < j ∑ a ij ( u i − u j ) ( log u i − log u j ) ≤ 0. (2)
Indeed, differentiation gives ∑ i p ˙ i log u i \sum_i \dot{p}_i \log u_i ∑ i p ˙ i log u i , because ∑ i p ˙ i = 0 \sum_i \dot{p}_i = 0 ∑ i p ˙ i = 0 . Pairing the two orientations of every edge gives the displayed sum. Each summand is nonnegative because the logarithm is increasing. Positive initial probabilities suffice; irreducibility also makes the law positive at every strictly positive time for a general initial law.
The dissipation has a geometric form. Define the logarithmic mean Λ ( x , y ) = ( x − y ) / ( log x − log y ) \Lambda(x,y) = (x-y)/(\log x-\log y) Λ ( x , y ) = ( x − y ) / ( log x − log y ) , with Λ ( x , x ) = x \Lambda(x,x)=x Λ ( x , x ) = x , and let e i e_i e i be the coordinate vectors. The positive semidefinite operator
K ( p ) = ∑ i < j a i j Λ ( u i , u j ) ( e i − e j ) ( e i − e j ) T K(p) = \sum_{i<j}a_{ij}\Lambda(u_i,u_j)(e_i-e_j)(e_i-e_j)^{\mathsf{T}} K ( p ) = i < j ∑ a ij Λ ( u i , u j ) ( e i − e j ) ( e i − e j ) T
satisfies p ˙ = − K ( p ) ∇ p D ( p ∥ π ) \dot{p}=-K(p)\nabla_pD(p\mathbin{\|}\pi) p ˙ = − K ( p ) ∇ p D ( p ∥ π ) . It acts in the simplex tangent space because K ( p ) 1 = 0 K(p)\mathbf{1}=0 K ( p ) 1 = 0 . Thus the transition network and its conductances fix a relaxation geometry as well as a stationary distribution.
For a thermal interpretation with fixed energies E i E_i E i and θ = k B T > 0 \theta=k_{\mathrm{B}}T>0 θ = k B T > 0 , set π i = Z − 1 e − E i / θ \pi_i=Z^{-1}e^{-E_i/\theta} π i = Z − 1 e − E i / θ . Then θ D ( p ∥ π ) = ∑ i p i E i + θ ∑ i p i log p i + θ log Z \theta D(p\mathbin{\|}\pi)=\sum_i p_iE_i+\theta\sum_i p_i\log p_i+\theta\log Z θ D ( p ∥ π ) = ∑ i p i E i + θ ∑ i p i log p i + θ log Z is excess Helmholtz free energy. Equation (2) is its isothermal relaxation law. Temperature and reservoir matching supply this physical meaning; for an arbitrary transition network the same equation is a probability identity. The energy values fix π \pi π , whereas the conductances also fix the time scale.
One variational identity for a reweighted law.
Let P \mathbb{P} P be the reference path law on a finite horizon. For a bounded dimensionless path score G G G , put Z G = E P e G Z_G = \mathbb{E}_{\mathbb{P}}e^G Z G = E P e G and define d P G = e G d P / Z G d\mathbb{P}_G = e^Gd\mathbb{P}/Z_G d P G = e G d P / Z G . For any Q ≪ P \mathbb{Q} \ll\mathbb{P} Q ≪ P with finite relative entropy,
D ( Q ∥ P ) − E Q G = D ( Q ∥ P G ) − log Z G . (3) D(\mathbb{Q}\Vert\mathbb{P})-\mathbb{E}_{\mathbb{Q}}G = D(\mathbb{Q}\Vert\mathbb{P}_G)-\log Z_G. \tag*{(3)} D ( Q ∥ P ) − E Q G = D ( Q ∥ P G ) − log Z G . (3)
This follows by substituting log ( d P G / d P ) = G − log Z G \log(d\mathbb{P}_G/d\mathbb{P}) = G-\log Z_G log ( d P G / d P ) = G − log Z G in the definition of relative entropy. Nonnegativity of relative entropy proves that P G \mathbb{P}_G P G is the unique minimizing law in (1). This standard variational identity underlies entropy projections and conditioned-process constructions [48] and [32].
For a state growth rate f i f_i f i and a positive terminal likelihood ℓ i \ell_i ℓ i , choose
G = ∫ 0 T f J t d t + log ℓ J T , B = L + diag f . (4) G = \int_0^T f_{J_t}\,\mathrm{d}t+\log\ell_{J_T}, \qquad B=L+\operatorname{diag}f. \tag*{(4)} G = ∫ 0 T f J t d t + log ℓ J T , B = L + diag f . (4)
Rates L i j L_{ij} L ij and f i f_i f i have units of inverse time; ℓ i \ell_i ℓ i is a dimensionless likelihood. The forward and backward equations are
n ( t ) = e B T t p 0 , h ( t ) = e B ( T − t ) ℓ , Z G = n ( t ) T h ( t ) . n(t)=e^{B^{\mathsf{T}}t}p_0,\qquad h(t)=e^{B(T-t)}\ell,\qquad Z_G=n(t)^{\mathsf{T}}h(t). n ( t ) = e B T t p 0 , h ( t ) = e B ( T − t ) ℓ , Z G = n ( t ) T h ( t ) .
The partition function is independent of the intermediate time because n ˙ = B T n \dot n=B^{\mathsf{T}}n n ˙ = B T n and h ˙ = − B h \dot h=-Bh h ˙ = − B h . The marginal of the fixed-horizon reweighted law is
p i G , T ( t ) = n i ( t ) h i ( t ) Z G , L i j G ( t ) = L i j h j ( t ) h i ( t ) ( i ≠ j ) . (5) p_i^{G,T}(t)=\frac{n_i(t)h_i(t)}{Z_G},\qquad L_{ij}^G(t)=L_{ij}\frac{h_j(t)}{h_i(t)}\quad(i\ne j). \tag*{(5)} p i G , T ( t ) = Z G n i ( t ) h i ( t ) , L ij G ( t ) = L ij h i ( t ) h j ( t ) ( i = j ) . (5)
To obtain the jump rate, condition on a transition i → j i\to j i → j during a short interval: its reference weight is L i j d t L_{ij}\,\mathrm{d}t L ij d t , and its remaining future weight changes from h i h_i h i to h j h_j h j . The diagonal is minus the sum of these off-diagonal rates. This is the finite-time Doob transform.
The initial law in this unrestricted variational problem is p 0 i h i ( 0 ) / Z G p_{0i}h_i(0)/Z_G p 0 i h i ( 0 ) / Z G . If the initial distribution must remain p 0 p_0 p 0 , the minimization is instead performed separately conditional on J 0 = i J_0=i J 0 = i . Its density is e G / h J 0 ( 0 ) e^G/h_{J_0}(0) e G / h J 0 ( 0 ) , its value is − ∑ i p 0 i log h i ( 0 ) -\sum_i p_{0i}\log h_i(0) − ∑ i p 0 i log h i ( 0 ) , and its jump rates still have the form in (5). The fixed-initial diffusion control problem uses this latter convention. This distinction preserves the initial condition when applying the path principle.
Population selection and information updating.
Suppose types change with generator L L L and type i i i has a constant net reproductive rate f i f_i f i . In a continuous-time multitype branching realization, the expected counts satisfy n ˙ = L T n + diag ( f ) n \dot n=L^{\mathsf{T}}n+\operatorname{diag}(f)n n ˙ = L T n + diag ( f ) n . Normalizing these expected counts gives
p i ( t ) = n i ( t ) ∑ j n j ( t ) , p ˙ i = ( L T p ) i + p i ( f i − f ˉ ) , f ˉ = ∑ j p j f j . (6) p_i(t)=\frac{n_i(t)}{\sum_j n_j(t)},\qquad\dot p_i=(L^{\mathsf{T}}p)_i+p_i(f_i-\bar f),\qquad\bar f=\sum_j p_jf_j. \tag*{(6)} p i ( t ) = ∑ j n j ( t ) n i ( t ) , p ˙ i = ( L T p ) i + p i ( f i − f ˉ ) , f ˉ = j ∑ p j f j . (6)
The subtraction of f ˉ \bar f f ˉ follows by differentiating the normalization; no population-level foresight is needed. For a finite random population, a ratio of expected counts is not the expectation of the random fraction. A deterministic population interpretation therefore uses an appropriate large-population limit or declares the expected-count normalization as the observable.
The selection part is ( diag p − p p T ) f (\operatorname{diag} p - pp^{\mathsf{T}})f ( diag p − p p T ) f . It is a gradient of f ˉ \bar{f} f ˉ in the Fisher geometry on the interior simplex when f f f is fixed. Its squared gradient norm is Var p ( f ) \operatorname{Var}_{p}(f) Var p ( f ) . This established connection between selection and information geometry is discussed in [90] and [96]. Mutation adds a separate transport contribution. In particular,
d f ˉ d t = p T L f + Var p ( f ) , d D ( p ∥ π ) d t = − D L ( p ) + Cov p ( f , log ( p / π ) ) , (7) \frac{d\bar{f}}{dt}=p^{\mathsf{T}}Lf+\operatorname{Var}_{p}(f), \qquad\frac{dD(p\Vert\pi)}{dt}=-D_{L}(p)+\operatorname{Cov}_{p}(f,\log(p/\pi)), \tag*{(7)} d t d f ˉ = p T L f + Var p ( f ) , d t d D ( p ∥ π ) = − D L ( p ) + Cov p ( f , log ( p / π )) , (7)
where D L D_{L} D L is the nonnegative sum in (2). Neither full derivative has a fixed sign. The common framework therefore gives balance laws appropriate to each process, rather than imposing one increasing function on every evolution.
For an observed channel output y y y , let ℓ i = P ( y ∣ J = i ) > 0 \ell_i=P(y\mid J=i)>0 ℓ i = P ( y ∣ J = i ) > 0 . Setting f = 0 f=0 f = 0 and weighting the terminal state gives
P ( J T = i ∣ y ) = ℓ i ( e L T T p 0 ) i ∑ j ℓ j ( e L T T p 0 ) j . P(J_{T}=i\mid y)=\frac{\ell_i(e^{L^{\mathsf{T}}T}p_0)_i}{\sum_j\ell_j(e^{L^{\mathsf{T}}T}p_0)_j}. P ( J T = i ∣ y ) = ∑ j ℓ j ( e L T T p 0 ) j ℓ i ( e L T T p 0 ) i .
This is Bayesian updating after prediction through the same channel environment. Relative likelihoods suffice; a common positive factor cancels. In a communication model, decoding, encoding and channel capacity require their specified input and output alphabets and constraints. The formula supplies the state-inference component.
For ℓ = 1 \ell=1 ℓ = 1 , the normalized population law in (6) equals the terminal marginal of the tilted law with horizon t t t . It is generally different from the intermediate marginal of a tilted law with a larger fixed horizon T T T , which contains the future factor h ( t ) h(t) h ( t ) . Thus endpoint selection and a selected history use the same operator but different observations.
A two-state calculation with a common parameter set
Take the transition and growth rates, in a chosen inverse-time unit,
L = ( − 1 / 5 1 / 5 4 / 5 − 4 / 5 ) , B = ( − 1 / 5 1 / 5 4 / 5 − 1 / 5 ) , π = ( 4 / 5 , 1 / 5 ) T , f = ( 0 , 3 / 5 ) T . (8) \begin{aligned}
L&=\begin{pmatrix}-1/5 & 1/5\\ 4/5 & -4/5\end{pmatrix}, \qquad B=\begin{pmatrix}-1/5 & 1/5\\ 4/5 & -1/5\end{pmatrix},\\
\pi&=(4/5,1/5)^{\mathsf{T}}, \qquad f=(0,3/5)^{\mathsf{T}}. \tag*{(8)}
\end{aligned} L π = ( − 1/5 4/5 1/5 − 4/5 ) , B = ( − 1/5 4/5 1/5 − 1/5 ) , = ( 4/5 , 1/5 ) T , f = ( 0 , 3/5 ) T . (8)
The unweighted relaxation rate is one. A thermal realization has E 2 − E 1 = θ log 4 E_2-E_1=\theta\log4 E 2 − E 1 = θ log 4 . For x = p 2 x=p_2 x = p 2 , the two evolutions are
x ˙ = 1 / 5 − x , x ˙ = 1 / 5 − ( 2 / 5 ) x − ( 3 / 5 ) x 2 . (9) \dot{x}=1/5-x,\qquad\dot{x}=1/5-(2/5)x-(3/5)x^2. \tag*{(9)} x ˙ = 1/5 − x , x ˙ = 1/5 − ( 2/5 ) x − ( 3/5 ) x 2 . (9)
The first tends to 1 / 5 1/5 1/5 . In the second,
x ( t ) − 1 / 3 x ( t ) + 1 = x ( 0 ) − 1 / 3 x ( 0 ) + 1 e − 4 t / 5 , \frac{x(t)-1/3}{x(t)+1}=\frac{x(0)-1/3}{x(0)+1}e^{-4t/5}, x ( t ) + 1 x ( t ) − 1/3 = x ( 0 ) + 1 x ( 0 ) − 1/3 e − 4 t /5 ,
so the selected fraction tends to 1 / 3 1/3 1/3 . Direct differentiation recovers the second equation and its initial value.
The matrix B B B has eigenvalues 1 / 5 1/5 1/5 and − 3 / 5 -3/5 − 3/5 , with positive right eigenvector h = ( 1 , 2 ) T h=(1,2)^{\mathrm{T}} h = ( 1 , 2 ) T for the first. The associated time-homogeneous driven generator is
L ^ = diag ( h ) − 1 ( B − 1 5 I ) diag ( h ) = ( − 2 / 5 2 / 5 2 / 5 − 2 / 5 ) . \widehat{L}=\operatorname{diag}(h)^{-1}\left(B-\frac{1}{5}I\right)\operatorname{diag}(h)=\begin{pmatrix}-2/5 & 2/5 \\ 2/5 & -2/5\end{pmatrix}. L = diag ( h ) − 1 ( B − 5 1 I ) diag ( h ) = ( − 2/5 2/5 2/5 − 2/5 ) .
Its stationary fraction is 1 / 2 1/2 1/2 . In a reversible reference model, the normalized population eigenvector is proportional to π i h i \pi_i h_i π i h i , whereas the driven stationary law is proportional to π i h i 2 \pi_i h_i^2 π i h i 2 . Indeed, π i L i j = π j L j i \pi_iL_{ij}=\pi_jL_{ji} π i L ij = π j L j i makes π i h i 2 L ^ i j = π j h j 2 L ^ j i \pi_i h_i^2\widehat{L}_{ij}=\pi_j h_j^2\widehat{L}_{ji} π i h i 2 L ij = π j h j 2 L j i . The three fractions 1 / 5 , 1 / 3 , 1 / 2 1/5,1/3,1/2 1/5 , 1/3 , 1/2 consequently refer to unweighted occupancy, selected population endpoints, and occupancy along a driven history. Their difference is predicted by the same specified data.
The example also displays the competing terms in (7) . At x = 9 / 10 x=9/10 x = 9/10 , the full mean fitness derivative is − 969 / 2500 < 0 -969/2500<0 − 969/2500 < 0 , despite the nonnegative selection variance. Starting at p = π p=\pi p = π , the first derivative of D ( p ∥ π ) D(p\|\pi) D ( p ∥ π ) is zero but its second derivative under selection is 36 / 625 > 0 36/625>0 36/625 > 0 . Mutation–selection therefore moves away from the bath equilibrium without contradicting the thermal relaxation identity for the unweighted process.
There is a direct parameter-transfer test. For arbitrary positive rates a = L 12 a=L_{12} a = L 12 , b = L 21 b=L_{21} b = L 21 , let s = a + b s=a+b s = a + b , ϖ = a / s \varpi=a/s ϖ = a / s , and use growth difference κ = f 2 − f 1 > 0 \kappa=f_2-f_1>0 κ = f 2 − f 1 > 0 , subtracting the common growth f 1 f_1 f 1 . The equilibrium covariance of the state-2 indicator is ϖ ( 1 − ϖ ) e − s t \varpi(1-\varpi)e^{-st} ϖ ( 1 − ϖ ) e − s t . Thus equilibrium fraction and correlation decay determine ϖ , s \varpi,s ϖ , s . With κ \kappa κ specified independently, the same operator gives
Δ = s 2 + 2 κ s ( 2 ϖ − 1 ) + κ 2 , λ = ( κ − s + Δ ) / 2 , x ∗ = λ / κ . (10) \Delta=\sqrt{s^2+2\kappa s(2\varpi-1)+\kappa^2},\qquad\lambda=(\kappa-s+\Delta)/2,\qquad x_*=\lambda/\kappa. \tag*{(10)} Δ = s 2 + 2 κ s ( 2 ϖ − 1 ) + κ 2 , λ = ( κ − s + Δ ) /2 , x ∗ = λ / κ . (10)
Here λ \lambda λ is the asymptotic expected-count growth rate and Δ \Delta Δ is the driven relaxation rate. These formulas follow from the characteristic polynomial of B B B . The normalized population equation is x ˙ = s ( ϖ − x ) + κ x ( 1 − x ) \dot{x}=s(\varpi-x)+\kappa x(1-x) x ˙ = s ( ϖ − x ) + κ x ( 1 − x ) , so x ∗ ∈ ( ϖ , 1 ) x_*\in(\varpi,1) x ∗ ∈ ( ϖ , 1 ) . Its implicit derivatives are
∂ ϖ x ∗ = s / Δ > 0 , ∂ s x ∗ = ( ϖ − x ∗ ) / Δ < 0 , ∂ κ x ∗ = x ∗ ( 1 − x ∗ ) / Δ > 0. \partial_{\varpi}x_*=s/\Delta>0,\qquad\partial_sx_*=(\varpi-x_*)/\Delta<0,\qquad\partial_{\kappa}x_*=x_*(1-x_*)/\Delta>0. ∂ ϖ x ∗ = s /Δ > 0 , ∂ s x ∗ = ( ϖ − x ∗ ) /Δ < 0 , ∂ κ x ∗ = x ∗ ( 1 − x ∗ ) /Δ > 0.
The same signs hold for λ \lambda λ , with ∂ s λ < 0 \partial_s\lambda<0 ∂ s λ < 0 . For s ∈ [ 0.98 , 1.02 ] s\in[0.98,1.02] s ∈ [ 0.98 , 1.02 ] , ϖ ∈ [ 0.195 , 0.205 ] \varpi\in[0.195,0.205] ϖ ∈ [ 0.195 , 0.205 ] and κ = 0.6 \kappa=0.6 κ = 0.6 , corner evaluation gives, to the displayed precision, λ ∈ [ 0.194277 , 0.205778 ] \lambda\in[0.194277,0.205778] λ ∈ [ 0.194277 , 0.205778 ] and x ∗ ∈ [ 0.323795 , 0.342963 ] x_*\in[0.323795,0.342963] x ∗ ∈ [ 0.323795 , 0.342963 ] . These are conditional propagated input ranges, not confidence intervals from observations.
A martingale constraint and a financial law
A financial model adds an economic constraint to probability selection. Let a stock have gross returns R i R_i R i , let the risk-free gross return be R f R_f R f , and let p i > 0 p_i>0 p i > 0 be reference probabilities. Minimize D ( q ∥ p ) D(q\|p) D ( q ∥ p ) subject to ∑ i q i = 1 \sum_iq_i=1 ∑ i q i = 1 and ∑ i q i R i = R f \sum_iq_iR_i=R_f ∑ i q i R i = R f . For a nonconstant return vector and R f R_f R f strictly between its minimum and maximum, the unique minimizer is
q i = p i e ζ R i ∑ j p j e ζ R j , ∑ i q i R i = R f . (11) q_i = \frac{p_i e^{\zeta R_i}}{\sum_j p_j e^{\zeta R_j}}, \qquad\sum_i q_iR_i = R_f. \tag*{(11)} q i = ∑ j p j e ζ R j p i e ζ R i , i ∑ q i R i = R f . (11)
The derivative of the exponential-family mean with respect to ζ \zeta ζ is Var q ( R ) > 0 \operatorname{Var}_q(R)>0 Var q ( R ) > 0 ; its limiting means are the extreme returns. This proves existence and uniqueness of ζ \zeta ζ , while the relative-entropy identity proves minimality.
For p = ( 0.8 , 0.2 ) p=(0.8,0.2) p = ( 0.8 , 0.2 ) , returns ( 0.9 , 1.2 ) (0.9,1.2) ( 0.9 , 1.2 ) , and R f = 1.02 R_f=1.02 R f = 1.02 , the constraint gives q = ( 0.6 , 0.4 ) q=(0.6,0.4) q = ( 0.6 , 0.4 ) and ζ = log ( 8 / 3 ) / 0.3 \zeta=\log(8/3)/0.3 ζ = log ( 8/3 ) /0.3 . A stock initially worth 100 has a one-period call payoff ( 0 , 20 ) (0,20) ( 0 , 20 ) at strike 100. In the frictionless two-state stock–bond market this payoff is replicable, and its price is 20 ( 0.4 ) / 1.02 = 400 / 51 20(0.4)/1.02=400/51 20 ( 0.4 ) /1.02 = 400/51 . The martingale measure is already unique by completeness. In an incomplete market, minimizing entropy is an additional selection criterion among admissible martingale measures; no-arbitrage alone does not require that criterion [91]. The economic constraints supply the interpretation of the law. They do not follow from a thermal reservoir.
Incomplete markets: what selects a price
The two-state example separates the martingale constraint from a thermal interpretation, but completeness leaves no remaining probability to select. An incomplete market exposes the role of the variational principle more clearly. The information projection and martingale interpretation use established methods [48] and [91]. The economic question is how to value a payoff that cannot be replicated using the traded stock and bond. No-arbitrage restricts its price; a specified reference law and a selection rule supply additional information. This is the point at which the common variational structure has a substantive modeling role.
Consider a one-period frictionless market with a bond of initial value one and terminal value R f > 0 R_f>0 R f > 0 , and a stock with initial price S 0 > 0 S_0>0 S 0 > 0 and terminal values s 1 , … , s n s_1,\ldots,s_n s 1 , … , s n . Every state has reference probability p i > 0 p_i>0 p i > 0 , with ∑ i p i = 1 \sum_i p_i=1 ∑ i p i = 1 . Put μ = R f S 0 \mu=R_fS_0 μ = R f S 0 and assume min i s i < μ < max i s i \min_i s_i<\mu<\max_i s_i min i s i < μ < max i s i . A strictly positive probability vector q q q satisfying ∑ i q i s i = μ \sum_i q_i s_i=\mu ∑ i q i s i = μ is an equivalent martingale measure: the discounted stock has the same initial value and terminal expectation. For a terminal payoff H = ( H 1 , … , H n ) H=(H_1,\ldots,H_n) H = ( H 1 , … , H n ) , it assigns the price R f − 1 ∑ i q i H i R_f^{-1}\sum_i q_iH_i R f − 1 ∑ i q i H i . Incomplete trading generally permits several such vectors and several prices for a nonreplicable payoff.
Proposition 1 (Entropy selection and its price sensitivity). Let the terminal stock values be nonconstant and let the preceding strict interior and positivity assumptions hold. On the closed feasible simplex
M μ = { q i ≥ 0 : ∑ i q i = 1 , ∑ i q i s i = μ } , \mathcal{M}_{\mu}=\left\{q_i\ge0:\ \sum_i q_i=1,\ \sum_i q_i s_i=\mu\right\}, M μ = { q i ≥ 0 : i ∑ q i = 1 , i ∑ q i s i = μ } ,
the unique minimizer of D ( q ∥ p ) = ∑ i q i log ( q i / p i ) D(q\Vert p)=\sum_i q_i\log(q_i/p_i) D ( q ∥ p ) = ∑ i q i log ( q i / p i ) is
q i μ = p i e λ ( μ ) s i Z ( λ ( μ ) ) , Z ( λ ) = ∑ i p i e λ s i , Z ′ ( λ ( μ ) ) Z ( λ ( μ ) ) = μ . (12) q_i^\mu=\frac{p_i e^{\lambda(\mu)s_i}}{Z(\lambda(\mu))},\qquad Z(\lambda)=\sum_i p_i e^{\lambda s_i},\qquad\frac{Z'(\lambda(\mu))}{Z(\lambda(\mu))}=\mu. \tag*{(12)} q i μ = Z ( λ ( μ )) p i e λ ( μ ) s i , Z ( λ ) = i ∑ p i e λ s i , Z ( λ ( μ )) Z ′ ( λ ( μ )) = μ . (12)
The multiplier λ ( μ ) \lambda(\mu) λ ( μ ) is finite and unique, and every q i μ q_i^\mu q i μ is positive. For every q ∈ M μ q\in\mathcal{M}_\mu q ∈ M μ ,
D ( q ∥ p ) = D ( q ∥ q μ ) + D ( q μ ∥ p ) . (13) D(q\Vert p)=D(q\Vert q^\mu)+D(q^\mu\Vert p). \tag*{(13)} D ( q ∥ p ) = D ( q ∥ q μ ) + D ( q μ ∥ p ) . (13)
Hold p p p , the terminal values s i s_i s i , the payoff H i H_i H i , and R f R_f R f fixed, and regard the target mean μ \mu μ as the variable. The selected price C H ( μ ) = R f − 1 ∑ i q i μ H i C_H(\mu)=R_f^{-1}\sum_i q_i^\mu H_i C H ( μ ) = R f − 1 ∑ i q i μ H i satisfies
d C H d μ = Cov q μ ( H , S 1 ) R f Var q μ ( S 1 ) . (14) \frac{\mathrm{d}C_H}{\mathrm{d}\mu}=\frac{\operatorname{Cov}_{q^\mu}(H,S_1)}{R_f\operatorname{Var}_{q^\mu}(S_1)}. \tag*{(14)} d μ d C H = R f Var q μ ( S 1 ) Cov q μ ( H , S 1 ) . (14)
For a call payoff H i = ( s i − K ) + H_i=(s_i-K)_+ H i = ( s i − K ) + , this derivative belongs to [ 0 , R f − 1 ] [0,R_f^{-1}] [ 0 , R f − 1 ] .
Proof. Define ψ ( λ ) = log Z ( λ ) \psi(\lambda)=\log Z(\lambda) ψ ( λ ) = log Z ( λ ) . Differentiating the finite sum gives
ψ ′ ( λ ) = ∑ i q i ( λ ) s i , ψ ′ ′ ( λ ) = ∑ i q i ( λ ) [ s i − ψ ′ ( λ ) ] 2 > 0. (15) \psi'(\lambda)=\sum_i q_i(\lambda)s_i,\qquad\psi''(\lambda)=\sum_i q_i(\lambda)[s_i-\psi'(\lambda)]^2>0. \tag*{(15)} ψ ′ ( λ ) = i ∑ q i ( λ ) s i , ψ ′′ ( λ ) = i ∑ q i ( λ ) [ s i − ψ ′ ( λ ) ] 2 > 0. (15)
Strict positivity follows because all tilted probabilities are positive and at least two stock values differ. As λ → − ∞ \lambda\to-\infty λ → − ∞ , terms with the smallest stock value dominate the finite sum, so ψ ′ ( λ ) → min i s i \psi'(\lambda)\to\min_i s_i ψ ′ ( λ ) → min i s i . As λ → + ∞ \lambda\to+\infty λ → + ∞ , the corresponding limit is max i s i \max_i s_i max i s i . The mean is continuous and strictly increasing; hence the intermediate value theorem gives exactly one finite multiplier with mean μ \mu μ .
For this multiplier, log ( q i μ / p i ) = λ ( μ ) s i − ψ ( λ ( μ ) ) \log(q_i^\mu/p_i)=\lambda(\mu)s_i-\psi(\lambda(\mu)) log ( q i μ / p i ) = λ ( μ ) s i − ψ ( λ ( μ )) . Therefore, for any feasible q q q , with $0\log 0=0,
D ( q ∥ p ) − D ( q ∥ q μ ) = ∑ i q i log q i μ p i = λ ( μ ) μ − ψ ( λ ( μ ) ) (16) D(q\Vert p)-D(q\Vert q^\mu)=\sum_i q_i\log\frac{q_i^\mu}{p_i}=\lambda(\mu)\mu-\psi(\lambda(\mu)) \tag*{(16)} D ( q ∥ p ) − D ( q ∥ q μ ) = i ∑ q i log p i q i μ = λ ( μ ) μ − ψ ( λ ( μ )) (16)
= D ( q μ ∥ p ) . (17) =D(q^\mu\Vert p). \tag*{(17)} = D ( q μ ∥ p ) . (17)
Relative entropy is nonnegative, with equality exactly when the two probability vectors coincide. This proves minimality and uniqueness on the entire feasible simplex, including its boundary.
Since ψ ′ ′ > 0 \psi''>0 ψ ′′ > 0 , the inverse function theorem gives λ ′ ( μ ) = 1 / Var q μ ( S 1 ) \lambda'(\mu)=1/\operatorname{Var}_{q^\mu}(S_1) λ ′ ( μ ) = 1/ Var q μ ( S 1 ) . Direct differentiation also gives ∂ λ q i = q i ( s i − E q S 1 ) \partial_\lambda q_i=q_i(s_i-\mathbb{E}_qS_1) ∂ λ q i = q i ( s i − E q S 1 ) , and thus ∂ λ E q H = Cov q ( H , S 1 ) \partial_\lambda\mathbb{E}_qH=\operatorname{Cov}_q(H,S_1) ∂ λ E q H = Cov q ( H , S 1 ) . Combining the two derivatives proves (14). Finally,
Cov q ( H , S 1 ) = 1 2 ∑ i , j q i q j ( H i − H j ) ( s i − s j ) . (18) \operatorname{Cov}_q(H,S_1)=\frac{1}{2}\sum_{i,j}q_iq_j(H_i-H_j)(s_i-s_j). \tag*{(18)} Cov q ( H , S 1 ) = 2 1 i , j ∑ q i q j ( H i − H j ) ( s i − s j ) . (18)
Var q ( S 1 ) = 1 2 ∑ i , j q i q j ( s i − s j ) 2 . (19) \operatorname{Var}_q(S_1)=\frac{1}{2}\sum_{i,j}q_iq_j(s_i-s_j)^2. \tag*{(19)} Var q ( S 1 ) = 2 1 i , j ∑ q i q j ( s i − s j ) 2 . (19)
The call function is increasing and one-Lipschitz, so every summand in the first line lies between zero and the corresponding summand in the second. Division by the positive variance proves the last assertion. □
This result supplies an exact measure of sensitivity within a declared model. The mean variation in (14) keeps terminal prices fixed. If μ = R f S 0 \mu= R_f S_0 μ = R f S 0 is varied through S 0 S_0 S 0 under these same conditions, the chain rule gives d C H / d S 0 = Cov q ( H , S 1 ) / Var q ( S 1 ) dC_H/dS_0 = \operatorname{Cov}_q(H,S_1)/\operatorname{Var}_q(S_1) d C H / d S 0 = Cov q ( H , S 1 ) / Var q ( S 1 ) . A model in which terminal stock values scale with S 0 S_0 S 0 has additional derivative terms and is a different comparison. Moreover, the multiplier has inverse-price units; equivalently one may use dimensionless returns and a dimensionless multiplier. A change of price units rescales the multiplier and leaves q μ q^\mu q μ unchanged.
There is also an informative limit on what a selected price can identify. For any real c c c , replace the reference law by p ~ i = p i e c s i / Z ( c ) \widetilde{p}_i = p_i e^{c s_i}/Z(c) p i = p i e c s i / Z ( c ) . The same selected law is obtained by replacing λ ( μ ) \lambda(\mu) λ ( μ ) with λ ( μ ) − c \lambda(\mu)-c λ ( μ ) − c . Therefore all prices selected by this criterion remain unchanged under this family of reference-law transformations. A pricing law alone cannot determine the reference law in every direction.
The directions that do affect the price have an explicit formula. Let a i a_i a i be fixed and set p i ( t ) = p i e t a i / ∑ j p j e t a j p_i(t) = p_i e^{t a_i}/\sum_j p_j e^{t a_j} p i ( t ) = p i e t a i / ∑ j p j e t a j . Keep μ \mu μ , the stock values, the payoff, and R f R_f R f fixed, and denote the corresponding selected law by q ( t ) q(t) q ( t ) . Differentiation at t = 0 t=0 t = 0 gives
d C H d t ∣ t = 0 = 1 R f [ Cov q μ ( H , a ) − Cov q μ ( H , S 1 ) Cov q μ ( S 1 , a ) Var q μ ( S 1 ) ] . (20) \left.\frac{dC_H}{dt}\right|_{t=0}
=
\frac{1}{R_f}
\left[
\operatorname{Cov}_{q^\mu}(H,a)
-
\frac{\operatorname{Cov}_{q^\mu}(H,S_1)\operatorname{Cov}_{q^\mu}(S_1,a)}
{\operatorname{Var}_{q^\mu}(S_1)}
\right]. \tag*{(20)} d t d C H t = 0 = R f 1 [ Cov q μ ( H , a ) − Var q μ ( S 1 ) Cov q μ ( H , S 1 ) Cov q μ ( S 1 , a ) ] . (20)
Indeed, the normalization in p ( t ) p(t) p ( t ) cancels from the tilted law, so q i ( t ) q_i(t) q i ( t ) is proportional to p i e t a i + λ ( t ) s i p_i e^{t a_i+\lambda(t)s_i} p i e t a i + λ ( t ) s i . Differentiating its mean constraint gives 0 = Cov q ( S 1 , a ) + λ ′ ( 0 ) Var q ( S 1 ) 0 = \operatorname{Cov}_q(S_1,a) + \lambda'(0)\operatorname{Var}_q(S_1) 0 = Cov q ( S 1 , a ) + λ ′ ( 0 ) Var q ( S 1 ) , which determines λ ′ ( 0 ) \lambda'(0) λ ′ ( 0 ) . Differentiating the payoff mean then proves (20). The bracket is the covariance remaining after the stock direction has been removed. It vanishes when a i a_i a i is affine in s i s_i s i , consistently with the exact invariance above. Thus the reference law matters through changes that remain visible after the traded-price constraint has been imposed.
An explicit three-state comparison. Take the synthetic inputs
S 0 = 100 , R f = 1.02 , ( s 1 , s 2 , s 3 ) = ( 80 , 100 , 130 ) , S_0 = 100,\qquad R_f = 1.02,\qquad(s_1,s_2,s_3) = (80,100,130), S 0 = 100 , R f = 1.02 , ( s 1 , s 2 , s 3 ) = ( 80 , 100 , 130 ) ,
p = ( 0.2 , 0.5 , 0.3 ) , H = ( 0 , 0 , 30 ) . (21) p = (0.2,0.5,0.3),\qquad H = (0,0,30). \tag*{(21)} p = ( 0.2 , 0.5 , 0.3 ) , H = ( 0 , 0 , 30 ) . (21)
Here E p S 1 = 105 E_p S_1 = 105 E p S 1 = 105 , whereas the martingale mean is 102. The reference law therefore cannot itself price the stock by discounted expectation. The entropy minimizer is its least relative-entropy adjustment subject to that pricing constraint.
Write t = q 3 t = q_3 t = q 3 . Normalization and the mean constraint give
q 1 = 3 2 t − 1 10 , q 2 = 11 10 − 5 2 t , 1 15 < t < 11 25 . (22) q_1 = \frac{3}{2}t-\frac{1}{10},\qquad
q_2 = \frac{11}{10}-\frac{5}{2}t,\qquad
\frac{1}{15}<t<\frac{11}{25}. \tag*{(22)} q 1 = 2 3 t − 10 1 , q 2 = 10 11 − 2 5 t , 15 1 < t < 25 11 . (22)
The strict inequalities enforce equivalence to the positive reference law. Hence the call admits the open no-arbitrage price interval
100 51 < C H < 220 17 , 1.9607843 < C H < 12.9411765. (23) \frac{100}{51}<C_H<\frac{220}{17},\qquad1.9607843<C_H<12.9411765. \tag*{(23)} 51 100 < C H < 17 220 , 1.9607843 < C H < 12.9411765. (23)
The endpoints belong to the closure of this interval; at either endpoint an equivalent measure is lost. The payoff is not replicable, because its values at stock prices 80 and 100 force any affine stock–bond replicating payoff to vanish everywhere, contradicting the value 30 at stock price 130.
The selected point can be computed through a scalar polynomial. Put x = e 10 λ > 0 x=e^{10\lambda}>0 x = e 10 λ > 0 , cancel the common factor e 80 λ e^{80\lambda} e 80 λ , and obtain
( q 1 , q 2 , q 3 ) = ( 2 , 5 x 2 , 3 x 5 ) 2 + 5 x 2 + 3 x 5 , 42 x 5 − 5 x 2 − 22 = 0. (24) (q_1,q_2,q_3)=\frac{(2,5x^2,3x^5)}{2+5x^2+3x^5},\qquad42x^5-5x^2-22=0. \tag*{(24)} ( q 1 , q 2 , q 3 ) = 2 + 5 x 2 + 3 x 5 ( 2 , 5 x 2 , 3 x 5 ) , 42 x 5 − 5 x 2 − 22 = 0. (24)
The polynomial has exactly one positive root: its derivative is 10 x ( 21 x 3 − 1 ) 10x(21x^3-1) 10 x ( 21 x 3 − 1 ) , so it first decreases from the negative value − 22 -22 − 22 and then increases strictly to infinity. Equivalently, uniqueness follows from the variance calculation above. The resulting values are
λ = − 0.00948723241569055 , (25) \lambda=-0.00948723241569055, \tag*{(25)} λ = − 0.00948723241569055 , (25)
q μ = ( 0.2499157104069353 , 0.5168071493217746 , 0.2332771402712902 ) , (26) q^\mu=(0.2499157104069353,0.5168071493217746,0.2332771402712902), \tag*{(26)} q μ = ( 0.2499157104069353 , 0.5168071493217746 , 0.2332771402712902 ) , (26)
C H = 6.86109236092030 , D ( q μ ∥ p ) = 0.0140872229460959 , (27) C_H=6.86109236092030,\qquad D(q^\mu\mathbin{\|}p)=0.0140872229460959, \tag*{(27)} C H = 6.86109236092030 , D ( q μ ∥ p ) = 0.0140872229460959 , (27)
Var q μ ( S 1 ) = 305.9157104069353 , d C H d μ ∣ μ = 102 = 0.627985355345821. (28) \operatorname{Var}_{q^\mu}(S_1)=305.9157104069353,\qquad\left.\frac{dC_H}{d\mu}\right|_{\mu=102}=0.627985355345821. \tag*{(28)} Var q μ ( S 1 ) = 305.9157104069353 , d μ d C H μ = 102 = 0.627985355345821. (28)
These are model calculations, not observed option prices. The scalar polynomial permits exact rational bracketing of its root and of the selected call price, in addition to floating-point evaluation.
Figure 1 shows the selection on the feasible martingale line and the price response when the target mean changes. The second panel holds the terminal values, payoff, reference probabilities and bond return fixed; it implements precisely the comparison in (14) .
Figure 1. Synthetic incomplete-market calculation using (21). Left: relative entropy along the feasible martingale line, parameterized by the call price; the point marks the unique entropy minimizer. Right: the selected call price at 41 target stock means from 85 to 125, with the point marking the original mean 102. The reference law and terminal values are held fixed throughout the second comparison. These curves describe the stated model and do not use observed option prices.
The example also makes the economic assumption visible. If the reference probabilities are replaced by any strictly positive feasible vector, that vector becomes the entropy minimizer because its relative entropy with respect to itself is zero. Thus different admissible reference laws can select different prices within (23) . The variational principle becomes empirically restrictive only after the reference law and its calibration are fixed independently of the target payoff price. It expresses a common constrained-selection operation; it supplies neither a thermal heat interpretation nor a unique market law without economic inputs.
This distinction leads naturally to the next model. Selection under a martingale constraint concerns the prices of traded risks, whereas a stationary process can also sustain probability currents. The cyclic example below explains why a fixed one-time distribution does not imply the absence of continuing change.
Cyclic change and an institutional interpretation
The same description also permits persistent cycles. On three states, give each clockwise transition rate 2 and each counterclockwise transition rate 1. The stationary law is uniform, yet the clockwise current on every edge is 1/3. For even state variables, its stationary forward–reverse path relative-entropy rate is
∑ { i , j } ( π i L i j − π j L j i ) log π i L i j π j L j i = log 2 \sum_{\{i,j\}}(\pi_iL_{ij}-\pi_jL_{ji})\log\frac{\pi_iL_{ij}}{\pi_jL_{ji}}=\log2 { i , j } ∑ ( π i L ij − π j L j i ) log π j L j i π i L ij = log 2
per chosen time unit. The sum has one orientation of each undirected edge. Stationarity therefore coexists with ongoing directed change. This model fits the path-selection principle but does not satisfy the detailed-balance assumption used in the thermal gradient calculation.
For an organizational application, the states may denote three institutional arrangements and the rates their empirically estimated replacement frequencies under fixed external conditions. A cyclic transition pattern is then described without assuming a global ranking of the arrangements. If resources, incentives or institutions alter the rates, they must be included as state variables or specified external inputs. The path entropy above quantifies statistical time asymmetry; a heat or welfare interpretation requires further physical or social definitions. This provides a conditional model for social evolution within the common structure, with its own measurable transitions.
Conservative motion and a matched reservoir
The finite-state example describes probabilities and selection. A continuous physical realization also exhibits the transport part. For mass one, oscillator frequency ω \omega ω , friction γ ≥ 0 \gamma\ge0 γ ≥ 0 , and reservoir energy θ > 0 \theta> 0 θ > 0 , consider
d q = p d t , d p = ( − ω 2 q − γ p ) d t + 2 γ θ d W t , H = ( p 2 + ω 2 q 2 ) / 2. (29) dq = p\,dt,\qquad dp = \left(-\omega^2q-\gamma p\right)\,dt+\sqrt{2\gamma\theta}\,dW_t,\qquad H=(p^2+\omega^2q^2)/2. \tag*{(29)} d q = p d t , d p = ( − ω 2 q − γ p ) d t + 2 γ θ d W t , H = ( p 2 + ω 2 q 2 ) /2. (29)
Units are fixed before setting the mass to one: q q q is displacement, p p p is momentum per unit mass, and H , θ H,\theta H , θ are energies per unit mass. Time and displacement can then be expressed in chosen units. The equilibrium covariance is Σ = diag ( θ / ω 2 , θ ) \Sigma=\operatorname{diag}(\theta/\omega^2,\theta) Σ = diag ( θ / ω 2 , θ ) . Writing the drift matrix as A = ( 0 1 − ω 2 − γ ) A=\begin{pmatrix}0&1\\-\omega^2&-\gamma\end{pmatrix} A = ( 0 − ω 2 1 − γ ) , the identity A Σ + Σ A T + diag ( 0 , 2 γ θ ) = 0 A\Sigma+\Sigma A^{T}+\operatorname{diag}(0,2\gamma\theta)=0 A Σ + Σ A T + diag ( 0 , 2 γ θ ) = 0 proves invariance of the Gaussian Gibbs law.
The Hamiltonian part ( p , − ω 2 q ) (p,-\omega^2q) ( p , − ω 2 q ) preserves energy and phase volume. Integration by parts for the damping and diffusion terms gives
d d t D ( ρ t ∥ π ) = − γ θ ∫ ρ t ∣ ∂ p log ( ρ t / π ) ∣ 2 d q d p , \frac{d}{dt}D(\rho_t\Vert\pi)=-\gamma\theta\int\rho_t\lvert\partial_p\log(\rho_t/\pi)\rvert^2\,dq\,dp, d t d D ( ρ t ∥ π ) = − γ θ ∫ ρ t ∣ ∂ p log ( ρ t / π ) ∣ 2 d q d p ,
for positive smooth densities with the stated boundary decay. Itô’s formula also gives d E H / d t = − γ E p 2 + γ θ d\mathbb{E}H/dt=-\gamma\mathbb{E}p^2+\gamma\theta d E H / d t = − γ E p 2 + γ θ . Thus the same γ \gamma γ governs damping and thermal injection. At γ = 0 \gamma=0 γ = 0 , both terms vanish and Hamilton’s equations remain. This conservative recovery follows from the matched reservoir scaling, rather than from suppressing noise alone.
For a small force added to the momentum equation, the position impulse response is R q p ( t ) = ( e A t ) 12 R_{qp}(t)=(e^{At})_{12} R q p ( t ) = ( e A t ) 12 . The equilibrium position covariance is C q q ( t ) = ( e A t Σ ) 11 C_{qq}(t)=(e^{At}\Sigma)_{11} C q q ( t ) = ( e A t Σ ) 11 . Since A Σ A\Sigma A Σ has first column ( 0 , − θ ) T (0,-\theta)^{T} ( 0 , − θ ) T ,
R q p ( t ) = − θ − 1 d d t C q q ( t ) , t > 0. (30) R_{qp}(t)=-\theta^{-1}\frac{d}{dt}C_{qq}(t),\qquad t>0. \tag*{(30)} R q p ( t ) = − θ − 1 d t d C q q ( t ) , t > 0. (30)
One parameter set therefore fixes mechanical response, equilibrium fluctuations and dissipation. The example makes the second organizing principle a measurable relation.
The scalar solution and matrix semigroup agree to 1.0 × 10 − 15 1.0\times10^{-15} 1.0 × 1 0 − 15 in the displayed two-state example. A separate numerical integration of its nonlinear equation agrees within 8.0 × 10 − 13 8.0\times10^{-13} 8.0 × 1 0 − 13 . Direct oscillator moment integration agrees with the matrix solution within 2.0 × 10 − 12 2.0\times10^{-12} 2.0 × 1 0 − 12 ; the fluctuation–response identity differs by less than 4.5 × 10 − 16 4.5\times10^{-16} 4.5 × 1 0 − 16 in double precision. These checks concern the synthetic finite models. Their purpose is to make the common construction calculable before pursuing its field and empirical applications.
The examples establish a precise division of work for the larger programme. Geometry specifies admissible variables and constraints; the reference dynamics specifies transport and fluctuations; weighting and constraints specify selection; observation determines which law is being compared. The field construction now supplies the geometric data on which those operations act. A physical application gains explanatory force when these data fix several responses together.
Figure 2. Synthetic representative models. Upper left: the two-state model with initial fraction 0.85 has different unweighted and selected population limits; the driven-history fraction is a third observable. Upper right: relative entropy to the bath law decreases for unweighted transitions, whereas selection permits a different balance. Below: the oscillator with ω = 1.3 \omega= 1.3 ω = 1.3 , γ = 0.4 \gamma= 0.4 γ = 0.4 and θ = 0.8 \theta= 0.8 θ = 0.8 satisfies the fluctuation–response relation. Units are chosen consistently within each model; the panels do not identify biological rates with mechanical frequencies.
The master configuration, action, and one path-space cost
Master configuration space and unified geometry
We now specify the objects on which the variational principle acts. Three spaces play different roles: the spacetime manifold carries the fields, the principal bundle describes their gauge transformations, and the configuration space carries the stochastic law. Keeping these spaces distinct is necessary for the later definitions of curvature, field-space gradients, and probability densities.
Definition 1 (Finite regulator). A finite-regulator realization specifies a finite volume or compact domain, its boundary conditions, and finitely many retained field coordinates. A lattice spacing a > 0 a > 0 a > 0 can regulate ultraviolet wave-lengths, while a spectral or Galerkin cutoff N < ∞ N < \infty N < ∞ can regulate the number of modes; a given realization need not use both. Gauge variables can be retained on compact link groups, represented on a specified local gauge slice, or treated on a physical quotient. The choice must preserve the identities used in the corresponding result; a globally smooth gauge slice is not assumed. The coordinate state space, measure, operator domains, and all constrained directions are part of the regulator data. A numerical time step Δ t \Delta t Δ t is an additional approximation and is not the regulator a a a or N N N . Removing a a a , N N N , the finite volume, and Δ t \Delta t Δ t requires separate estimates, with an explicitly stated order of limits.
In particular, a finite-dimensional Gibbs density is a classical probability law on the declared bosonic coordinates. It is not a probability density on Grassmann-valued fermion variables. Incorporating fermions into a positive Euclidean physical-state construction requires the additional steps described in Section 12.1.
The microscopic configuration
Let M M M be a smooth four-dimensional manifold used as the mathematical starting point. No physical spacetime metric is assumed as an independent microscopic field. A non-degenerate coframe contained in the Cartan connection will induce the Euclidean metric used by the geometric sector. In finite-dimensional or Galerkin calculations one may, before such a coframe is selected, introduce an auxiliary positive reference metric g ˉ \bar{g} g ˉ and its volume form vol g ˉ \operatorname{vol}_{\bar{g}} vol g ˉ to define Sobolev norms, regulators, and temporary analytic estimates. Every formula involving the physical curvature or the heat-kernel coefficients is to be read with the induced metric after the coframe exists; formulas proved on the regulator are statements about that regularized realization. A global physical Lorentzian completion, if it exists, must be obtained by a separately justified analytic continuation together with a proof of causality and unitarity. This global completion is stronger than the local continued Euler equations recovered below. We do not identify Euclidean stochastic time with physical time.
Let G ~ \widetilde{G} G be a Lie group with Lie algebra g \mathfrak{g} g , let H ⊂ G ~ H \subset\widetilde{G} H ⊂ G be a closed subgroup with Lie algebra h \mathfrak{h} h , and let P H → M P_H \to M P H → M be a principal H H H -bundle. Because the translational generators are represented by a coframe, the master field is a Cartan connection A ∈ Ω 1 ( P H , g ) A \in\Omega^1(P_H,\mathfrak{g}) A ∈ Ω 1 ( P H , g ) , rather than an ordinary principal connection whose values are confined to h \mathfrak{h} h . Thus, in addition to H H H -equivariance and reproduction of fundamental vertical vector fields, a Cartan connection is a pointwise linear isomorphism T p P H → g T_pP_H \to\mathfrak{g} T p P H → g . These are the standard geometric requirements in gauge-geometric formulations of gravity [152], [118], and [122]. A concrete model must specify ( G ~ , H ) (\widetilde{G},H) ( G , H ) and prove that the associated bundle exists globally; until then, the formulas below are local expressions. The master algebra is required to admit a representation in which spacetime translations, local rotations/boosts after continuation, and internal transformations occur as components of one algebraic structure. Explicitly, for h ∈ H h \in H h ∈ H and ξ ∈ h \xi\in\mathfrak{h} ξ ∈ h , the Cartan axioms are
( R h ) ∗ A = Ad ( h − 1 ) A , A ( ξ # ) = ξ , A p : T p P H ⟶ g is an isomorphism. (31) (R_h)^*A = \operatorname{Ad}(h^{-1})A,\qquad A(\xi^\#)=\xi,\qquad A_p:T_pP_H\longrightarrow\mathfrak{g}\text{ is an isomorphism.} \tag*{(31)} ( R h ) ∗ A = Ad ( h − 1 ) A , A ( ξ # ) = ξ , A p : T p P H ⟶ g is an isomorphism. (31)
For a Cartan geometry of type ( G ~ , H ) (\widetilde{G},H) ( G , H ) , these axioms imply dim g = dim P H = dim M + dim H \dim\mathfrak{g}=\dim P_H=\dim M+\dim H dim g = dim P H = dim M + dim H and identify g / h \mathfrak{g}/\mathfrak{h} g / h with the tangent model of M M M . In the component notation used below we assume that the classes of the P a P_a P a form a basis of g / h \mathfrak{g}/\mathfrak{h} g / h , while the J a b J_{ab} J ab and T A T_A T A span the selected subalgebra h \mathfrak{h} h . If a reductive enlargement is used, it must be declared as the new isotropy algebra before taking the quotient; generators outside that declared isotropy algebra cannot be counted simultaneously as its basis elements. If a concrete model assigns any of these generators differently, the decomposition and all dimension counts must be changed accordingly. The Cartan-connection axioms used here are standard; see, for example, [145]. We now write the master algebra abstractly.
g = span { P a , J a b , T A } , (32) \mathfrak{g}=\operatorname{span}\{P_a,J_{ab},T_A\}, \tag*{(32)} g = span { P a , J ab , T A } , (32)
without assuming at this stage that the algebra is a direct sum. The brackets are
[ P a , P b ] = c a b c P c + c a b c d J c d + c a b A T A , (33) [P_a,P_b]=c_{ab}^{\ \ c}P_c+c_{ab}^{\ \ cd}J_{cd}+c_{ab}^{\ \ A}T_A, \tag*{(33)} [ P a , P b ] = c ab c P c + c ab c d J c d + c ab A T A , (33)
[ J a b , P c ] = c a b , c d P d , [ T A , P b ] = c A b c P c , (34) [J_{ab},P_c]=c_{ab,c}^{\ \ \ d}P_d,\qquad[T_A,P_b]=c_{Ab}^{\ \ c}P_c, \tag*{(34)} [ J ab , P c ] = c ab , c d P d , [ T A , P b ] = c A b c P c , (34)
and similarly for the remaining generators. The coefficients are subject to the Jacobi identities. For a reductive splitting g = h ⊕ m \mathfrak{g}=\mathfrak{h}\oplus\mathfrak{m} g = h ⊕ m , the defining condition is [ h , m ] ⊆ m [\mathfrak{h},\mathfrak{m}]\subseteq\mathfrak{m} [ h , m ] ⊆ m . It does not in general force c a b c = 0 c_{ab}^{\ \ c}=0 c ab c = 0 . The additional condition [ m , m ] ⊆ h [\mathfrak{m},\mathfrak{m}]\subseteq\mathfrak{h} [ m , m ] ⊆ h gives a symmetric reductive pair and does force that coefficient to vanish. Keeping c a b c c_{ab}^{\ \ c} c ab c in the general curvature formula records the translational curvature term in a nonsymmetric reductive realization. In a nonreductive realization the chosen complement need not transform invariantly, and the corresponding local coframe splitting requires a separate formulation.
At least one nonzero mixed structure constant is a useful diagnostic of microscopic coupling, but it is neither necessary when the invariant pairing or field-space metric has cross-components nor sufficient for irreducibility. If we impose this diagnostic as a model-building requirement, it reads
at least one admissible mixed bracket in the master algebra is nonzero. (35) \text{at least one admissible mixed bracket in the master algebra is nonzero.} \tag*{(35)} at least one admissible mixed bracket in the master algebra is nonzero. (35)
Vanishing mixed brackets and an orthogonal invariant pairing imply factorization of these algebraic data, but not of the complete theory. A shared metric, scalar, reference measure, or mobility can still couple the factors. Reducibility requires factorization of the entire dynamical package in Definition 2.
After choosing a local section of P H P_H P H , the master connection is represented by
A = ℓ − 1 e a P a + 1 2 ω a b J a b + A A T A , (36) A = \ell^{-1} e^a P_a + \frac{1}{2}\omega^{ab}J_{ab} + A^A T_A, \tag*{(36)} A = ℓ − 1 e a P a + 2 1 ω ab J ab + A A T A , (36)
where ℓ > 0 \ell> 0 ℓ > 0 is a reference length, e a e^a e a is a coframe, ω a b = − ω b a \omega^{ab} = -\omega^{ba} ω ab = − ω ba is a connection component, and A A A^A A A are internal components. (36) is not yet a decomposition of four forces; it is the component expression of one connection in a chosen basis. All forms in the action and curvature formulas below are the pullbacks of these Cartan data to the base manifold through the chosen local section.
Metric status and the meaning of the master geometry
The metric used in the Euclidean action must be distinguished from the field-space metric used to define gradients and mobilities. On the non-degenerate coframe branch, the Cartan component e e e induces a spacetime metric by
g e : = δ a b e a ⊗ e b , vol g e : = e 1 ∧ e 2 ∧ e 3 ∧ e 4 . (37) g_e := \delta_{ab} e^a \otimes e^b,\qquad\operatorname{vol}_{g_e} := e^1 \wedge e^2 \wedge e^3 \wedge e^4. \tag*{(37)} g e := δ ab e a ⊗ e b , vol g e := e 1 ∧ e 2 ∧ e 3 ∧ e 4 . (37)
Here an orientation and a Euclidean internal frame have been fixed. In the induced-metric realization the isotropy action on the coframe quotient is required to preserve the Euclidean form: if ρ H \rho_H ρ H denotes the action on g / h ≃ R 4 \mathfrak{g}/\mathfrak{h} \simeq\mathbb{R}^4 g / h ≃ R 4 , then ρ H ( h ) T ρ H ( h ) = I 4 \rho_H(h)^{\mathsf{T}}\rho_H(h) = I_4 ρ H ( h ) T ρ H ( h ) = I 4 and det ρ H ( h ) = 1 \det\rho_H(h) = 1 det ρ H ( h ) = 1 for every admitted h ∈ H h \in H h ∈ H . Equivalently, for the connected component of H H H , in an infinitesimal basis one requires the symmetric part of each generator on the coframe indices to vanish. Without this condition the displayed metric depends on the choice of local isotropy frame and is not a gauge-invariant metric. A concrete model may replace this condition by an explicitly specified compensating field, but that field must then be included in the configuration and its variation. In the induced-metric realization, g = g e g = g_e g = g e and the coframe is a component of A A A that is varied together with the other master variables. The non-degeneracy assumption means that e x : T x M → R 4 e_x : T_xM \to\mathbb{R}^4 e x : T x M → R 4 is an isomorphism on every point of the branch under consideration; it is not a consequence of the stochastic notation.
For a finite-dimensional regulator one may instead use a fixed reference metric g 0 g_0 g 0 in the Hodge operator and volume form. This is the background-regularized realization ; in that case the action and all heat-kernel coefficients are defined relative to g 0 g_0 g 0 , and no claim that the spacetime metric is generated by the master dynamics is made. Whenever the symbol g g g is used without a subscript below, the intended realization must therefore be stated: g = g e g = g_e g = g e when the coframe is dynamical, and g = g 0 g = g_0 g = g 0 in a fixed-background calculation. In particular, Appendix A freezes e e e , g g g , and vol g \operatorname{vol}_g vol g while deriving only the connection component of the Euler–Lagrange equation; the full induced-metric theory requires the additional coframe and volume variations. The notation G \mathbb{G} G for the positive field-space metric is independent of the spacetime metric g g g , although a concrete realization may construct G \mathbb{G} G from g g g after gauge fixing.
The phrase “master geometry” consequently refers to the tuple ( P H , A , e , ω , g e ) (P_H, A, e, \omega, g_e) ( P H , A , e , ω , g e ) on the induced branch, or to ( P H , A , g 0 ) (P_H, A, g_0) ( P H , A , g 0 ) in a background-regularized branch. These two realizations share the algebraic curvature formulas, but only the first can support a claim that the metric itself is part of the dynamical master configuration. A Lorentzian metric, if obtained by continuation, requires a separate choice of real form and a proof of the corresponding analytic, causal, and unitarity properties.
Choose a basis { T I } = { P a , J a b , T A } \{T_I\} = \{P_a, J_{ab}, T_A\} { T I } = { P a , J ab , T A } , write A = A I T I A = A^I T_I A = A I T I , and denote the structure constants by [ T I , T J ] = C I J K T K [T_I,T_J] = C_{IJ}^{\ \ K}T_K [ T I , T J ] = C I J K T K . The master curvature is
F = d A + A ∧ A . (38) F = \mathrm{d}A + A \wedge A. \tag*{(38)} F = d A + A ∧ A . (38)
In components,
F = F I T I , F K = d A K + 1 2 C I J K A I ∧ A J . (39) F = F^I T_I,\qquad F^K = \mathrm{d}A^K + \frac{1}{2}C_{IJ}^{\ \ K}A^I \wedge A^J. \tag*{(39)} F = F I T I , F K = d A K + 2 1 C I J K A I ∧ A J . (39)
This component identity is exact and includes every gravitational–internal mixed term through the same structure constants. In the adapted local notation of (36), one has, for example,
T a = ℓ ( d A P a + 1 2 C I J P a A I ∧ A J ) , (40) T^a = \ell\left(\mathrm{d}A^{P_a} + \frac{1}{2}C_{IJ}^{\ \ P_a}A^I \wedge A^J\right), \tag*{(40)} T a = ℓ ( d A P a + 2 1 C I J P a A I ∧ A J ) , (40)
R c d = 2 ( d A J c d + 1 2 C I J J c d A I ∧ A J ) , (41) R^{cd} = 2\left(\mathrm{d}A^{J_{cd}} + \frac{1}{2}C_{IJ}^{\ \ J_{cd}}A^I \wedge A^J\right), \tag*{(41)} R c d = 2 ( d A J c d + 2 1 C I J J c d A I ∧ A J ) , (41)
F A = d A T A + 1 2 C I J T A A I ∧ A J . (42) F^A = \mathrm{d}A^{T_A} + \frac{1}{2}C_{IJ}^{\ \ T_A}A^I \wedge A^J. \tag*{(42)} F A = d A T A + 2 1 C I J T A A I ∧ A J . (42)
If the selected algebra has the usual Lorentz brackets, the first two lines contain d e a + ω b a ∧ e b \mathrm{d}e^a + \omega^a_{\ b} \wedge e^b d e a + ω b a ∧ e b and d ω c d + ω e c ∧ ω e d \mathrm{d}\omega^{cd} + \omega^c_{\ e} \wedge\omega^{ed} d ω c d + ω e c ∧ ω e d ; the remaining terms are not omitted but are exactly the corresponding C I J K C_{IJ}^{\ \ K} C I J K -contributions. In particular, the coefficient of the e a ∧ e b e^a \wedge e^b e a ∧ e b term in R c d R^{cd} R c d is fixed by C P a P b J c d ℓ − 2 C_{P_aP_b}^{\ \ J_{cd}}\ell^{-2} C P a P b J c d ℓ − 2 under the normalization in (39); it is present precisely when [ P a , P b ] [P_a, P_b] [ P a , P b ] has the displayed J c d J_{cd} J c d -component.
Matter, order parameter, and information as one state variable
Let E Ψ → M E_{\Psi} \to M E Ψ → M be the representation bundle for the complete matter multiplet Ψ \Psi Ψ , and let E H → M E_{H} \to M E H → M be a representation bundle in which a possible scalar order parameter Φ H \Phi_{H} Φ H takes values. Let E χ → M E_{\chi} \to M E χ → M be the specified information-state bundle, possibly the trivial real line bundle, and let χ \chi χ be a section of E χ E_{\chi} E χ . Let E σ : = M × R E_{\sigma} := M \times\mathbb{R} E σ := M × R carry the dimensionless shared scalar coordinate σ \sigma σ used below; its covariant differential is simply d σ \mathrm{d}\sigma d σ . The scalar is a genuine configuration coordinate whenever the joint candidate is used, not a parameter silently added after variation. The information state is not assumed to be a new fundamental force. It is an auxiliary or potentially fundamental variable whose physical status must be determined by the selection problem. Its inclusion is motivated by the fact that probability, entropy, and state preparation enter the stochastic principle. The order parameter is written explicitly because its covariant derivative and its representation are needed in the conditional electroweak statement below; omitting it from the primitive configuration would make that statement ill-typed.
Define the field configuration
X = ( A , Ψ , Φ H , σ , χ ) , (43) \mathsf{X} = (A, \Psi, \Phi_{H}, \sigma, \chi), \tag*{(43)} X = ( A , Ψ , Φ H , σ , χ ) , (43)
The spacetime geometry is derived from the translational component of A A A , rather than appended as a fifth independent interaction. The precise induced metric, the alternative fixed-background regularization, and the scope of the fixed-coframe variation are given in Subsection 3.3 and equation (37) . This cross-reference is important: the regularized stochastic identities do not by themselves assert that a heat-kernel calculation creates a metric from no geometric data.
Let X \mathcal{X} X denote the chosen configuration manifold (finite dimensional or rigorously regularized when analytic identities are used) containing these quintuples. We reserve μ i n f o \mu_{\mathrm{info}} μ info for the information-state parameter entering I \mathcal{I} I , while the evolving probability law of X t \mathsf{X}_{t} X t is denoted by μ t \mu_{t} μ t (or by its density p t p_{t} p t when a reference volume exists). Thus the probability law is not confused with a field coordinate, and X \mathcal{X} X contains the geometric and matter/order-parameter variables themselves. In particular, μ t \mu_{t} μ t is not a variational component of X \mathsf{X} X in the present formulation; it enters through the probability and free-energy layer. Identifying μ t \mu_{t} μ t with a component of the primitive state would produce a McKean–Vlasov problem and would require a separate variational and Fokker–Planck analysis.
A master covariant derivative is defined by
D = d + ρ r e p ( A ) , (44) \mathbb{D} = \mathrm{d} + \rho_{\mathrm{rep}}(A), \tag*{(44)} D = d + ρ rep ( A ) , (44)
where ρ r e p \rho_{\mathrm{rep}} ρ rep denotes the relevant representation. Since E E E is associated to the principal H H H -bundle, this notation presupposes either that the representation extends from H H H to the selected action of the full Cartan algebra g \mathfrak{g} g , or that $\mathbb{D is explicitly understood to use only the h \mathfrak{h} h -component of A A A . A concrete model must make this choice before the covariant derivatives below are defined. Thus
D Ψ , D Φ H , D χ (45) \mathbb{D}\Psi,\qquad\mathbb{D}\Phi_H,\qquad\mathbb{D}\chi \tag*{(45)} D Ψ , D Φ H , D χ (45)
are components of one covariant differential rather than separately postulated kinetic terms. The representation used for E H E_H E H must be specified (or extended from H H H to g \mathfrak{g} g ) before D Φ H \mathbb{D}\Phi_H D Φ H is defined.
The single master action
The geometric construction supplies curvature and covariant derivatives, but a scalar action also requires pairings and normalizations. We state these data before forming the extended norm. Their positivity controls the Euclidean action; their gauge equivariance ensures that the value of the action is independent of the chosen local representative.
Invariant pairing
Let ⟨ ⋅ , ⋅ ⟩ g \langle\cdot,\cdot\rangle_{\mathfrak{g}} ⟨ ⋅ , ⋅ ⟩ g be an invariant bilinear form on the master algebra:
⟨ [ X , Y ] , Z ⟩ g + ⟨ Y , [ X , Z ] ⟩ g = 0 ∀ X , Y , Z ∈ g . (46) \langle[X,Y],Z\rangle_{\mathfrak{g}}+\langle Y,[X,Z]\rangle_{\mathfrak{g}}=0\qquad\forall X,Y,Z\in\mathfrak{g}. \tag*{(46)} ⟨[ X , Y ] , Z ⟩ g + ⟨ Y , [ X , Z ] ⟩ g = 0 ∀ X , Y , Z ∈ g . (46)
In the basis { T I } \{T_I\} { T I } , write ⟨ T I , T J ⟩ g = h I J \langle T_I,T_J\rangle_{\mathfrak{g}}=h_{IJ} ⟨ T I , T J ⟩ g = h I J . For a general master algebra, the invariant form may contain mixed blocks, but this is not automatic. Distinct simple ideals in a direct-product algebra are orthogonal for every invariant bilinear form, even when the ideals are isomorphic. Indeed, if X = [ U , V ] X=[U,V] X = [ U , V ] belongs to one simple ideal and Y Y Y belongs to another, invariance gives ⟨ [ U , V ] , Y ⟩ g = − ⟨ V , [ U , Y ] ⟩ g = 0 \langle[U,V],Y\rangle_{\mathfrak{g}}=-\langle V,[U,Y]\rangle_{\mathfrak{g}}=0 ⟨[ U , V ] , Y ⟩ g = − ⟨ V , [ U , Y ] ⟩ g = 0 ; every element of a simple ideal is a sum of such commutators. The same argument makes every perfect ideal orthogonal to the centre. Mixed invariant pairings within an abelian centre remain possible. Consequently, for the explicit product candidate in § 5.4 , the gravitational–internal coupling is carried by the shared order parameter and the coupled field-space mobility, not by an assumed cross-pairing between the simple Lie-algebra factors. Whether the full extended pairing is block diagonal is determined by its representations and shared field dependence. Symmetry reduction can induce further block structure, but is not required for an invariant orthogonal decomposition already present microscopically. For a non-compact real form, an invariant form is generally indefinite; consequently, the positivity assumption used for the Euclidean stochastic theory entails a specified Euclidean/compact real slice (or another explicitly defined positive regularization) and cannot be transferred to the Lorentzian continuation without a separate argument.
Let G \mathsf{G} G be the positive field-space metric on the bosonic variables, let G H \mathsf{G}_H G H denote the positive pairing on the scalar order-parameter sector, and let K \mathsf{K} K be the specified regularized fermionic pairing. Let P μ i n f o \mathcal{P}_{\mu_{\mathrm{info}}} P μ info be a μ i n f o \mu_{\mathrm{info}} μ info -dependent Hilbert bundle over M M M , obtained from a specified realization of the tangent space to the probability-law manifold at μ i n f o \mu_{\mathrm{info}} μ info . Let E H , 0 H \mathcal{E}^{\mathsf{H}}_{H,0} E H , 0 H be an optional algebraic auxiliary bundle. If a candidate is to generate a scalar potential from the one-norm principle, it must supply a specified map J H : E H × R → E H , 0 \mathcal{J}_H:E_H \times\mathbb{R} \to\mathcal{E}_{H,0} J H : E H × R → E H , 0 ; if no such bundle and map are present, no scalar potential is claimed to follow from the master action. We then set
E e x t : = ( Λ 2 T ∗ M ⊗ g P H ) ⊕ ( T ∗ M ⊗ E Ψ ) ⊕ ( T ∗ M ⊗ E H ) ⊕ ( T ∗ M ⊗ E σ ) ⊕ ( T ∗ M ⊗ E χ ) ⊕ E H , 0 ⊕ J μ i n f o , (47) \mathcal{E}_{\mathrm{ext}} := (\Lambda^2 T^*M \otimes\mathfrak{g}_{PH}) \oplus(T^*M \otimes E_\Psi) \oplus(T^*M \otimes E_H)
\oplus(T^*M \otimes E_\sigma) \oplus(T^*M \otimes E_\chi) \oplus\mathcal{E}_{H,0} \oplus\mathcal{J}_{\mu_{\mathrm{info}}}, \tag*{(47)} E ext := ( Λ 2 T ∗ M ⊗ g P H ) ⊕ ( T ∗ M ⊗ E Ψ ) ⊕ ( T ∗ M ⊗ E H ) ⊕ ( T ∗ M ⊗ E σ ) ⊕ ( T ∗ M ⊗ E χ ) ⊕ E H , 0 ⊕ J μ info , (47)
Here g P H : = P H × Ad ∣ H g \mathfrak{g}_{PH}:=P_H \times_{\operatorname{Ad}|_H} \mathfrak{g} g P H := P H × Ad ∣ H g ; the extended curvature is a section of this bundle. This distinguishes a bundle from its space of differential-form sections. The restrictions of h \mathbf{h} h to the bosonic, scalar, and regularized fermionic sectors are built from G \mathbf{G} G , G H \mathbf{G}_H G H , and K \mathbf{K} K , respectively. The seven entries in this direct sum have different form degrees and fibre types; their addition below is therefore shorthand for the ordered tuple, not an untyped sum. We fix a continuous graded bilinear form h \mathbf{h} h on E e x t \mathcal{E}_{\mathrm{ext}} E ext , including the required Hodge maps and reference scales, and assume that it is non-degenerate and positive on the chosen Euclidean real slice. Without this assumption the expression is not a norm and neither boundedness below nor a Gibbs measure follows. For Grassmann-valued matter variables, this positivity statement refers to the associated regularized quadratic form after the fermionic integration; it is not a claim that a Grassmann algebra carries an ordinary pointwise norm. In particular, J ( μ i n f o ) \mathcal{J}(\mu_{\mathrm{info}}) J ( μ info ) is required to be represented as a section over M M M , not merely as an abstract functional derivative. At the abstract level used here, J \mathcal{J} J is an unspecified map from the admissible probability-law manifold into this section bundle; a concrete realization must define that map (for example, by specifying an information functional and its variational derivative). No particular information functional or gradient representation of J \mathcal{J} J is assumed here. The preceding section-valued requirement is part of the data, not a consequence of notation. Let P \mathcal{P} P denote the chosen manifold of admissible probability laws on the regularized configuration space. More explicitly, a realization must provide an evaluation or pullback map
E μ i n f o : T μ i n f o P ⟶ Γ ( M , J μ i n f o ) . (48) \mathcal{E}_{\mu_{\mathrm{info}}}:T_{\mu_{\mathrm{info}}}\mathcal{P}\longrightarrow\Gamma(M,\mathcal{J}_{\mu_{\mathrm{info}}}). \tag*{(48)} E μ info : T μ info P ⟶ Γ ( M , J μ info ) . (48)
and J \mathcal{J} J denotes a specified map whose value is interpreted through E μ i n f o \mathcal{E}_{\mu_{\mathrm{info}}} E μ info . Without such a map the last component is only a formal ordered entry and the master action is not yet a well-typed geometric functional.
We also require the pairing used in the action to be equivariant under every gauge transformation admitted by the chosen bundle. If R g R_g R g is its induced action on the extended bundle, the required condition is
h ( R g U , R g V ) = h ( U , V ) for all admissible g and U , V ∈ E e x t . (49) \mathbf{h}(R_gU,R_gV)=\mathbf{h}(U,V) \qquad\text{for all admissible } g \text{ and } U,V\in\mathcal{E}_{\mathrm{ext}}. \tag*{(49)} h ( R g U , R g V ) = h ( U , V ) for all admissible g and U , V ∈ E ext . (49)
Thus the Hodge operators, reference scales, and mixed blocks entering h \mathbf{h} h must be H H H -equivariant (or equivariant under a declared extension of the action). The Lie-algebra identity (46) alone does not imply (49) for the full extended tuple.
The same requirement must be imposed on every map used to form that tuple. If R h H R_h^H R h H and R h 0 R_h^0 R h 0 denote the actions on E H E_H E H and E H , 0 \mathcal{E}_{H,0} E H , 0 , respectively, then
L H ( R h H Φ , σ ) = R h 0 L H ( Φ , σ ) , h ∈ H . (50) \mathcal{L}_H\left(R_h^H\Phi,\sigma\right)=R_h^0\mathcal{L}_H(\Phi,\sigma),\qquad h\in H. \tag*{(50)} L H ( R h H Φ , σ ) = R h 0 L H ( Φ , σ ) , h ∈ H . (50)
Here σ \sigma σ is a gauge singlet, as specified in the configuration. The information source must either be gauge invariant, R h I ( μ i n f o ) = I ( μ i n f o ) R_h\mathcal{I}(\mu_{\mathrm{info}})=\mathcal{I}(\mu_{\mathrm{info}}) R h I ( μ info ) = I ( μ info ) , or transform as a declared spurion. If h \mathbf{h} h depends on the field, its condition is the corresponding equivariance identity h X ( R h U , R h V ) = h X ( U , V ) \mathbf{h}_{X}(R_hU,R_hV)=\mathbf{h}_{X}(U,V) h X ( R h U , R h V ) = h X ( U , V ) . These conditions are part of the definition of a gauge-covariant realization; they do not follow from positivity of the fibre pairing. We now define
F M e x t : = ( F , ℓ Ψ − 1 D Ψ , ℓ H − 1 D Φ H , ℓ σ − 1 d σ , ℓ χ − 1 D χ , ℓ V − 2 L H ( Φ H , σ ) , ℓ μ − 1 I ( μ i n f o ) ) ∈ Γ ( M , E e x t ) . (51) \mathbf{F}_{M}^{\mathrm{ext}}:=\left(\mathbf{F},\ell_{\Psi}^{-1}\mathsf{D}\Psi,\ell_{H}^{-1}\mathsf{D}\Phi_{H},\ell_{\sigma}^{-1}d\sigma,\ell_{\chi}^{-1}\mathsf{D}\chi,\ell_{V}^{-2}\mathcal{L}_{H}(\Phi_{H},\sigma),\ell_{\mu}^{-1}\mathcal{I}(\mu_{\mathrm{info}})\right)\in\Gamma(M,\mathcal{E}_{\mathrm{ext}}). \tag*{(51)} F M ext := ( F , ℓ Ψ − 1 D Ψ , ℓ H − 1 D Φ H , ℓ σ − 1 d σ , ℓ χ − 1 D χ , ℓ V − 2 L H ( Φ H , σ ) , ℓ μ − 1 I ( μ info ) ) ∈ Γ ( M , E ext ) . (51)
where ℓ Ψ , ℓ H , ℓ σ , ℓ χ , ℓ V , ℓ μ \ell_{\Psi},\ell_{H},\ell_{\sigma},\ell_{\chi},\ell_{V},\ell_{\mu} ℓ Ψ , ℓ H , ℓ σ , ℓ χ , ℓ V , ℓ μ are reference scales introduced only to make the components dimensionally commensurate; ℓ μ \ell_{\mu} ℓ μ is associated with the fixed information parameter μ i n f o \mu_{\mathrm{info}} μ info . The algebraic component L H ( Φ H , σ ) \mathcal{L}_{H}(\Phi_{H},\sigma) L H ( Φ H , σ ) is the precise place at which a Higgs-type potential may enter the one-norm construction. For example, after a specified reduction of the E H , 0 \mathcal{E}_{H,0} E H , 0 block, its diagonal contribution can define a function V H ( Φ H ) V_{H}(\Phi_{H}) V H ( Φ H ) ; this is not an independently added scalar potential. If the candidate supplies no such map, the later Higgs discussion is only a target condition and no potential is derived. Because μ i n f o \mu_{\mathrm{info}} μ info is fixed in the present formulation, the last entry is a prescribed source section rather than an independently stochastic coordinate. It can influence the drift only through the chosen pairing and its cross-blocks with the varied fields; if the pairing is block diagonal and the source is field-independent, its contribution is an additive constant and does not alter the field equations. A dynamical information law would require promoting μ i n f o \mu_{\mathrm{info}} μ info to the configuration and rederiving the stochastic and equilibrium identities. The construction is formal until the Hilbert realization, the reference measure, and the pairing h \mathbf{h} h are supplied. The stochastic and Gibbs arguments below are understood on a bosonic finite-dimensional or regularized realization (for example, after the fermionic variables have been integrated out); no Itô diffusion on a raw Grassmann algebra is asserted. The positive Euclidean norm of a component such as D Ψ \mathsf{D}\Psi D Ψ produces only a quadratic second-order form of schematic type ⟨ D Ψ , K D Ψ ⟩ \langle\mathsf{D}\Psi,K\mathsf{D}\Psi\rangle ⟨ D Ψ , K D Ψ ⟩ . It does not, without an additional construction, equal the first-order chiral Dirac functional \int\overline{\Psi}\,\slashed{D}\,\Psi . Obtaining the latter requires a specified Grassmann functional integral, a Dirac operator and its adjoint, a fermion measure, the determinant or Pfaffian phase, and a regulator for chiral and gauge anomalies. Thus any occurrence of Ψ ‾ D Ψ \overline{\Psi}\mathsf{D}\Psi Ψ D Ψ in a matched effective target is a conditional fermionic interface, not a consequence of positivity of the bosonic master norm alone.
The master action is therefore the single norm-square
S M [ X ] = 1 4 ∫ M ⟨ F M e x t , F M e x t ⟩ h vol g (52) \mathcal{S}_{\mathrm{M}}[X] = \frac{1}{4}\int_{M}\left\langle\mathbf{F}^{\mathrm{ext}}_{\mathrm{M}},\mathbf{F}^{\mathrm{ext}}_{\mathrm{M}}\right\rangle_{\mathbf{h}} \operatorname{vol}_{g} \tag*{(52)} S M [ X ] = 4 1 ∫ M ⟨ F M ext , F M ext ⟩ h vol g (52)
Here “one pairing” means the single graded pairing h \mathbf{h} h on E e x t \mathcal{E}_{\mathrm{ext}} E ext . The objects h I J h_{IJ} h I J , G \mathbf{G} G , K \mathbf{K} K , and the reduced κ A B \kappa_{AB} κ A B are its restrictions or induced blocks after the relevant geometric structures have been fixed; they are not independent action functionals. Any order-parameter potential is, by definition, included as a component of the extended curvature in E e x t \mathcal{E}_{\mathrm{ext}} E ext ; introducing an independent additive potential would violate the one-norm principle. Equation (52) is the fundamental definition used in the conceptual formulation. Appendix A derives the fixed-background connection variation; § 5.1 treats the metric and Hodge dependence associated with a dynamical coframe. A complete Euler–Lagrange system also includes the matter and order-parameter variations and any field dependence of the pairing. The information-state parameter is fixed, so there is no probability-law Euler–Lagrange equation in this formulation. If the Hodge maps in h \mathbf{h} h are built from the induced metric g [ X ] g[X] g [ X ] , then h \mathbf{h} h is field-dependent in the full problem; freezing it is an explicitly stated fixed-background slice. The primitive action is the squared norm of an extended master curvature. Its components and invariant pairing are part of the primitive data. After those data are fixed, expansion determines the displayed sector terms. This representation alone does not determine independently variable block weights or potentials: an arbitrary nonnegative sum can also be represented as a norm on a direct sum. A substantive unification claim therefore requires constraints on the primitive data, a nonfactorization result for the complete dynamics, and jointly compatible physical projections. The evolving probability law enters through the stochastic and free-energy layers below.
The extended-curvature formulation also imposes that the order-parameter potential be generated by the same master data. No independent sector-specific potential is introduced at the microscopic level. In a concrete realization, the scalar potential must arise from the internal components, algebraic constraints, or auxiliary directions of F M e x t \mathbf{F}^{\mathrm{ext}}_{\mathrm{M}} F M ext . The observed masses and couplings must consequently arise through vacuum selection and matching.
In the stochastic and equilibrium sections, μ i n f o \mu_{\mathrm{info}} μ info in I ( μ i n f o ) I(\mu_{\mathrm{info}}) I ( μ info ) is held fixed as an information-state parameter, and its dependence is suppressed in the notation S M [ X ] \mathcal{S}_{\mathrm{M}}[X] S M [ X ] . If μ i n f o \mu_{\mathrm{info}} μ info is instead identified with the evolving law μ t \mu_t μ t of X t X_t X t , the model becomes a McKean–Vlasov system and the Markov Fokker–Planck and Gibbs statements below require a separate derivation; that nonlinear-law case is not asserted here.
Factorization of the complete dynamics
The distinction can be stated formally. Suppose the configuration manifold were a product
X = X g × X i (53) \mathcal{X} = \mathcal{X}_g \times\mathcal{X}_i \tag*{(53)} X = X g × X i (53)
and suppose both the invariant pairing and the action factorized:
h = h g ⊕ h i , S M = S g + S i . (54) h = h_g \oplus h_i,\qquad S_{\mathrm{M}} = S_g + S_i. \tag*{(54)} h = h g ⊕ h i , S M = S g + S i . (54)
If, in addition, the reference measure is a product and the mobility and all shared field dependence split into the same independent factors, the associated dynamics factorizes. A block decomposition of the action alone does not give that conclusion. The irreducibility requirement below therefore concerns the complete configuration, action, measure, and stochastic data.
Definition 2 (Microscopic irreducible unification). A master model is called microscopically irreducible if its complete data package admits no nontrivial product decomposition into two independent configuration spaces, reference measures, actions, admissible density classes, and Markov generators. A decomposition of the Lie algebra into orthogonal invariant ideals is only a diagnostic for such a factorization; it is neither necessary nor sufficient. In particular, a simple master algebra can still be accompanied by an independent neutral field, while a direct-product algebra can remain coupled through a shared scalar, metric, measure, or mobility. The four low-energy sectors are said to be emergent if their decoupled effective actions are obtained only after a mathematically specified limiting or symmetry-breaking operation.
This definition is intentionally stronger than simply writing a larger Lagrangian. For later reference, collect the microscopic data into the single object
Γ : = ( G ~ , H , P H , R , h , E e x t , U H , I , X Γ , ν Γ , G Γ , M Γ , m Γ , μ i n f o ) . (55) \Gamma:= \left(\widetilde{G}, H, P_H, \mathcal{R}, \mathbf{h}, \mathcal{E}_{\mathrm{ext}}, \mathcal{U}_H, \mathcal{I}, \mathcal{X}_{\Gamma}, \nu_{\Gamma}, \mathcal{G}_{\Gamma}, \mathcal{M}_{\Gamma}, m_{\Gamma}, \mu_{\mathrm{info}}\right). \tag*{(55)} Γ := ( G , H , P H , R , h , E ext , U H , I , X Γ , ν Γ , G Γ , M Γ , m Γ , μ info ) . (55)
Here R \mathcal{R} R denotes all representation data, including the matter and order-parameter representations; X Γ \mathcal{X}_{\Gamma} X Γ and ν Γ \nu_{\Gamma} ν Γ are the candidate configuration space and reference-measure class; and m Γ m_{\Gamma} m Γ records whether the metric is induced by the coframe or held fixed as a regulator. Across the abstract candidate class, the residual group, representation content, and coupling parameters may vary with Γ \Gamma Γ ; they are not universal conclusions of the notation. In the worked candidate Γ 0 \Gamma_0 Γ 0 , the internal group and the displayed matter representations are explicit inputs. They become predictions only if independent selection, anomaly, and matching arguments determine them from a previously fixed admissible class. The irreducibility requirement is a requirement on the entire package Γ \Gamma Γ , not on a Lie bracket alone: a nonzero mixed bracket is neither necessary nor sufficient if the pairing, representation maps, or mobility still split into independent blocks.
The extended norm and its component expansion
The distinction can be made algebraically. Let F M e x t \mathbf{F}^{\mathrm{ext}}_{\mathrm{M}} F M ext be a vector in an extended bundle E \mathcal{E} E with invariant metric h \mathbf{h} h . The only primitive scalar is
ϵ M = 1 4 ⟨ F M e x t , F M e x t ⟩ h . (56) \epsilon_{\mathrm{M}} = \frac{1}{4}\left\langle\mathbf{F}^{\mathrm{ext}}_{\mathrm{M}}, \mathbf{F}^{\mathrm{ext}}_{\mathrm{M}}\right\rangle_{\mathbf{h}}. \tag*{(56)} ϵ M = 4 1 ⟨ F M ext , F M ext ⟩ h . (56)
A decomposition E = E 1 ⊕ ⋯ ⊕ E k \mathcal{E}=E_1\oplus\cdots\oplus E_k E = E 1 ⊕ ⋯ ⊕ E k expands this contraction into its diagonal and permitted mixed terms. This coordinate expansion does not by itself establish either independence or coupling. Mixed components are allowed only when the representation and invariance conditions permit them; distinct simple ideals are already orthogonal for an invariant Lie-algebra pairing. Shared field dependence can nevertheless couple orthogonal blocks. Sector separation is an emergent property only when a specified limiting argument establishes the corresponding factorization of the effective dynamics. For example, on two compact simple ideals the positive invariant pairing can have the form
h a l g = α ( − B s u ( 2 ) ) ⊕ β ( − B s u ( 3 ) ) , α , β > 0 , (57) \mathbf{h}_{\mathrm{alg}}=\alpha(-B_{\mathfrak{su}(2)})\oplus\beta(-B_{\mathfrak{su}(3)}),\qquad\alpha,\beta>0, \tag*{(57)} h alg = α ( − B su ( 2 ) ) ⊕ β ( − B su ( 3 ) ) , α , β > 0 , (57)
where the B B B ’s are the Killing forms. After fixed generator normalizations, the ratio of the two curvature coefficients is α / β \alpha/\beta α / β , which is unrestricted by positivity alone. A common multiplier w ( σ ) w(\sigma) w ( σ ) cancels from that ratio. Accordingly, the extended norm provides a common variational formulation; relations among its independent weights require further microscopic constraints or a solved selection problem.
The common path-space cost functional
The action determines variational forces once a mobility and a reference measure have been chosen. A diffusion then adds fluctuations with a specified covariance. The control cost measures the additional forcing needed to change that diffusion’s drift. This construction connects the geometric action to probability without differentiating Brownian sample paths.
The normalization in this stochastic realization must be stated separately from that of a Lorentzian mechanical action. If an unnormalized geometric functional I M \mathcal{I}_{\mathrm{M}} I M has action units, choose an action unit I r e f > 0 I_{\mathrm{ref}}>0 I ref > 0 and an energy unit E r e f > 0 E_{\mathrm{ref}}>0 E ref > 0 , and use
S ˉ M : = I M / I r e f , U M : = E r e f S ˉ M , S M ≡ U M , θ = E r e f θ ˉ . (58) \bar{S}_{\mathrm{M}}:=\mathcal{I}_{\mathrm{M}}/I_{\mathrm{ref}},\qquad U_{\mathrm{M}}:=E_{\mathrm{ref}}\bar{S}_{\mathrm{M}},\qquad S_{\mathrm{M}}\equiv U_{\mathrm{M}},\qquad\theta=E_{\mathrm{ref}}\bar{\theta}. \tag*{(58)} S ˉ M := I M / I ref , U M := E ref S ˉ M , S M ≡ U M , θ = E ref θ ˉ . (58)
Here I M \mathcal{I}_{\mathrm{M}} I M has action units, S ˉ M \bar{S}_{\mathrm{M}} S ˉ M and θ ˉ \bar{\theta} θ ˉ are dimensionless, and U M U_{\mathrm{M}} U M , S M S_{\mathrm{M}} S M , and θ \theta θ have energy units whenever dimensional stochastic variables are used. We retain S M S_{\mathrm{M}} S M in the existing variational identities as the energy-normalized notation. A dimensionless version is always obtained by dividing it by E r e f E_{\mathrm{ref}} E ref , not merely by setting ℏ = c = 1 \hbar=c=1 ℏ = c = 1 . Thus S M / θ S_{\mathrm{M}}/\theta S M / θ is dimensionless. The reference factors may be included in the pairing in (52); the subsequent stochastic equations use this energy-normalized functional. The ratio I r e f / E r e f I_{\mathrm{ref}}/E_{\mathrm{ref}} I ref / E ref is a unit conversion, not a derived material clock, and I r e f I_{\mathrm{ref}} I ref is not identified with Planck’s constant without additional quantum matching. The physical Lorentzian action, whose stationary variation gives the mechanical equations, retains its own normalization. A residual control cost with energy units cannot be identified with it merely by using the same word action . Appendix E gives the component units, the conversion of the noise and generator, and the corresponding transformation of the path and terminal costs.
Master mobility and stochastic dynamics
All stochastic and equilibrium identities in this section and in § 4.1 are understood on a fixed physical configuration slice X p h y s \mathcal{X}_{\mathrm{phys}} X phys , or on the quotient by gauge orbits. In a gauge-fixed realization the reference measure includes the stated Faddeev–Popov (or equivalent BRST/BV) factor; on a quotient it is the corresponding quotient measure. The raw unreduced field space is used only as notation for the master variables. A probability construction requires a positive reference measure on that slice or quotient; a signed Faddeev–Popov determinant or a Grassmann-valued BRST integral is not by itself such a measure. Removing gauge zero modes is necessary in these realizations, but does not prove confinement, finiteness of the partition function, or invertibility on all remaining directions.
Let X \mathcal{X} X be a separable Hilbert realization of the configuration space, and let M ( X ) : T ∗ X → T X \mathrm{M}(\mathcal{X}):T^{*}\mathcal{X}\to T\mathcal{X} M ( X ) : T ∗ X → T X be a mobility operator with a specified domain. Assume that it is symmetric and nonnegative with respect to the field-space metric (strict positivity is imposed whenever an ordinary inverse is used). In the finite-dimensional or Galerkin realization used for the identities below, the stochastic equation is interpreted in the Itô sense. The variational derivative δ S M / δ X \delta S_{\mathrm{M}}/\delta\mathcal{X} δ S M / δ X is a cotangent vector. Whenever self-adjointness is invoked, the field-space metric supplies the cotangent–tangent identification; equivalently, if G \mathrm{G} G is the field-space metric, the gradient is grad G S M = G − 1 δ S M / δ X \operatorname{grad}_{\mathrm{G}} S_{\mathrm{M}}=\mathrm{G}^{-1}\delta S_{\mathrm{M}}/\delta\mathcal{X} grad G S M = G − 1 δ S M / δ X . To keep the types explicit, write ♭ G : T X → T ∗ X \flat_{\mathrm{G}}:T\mathcal{X}\to T^{*}\mathcal{X} ♭ G : T X → T ∗ X for this metric identification and define the induced tangent mobility
M ^ : = M ∘ ♭ G : T X → T X , ⟨ u , M ^ v ⟩ G = ⟨ M ^ u , v ⟩ G , ⟨ u , M ^ u ⟩ G ≥ 0. (59) \widehat{\mathrm{M}}:=\mathrm{M}\circ\flat_{\mathrm{G}}:T\mathcal{X}\to T\mathcal{X},\qquad\langle u,\widehat{\mathrm{M}}v\rangle_{\mathrm{G}}=\langle\widehat{\mathrm{M}}u,v\rangle_{\mathrm{G}},\qquad\langle u,\widehat{\mathrm{M}}u\rangle_{\mathrm{G}}\ge0. \tag*{(59)} M := M ∘ ♭ G : T X → T X , ⟨ u , M v ⟩ G = ⟨ M u , v ⟩ G , ⟨ u , M u ⟩ G ≥ 0. (59)
Thus expressions containing a differential, such as M d f \mathrm{M}\,df M df , use the cotangent-to-tangent map, whereas expressions containing a Riemannian gradient, such as M ^ ∇ X f \widehat{\mathrm{M}}\,\nabla_{\mathcal{X}}f M ∇ X f , use the induced tangent operator. In coordinates the cotangent-to-tangent map has components M i j M^{ij} M ij , whereas the tangent endomorphism has components M ^ i j = M i k G k j \widehat{M}^{i}{}_{j}=M^{ik}G_{kj} M i j = M ik G k j . Their matrices coincide only in an orthonormal affine chart. The displayed Itô equation below uses such a fixed Hilbert or finite-dimensional affine realization; on a curved configuration space the generator, rather than an uncorrected vector drift, specifies the intrinsic diffusion. To make the covariance convention independent of coordinates, let U U U be the noise Hilbert space and choose a noise frame E X : U → T X X \mathcal{E}_{X}:U\to T_{X}\mathcal{X} E X : U → T X X satisfying, in a coordinate chart, E X E X T = G − 1 \mathcal{E}_{X}\mathcal{E}_{X}^{\mathsf{T}}=G^{-1} E X E X T = G − 1 . Let M ^ G 1 / 2 \widehat{M}_{G}^{1/2} M G 1/2 denote the positive square root with respect to ⟨ ⋅ , ⋅ ⟩ G \langle\cdot,\cdot\rangle_{G} ⟨ ⋅ , ⋅ ⟩ G . We use 2 θ M ^ G 1 / 2 E X \sqrt{2\theta}\widehat{M}_{G}^{1/2}\mathcal{E}_{X} 2 θ M G 1/2 E X as the noise coefficient. Its coordinate covariance is 2 θ M i j 2\theta M^{ij} 2 θ M ij , while its field-space covariance is 2 θ M ^ 2\theta\widehat{M} 2 θ M when the adjoint is taken with respect to G G G . In finite-dimensional coordinates, putting R = M ^ G 1 / 2 R=\widehat{M}_{G}^{1/2} R = M G 1/2 makes this verification explicit:
R T G = G R , R 2 = M G , σ σ T = 2 θ R G − 1 R T = 2 θ R 2 G − 1 = 2 θ M . (60) R^{\mathsf{T}}G=GR,\qquad R^{2}=MG,\qquad\sigma\sigma^{\mathsf{T}}=2\theta RG^{-1}R^{\mathsf{T}}=2\theta R^{2}G^{-1}=2\theta M. \tag*{(60)} R T G = GR , R 2 = M G , σ σ T = 2 θ R G − 1 R T = 2 θ R 2 G − 1 = 2 θ M . (60)
The first equality expresses self-adjointness in the field-space metric. It explains why a coordinate-standard Wiener process cannot be multiplied by the matrix R R R alone in a nonorthonormal chart.
Write B \mathbb{B} B for the complete reference drift. Its dependence on θ \theta θ is suppressed except when a small-noise limit is taken. On the reversible coordinate realization relative to ν ( d x ) = m ν ( x ) d x \nu(dx)=m_{\nu}(x)dx ν ( d x ) = m ν ( x ) d x , it is
B i ( x ) = − M i j ( x ) ∂ j S M ( x ) + θ β ν i ( x ) , β ν i ( x ) = 1 m ν ( x ) ∂ j ( m ν ( x ) M i j ( x ) ) . (61) \mathbb{B}^{i}(x)=-M^{ij}(x)\partial_{j}S_{\mathrm{M}}(x)+\theta\beta_{\nu}^{i}(x),\qquad\beta_{\nu}^{i}(x)=\frac{1}{m_{\nu}(x)}\partial_{j}\bigl(m_{\nu}(x)M^{ij}(x)\bigr). \tag*{(61)} B i ( x ) = − M ij ( x ) ∂ j S M ( x ) + θ β ν i ( x ) , β ν i ( x ) = m ν ( x ) 1 ∂ j ( m ν ( x ) M ij ( x ) ) . (61)
Section 4.1 derives this correction from the forward equation. With constant mobility and a flat reference density the correction vanishes and the drift reduces to − M δ S M / δ X -M\,\delta S_{\mathrm{M}}/\delta X − M δ S M / δ X . The general stochastic equation is
d X t = B ( X t ) d t + 2 θ M ^ G 1 / 2 ( X t ) E X t d W t . (62) dX_{t}=\mathbb{B}(X_{t})\,dt+\sqrt{2\theta}\widehat{M}_{G}^{1/2}(X_{t})\mathcal{E}_{X_{t}}\,dW_{t}. \tag*{(62)} d X t = B ( X t ) d t + 2 θ M G 1/2 ( X t ) E X t d W t . (62)
where W t W_{t} W t is a cylindrical Wiener process on U U U and θ > 0 \theta>0 θ > 0 is the fluctuation scale. An infinite-dimensional strong Itô equation with values in the state Hilbert space requires the displayed composite coefficient to be an admissible Hilbert–Schmidt, or more generally radonifying, noise coefficient. A mild equation may instead be driven by cylindrical noise when its linear evolution family supplies enough smoothing. For example, in the autonomous case with linear generator − A -A − A and noise coefficient Σ \Sigma Σ , the stochastic convolution is state-space valued under the sufficient condition
∫ 0 T ∥ e − s A Σ ∥ L 2 ( U , H ) 2 d s < ∞ for every T < ∞ . (63) \int_{0}^{T}\left\|e^{-sA}\Sigma\right\|_{L_{2}(U,H)}^{2}\,ds<\infty\qquad\text{for every }T<\infty. \tag*{(63)} ∫ 0 T e − s A Σ L 2 ( U , H ) 2 d s < ∞ for every T < ∞. (63)
The nonlinear exponential realization in Section 10.22 uses this mild convention: its unsmoothed identity covariance is cylindrical, whereas its stochastic convolution is H H H -valued. In either formulation the drift must be defined on its declared domain; these analytic properties are not consequences of the notation. The exact Gibbs and dissipation identities use the reversible generator associated with (61). In infinite dimension the coordinate divergence is replaced by a justified generator or Dirichlet-form construction relative to the specified reference measure.
Equation (62) specifies the stochastic law. Existence and uniqueness require the hypotheses in Section 7.5. In the induced-metric realization one must also prove that the nondegenerate coframe branch is invariant, or stop the process at its boundary and specify boundary conditions. This is a substantive analytic requirement: e t = e 0 + 2 θ W t e_t=e_0+\sqrt{2\theta}W_t e t = e 0 + 2 θ W t , with e 0 > 0 e_0>0 e 0 > 0 , hits zero almost surely in one dimension. Nondegeneracy at the initial time therefore cannot supply boundary nonattainment for an arbitrary coframe diffusion. Gravity and gauge fields receive projections of a common noise operator. Whether those projections are statistically correlated is determined by the corresponding covariance blocks, as examined in Section 11.7.
3.11. Definition of the unified cost. For an absolutely continuous skeleton path x : [ 0 , T ] → X x:[0,T]\to\mathcal{X} x : [ 0 , T ] → X in the fixed affine realization, define the deterministic skeleton control cost. The velocity and drift are compared in that realization. In a finite-dimensional or uniformly elliptic realization, M ^ ( x ) − 1 \widehat{M}(x)^{-1} M ( x ) − 1 is the ordinary inverse; in a degenerate realization we write M ^ ( x ) † \widehat{M}(x)^{\dagger} M ( x ) † for the Moore–Penrose inverse on the tangent range and impose
x ˙ t − B ( x t ) ∈ Ran M ^ ( x t ) for a.e. t . (64) \dot{x}_t-\mathbb{B}(x_t)\in\operatorname{Ran}\widehat{M}(x_t)\quad\text{for a.e. }t. \tag*{(64)} x ˙ t − B ( x t ) ∈ Ran M ( x t ) for a.e. t . (64)
If this range constraint fails, the cost is defined to be + ∞ +\infty + ∞ . Here ∥ v ∥ M ^ − 1 2 \lVert v\rVert_{\widehat{M}^{-1}}^2 ∥ v ∥ M − 1 2 (respectively ∥ v ∥ M ^ † 2 \lVert v\rVert_{\widehat{M}^{\dagger}}^2 ∥ v ∥ M † 2 ) means ⟨ v , M ^ − 1 v ⟩ g X \langle v,\widehat{M}^{-1}v\rangle_{g_{\mathcal{X}}} ⟨ v , M − 1 v ⟩ g X (respectively ⟨ v , M ^ † v ⟩ g X \langle v,\widehat{M}^{\dagger}v\rangle_{g_{\mathcal{X}}} ⟨ v , M † v ⟩ g X ) in the tangent-space pairing. In every finite-dimensional realization used below, g X g_{\mathcal{X}} g X is the restriction of the field-space metric G G G ; it is not an additional independent metric. This convention avoids silently applying a cotangent operator to a tangent vector.
J T [ x ] : = 1 4 θ ∫ 0 T ∥ x ˙ t − B ( x t ) ∥ M ^ ( x t ) − 1 2 d t . (65) \mathcal{J}_T[x]:=\frac{1}{4\theta}\int_0^T\lVert\dot{x}_t-\mathbb{B}(x_t)\rVert_{\widehat{M}(x_t)^{-1}}^2\,\mathrm{d}t. \tag*{(65)} J T [ x ] := 4 θ 1 ∫ 0 T ∥ x ˙ t − B ( x t ) ∥ M ( x t ) − 1 2 d t . (65)
This formula defines the skeleton cost in finite dimension or on a closed noise range. In a uniformly elliptic realization the Moore–Penrose inverse is the ordinary inverse. In infinite dimension the correct finite-energy domain is generally larger than Ran M ^ \operatorname{Ran}\widehat{M} Ran M . For a bounded nonnegative covariance operator, define
∥ v ∥ M ^ † 2 : = inf M ^ 1 / 2 z = v ∥ z ∥ 2 = ∥ M ^ † / 2 v ∥ 2 , v ∈ Ran M ^ 1 / 2 . (66) \lVert v\rVert_{\widehat{M}^{\dagger}}^2:=\inf_{\widehat{M}^{1/2}z=v}\lVert z\rVert^2=\lVert\widehat{M}^{\dagger/2}v\rVert^2,\qquad v\in\operatorname{Ran}\widehat{M}^{1/2}. \tag*{(66)} ∥ v ∥ M † 2 := M 1/2 z = v inf ∥ z ∥ 2 = ∥ M † /2 v ∥ 2 , v ∈ Ran M 1/2 . (66)
The infimum over an empty set is + ∞ +\infty + ∞ . The right-hand side is a quadratic-form value, so it does not require v v v to lie in the operator domain of M ^ † \widehat{M}^{\dagger} M † . In this setting (64) is replaced by membership in Ran M ^ 1 / 2 \operatorname{Ran}\widehat{M}^{1/2} Ran M 1/2 , together with square integrability of the minimum-norm control. For example, on ℓ 2 \ell^2 ℓ 2 , if M ^ e n = n − 2 e n \widehat{M}e_n=n^{-2}e_n M e n = n − 2 e n and v n = n − 2 v_n=n^{-2} v n = n − 2 , then ∑ n n 2 v n 2 < ∞ \sum_n n^2v_n^2<\infty ∑ n n 2 v n 2 < ∞ , whereas ∑ n n 4 v n 2 = ∞ \sum_n n^4v_n^2=\infty ∑ n n 4 v n 2 = ∞ . Thus v v v has finite control energy but does not belong to Ran M ^ \operatorname{Ran}\widehat{M} Ran M . Finite-dimensional range notation must therefore not be carried to the continuum without this distinction.
The functional uses the common master action and mobility, with the interaction sectors represented by components of x x x , M \mathrm{M} M , and S M \mathrm{S}_{\mathrm{M}} S M . Sharing this functional does not independently fix the relative coefficients or the physical projections of those components. The functional is defined on absolutely continuous skeletons. It does not require a derivative of a stochastic sample path: nondegenerate diffusion paths are not absolutely continuous, while degenerate realizations can include deterministic directions.
The zero-cost condition follows from positivity on the noise range:
J T [ x ] = 0 ⟺ x ˙ t = B ( x t ) for a.e. t , (67) \mathcal{J}_{T}[x] = 0 \quad\Longleftrightarrow\quad\dot{x}_{t} = \mathrm{B}(x_{t}) \quad\text{for a.e. } t, \tag*{(67)} J T [ x ] = 0 ⟺ x ˙ t = B ( x t ) for a.e. t , (67)
provided the norm is strictly positive on the relevant tangent space; in the degenerate case this equivalence is understood within the range-constrained class specified above. Uniqueness of a zero-cost path additionally requires a fixed initial condition and uniqueness for the drift equation. Positivity of the cost identifies its zeros; it does not prove uniqueness for an ordinary differential equation with an irregular drift.
Thus the zero-cost dynamics follows the complete reference drift. It is a gradient flow when β ν = 0 \beta_{\nu}=0 β ν = 0 ; at positive noise scale a general reversible coordinate drift also includes θ β ν \theta\beta_{\nu} θ β ν . Stationary points of the uncorrected gradient flow solve the Euler–Lagrange equations only when the mobility is injective on the physical variations. A degenerate mobility can annihilate a nonzero action differential, so that implication needs the stated injectivity or an appropriate restriction of the variation space. The stochastic time describes relaxation in configuration space; its relation to physical Lorentzian time is a separate construction.
At positive θ \theta θ , the skeleton formula also requires its coordinate convention. For a smooth time-independent change y = Υ ( x ) y=\Upsilon(x) y = Υ ( x ) , the Itô transformation rule is
b y α = ∂ i Υ α b i + θ M i j ∂ i ∂ j Υ α , y ˙ α − b y α = ∂ i Υ α ( x ˙ i − b i ) − θ M i j ∂ i ∂ j Υ α . (68) b_{y}^{\alpha}=\partial_{i}\Upsilon^{\alpha}b^{i}+\theta M^{ij}\partial_{i}\partial_{j}\Upsilon^{\alpha},\qquad
\dot{y}^{\alpha}-b_{y}^{\alpha}=\partial_{i}\Upsilon^{\alpha}(\dot{x}^{i}-b^{i})-\theta M^{ij}\partial_{i}\partial_{j}\Upsilon^{\alpha}. \tag*{(68)} b y α = ∂ i Υ α b i + θ M ij ∂ i ∂ j Υ α , y ˙ α − b y α = ∂ i Υ α ( x ˙ i − b i ) − θ M ij ∂ i ∂ j Υ α . (68)
The last term shows that the finite-noise residual obtained from the coordinate Itô drift is not a vector under a general nonlinear reparameterization. Equation (65) therefore refers to the fixed affine or declared chart realization, rather than to an unspecified coordinate-independent functional on smooth curves. For a concrete example, let d X t = 2 θ d W t \mathrm{d}X_{t}=\sqrt{2\theta}\,\mathrm{d}W_{t} d X t = 2 θ d W t on R \mathbb{R} R , and set Y t = e X t Y_{t}=e^{X_{t}} Y t = e X t . Itô’s formula gives d Y t = θ Y t d t + 2 θ Y t d W t \mathrm{d}Y_{t}=\theta Y_{t}\,\mathrm{d}t+\sqrt{2\theta}Y_{t}\,\mathrm{d}W_{t} d Y t = θ Y t d t + 2 θ Y t d W t . The constant paths x t = 0 x_{t}=0 x t = 0 and y t = 1 y_{t}=1 y t = 1 describe the same curve in the two coordinates, but their costs computed directly from the two Itô drifts are
J T x [ 0 ] = 0 , J T y [ 1 ] = 1 4 θ ∫ 0 T ( 0 − θ ) 2 1 2 d t = θ T 4 . (69) \mathcal{J}_{T}^{x}[0]=0,\qquad\mathcal{J}_{T}^{y}[1]=\frac{1}{4\theta}\int_{0}^{T}\frac{(0-\theta)^{2}}{1^{2}}\,\mathrm{d}t=\frac{\theta T}{4}. \tag*{(69)} J T x [ 0 ] = 0 , J T y [ 1 ] = 4 θ 1 ∫ 0 T 1 2 ( 0 − θ ) 2 d t = 4 θ T . (69)
The discrepancy vanishes at the leading small-noise scale but is nonzero at every fixed θ > 0 \theta>0 θ > 0 and T > 0 T>0 T > 0 . The reference generator and the relative entropy of diffusion path laws are intrinsic. In particular, the difference between controlled and reference drifts transforms as v y = D Υ v x v_y = D\Upsilon v_x v y = D Υ v x , because the common Itô correction cancels. This is the drift difference used in the Girsanov calculation below. A finite-noise geometric skeleton on a curved configuration manifold requires a further specified connection or equivalent stochastic convention; the leading small-noise rate in § 4.5 is unaffected by this order-θ \theta θ term.
Relation to relative entropy
Consider a controlled process
d X t u = B ( X t u ) d t + B 0 ( X t u ) u t d t + σ ( X t u ) d W t u , (70) \mathrm{d}X_t^u = \mathbb{B}(X_t^u)\,\mathrm{d}t + \mathbb{B}_0(X_t^u)u_t\,\mathrm{d}t + \sigma(X_t^u)\,\mathrm{d}W_t^u , \tag*{(70)} d X t u = B ( X t u ) d t + B 0 ( X t u ) u t d t + σ ( X t u ) d W t u , (70)
where B \mathbb{B} B denotes the complete reference Itô drift. For constant mobility and a flat reference density, B = − M δ S M / δ X \mathbb{B}=-M\delta S_M/\delta X B = − M δ S M / δ X ; for the general reversible realization use (61), derived again in (80). The residual in (65) therefore always uses the drift of the process being controlled. In either case, B 0 \mathbb{B}_0 B 0 is a prescribed control operator, and σ σ ∗ = 2 θ M ^ \sigma\sigma^*=2\theta\widehat{M} σ σ ∗ = 2 θ M . Here W u W^u W u is Brownian under the controlled law P u \mathbb{P}^u P u , whereas W W W denotes the Brownian motion under the reference law P 0 \mathbb{P}^0 P 0 . Assume that the two laws have the same initial distribution, that v t : = B 0 ( X t u ) u t v_t:=\mathbb{B}_0(X_t^u)u_t v t := B 0 ( X t u ) u t is a progressively measurable functional of the canonical state path in its own completed filtration under both laws, lies in the range of σ ( X t u ) \sigma(X_t^u) σ ( X t u ) , and yields an exponential martingale satisfying Novikov’s condition. The process v t v_t v t , and hence σ † v t \sigma^\dagger v_t σ † v t , must be defined on the canonical path space under both laws; evaluating it only on a controlled weak solution would not by itself define the Radon–Nikodym density under the reference law. In the finite-dimensional realization, or more generally when σ \sigma σ has closed range, let σ † \sigma^\dagger σ † denote the Moore–Penrose inverse on that range. The control coordinate u t u_t u t and the Brownian shift are not the same object: set u ^ t : = σ † v t \widehat{u}_t:=\sigma^\dagger v_t u t := σ † v t . If B 0 \mathbb{B}_0 B 0 has a kernel, different control coordinates can produce the same drift increment v t v_t v t ; the relative entropy depends only on v t v_t v t and uses the minimum-norm shift u ^ t \widehat{u}_t u t . Assume also uniqueness in law for the controlled martingale problem and E u ∫ 0 T ∥ u ^ t ∥ 2 d t < ∞ \mathbb{E}^u\int_0^T\|\widehat{u}_t\|^2\mathrm{d}t<\infty E u ∫ 0 T ∥ u t ∥ 2 d t < ∞ . The first assumption identifies the tilted reference law with the proposed controlled law; the second justifies taking expectation of its log density. Because the drift increment is measurable in the state-path filtration, the density is itself a state-path functional, as shown in Appendix C. Girsanov’s theorem (see [105]) then yields
KL ( P u ∥ P 0 ) = 1 2 E u ∫ 0 T ∥ σ † v t ∥ 2 d t . (71) \operatorname{KL}(\mathbb{P}^u\Vert\mathbb{P}^0)=\frac{1}{2}\mathbb{E}^u\int_0^T\|\sigma^\dagger v_t\|^2\mathrm{d}t . \tag*{(71)} KL ( P u ∥ P 0 ) = 2 1 E u ∫ 0 T ∥ σ † v t ∥ 2 d t . (71)
Define the controlled-law cost by
C T ( u ) : = 1 2 E u ∫ 0 T ∥ σ † v t ∥ 2 d t . (72) C_T(u):=\frac{1}{2}\mathbb{E}^u\int_0^T\|\sigma^\dagger v_t\|^2\mathrm{d}t . \tag*{(72)} C T ( u ) := 2 1 E u ∫ 0 T ∥ σ † v t ∥ 2 d t . (72)
If a deterministic absolutely continuous skeleton x x x satisfies x ˙ t = B ( x t ) + v t \dot{x}_t=\mathbb{B}(x_t)+v_t x ˙ t = B ( x t ) + v t , then the corresponding quadratic control energy is exactly the skeleton functional J T [ x ] \mathcal{J}_T[x] J T [ x ] after using σ σ ∗ = 2 θ M ^ \sigma\sigma^*=2\theta\widehat{M} σ σ ∗ = 2 θ M . For a genuinely random adapted control, C T C_T C T is the expectation of that energy under P u \mathbb{P}^{u} P u . It is not legitimate to identify v t v_t v t with the derivative of a sample diffusion path.
KL ( P u ∥ P 0 ) = C T ( u ) . (73) \operatorname{KL}(\mathbb{P}^{u}\Vert\mathbb{P}^{0}) = C_T(u). \tag*{(73)} KL ( P u ∥ P 0 ) = C T ( u ) . (73)
Therefore relative entropy is not an additional physical cost in this formulation. It is a probabilistic representation of the same path-deviation cost under the reference stochastic dynamics; the finite-dimensional Radon–Nikodym calculation is given in Appendix C.
The equality concerns path-space relative entropy for the specified controlled dynamics. It does not identify this entropy with the state-space free energy introduced below: the two functionals have different domains. If controls use randomness not observable from the state path, projection from the enlarged probability space gives only the data-processing bound
KL ( Law ( X u ) ∥ Law ( X 0 ) ) ≤ 1 2 E u ∫ 0 T ∥ u ^ t ∥ 2 d t . (74) \operatorname{KL}\bigl(\operatorname{Law}(X^{u})\Vert\operatorname{Law}(X^{0})\bigr) \leq\frac{1}{2}\mathbb{E}^{u}\int_{0}^{T}\lVert\widehat{u}_{t}\rVert^{2}\,\mathrm{d}t. \tag*{(74)} KL ( Law ( X u ) ∥ Law ( X 0 ) ) ≤ 2 1 E u ∫ 0 T ∥ u t ∥ 2 d t . (74)
Equality requires measurability of the likelihood ratio in the retained state path. This condition specifies exactly when the control energy is a representation of path entropy.
Equilibrium, limits, and symmetry breaking
Equilibrium generated by the same principle.
The path-space calculation compares controlled trajectories. Equilibrium instead concerns probability distributions at a fixed stochastic time. The link between them is the chosen reversible generator: it uses the same action and mobility, while its divergence correction depends on the reference measure. Deriving the forward equation makes that dependence explicit. Here “reversible generator” means self-adjointness in the equilibrium L 2 L^{2} L 2 -space, equivalently detailed balance for this overdamped branch. It does not mean that the diffusion coefficient vanishes. Later, b r e v = L ∇ E b_{\mathrm{rev}}=\mathsf{L}\nabla E b rev = L ∇ E denotes the conservative part of a parity-aware reversible–irreversible decomposition. These uses are consistent: the present overdamped generator has no conservative anti-self-adjoint part, while the later generator may have one whose sign is reversed by the time-reversal parity involution.
Fokker–Planck equation.
Fix a finite-dimensional physical configuration manifold and a positive reference measure ν \nu ν . In a chart write ν ( d x ) = m ν ( x ) d x \nu(\mathrm{d}x)=m_{\nu}(x)\,\mathrm{d}x ν ( d x ) = m ν ( x ) d x , where m ν m_{\nu} m ν is smooth and positive. The density p t p_t p t is always taken with respect to ν \nu ν ; its density with respect to coordinate Lebesgue measure is ρ t = m ν p t \rho_t=m_{\nu}p_t ρ t = m ν p t . These two densities have different forward equations. For a vector field V V V , the reference divergence is
div ν V : = m ν − 1 ∂ i ( m ν V i ) . (75) \operatorname{div}_{\nu}V := m_{\nu}^{-1}\partial_i(m_{\nu}V^i). \tag*{(75)} div ν V := m ν − 1 ∂ i ( m ν V i ) . (75)
Here and below repeated coordinate indices are summed. For a scalar f f f , ∇ j f \nabla_j f ∇ j f means ∂ j f \partial_j f ∂ j f , whereas ∇ X f \nabla_X f ∇ X f is the field-space metric gradient. The contravariant mobility M i j M^{ij} M ij acts on the covector d f \mathrm{d}f d f .
Consider the coordinate Itô equation d X i = b i d t + σ α i d W α \mathrm{d}X^i=b^i\,\mathrm{d}t+\sigma^i_{\alpha}\,\mathrm{d}W^\alpha d X i = b i d t + σ α i d W α , with a i j = ∑ α σ α i σ α j = 2 θ M i j a^{ij}=\sum_\alpha\sigma^i_\alpha\sigma^j_\alpha=2\theta M^{ij} a ij = ∑ α σ α i σ α j = 2 θ M ij . Itô’s formula gives
L f = b i ∂ i f + θ M i j ∂ i ∂ j f . (76) Lf=b^i\partial_i f+\theta M^{ij}\partial_i\partial_j f. \tag*{(76)} L f = b i ∂ i f + θ M ij ∂ i ∂ j f . (76)
For a smooth compactly supported test function f f f , two integrations by parts against d x \mathrm{d}x d x , or the corresponding boundary conditions, give
d d t ∫ f p t m ν d x = ∫ ( L f ) p t m ν d x = ∫ f { − ∂ i ( b i m ν p t ) + θ ∂ i ∂ j ( M i j m ν p t ) } d x . (77) \frac{\mathrm{d}}{\mathrm{d}t}\int f p_t m_\nu\,\mathrm{d}x=\int(Lf)p_t m_\nu\,\mathrm{d}x
=\int f\left\{-\partial_i(b^i m_\nu p_t)+\theta\partial_i\partial_j(M^{ij}m_\nu p_t)\right\}\,\mathrm{d}x. \tag*{(77)} d t d ∫ f p t m ν d x = ∫ ( L f ) p t m ν d x = ∫ f { − ∂ i ( b i m ν p t ) + θ ∂ i ∂ j ( M ij m ν p t ) } d x . (77)
Consequently the adjoint relative to ν \nu ν is
∂ t p t = − 1 m ν ∂ i ( m ν b i p t ) + θ m ν ∂ i ∂ j ( m ν M i j p t ) . (78) \partial_t p_t=-\frac{1}{m_\nu}\partial_i(m_\nu b^i p_t)+\frac{\theta}{m_\nu}\partial_i\partial_j(m_\nu M^{ij}p_t). \tag*{(78)} ∂ t p t = − m ν 1 ∂ i ( m ν b i p t ) + m ν θ ∂ i ∂ j ( m ν M ij p t ) . (78)
For comparison, retaining only the gradient part of (61) gives the uncorrected coordinate drift b i = − M i j ∂ j S M b^i=-M^{ij}\partial_j S_{\mathrm M} b i = − M ij ∂ j S M and the forward equation
∂ t p t = 1 m ν ∂ i ( m ν p t M i j ∂ j S M ) + θ m ν ∂ i ∂ j ( m ν M i j p t ) . (79) \partial_t p_t=\frac{1}{m_\nu}\partial_i(m_\nu p_t M^{ij}\partial_j S_{\mathrm M})+\frac{\theta}{m_\nu}\partial_i\partial_j(m_\nu M^{ij}p_t). \tag*{(79)} ∂ t p t = m ν 1 ∂ i ( m ν p t M ij ∂ j S M ) + m ν θ ∂ i ∂ j ( m ν M ij p t ) . (79)
The second term contains one factor m ν m_\nu m ν inside the two coordinate derivatives. It cannot be replaced by iterating the vector divergence on the components of a rank-two tensor. Equation (78) is the density form of the usual forward equation [137].
Define in this chart
b r e v i = − M i j ∂ j S M + θ β ν i , b^i_{\mathrm{rev}}=-M^{ij}\partial_j S_{\mathrm M}+\theta\beta^i_\nu, b rev i = − M ij ∂ j S M + θ β ν i ,
β ν i : = 1 m ν ∂ j ( m ν M i j ) = ∂ j M i j + M i j ∂ j log m ν . (80) \beta^i_\nu:=\frac{1}{m_\nu}\partial_j(m_\nu M^{ij})=\partial_jM^{ij}+M^{ij}\partial_j\log m_\nu. \tag*{(80)} β ν i := m ν 1 ∂ j ( m ν M ij ) = ∂ j M ij + M ij ∂ j log m ν . (80)
An Itô drift transforms with a second-derivative correction under changes of coordinates, so β ν i \beta^i_\nu β ν i alone is not asserted to be a vector. The invariant object is the divergence-form generator
L r e v f = θ e S M / θ m ν ∂ i ( m ν e − S M / θ M i j ∂ j f ) . (81) L_{\mathrm{rev}}f=\frac{\theta e^{S_{\mathrm M}/\theta}}{m_\nu}\partial_i\left(m_\nu e^{-S_{\mathrm M}/\theta}M^{ij}\partial_j f\right). \tag*{(81)} L rev f = m ν θ e S M / θ ∂ i ( m ν e − S M / θ M ij ∂ j f ) . (81)
Expanding the derivative, and using M i j = M j i M^{ij}=M^{ji} M ij = M j i , recovers (76) with the drift in (80). This also specifies the equation on overlapping charts.
For clarity, the cancellation in the forward equation is
∂ j ( m ν M i j p t ) = p t ∂ j ( m ν M i j ) + m ν M i j ∂ j p t , \partial_j(m_\nu M^{ij}p_t)=p_t\partial_j(m_\nu M^{ij})+m_\nu M^{ij}\partial_jp_t, ∂ j ( m ν M ij p t ) = p t ∂ j ( m ν M ij ) + m ν M ij ∂ j p t ,
− m ν b r e v i p t + θ ∂ j ( m ν M i j p t ) = m ν M i j ( p t ∂ j S M + θ ∂ j p t ) . (82) -m_\nu b^i_{\mathrm{rev}}p_t+\theta\partial_j(m_\nu M^{ij}p_t)=m_\nu M^{ij}(p_t\partial_jS_{\mathrm M}+\theta\partial_jp_t). \tag*{(82)} − m ν b rev i p t + θ ∂ j ( m ν M ij p t ) = m ν M ij ( p t ∂ j S M + θ ∂ j p t ) . (82)
Applying m ν − 1 ∂ i m_{\nu}^{-1}\partial_i m ν − 1 ∂ i now yields
∂ t p t = div ν [ p t M ^ ∇ X S M + θ M ^ ∇ X p t ] . (83) \partial_t p_t=\operatorname{div}_{\nu}\left[p_t\widehat{M}\nabla_X S_M+\theta\widehat{M}\nabla_Xp_t\right]. \tag*{(83)} ∂ t p t = div ν [ p t M ∇ X S M + θ M ∇ X p t ] . (83)
No constancy assumption on the mobility is needed for this identity. If the uncorrected coordinate drift is retained, this flux has the same generator only when β ν = 0 \beta_{\nu}=0 β ν = 0 in the specified chart. A sufficient case is constant mobility and constant m ν m_{\nu} m ν in affine coordinates. In that case,
∂ t p t = div ν [ M ^ ( p t ∇ X S M + θ ∇ X p t ) ] . (84) \partial_t p_t=\operatorname{div}_{\nu}\left[\widehat{M}\left(p_t\nabla_XS_M+\theta\nabla_Xp_t\right)\right]. \tag*{(84)} ∂ t p t = div ν [ M ( p t ∇ X S M + θ ∇ X p t ) ] . (84)
Define the free energy relative to a fixed positive, sigma-finite reference measure ν \nu ν . The units of this reference are fixed so that its densities have a dimensionless logarithm. This does not require ν \nu ν to have finite mass: Lebesgue measure on a noncompact configuration space cannot be normalized by division by its total mass. Replacing ν \nu ν by a nonconstant weighted measure changes the model unless the action is changed by the compensating logarithmic term. For comparison of different microscopic candidates, Section 6.1 imposes a common normalization separately. If gauge symmetries are present, ν \nu ν must be a positive measure on a well-defined quotient or on a specified gauge-fixed domain with a nonnegative Jacobian. A formal signed Faddeev–Popov determinant does not define a probability reference measure. These choices do not themselves imply 0 < Z < ∞ 0<Z<\infty 0 < Z < ∞ . For normalized densities p , q p,q p , q with p ≪ q p\ll q p ≪ q , we use
KL ( p ∥ q ) : = ∫ X p ( X ) log p ( X ) q ( X ) ν ( d X ) . (85) \operatorname{KL}(p\|q):=\int_{\mathcal X}p(\mathrm{X})\log\frac{p(\mathrm{X})}{q(\mathrm{X})}\,\nu(\mathrm{dX}). \tag*{(85)} KL ( p ∥ q ) := ∫ X p ( X ) log q ( X ) p ( X ) ν ( dX ) . (85)
F [ p ] = ∫ X p ( X ) S M ( X ) ν ( d X ) + θ ∫ X p log p ν ( d X ) . (86) \mathcal{F}[p]=\int_{\mathcal X}p(\mathrm{X})S_M(\mathrm{X})\,\nu(\mathrm{dX})+\theta\int_{\mathcal X}p\log p\,\nu(\mathrm{dX}). \tag*{(86)} F [ p ] = ∫ X p ( X ) S M ( X ) ν ( dX ) + θ ∫ X p log p ν ( dX ) . (86)
In this formulation, F \mathcal{F} F is not a second cost added to J T \mathcal{J}_T J T . It is the Lyapunov functional induced by the same master action and the entropy of the probability state. As a sum of two ordinary integrals, (86) is defined here on densities satisfying ∫ p ∣ S M ∣ d ν + ∫ p ∣ log p ∣ d ν < ∞ \int p|S_M|\,\mathrm{d}\nu+\int p|\log p|\,\mathrm{d}\nu<\infty ∫ p ∣ S M ∣ d ν + ∫ p ∣ log p ∣ d ν < ∞ . If 0 < Z < ∞ 0<Z<\infty 0 < Z < ∞ , its relative extension is F r e l [ p ] = − θ log Z + θ KL ( p ∥ p ∗ ) \mathcal{F}_{\mathrm{rel}}[p]=-\theta\log Z+\theta\operatorname{KL}(p\|p_*) F rel [ p ] = − θ log Z + θ KL ( p ∥ p ∗ ) . This agrees with the separate-integral definition on that domain but also applies when the individual terms cannot be separated. In particular, finite relative entropy alone does not justify subtracting two infinite energy or entropy terms.
Dissipation identity
Theorem 2 (Free-energy dissipation). *Suppose that X \mathcal{X} X is connected and the symmetric mobility is positive definite. Let the density solve (83); this includes the uncorrected equation when β ν = 0 \beta_{\nu}=0 β ν = 0 . Assume that p t p_t p t is smooth, strictly positive, normalized, and has finite free energy and finite displayed dissipation. Assume also that differentiation under the integrals and the integration by parts used below are valid, with zero boundary flux and vanishing boundary
pairings. These conditions hold, for example, for smooth positive solutions on a compact manifold without boundary, or with suitable decay and domination on a noncompact realization. Then
d d t F [ p t ] = − ∫ X p t ⟨ ∇ X ( S M + θ log p t ) , M ^ ∇ X ( S M + θ log p t ) ⟩ ν ( d X ) ≤ 0. (87) \frac{\mathrm{d}}{\mathrm{d}t}\mathcal{F}[p_t] = -\int_{\mathcal{X}} p_t \left\langle\nabla_{\mathsf{X}}\left(\mathcal{S}_{\mathrm{M}}+\theta\log p_t\right),\widehat{\mathrm{M}}\nabla_{\mathsf{X}}\left(\mathcal{S}_{\mathrm{M}}+\theta\log p_t\right)\right\rangle\nu(\mathrm{dX}) \le0. \tag*{(87)} d t d F [ p t ] = − ∫ X p t ⟨ ∇ X ( S M + θ log p t ) , M ∇ X ( S M + θ log p t ) ⟩ ν ( dX ) ≤ 0. (87)
Equality holds if and only if
∇ X ( S M + θ log p t ) = 0 (88) \nabla_{\mathsf{X}}\left(\mathcal{S}_{\mathrm{M}}+\theta\log p_t\right)=0 \tag*{(88)} ∇ X ( S M + θ log p t ) = 0 (88)
where p t > 0 p_t>0 p t > 0 .
Proof. Differentiate (86) and use conservation of total probability. The term θ ∫ p t ν ( d X ) \theta\int p_t\,\nu(\mathrm{dX}) θ ∫ p t ν ( dX ) contributes zero through the constant part of the first variation because ∫ ∂ t p t ν ( d X ) = 0 \int\partial_t p_t\,\nu(\mathrm{dX})=0 ∫ ∂ t p t ν ( dX ) = 0 . More explicitly, since
δ F δ p = S M + θ ( 1 + log p ) , (89) \frac{\delta\mathcal{F}}{\delta p}=\mathcal{S}_{\mathrm{M}}+\theta(1+\log p), \tag*{(89)} δ p δ F = S M + θ ( 1 + log p ) , (89)
we obtain
d d t F [ p t ] = ∫ ( S M + θ log p t ) ∂ t p t ν ( d X ) . (90) \frac{\mathrm{d}}{\mathrm{d}t}\mathcal{F}[p_t]=\int\left(\mathcal{S}_{\mathrm{M}}+\theta\log p_t\right)\partial_t p_t\,\nu(\mathrm{dX}). \tag*{(90)} d t d F [ p t ] = ∫ ( S M + θ log p t ) ∂ t p t ν ( dX ) . (90)
Set H t = S M + θ log p t H_t=\mathcal{S}_{\mathrm{M}}+\theta\log p_t H t = S M + θ log p t . Equation (83) is ∂ t p t = div ν ( p t M ^ ∇ X H t ) \partial_t p_t=\operatorname{div}_{\nu}(p_t\widehat{\mathrm{M}}\nabla_{\mathsf{X}}H_t) ∂ t p t = div ν ( p t M ∇ X H t ) . The assumed boundary or decay condition gives the integration-by-parts identity ∫ f div ν V ν ( d X ) = − ∫ ⟨ ∇ X f , V ⟩ ν ( d X ) \int f\,\operatorname{div}_{\nu}V\,\nu(\mathrm{dX})=-\int\langle\nabla_{\mathsf{X}}f,V\rangle\,\nu(\mathrm{dX}) ∫ f div ν V ν ( dX ) = − ∫ ⟨ ∇ X f , V ⟩ ν ( dX ) . Applying it with f = H t f=H_t f = H t and V = p t M ^ ∇ X H t V=p_t\widehat{\mathrm{M}}\nabla_{\mathsf{X}}H_t V = p t M ∇ X H t gives
d d t F [ p t ] = − ∫ ⟨ ∇ X ( S M + θ log p t ) , M ^ ∇ X ( S M + θ log p t ) ⟩ p t ν ( d X ) . (91) \frac{\mathrm{d}}{\mathrm{d}t}\mathcal{F}[p_t]=-\int\left\langle\nabla_{\mathsf{X}}\left(\mathcal{S}_{\mathrm{M}}+\theta\log p_t\right),\widehat{\mathrm{M}}\nabla_{\mathsf{X}}\left(\mathcal{S}_{\mathrm{M}}+\theta\log p_t\right)\right\rangle p_t\,\nu(\mathrm{dX}). \tag*{(91)} d t d F [ p t ] = − ∫ ⟨ ∇ X ( S M + θ log p t ) , M ∇ X ( S M + θ log p t ) ⟩ p t ν ( dX ) . (91)
Positive definiteness of M ^ \widehat{\mathrm{M}} M proves the inequality. Equality is equivalent to vanishing of the quadratic form almost everywhere, which gives the stated condition. □
Equilibrium measure
For constant mobility satisfying the compatibility condition in the preceding theorem, or for the state-dependent Itô equation with the reversible divergence correction (80), the equilibrium density satisfies
p ∗ ( X ) = Z − 1 exp [ − S M ( X ) θ ] , (92) p_*(\mathsf{X})=Z^{-1}\exp\left[-\frac{\mathcal{S}_{\mathrm{M}}(\mathsf{X})}{\theta}\right], \tag*{(92)} p ∗ ( X ) = Z − 1 exp [ − θ S M ( X ) ] , (92)
provided the displayed integrals are well defined (or are interpreted through a common lower-semicontinuous extension), and
Z = ∫ X exp [ − S M ( X ) θ ] ν ( d X ) < ∞ . (93) Z=\int_{\mathcal{X}}\exp\left[-\frac{\mathcal{S}_{\mathrm{M}}(\mathsf{X})}{\theta}\right]\nu(\mathrm{dX})<\infty. \tag*{(93)} Z = ∫ X exp [ − θ S M ( X ) ] ν ( dX ) < ∞. (93)
In addition, when F [ p ∗ ] \mathcal{F}[p_*] F [ p ∗ ] is treated as a finite ordinary number, S M p ∗ S_{\mathrm{M}}p_* S M p ∗ and p ∗ log p ∗ p_* \log p_* p ∗ log p ∗ are required to be integrable. Moreover,
F [ p ] − F [ p ∗ ] = θ K L ( p ∥ p ∗ ) . (94) \mathcal{F}[p]-\mathcal{F}[p_*]=\theta\,\mathrm{KL}(p\Vert p_*). \tag*{(94)} F [ p ] − F [ p ∗ ] = θ KL ( p ∥ p ∗ ) . (94)
Equation (94) shows why entropy and action combine naturally at equilibrium, while the path cost remains a single object. Once the reference measure, action, fluctuation scale, and reversible generator have all been fixed, this identity determines the equilibrium density. The action alone does not determine the reference measure. The algebraic Gibbs identity is recorded in Appendix B.
Theorem 3 (Equilibrium as the unique zero-dissipation state). Assume X \mathcal{X} X is connected, 0 < Z < ∞ 0<Z<\infty 0 < Z < ∞ , the integrability conditions stated above hold, and the positive definite mobility and reversible dynamics satisfy the hypotheses of the preceding theorem. Assume also that the admissible probability densities are normalized, absolutely continuous with respect to the reference volume, and have finite relative entropy with respect to p ∗ p_* p ∗ . Then p ∗ p_* p ∗ is the unique minimizer of F \mathcal{F} F up to equality almost everywhere, and p t = p ∗ p_t=p_* p t = p ∗ is the unique zero-dissipation equilibrium among smooth positive admissible densities for which the dissipation identity applies. This assertion does not by itself exclude singular invariant measures or imply convergence to equilibrium.
Proof. Let r = p / p ∗ r=p/p_* r = p / p ∗ , defined almost everywhere with respect to π ∗ ( d x ) = p ∗ ( x ) ν ( d x ) \pi_*(\mathrm{d}x)=p_*(x)\nu(\mathrm{d}x) π ∗ ( d x ) = p ∗ ( x ) ν ( d x ) . Both densities are normalized, so ∫ r d π ∗ = 1 \int r\,\mathrm{d}\pi_*=1 ∫ r d π ∗ = 1 . The scalar function φ ( r ) = r log r − r + 1 \varphi(r)=r\log r-r+1 φ ( r ) = r log r − r + 1 , extended by φ ( 0 ) = 1 \varphi(0)=1 φ ( 0 ) = 1 , is nonnegative and vanishes only at r = 1 r=1 r = 1 . Consequently,
K L ( p ∥ p ∗ ) = ∫ φ ( r ) d π ∗ ≥ 0. (95) \mathrm{KL}(p\Vert p_*)=\int\varphi(r)\,\mathrm{d}\pi_* \ge0. \tag*{(95)} KL ( p ∥ p ∗ ) = ∫ φ ( r ) d π ∗ ≥ 0. (95)
Together with (94) , this proves the minimum and its uniqueness. For a smooth positive zero-dissipation density, positive definiteness of the mobility makes ∇ X ( S M + θ log p ) = 0 \nabla_{\mathcal{X}}(S_{\mathrm{M}}+\theta\log p)=0 ∇ X ( S M + θ log p ) = 0 . Connectedness makes this scalar function a constant, say c c c , and hence p = e c / θ e − S M / θ p=e^{c/\theta}e^{-S_{\mathrm{M}}/\theta} p = e c / θ e − S M / θ . Normalization gives e c / θ = Z − 1 e^{c/\theta}=Z^{-1} e c / θ = Z − 1 , so p = p ∗ p=p_* p = p ∗ . Conversely this density makes the flux vanish and therefore is stationary and has zero dissipation under the stated boundary conditions. □ \square □
A stochastic meaning of minimum action
The phrase “minimum action” requires care. The stochastic formulation does not identify every physical realization with a pointwise minimizer of the classical Euclidean action. The mathematically precise statement is that the stochastic path measure is centered on the master drift and the relative cost of a controlled path is the squared Cameron–Martin displacement from that drift. With a prescribed initial condition and no terminal constraint, the deterministic master trajectory is the unique zero-cost path among range-constrained admissible skeletons whenever the drift equation is unique. If endpoints or other path constraints are imposed, the minimizer generally has positive cost and must solve a separate constrained variational problem.
Recall that B \mathbb{B} B denotes the complete drift, including the reference correction in a reversible realization. Define its gradient part by
B g r a d ( X ) = − M ( X ) δ S M δ X . (96) \mathbb{B}_{\mathrm{grad}}(X)=-M(X)\frac{\delta S_M}{\delta X}. \tag*{(96)} B grad ( X ) = − M ( X ) δ X δ S M . (96)
For an absolutely continuous skeleton x x x and positive definite tangent mobility, the cost is
J T [ x ] = 1 4 θ ∫ 0 T ∥ x ˙ t − B ( x t ) ∥ M ^ ( x t ) − 1 2 d t . (97) \mathcal{J}_T[x]=\frac{1}{4\theta}\int_0^T \lVert\dot{x}_t-\mathbb{B}(x_t)\rVert_{\widehat{M}(x_t)^{-1}}^2\,\mathrm{d}t. \tag*{(97)} J T [ x ] = 4 θ 1 ∫ 0 T ∥ x ˙ t − B ( x t ) ∥ M ( x t ) − 1 2 d t . (97)
Its zero set consists of solutions of x ˙ = B ( x ) \dot{x}=\mathbb{B}(x) x ˙ = B ( x ) . This is a statement about paths with a fixed initial value. It does not say that every configuration visited by such a path is a stationary point of the Euclidean action.
The distinction matters even on a compact state space. On the circle, take S M = 0 S_M=0 S M = 0 , m ν = 1 m_\nu=1 m ν = 1 , and M ( x ) = 2 + sin x M(x)=2+\sin x M ( x ) = 2 + sin x . Then B g r a d = 0 \mathbb{B}_{\mathrm{grad}}=0 B grad = 0 , whereas the reversible drift is B ( x ) = θ cos x \mathbb{B}(x)=\theta\cos x B ( x ) = θ cos x . For the constant path x t = 0 x_t=0 x t = 0 ,
J T [ x ] = 1 4 θ ∫ 0 T θ 2 2 d t = θ T 8 . (98) \mathcal{J}_T[x]=\frac{1}{4\theta}\int_0^T\frac{\theta^2}{2}\,\mathrm{d}t=\frac{\theta T}{8}. \tag*{(98)} J T [ x ] = 4 θ 1 ∫ 0 T 2 θ 2 d t = 8 θ T . (98)
The cost would be zero for the different diffusion obtained by discarding the reference correction. The positive value therefore records a difference between two reference dynamics, rather than a failure of the Gibbs identity.
For a finite-dimensional approximation with orthonormal affine coordinates x ∈ R n x\in\mathbb{R}^n x ∈ R n , suppose that M ^ ( x ) \widehat{M}(x) M ( x ) is uniformly positive definite on the region under consideration. The degenerate case requires the corresponding range constraint and a separate large-deviation theorem. To state the small-noise limit, write b θ = b 0 + θ β ν b_\theta=b_0+\theta\beta_\nu b θ = b 0 + θ β ν , with b 0 = B g r a d b_0=\mathbb{B}_{\mathrm{grad}} b 0 = B grad , and let J T θ \mathcal{J}_T^\theta J T θ emphasize the noise-scale dependence of the cost. Under the regularity and exponential-tightness assumptions of the Freidlin–Wentzell principle, the limiting rate functional is
I T [ x ] = 1 4 ∫ 0 T ⟨ x ˙ − b 0 ( x ) , M ^ ( x ) − 1 ( x ˙ − b 0 ( x ) ) ⟩ d t . (99) I_T[x]=\frac{1}{4}\int_0^T\langle\dot{x}-b_0(x),\widehat{M}(x)^{-1}(\dot{x}-b_0(x))\rangle\,\mathrm{d}t. \tag*{(99)} I T [ x ] = 4 1 ∫ 0 T ⟨ x ˙ − b 0 ( x ) , M ( x ) − 1 ( x ˙ − b 0 ( x ))⟩ d t . (99)
For a path on which all the following integrals are finite, direct expansion of the square gives
θ J T θ [ x ] = I T [ x ] − θ 2 ∫ 0 T ⟨ x ˙ − b 0 ( x ) , M ^ ( x ) − 1 β ν ( x ) ⟩ d t + θ 2 4 ∫ 0 T ⟨ β ν ( x ) , M ^ ( x ) − 1 β ν ( x ) ⟩ d t . (100) \theta\mathcal{J}_T^\theta[x]=I_T[x]-\frac{\theta}{2}\int_0^T\langle\dot{x}-b_0(x),\widehat{M}(x)^{-1}\beta_\nu(x)\rangle\,\mathrm{d}t
+\frac{\theta^2}{4}\int_0^T\langle\beta_\nu(x),\widehat{M}(x)^{-1}\beta_\nu(x)\rangle\,\mathrm{d}t. \tag*{(100)} θ J T θ [ x ] = I T [ x ] − 2 θ ∫ 0 T ⟨ x ˙ − b 0 ( x ) , M ( x ) − 1 β ν ( x )⟩ d t + 4 θ 2 ∫ 0 T ⟨ β ν ( x ) , M ( x ) − 1 β ν ( x )⟩ d t . (100)
In particular, a vanishing reference correction gives the exact identity I T = θ J T θ I_T=\theta\mathcal{J}_T^\theta I T = θ J T θ . For an individual path the two correction integrals can also cancel even when β ν ≠ 0 \beta_\nu\ne0 β ν = 0 ; no converse is asserted on that basis. In the general reversible case θ J T θ [ x ] → I T [ x ] \theta J_T^\theta[x] \to I_T[x] θ J T θ [ x ] → I T [ x ] on this fixed-path domain. This algebraic convergence alone does not prove a path-space large-deviation principle; tightness, coefficient control, and the initial-law hypotheses are also required. Under those hypotheses, probabilities have the logarithmic asymptotics P ( X θ ≈ x ) ≍ exp [ − I T [ x ] / θ ] \mathbb{P}(X^\theta\approx x) \asymp\exp[-I_T[x]/\theta] P ( X θ ≈ x ) ≍ exp [ − I T [ x ] / θ ] [72]. Here the notation concerns sufficiently small path neighborhoods at logarithmic scale, not a pointwise density. The infinite-dimensional extension requires the appropriate Cameron–Martin geometry and its own compactness estimates.
Macroscopic and microscopic limits of the same system
The preceding rate functional concerns rare paths as the noise amplitude vanishes. Two other limits clarify how the same stochastic description can produce deterministic collective motion and local Gaussian fluctuations. They use different rescalings and should therefore be stated separately.
Macroscopic deterministic limit
The following equations are admissible reduced models of the master system, not automatic consequences of an abstract projection. To make the required relationship explicit, choose finite-dimensional embeddings ι N : X N → X \iota_N:\mathcal{X}_N \to\mathcal{X} ι N : X N → X and projections Π N : T X → T X N \Pi_N:T\mathcal{X} \to T\mathcal{X}_N Π N : T X → T X N with Π N ι N ∗ = Id \Pi_N\iota_{N*}=\operatorname{Id} Π N ι N ∗ = Id on the retained modes. After a common nondimensionalization, assume
Φ N = S M ∘ ι N , M Q N ( q ) : = Π N M ^ ( ι N q ) Π N ∗ , σ Q , N σ Q , N ∗ = 2 θ N M Q N . (101) \Phi_N=\mathcal{S}_M\circ\iota_N,\qquad M_Q^N(q):=\Pi_N\widehat{M}(\iota_Nq)\Pi_N^*,\qquad\sigma_{Q,N}\sigma_{Q,N}^*=\frac{2\theta}{N}M_Q^N. \tag*{(101)} Φ N = S M ∘ ι N , M Q N ( q ) := Π N M ( ι N q ) Π N ∗ , σ Q , N σ Q , N ∗ = N 2 θ M Q N . (101)
The adjoint Π N ∗ \Pi_N^* Π N ∗ is the metric adjoint, with cotangent inputs identified through the previously specified Riesz maps. The last relation is the collective-averaging scaling hypothesis; it does not follow from projection alone. We assume that the state-dependent matrices M Q N M_Q^N M Q N converge locally uniformly to a limiting mobility M Q M_Q M Q on the common coordinate domain. In the convergence argument below, all collective coordinates are identified with one fixed finite-dimensional domain in R d \mathbb{R}^d R d , independently of N N N . This assumption concerns the retained collective variables; the dimension of the underlying microscopic regulator may still increase with N N N . For the tangent limit below, an analogous choice is Q = Π Q X Q=\Pi_Q\mathcal{X} Q = Π Q X , b = Π Q B b=\Pi_Q B b = Π Q B , and σ Q = Π Q σ \sigma_Q=\Pi_Q\sigma σ Q = Π Q σ on the retained subspace. If no maps satisfying these conditions are supplied by a concrete realization, the displayed reduced equations remain specified SDE examples rather than consequences of the master action. Even the restriction of the action and compression of the mobility do not, on their own, identify the reduced drift. An exact projection must also satisfy, on the retained branch,
Π N B ( ι N q ) = − M Q N ( q ) ∇ Q Φ N ( q ) + r N p r o j ( q ) , (102) \Pi_N\mathbb{B}(\iota_Nq)=-M_Q^N(q)\nabla_Q\Phi_N(q)+r_N^{\mathrm{proj}}(q), \tag*{(102)} Π N B ( ι N q ) = − M Q N ( q ) ∇ Q Φ N ( q ) + r N proj ( q ) , (102)
where the defect is zero for an exact closure and must vanish in the declared scaling for an asymptotic closure. Reference-drift corrections and the influence of discarded variables are included in this defect unless retained explicitly.
For example, take ι ( q ) = ( q , 0 ) \iota(q)=(q,0) ι ( q ) = ( q , 0 ) , Π ( x 1 , x 2 ) = x 1 \Pi(x_1,x_2)=x_1 Π ( x 1 , x 2 ) = x 1 , S ( x 1 , x 2 ) = ( x 2 + 1 ) 2 / 2 S(x_1,x_2)=(x_2+1)^2/2 S ( x 1 , x 2 ) = ( x 2 + 1 ) 2 /2 , and
M = ( 1 ζ ζ 1 ) , ∣ ζ ∣ < 1 , Φ ( q ) = 1 2 , M Q = 1. (103) M=\begin{pmatrix} 1 & \zeta\\ \zeta& 1 \end{pmatrix}, \qquad|\zeta|<1, \qquad\Phi(q)=\frac{1}{2}, \qquad M_Q=1. \tag*{(103)} M = ( 1 ζ ζ 1 ) , ∣ ζ ∣ < 1 , Φ ( q ) = 2 1 , M Q = 1. (103)
The restricted-action drift is zero, whereas the projected master drift on ι ( q ) \iota(q) ι ( q ) is − Π M ∇ S = − ζ -\Pi M\nabla S=-\zeta − Π M ∇ S = − ζ . The off-diagonal mobility transmits a force from the discarded direction. Thus the reduced SDE below includes a closure assumption, not just a change of notation. Its convergence proof applies when that assumption holds, or when a vanishing projection defect is added to the drift residual controlled in (109).
Introduce a collective variable Q N Q^N Q N and consider
d Q t N = − M Q N ( Q t N ) ∇ Q Φ N ( Q t N ) d t + 2 θ N M Q N ( Q t N ) 1 / 2 d B t . (104) \mathrm{d}Q_t^N=-M_Q^N(Q_t^N)\nabla_Q\Phi_N(Q_t^N)\mathrm{d}t+\sqrt{\frac{2\theta}{N}}M_Q^N(Q_t^N)^{1/2}\mathrm{d}B_t. \tag*{(104)} d Q t N = − M Q N ( Q t N ) ∇ Q Φ N ( Q t N ) d t + N 2 θ M Q N ( Q t N ) 1/2 d B t . (104)
Assume that every M Q N M_Q^N M Q N and its limit M Q M_Q M Q are symmetric nonnegative, so that the displayed square roots are their positive matrix square roots. Fix T < ∞ T<\infty T < ∞ . Assume that the paths Q N Q^N Q N and the solution Q Q Q of the limiting equation remain in one common compact set K K K on [ 0 , T ] [0,T] [ 0 , T ] , that Q 0 N → Q 0 Q_0^N\to Q_0 Q 0 N → Q 0 in probability, and that the mobilities, their square roots, and ∇ Φ N \nabla\Phi_N ∇ Φ N converge locally uniformly on the common coordinate domain. Assume that the limiting equation is locally well posed with a unique solution. The uniform trace and Lipschitz bounds used below are also imposed on K K K . If compact containment is available only after localization, the following estimates first apply to stopped processes; removing the stopping times requires exit probabilities tending to zero uniformly in N N N .
Then the martingale part obeys
sup 0 ≤ t ≤ T ∥ 2 θ N ∫ 0 t M Q N ( Q s N ) 1 / 2 d B s ∥ ⟶ 0 (105) \sup_{0\le t\le T}\left\|\sqrt{\frac{2\theta}{N}}\int_0^t M_Q^N(Q_s^N)^{1/2}\,\mathrm{d}B_s\right\|\longrightarrow0 \tag*{(105)} 0 ≤ t ≤ T sup N 2 θ ∫ 0 t M Q N ( Q s N ) 1/2 d B s ⟶ 0 (105)
in probability, and every weakly convergent subsequence satisfies
Q ˙ t = − M Q ( Q t ) ∇ Q Φ ( Q t ) , Q ( 0 ) = Q 0 . (106) \dot{Q}_t=-M_Q(Q_t)\nabla_Q\Phi(Q_t), \qquad Q(0)=Q_0. \tag*{(106)} Q ˙ t = − M Q ( Q t ) ∇ Q Φ ( Q t ) , Q ( 0 ) = Q 0 . (106)
Indeed, in integral form
Q t N = Q 0 N − ∫ 0 t M Q N ( Q s N ) ∇ Q Φ N ( Q s N ) d s + Z t N , Z t N : = 2 θ N ∫ 0 t M Q N ( Q s N ) 1 / 2 d B s . (107) \begin{aligned}
Q_t^N&=Q_0^N-\int_0^t M_Q^N(Q_s^N)\nabla_Q\Phi_N(Q_s^N)\,\mathrm{d}s+Z_t^N,\\
Z_t^N&:=\sqrt{\frac{2\theta}{N}}\int_0^t M_Q^N(Q_s^N)^{1/2}\,\mathrm{d}B_s. \tag*{(107)}
\end{aligned} Q t N Z t N = Q 0 N − ∫ 0 t M Q N ( Q s N ) ∇ Q Φ N ( Q s N ) d s + Z t N , := N 2 θ ∫ 0 t M Q N ( Q s N ) 1/2 d B s . (107)
If sup N tr M Q N ≤ C \sup_N \operatorname{tr} M_Q^N \le C sup N tr M Q N ≤ C on the common compact region, the Burkholder–Davis–Gundy inequality gives
E sup t ≤ T ∥ Z t N ∥ 2 ≤ 8 θ N E ∫ 0 T tr M Q N ( Q s N ) d s ≤ 8 θ C T N . (108) \mathbb{E}\sup_{t\le T}\lVert Z_t^N\rVert^2 \le\frac{8\theta}{N}\mathbb{E}\int_0^T \operatorname{tr} M_Q^N(Q_s^N)\,\mathrm{d}s \le\frac{8\theta CT}{N}. \tag*{(108)} E t ≤ T sup ∥ Z t N ∥ 2 ≤ N 8 θ E ∫ 0 T tr M Q N ( Q s N ) d s ≤ N 8 θ C T . (108)
Hence Z N → 0 Z^N \to0 Z N → 0 in probability in the uniform topology. Local uniform convergence of the drifts and uniqueness of the limiting ordinary differential equation then imply, by the usual Gronwall estimate applied to (107) , convergence of every tight subsequence to (106) . For completeness, let b N ( q ) = − M Q N ( q ) ∇ Q Φ N ( q ) b_N(q)=-M_Q^N(q)\nabla_Q\Phi^N(q) b N ( q ) = − M Q N ( q ) ∇ Q Φ N ( q ) and b ( q ) = − M Q ( q ) ∇ Q Φ ( q ) b(q)=-M_Q(q)\nabla_Q\Phi(q) b ( q ) = − M Q ( q ) ∇ Q Φ ( q ) . On the common compact region assume sup N Lip ( b N ) ≤ L \sup_N \operatorname{Lip}(b_N)\le L sup N Lip ( b N ) ≤ L and r N : = sup q ∥ b N ( q ) − b ( q ) ∥ → 0 r_N:=\sup_q\lVert b_N(q)-b(q)\rVert\to0 r N := sup q ∥ b N ( q ) − b ( q )∥ → 0 . If Q Q Q solves (106) , subtraction of the two mild equations gives
sup s ≤ t ∥ Q s N − Q s ∥ ≤ ∥ Q 0 N − Q 0 ∥ + sup s ≤ t ∥ Z s N ∥ + r N t + L ∫ 0 t sup u ≤ s ∥ Q u N − Q u ∥ d s . (109) \sup_{s\le t}\lVert Q_s^N-Q_s\rVert\le\lVert Q_0^N-Q_0\rVert+\sup_{s\le t}\lVert Z_s^N\rVert+r_Nt+L\int_0^t\sup_{u\le s}\lVert Q_u^N-Q_u\rVert\,\mathrm{d}s. \tag*{(109)} s ≤ t sup ∥ Q s N − Q s ∥ ≤ ∥ Q 0 N − Q 0 ∥ + s ≤ t sup ∥ Z s N ∥ + r N t + L ∫ 0 t u ≤ s sup ∥ Q u N − Q u ∥ d s . (109)
Gronwall’s lemma consequently yields
sup t ≤ T ∥ Q t N − Q t ∥ ≤ e L T ( ∥ Q 0 N − Q 0 ∥ + sup t ≤ T ∥ Z t N ∥ + r N T ) ⟶ 0 (110) \sup_{t\le T}\lVert Q_t^N-Q_t\rVert\le e^{LT}\left(\lVert Q_0^N-Q_0\rVert+\sup_{t\le T}\lVert Z_t^N\rVert+r_NT\right)\longrightarrow0 \tag*{(110)} t ≤ T sup ∥ Q t N − Q t ∥ ≤ e L T ( ∥ Q 0 N − Q 0 ∥ + t ≤ T sup ∥ Z t N ∥ + r N T ) ⟶ 0 (110)
in probability. These explicit Lipschitz and drift-convergence hypotheses are one sufficient set of conditions behind the subsequence statement. Thus collective averaging removes the stochastic forcing. The limiting trajectory is deterministic when Q 0 Q_0 Q 0 is deterministic; for random Q 0 Q_0 Q 0 , it is the deterministic ODE flow applied to that random initial state, and remains random only through its initial condition.
Proposition 4 (Conditional macroscopic equilibrium convergence). Fix θ > 0 \theta>0 θ > 0 and the undriven reversible branch (83) . Assume the connectedness, positive-definite mobility, boundary, and integrability hypotheses of (87) , and write π θ ( d x ) = p ∗ ( x ) ν ( d x ) \pi_\theta(\mathrm{d}x)=p_*(x)\nu(\mathrm{d}x) π θ ( d x ) = p ∗ ( x ) ν ( d x ) , with p ∗ p_* p ∗ given by (92) . For densities relative to ν \nu ν , define
F θ ( p ) : = θ KL ( p ∥ p ∗ ) , \mathcal{F}_\theta(p):=\theta\operatorname{KL}(p\mathbin\Vert p_*), F θ ( p ) := θ KL ( p ∥ p ∗ ) , D θ ( p ) : = θ 2 ∫ p ⟨ ∇ log ( p / p ∗ ) , M ^ ∇ log ( p / p ∗ ) ⟩ d ν . (111) \mathcal{D}_\theta(p):=\theta^2\int p\left\langle\nabla\log(p/p_*),\widehat{M}\nabla\log(p/p_*)\right\rangle\,\mathrm{d}\nu. \tag*{(111)} D θ ( p ) := θ 2 ∫ p ⟨ ∇ log ( p / p ∗ ) , M ∇ log ( p / p ∗ ) ⟩ d ν . (111) Assume that the Fokker–Planck semigroup is continuous on the closure K K K of the forward density orbit in a declared metrizable topology, that K K K is compact and consists of smooth positive admissible densities, and that F θ \mathcal{F}_\theta F θ and D θ \mathcal{D}_\theta D θ are continuous on K K K . Require the integrated free-energy identity to hold for every orbit starting in K K K . Compact free-energy sublevels are one sufficient source of precompactness, but do not replace these continuity and admissibility requirements. Then
p t ⟶ p ∗ in the declared topology as t → ∞ . (112) p_t\longrightarrow p_*
\quad\text{in the declared topology as }t\to\infty. \tag*{(112)} p t ⟶ p ∗ in the declared topology as t → ∞. (112)
In particular, this implies weak convergence of probability measures if the topology dominates weak convergence. Separately, suppose that the dissipation identity holds along the orbit and that a logarithmic Sobolev estimate is available with the convention F θ ( p ) ≤ C L S D θ ( p ) \mathcal{F}_{\theta}(p) \le C_{\mathrm{LS}}\mathcal{D}_{\theta}(p) F θ ( p ) ≤ C LS D θ ( p ) , where 0 < C L S < ∞ 0 < C_{\mathrm{LS}} < \infty 0 < C LS < ∞ . This quantitative conclusion does not require the compact-orbit argument:
F θ ( p t ) ≤ e − t / C L S F θ ( p 0 ) . (113) \mathcal{F}_{\theta}(p_t) \le e^{-t/C_{\mathrm{LS}}}\mathcal{F}_{\theta}(p_0). \tag*{(113)} F θ ( p t ) ≤ e − t / C LS F θ ( p 0 ) . (113)
Without these compactness, uniqueness, and coercive-dissipation hypotheses, Gibbs invariance or minimum cost alone does not imply attraction. The conclusion is therefore a conditional statement about convergence of the law, not a claim that every microscopic coordinate becomes static.
Proof. Equations (87) and (94) give F ˙ θ = − D θ ≤ 0 \dot{\mathcal{F}}_{\theta}=-\mathcal{D}_{\theta}\le0 F ˙ θ = − D θ ≤ 0 . The nonnegative free energy therefore has a limit L L L . Compactness of K K K gives an omega-limit point q q q : along some t n → ∞ t_n\to\infty t n → ∞ , p t n → q p_{t_n}\to q p t n → q . Continuity gives F θ ( q ) = L \mathcal{F}_{\theta}(q)=L F θ ( q ) = L . For any fixed s ≥ 0 s\ge0 s ≥ 0 , semigroup continuity gives p t n + s → P s ∗ q p_{t_n+s}\to P_s^*q p t n + s → P s ∗ q , and hence F θ ( P s ∗ q ) = L \mathcal{F}_{\theta}(P_s^*q)=L F θ ( P s ∗ q ) = L . Apply the integrated identity to this limiting orbit:
0 = F θ ( q ) − F θ ( P s ∗ q ) = ∫ 0 s D θ ( P r ∗ q ) d r . 0=\mathcal{F}_{\theta}(q)-\mathcal{F}_{\theta}(P_s^*q)=\int_0^s\mathcal{D}_{\theta}(P_r^*q)\,\mathrm{d}r. 0 = F θ ( q ) − F θ ( P s ∗ q ) = ∫ 0 s D θ ( P r ∗ q ) d r .
Continuity and nonnegativity of the integrand imply D θ ( q ) = 0 \mathcal{D}_{\theta}(q)=0 D θ ( q ) = 0 . Positive definiteness and strict positivity of q q q give ∇ log ( q / p ∗ ) = 0 \nabla\log(q/p_*)=0 ∇ log ( q / p ∗ ) = 0 ; connectedness and normalization give q = p ∗ q=p_* q = p ∗ . Thus every omega-limit point is p ∗ p_* p ∗ . If the orbit did not converge, a sequence staying outside a neighborhood of p ∗ p_* p ∗ would have a different limit point by compactness, a contradiction. This proves (112) . For the quantitative assertion, the stated logarithmic Sobolev bound gives F ˙ θ ≤ − F θ / C L S \dot{\mathcal{F}}_{\theta}\le-\mathcal{F}_{\theta}/C_{\mathrm{LS}} F ˙ θ ≤ − F θ / C LS . Multiplying by e t / C L S e^{t/C_{\mathrm{LS}}} e t / C LS and integrating proves (113) . □ \square □
Here D θ \mathcal{D}_{\theta} D θ is relative-free-energy dissipation for the specified undriven reversible branch. Its identification with physical entropy production requires reservoir calibration; it is not the stationary path entropy rate of an arbitrary driven diffusion. On a degenerate energy-conserving branch, connectedness, zero-set uniqueness, and coercivity must instead be established on the chosen invariant energy leaf. Neither statement implies global attraction across disconnected components.
Remark 5 (An explicit nonempty class of convergence hypotheses). In a regulated affine chart, take ν = d x \nu=\mathrm{d}x ν = d x , S M ( x ) = k ∣ x ∣ 2 / 2 \mathcal{S}_{\mathrm{M}}(x)=k|x|^2/2 S M ( x ) = k ∣ x ∣ 2 /2 , and M ^ = m I \hat{M}=mI M ^ = m I , with k , m > 0 k,m>0 k , m > 0 . The process and its invariant covariance are
d X t = − γ X t d t + 2 θ m d W t , γ = m k , C ∗ = ( θ / k ) I . (114) \mathrm{d}X_t=-\gamma X_t\,\mathrm{d}t+\sqrt{2\theta m}\,\mathrm{d}W_t,\qquad\gamma=mk,\qquad C_*=(\theta/k)I. \tag*{(114)} d X t = − γ X t d t + 2 θ m d W t , γ = mk , C ∗ = ( θ / k ) I . (114) The Gaussian logarithmic Sobolev inequality [85] gives
KL ( p ∥ p ∗ ) ≤ θ 2 k ∫ p ∣ ∇ log ( p / p ∗ ) ∣ 2 d x , F θ ( p ) ≤ 1 2 γ D θ ( p ) . (115) \operatorname{KL}(p\Vert p_*)\le\frac{\theta}{2k}\int p\left|\nabla\log(p/p_*)\right|^2\mathrm{d}x,\qquad\mathcal{F}_{\theta}(p)\le\frac{1}{2\gamma}\mathcal{D}_{\theta}(p). \tag*{(115)} KL ( p ∥ p ∗ ) ≤ 2 k θ ∫ p ∣ ∇ log ( p / p ∗ ) ∣ 2 d x , F θ ( p ) ≤ 2 γ 1 D θ ( p ) . (115)
For an initial Gaussian with mean a a a and covariance C ∗ C_\ast C ∗ , the mean is e − γ t a e^{-\gamma t}a e − γ t a , the covariance remains C ∗ C_\ast C ∗ , and direct integration yields
F θ ( p t ) = k ∣ a ∣ 2 2 e − 2 γ t , D θ ( p t ) = m k 2 ∣ a ∣ 2 e − 2 γ t = 2 γ F θ ( p t ) . (116) F_\theta(p_t)=\frac{k|a|^2}{2}e^{-2\gamma t},\qquad D_\theta(p_t)=mk^2|a|^2e^{-2\gamma t}=2\gamma F_\theta(p_t). \tag*{(116)} F θ ( p t ) = 2 k ∣ a ∣ 2 e − 2 γ t , D θ ( p t ) = m k 2 ∣ a ∣ 2 e − 2 γ t = 2 γ F θ ( p t ) . (116)
Thus the rate in (113) is attained with C L S = 1 / ( 2 γ ) C_{\mathrm{LS}}=1/(2\gamma) C LS = 1/ ( 2 γ ) . Pinsker’s inequality also gives ∥ p t − p ∗ ∥ L 1 ( ν ) ≤ 2 F θ ( p 0 ) / θ e − γ t \|p_t-p_\ast\|_{L^1(\nu)}\le\sqrt{2F_\theta(p_0)/\theta}\,e^{-\gamma t} ∥ p t − p ∗ ∥ L 1 ( ν ) ≤ 2 F θ ( p 0 ) / θ e − γ t . This is a finite-regulator realization of probabilistic attraction, not a derivation of a Lorentzian particle mass or of gravitational collapse.
Microscopic tangent limit
The opposite statement must be formulated carefully. It is not correct to say without qualification that the physical drift “becomes zero” microscopically. The correct statement concerns the rescaled tangent process. Let
Y t h = Q h t − Q 0 h . (117) Y_t^h=\frac{Q_{ht}-Q_0}{\sqrt{h}}. \tag*{(117)} Y t h = h Q h t − Q 0 . (117)
Suppose Q 0 Q_0 Q 0 is deterministic and
d Q t = b ( Q t ) d t + σ Q ( Q t ) d W t (118) \mathrm{d}Q_t=b(Q_t)\mathrm{d}t+\sigma_Q(Q_t)\mathrm{d}W_t \tag*{(118)} d Q t = b ( Q t ) d t + σ Q ( Q t ) d W t (118)
and b , σ Q b,\sigma_Q b , σ Q are locally bounded and continuous at Q 0 Q_0 Q 0 . Then
Y t h = h ∫ 0 t b ( Q h s ) d s + 1 h ∫ 0 h t σ Q ( Q s ) d W s . (119) Y_t^h=\sqrt{h}\int_0^t b(Q_{hs})\mathrm{d}s+\frac{1}{\sqrt{h}}\int_0^{ht}\sigma_Q(Q_s)\mathrm{d}W_s. \tag*{(119)} Y t h = h ∫ 0 t b ( Q h s ) d s + h 1 ∫ 0 h t σ Q ( Q s ) d W s . (119)
If sup s ≤ h T ∥ b ( Q s ) ∥ ≤ C b \sup_{s\le hT}\|b(Q_s)\|\le C_b sup s ≤ h T ∥ b ( Q s ) ∥ ≤ C b on the localization event, then
sup t ≤ T ∥ h ∫ 0 t b ( Q h s ) d s ∥ ≤ C b T h ⟶ 0. (120) \sup_{t\le T}\left\|\sqrt{h}\int_0^t b(Q_{hs})\mathrm{d}s\right\|\le C_bT\sqrt{h}\longrightarrow0. \tag*{(120)} t ≤ T sup h ∫ 0 t b ( Q h s ) d s ≤ C b T h ⟶ 0. (120)
Thus the drift term vanishes in probability. The second term has quadratic variation
⟨ 1 h ∫ 0 h ⋅ σ Q ( Q s ) d W s ⟩ t = 1 h ∫ 0 h t σ Q ( Q s ) σ Q ( Q s ) ∗ d s ⟶ t σ Q ( Q 0 ) σ Q ( Q 0 ) ∗ (121) \left\langle\frac{1}{\sqrt{h}}\int_0^{h\cdot}\sigma_Q(Q_s)\mathrm{d}W_s\right\rangle_t
=\frac{1}{h}\int_0^{ht}\sigma_Q(Q_s)\sigma_Q(Q_s)^\ast\mathrm{d}s
\longrightarrow t\sigma_Q(Q_0)\sigma_Q(Q_0)^\ast \tag*{(121)} ⟨ h 1 ∫ 0 h ⋅ σ Q ( Q s ) d W s ⟩ t = h 1 ∫ 0 h t σ Q ( Q s ) σ Q ( Q s ) ∗ d s ⟶ t σ Q ( Q 0 ) σ Q ( Q 0 ) ∗ (121)
and its jumps vanish because the martingale is continuous. Since continuity of Q Q Q at the deterministic initial point gives sup s ≤ h T ∥ Q s − Q 0 ∥ → 0 \sup_{s\le hT}\|Q_s-Q_0\|\to0 sup s ≤ h T ∥ Q s − Q 0 ∥ → 0 in probability, the displayed quadratic-variation convergence holds in probability. The martingale functional central-limit theorem therefore gives the Gaussian limit
σ Q ( Q 0 ) W t . (122) \sigma_Q(Q_0)W_t. \tag*{(122)} σ Q ( Q 0 ) W t . (122)
Hence
Y h ⇒ σ Q ( Q 0 ) W ( h ↓ 0 ) . (123) Y^h\Rightarrow\sigma_Q(Q_0)W\qquad(h\downarrow0). \tag*{(123)} Y h ⇒ σ Q ( Q 0 ) W ( h ↓ 0 ) . (123)
The drift disappears only after the microscopic tangent rescaling. This distinction is essential for mathematical correctness.
Time reversal is a property of path laws, not of one coefficient
The preceding limits separate a deterministic collective trajectory from a Gaussian tangent fluctuation. They do not, by themselves, classify either limit as reversible or irreversible. Four notions must be kept distinct:
(i) a complete ordinary differential equation may be solved both forward and backward, so its flow maps are invertible;
(ii) a deterministic equation is physically time-reversal covariant only after a parity involution on its variables has been specified;
(iii) a stationary stochastic process is statistically reversible when its entire path law is invariant under the corresponding reversal;
(iv) thermodynamic irreversibility is quantified by a positive forward-versus-reversed path-space relative entropy, or equivalently by an irreversible probability current under the hypotheses below.
These distinctions and their path-space formulations follow the standard stochastic-thermodynamic treatment of diffusion reversal and odd-parity variables [31], [146], and [144]. The first statement is a property of uniqueness for an ODE. It is not the same as the second statement. In particular, an invertible gradient-flow map can still select a preferred time direction through a strict Lyapunov function.
Let Θ \Theta Θ be a smooth involution of the finite-regulator state space. Positions and scalar amplitudes are normally even, while momenta and magnetic variables may be odd. For a path ω ∈ C ( [ 0 , T ] ; X ) \omega\in C([0,T];\mathcal{X}) ω ∈ C ([ 0 , T ] ; X ) , define
( T ω ) ( t ) : = Θ ω ( T − t ) , 0 ≤ t ≤ T . (124) (\mathcal{T}\omega)(t) := \Theta\omega(T-t), \qquad0 \le t \le T. \tag*{(124)} ( T ω ) ( t ) := Θ ω ( T − t ) , 0 ≤ t ≤ T . (124)
If P π [ 0 , T ] \mathbb{P}_{\pi}^{[0,T]} P π [ 0 , T ] is the stationary path law with invariant density π \pi π , its finite-time arrow is measured by
A T ( π ) : = D K L ( P π [ 0 , T ] ∥ P π [ 0 , T ] ∘ T − 1 ) ≥ 0. (125) \mathcal{A}_T(\pi) := D_{\mathrm{KL}}\left(\mathbb{P}_{\pi}^{[0,T]} \,\middle\|\, \mathbb{P}_{\pi}^{[0,T]} \circ\mathcal{T}^{-1}\right) \ge0. \tag*{(125)} A T ( π ) := D KL ( P π [ 0 , T ] P π [ 0 , T ] ∘ T − 1 ) ≥ 0. (125)
This definition has an operational meaning: it is the information available for distinguishing a stationary movie from the same movie run backward with all odd variables parity reversed. Statistical reversibility is exactly A T ( π ) = 0 \mathcal{A}_T(\pi) = 0 A T ( π ) = 0 for every T > 0 T > 0 T > 0 .
Proposition 6 (Coefficient limits do not determine time reversal). Let
d X t ε = b ε ( X t ε ) d t + 2 D ε ( X t ε ) d W t (126) \mathrm{d}X_t^\varepsilon= b_\varepsilon(X_t^\varepsilon)\,\mathrm{d}t + \sqrt{2D_\varepsilon(X_t^\varepsilon)}\,\mathrm{d}W_t \tag*{(126)} d X t ε = b ε ( X t ε ) d t + 2 D ε ( X t ε ) d W t (126) be a smooth, conservative diffusion on a torus, or on R d \mathbb{R}^d R d with the decay and generator-domain conditions needed for the integrations by parts below. Assume that the stationary forward and reversed martingale problems are unique in law, their Girsanov density is a true martingale, and the expected drift-change quadratic energy is finite on each finite interval. These conditions hold for
smooth uniformly elliptic coefficients on a torus. A reflected-domain version additionally requires compatible forward and reversed boundary realizations; zero probability flux alone does not specify them. Suppose first that all variables are even, D ε = D ε T ≻ 0 D_\varepsilon= D_\varepsilon^{\mathsf T} \succ0 D ε = D ε T ≻ 0 , and π ε > 0 \pi_\varepsilon> 0 π ε > 0 is a smooth invariant density. Define
J ε , i [ ρ ] : = b ε , i ρ − ∂ j ( D ε , i j ρ ) . (127) J_{\varepsilon,i}[\rho] := b_{\varepsilon,i}\rho- \partial_j(D_{\varepsilon,ij}\rho). \tag*{(127)} J ε , i [ ρ ] := b ε , i ρ − ∂ j ( D ε , ij ρ ) . (127)
Then the drift of the stationary reversed process is
b ε , i R = − b ε , i + 2 π ε − 1 ∂ j ( D ε , i j π ε ) , b ε − b ε R = 2 J ε [ π ε ] π ε . (128) b_{\varepsilon,i}^{R} = -b_{\varepsilon,i} + 2\pi_\varepsilon^{-1}\partial_j(D_{\varepsilon,ij}\pi_\varepsilon), \qquad b_\varepsilon- b_\varepsilon^{R} = \frac{2J_\varepsilon[\pi_\varepsilon]}{\pi_\varepsilon}. \tag*{(128)} b ε , i R = − b ε , i + 2 π ε − 1 ∂ j ( D ε , ij π ε ) , b ε − b ε R = π ε 2 J ε [ π ε ] . (128)
Consequently, the following conditions are equivalent:
P π ε [ 0 , T ] = P π ε [ 0 , T ] ∘ T T − 1 for every T ⟺ J ε [ π ε ] = 0 ⟺ L ε = L ε † , π ε . (129) \mathbb{P}_{\pi_\varepsilon}^{[0,T]} = \mathbb{P}_{\pi_\varepsilon}^{[0,T]} \circ\mathcal{T}_T^{-1}\ \text{for every }T \quad\Longleftrightarrow\quad J_\varepsilon[\pi_\varepsilon] = 0 \quad\Longleftrightarrow\quad\mathcal{L}_\varepsilon= \mathcal{L}_\varepsilon^{\dagger,\pi_\varepsilon}. \tag*{(129)} P π ε [ 0 , T ] = P π ε [ 0 , T ] ∘ T T − 1 for every T ⟺ J ε [ π ε ] = 0 ⟺ L ε = L ε † , π ε . (129)
Moreover, the stationary entropy-production rate is
epr ε : = lim T → ∞ 1 T A T ( π ε ) = ∫ X J ε [ π ε ] T D ε − 1 J ε [ π ε ] π ε d x ≥ 0. (130) \operatorname{epr}_\varepsilon:= \lim_{T\to\infty}\frac{1}{T}\mathcal{A}_T(\pi_\varepsilon) = \int_{\mathcal{X}} \frac{J_\varepsilon[\pi_\varepsilon]^{\mathsf T}D_\varepsilon^{-1}J_\varepsilon[\pi_\varepsilon]}{\pi_\varepsilon}\,dx \ge0. \tag*{(130)} epr ε := T → ∞ lim T 1 A T ( π ε ) = ∫ X π ε J ε [ π ε ] T D ε − 1 J ε [ π ε ] d x ≥ 0. (130)
For a positive semidefinite diffusion matrix, the same identity holds with D ε † D_\varepsilon^\dagger D ε † only when the forward–reverse drift difference lies in Ran D ε \operatorname{Ran}D_\varepsilon Ran D ε , its pseudoinverse energy is finite, the relevant range is measurable with controlled rank, the Girsanov exponential is a true martingale, and the two martingale problems are unique in law. Otherwise the path laws need not be mutually absolutely continuous.
If D ε → 0 D_\varepsilon\to0 D ε → 0 and b ε → b 0 b_\varepsilon\to b_0 b ε → b 0 , the limiting ODE is time-reversal covariant under a smooth involution Θ \Theta Θ only if
d Θ x b 0 ( x ) = − b 0 ( Θ x ) . (131) \mathrm{d}\Theta_x b_0(x) = -b_0(\Theta x). \tag*{(131)} d Θ x b 0 ( x ) = − b 0 ( Θ x ) . (131)
If instead b ε → 0 b_\varepsilon\to0 b ε → 0 and D ε → D 0 D_\varepsilon\to D_0 D ε → D 0 , the limiting pure diffusion is reversible with respect to a positive density π 0 \pi_0 π 0 precisely when
∂ j ( D 0 , i j π 0 ) = 0. (132) \partial_j(D_{0,ij}\pi_0) = 0. \tag*{(132)} ∂ j ( D 0 , ij π 0 ) = 0. (132)
Thus neither D ε → 0 D_\varepsilon\to0 D ε → 0 nor b ε → 0 b_\varepsilon\to0 b ε → 0 , considered alone, determines time-reversal symmetry.
The limits in this proposition are singular at the level of path measures. For example, a nondegenerate diffusion law may be mutually singular with the law of its deterministic limit, so the finite-diffusion relative-entropy rate need not converge to the relative entropy of the limiting process. The criteria above are therefore applied before the singular limit, or together with a separately proved weak-convergence, large-deviation, or Γ \Gamma Γ -convergence statement. No continuity of the arrow functional in the diffusion parameter is assumed.
Proof. The stationary forward equation is ∂ t ρ = − ∂ i J ε , i [ ρ ] \partial_t \rho= -\partial_i J_{\varepsilon,i}[\rho] ∂ t ρ = − ∂ i J ε , i [ ρ ] . Integrating the generator by parts in L 2 ( π ε ) L^2(\pi_\varepsilon) L 2 ( π ε ) gives the adjoint drift in (128) . Hence the forward and reversed generators agree exactly when the stationary current vanishes, which proves (129) . The two processes have the same covariance 2 D ε 2D_\varepsilon 2 D ε . Girsanov’s formula and (128) therefore give
1 2 ∫ ( b ε − b ε R ) T ( 2 D ε ) − 1 ( b ε − b ε R ) π ε d x = ∫ J ε [ π ε ] T D ε − 1 J ε [ π ε ] π ε d x . \frac{1}{2}\int(b_\varepsilon-b_\varepsilon^{R})^{\mathsf{T}}(2D_\varepsilon)^{-1}(b_\varepsilon-b_\varepsilon^{R})\pi_\varepsilon\,\mathrm{d}x=\int\frac{J_\varepsilon[\pi_\varepsilon]^{\mathsf{T}}D_\varepsilon^{-1}J_\varepsilon[\pi_\varepsilon]}{\pi_\varepsilon}\,\mathrm{d}x. 2 1 ∫ ( b ε − b ε R ) T ( 2 D ε ) − 1 ( b ε − b ε R ) π ε d x = ∫ π ε J ε [ π ε ] T D ε − 1 J ε [ π ε ] d x .
Stationarity removes the endpoint relative-entropy contribution, proving (130) . For the deterministic limit, if x ( t ) x(t) x ( t ) solves x ˙ = b 0 ( x ) \dot{x}=b_0(x) x ˙ = b 0 ( x ) , differentiation of y ( t ) = Θ x ( − t ) y(t)=\Theta x(-t) y ( t ) = Θ x ( − t ) shows that y y y solves the same equation exactly when (131) holds. Finally, setting b 0 = 0 b_0=0 b 0 = 0 in (127) proves (132) . □
For odd variables, assume that Θ \Theta Θ preserves the state domain and its boundary conditions, Θ # π = π \Theta_{\#}\pi=\pi Θ # π = π , the generator domains are mapped into each other, and the diffusion is parity covariant:
D ( Θ x ) = d Θ x D ( x ) d Θ x T . D(\Theta x)=\mathrm{d}\Theta_xD(x)\mathrm{d}\Theta_x^{\mathsf{T}}. D ( Θ x ) = d Θ x D ( x ) d Θ x T .
Let U Θ f = f ∘ Θ U_\Theta f=f\circ\Theta U Θ f = f ∘ Θ . The parity-corrected detailed-balance condition is
L † , π = U Θ L U Θ . (133) \mathcal{L}^{\dagger,\pi}=U_\Theta\mathcal{L}U_\Theta. \tag*{(133)} L † , π = U Θ L U Θ . (133)
If
L = A + S , A † , π = − A , S † , π = S , (134) \mathcal{L}=\mathcal{A}+\mathcal{S},\qquad\mathcal{A}^{\dagger,\pi}=-\mathcal{A},\qquad\mathcal{S}^{\dagger,\pi}=\mathcal{S}, \tag*{(134)} L = A + S , A † , π = − A , S † , π = S , (134)
and
U Θ A U Θ = − A , U Θ S U Θ = S , (135) U_\Theta\mathcal{A}U_\Theta=-\mathcal{A},\qquad U_\Theta\mathcal{S}U_\Theta=\mathcal{S}, \tag*{(135)} U Θ A U Θ = − A , U Θ S U Θ = S , (135)
then (133) follows. This operator identity reconciles two uses of the term reversible in this paper. The overdamped Gibbs generator (81) has A = 0 \mathcal{A}=0 A = 0 and is itself self-adjoint. In the later reversible–irreversible decomposition, L ∇ E \mathcal{L}\nabla E L ∇ E supplies the conservative anti-self-adjoint transport, whose sign is reversed by Θ \Theta Θ , while the mobility and diffusion form the symmetric sector.
The irreversible-current formula can also be made explicit when odd variables are present. Its statement needs more than an algebraic split of the drift.
Proposition 7 (Entropy rate with linear time-reversal parity). Consider a stationary, nonexplosive diffusion with generator L f = b ⋅ ∇ f + D i j ∂ i ∂ j f \mathcal{L}f=b\cdot\nabla f+D_{ij}\partial_i\partial_j f L f = b ⋅ ∇ f + D ij ∂ i ∂ j f and smooth positive invariant coordinate density π \pi π . Its state space is R d \mathbb{R}^d R d with sufficient decay for integration by parts, or a torus. Let Θ x = P x \Theta x=Px Θ x = P x , where P P P is an orthogonal involution preserving the domain and, on a torus, the period lattice. Assume
P 2 = I , π ( P x ) = π ( x ) , D ( P x ) = P D ( x ) P T , D ( x ) ≻ 0. (136) P^2=I,\qquad\pi(Px)=\pi(x),\qquad D(Px)=PD(x)P^{\mathsf{T}},\qquad D(x)\succ0. \tag*{(136)} P 2 = I , π ( P x ) = π ( x ) , D ( P x ) = P D ( x ) P T , D ( x ) ≻ 0. (136)
Assume that the forward and reversed martingale problems are unique in law, the Girsanov density relating them is a true martingale, and its expected quadratic energy is finite on every finite interval. Define
c i ( x ) : = π ( x ) − 1 ∂ j ( D i j ( x ) π ( x ) ) , b i r r ( x ) : = 1 2 ( b ( x ) + P b ( P x ) ) , J i r r , i ( x ) : = π ( x ) b i r r , i ( x ) − ∂ j ( D i j ( x ) π ( x ) ) . (137) \begin{aligned}
c_i(x) &:= \pi(x)^{-1}\partial_j\bigl(D_{ij}(x)\pi(x)\bigr), \qquad b_{\mathrm{irr}}(x) := \tfrac{1}{2}\bigl(b(x)+P b(Px)\bigr),\\
J_{\mathrm{irr},i}(x) &:= \pi(x)b_{\mathrm{irr},i}(x)-\partial_j\bigl(D_{ij}(x)\pi(x)\bigr). \tag*{(137)}
\end{aligned} c i ( x ) J irr , i ( x ) := π ( x ) − 1 ∂ j ( D ij ( x ) π ( x ) ) , b irr ( x ) := 2 1 ( b ( x ) + P b ( P x ) ) , := π ( x ) b irr , i ( x ) − ∂ j ( D ij ( x ) π ( x ) ) . (137)
The parity-reversed path law has drift
b R Θ ( x ) = − P b ( P x ) + 2 c ( x ) , b ( x ) − b R Θ ( x ) = 2 J i r r ( x ) / π ( x ) . (138) b^{R\Theta}(x)=-P b(Px)+2c(x), \qquad b(x)-b^{R\Theta}(x)=2J_{\mathrm{irr}}(x)/\pi(x). \tag*{(138)} b R Θ ( x ) = − P b ( P x ) + 2 c ( x ) , b ( x ) − b R Θ ( x ) = 2 J irr ( x ) / π ( x ) . (138)
Consequently, for the path-law relative entropy A T ( π ) \mathcal{A}_T(\pi) A T ( π ) in (125) ,
A T ( π ) T = ∫ X J i r r T D − 1 J i r r π d x ≥ 0. (139) \frac{\mathcal{A}_T(\pi)}{T}=\int_{\mathcal{X}}\frac{J_{\mathrm{irr}}^{\mathsf{T}}D^{-1}J_{\mathrm{irr}}}{\pi}\,dx\ge0. \tag*{(139)} T A T ( π ) = ∫ X π J irr T D − 1 J irr d x ≥ 0. (139)
The equality holds for every T > 0 T>0 T > 0 , not only asymptotically, under these stationary hypotheses.
Proof. The ordinary stationary reverse has drift b R = − b + 2 c b^R=-b+2c b R = − b + 2 c , by (128) . Parity covariance and invariance of π \pi π imply c ( P x ) = P c ( x ) c(Px)=Pc(x) c ( P x ) = P c ( x ) . Since P P P is linear, the pushforward drift is P b R ( P x ) Pb^R(Px) P b R ( P x ) , with no additional Itô Hessian term; this gives (138) . The reversed law starts with Θ # π = π \Theta_{\#}\pi=\pi Θ # π = π , so there is no initial relative-entropy term. Both laws have covariance 2 D 2D 2 D . Taking the controlled expectation of the log Girsanov density gives
A T ( π ) = 1 4 E ∫ 0 T ( b − b R Θ ) T D − 1 ( b − b R Θ ) ( X t ) d t = T ∫ X J i r r T D − 1 J i r r π d x . (140) \mathcal{A}_T(\pi)=\frac{1}{4}\mathbb{E}\int_0^T\bigl(b-b^{R\Theta}\bigr)^{\mathsf{T}}D^{-1}\bigl(b-b^{R\Theta}\bigr)(X_t)\,dt=T\int_{\mathcal{X}}\frac{J_{\mathrm{irr}}^{\mathsf{T}}D^{-1}J_{\mathrm{irr}}}{\pi}\,dx. \tag*{(140)} A T ( π ) = 4 1 E ∫ 0 T ( b − b R Θ ) T D − 1 ( b − b R Θ ) ( X t ) d t = T ∫ X π J irr T D − 1 J irr d x . (140)
The first factor is 1 2 ( 2 D ) − 1 \tfrac{1}{2}(2D)^{-1} 2 1 ( 2 D ) − 1 ; substitution of the drift difference cancels the factor 1 4 \tfrac{1}{4} 4 1 . Stationarity supplies the last equality. Positive definiteness of D D D gives nonnegativity. □
For degenerate D D D , replacement by D † D^\dagger D † requires the range, measurability, finite-energy, and absolute-continuity conditions following (130) . Zero probability flux alone does not specify a boundary diffusion: a reflected version additionally needs the same conormal reflection and compatible generator domains for both path laws. A nonlinear parity map would introduce an Itô Hessian correction and is not covered by Proposition 7.
For example, underdamped Langevin dynamics with constant particle mass m p > 0 m_p>0 m p > 0 is
d q t = p t / m p d t , d p t = − ∇ V ( q t ) d t − γ p t d t + 2 γ m p θ d W t . (141) dq_t=p_t/m_p\,dt,\qquad dp_t=-\nabla V(q_t)\,dt-\gamma p_t\,dt+\sqrt{2\gamma m_p\theta}\,dW_t. \tag*{(141)} d q t = p t / m p d t , d p t = − ∇ V ( q t ) d t − γ p t d t + 2 γ m p θ d W t . (141)
Assume γ , θ > 0 \gamma,\theta>0 γ , θ > 0 , a normalizable Gibbs density, appropriate periodic or confining position data, and well-posed stationary dynamics. Then Θ ( q , p ) = ( q , − p ) \Theta(q,p)=(q,-p) Θ ( q , p ) = ( q , − p ) and π ∝ exp [ − ( p 2 / ( 2 m p ) + V ( q ) ) / θ ] \pi\propto\exp[-(p^2/(2m_p)+V(q))/\theta] π ∝ exp [ − ( p 2 / ( 2 m p ) + V ( q )) / θ ] . Its diffusion matrix is degenerate, but direct calculation gives b i r r = ( 0 , − γ p ) = c b_{\mathrm{irr}}=(0,-\gamma p)=c b irr = ( 0 , − γ p ) = c . Thus J i r r = 0 J_{\mathrm{irr}}=0 J irr = 0 and the parity-corrected stationary entropy rate is zero, although the Hamiltonian transport current π ( p / m p , − ∇ V ) \pi(p/m_{\mathrm{p}},-\nabla V) π ( p / m p , − ∇ V ) need not vanish. This example demonstrates why the total current cannot be substituted for the irreversible current in the presence of momenta.
The two limits above now have an exact interpretation. Along the macroscopic gradient-flow limit (106) ,
d d t Φ ( Q t ) = − ⟨ ∇ Q Φ ( Q t ) , M Q ( Q t ) ∇ Q Φ ( Q t ) ⟩ ≤ 0. (142) \frac{\mathrm{d}}{\mathrm{d}t}\Phi(Q_t)=-\langle\nabla_Q\Phi(Q_t),M_Q(Q_t)\nabla_Q\Phi(Q_t)\rangle\leq0. \tag*{(142)} d t d Φ ( Q t ) = − ⟨ ∇ Q Φ ( Q t ) , M Q ( Q t ) ∇ Q Φ ( Q t )⟩ ≤ 0. (142)
For ordinary reversal, Θ = Id \Theta=\operatorname{Id} Θ = Id , condition (131) would require b 0 = − b 0 b_0=-b_0 b 0 = − b 0 . Except at a stationary point, the deterministic gradient flow is therefore dissipative even though its martingale term has disappeared. Noise suppression establishes determinism, not physical reversibility.
Conversely, periodize a constant version of the tangent diffusion (123) on a torus and put
D 0 ≔ 1 2 σ Q ( Q 0 ) σ Q ( Q 0 ) ∗ . D_0\coloneqq\frac{1}{2}\sigma_Q(Q_0)\sigma_Q(Q_0)^{*}. D 0 : = 2 1 σ Q ( Q 0 ) σ Q ( Q 0 ) ∗ .
Its uniform invariant density has zero current and its stationary path law is reversible. If it is prepared instead with a nonuniform density, then
J [ ρ t ] = − D 0 ∇ ρ t , J[\rho_t]=-D_0\nabla\rho_t, J [ ρ t ] = − D 0 ∇ ρ t ,
d d t D K L ( ρ t ∥ π u n i f ) = − ∫ ρ t ⟨ ∇ log ρ t π u n i f , D 0 ∇ log ρ t π u n i f ⟩ d x ≤ 0. (143) \frac{\mathrm{d}}{\mathrm{d}t}D_{\mathrm{KL}}(\rho_t\mathbin\Vert\pi_{\mathrm{unif}})=-\int\rho_t\left\langle\nabla\log\frac{\rho_t}{\pi_{\mathrm{unif}}},D_0\nabla\log\frac{\rho_t}{\pi_{\mathrm{unif}}}\right\rangle\mathrm{d}x\leq0. \tag*{(143)} d t d D KL ( ρ t ∥ π unif ) = − ∫ ρ t ⟨ ∇ log π unif ρ t , D 0 ∇ log π unif ρ t ⟩ d x ≤ 0. (143)
On a finite interval [ 0 , T ] [0,T] [ 0 , T ] , the drift of the process viewed backward is
b R ( s , x ) = 2 D 0 ∇ log ρ T − s ( x ) . (144) b^{R}(s,x)=2D_0\nabla\log\rho_{T-s}(x). \tag*{(144)} b R ( s , x ) = 2 D 0 ∇ log ρ T − s ( x ) . (144)
Thus a driftless forward equation can exhibit a transient arrow from a nonequilibrium initial law, while the same diffusion is exactly reversible at uniform equilibrium.
The intuitive statement that the macroscopic branch is reversible can be made correct only through a matched scaling. Consider
d X t α = [ b c o n s ( X t α ) + α b d i s s ( X t α ) ] d t + 2 α D ( X t α ) d W t , α ↓ 0. (145) \mathrm{d}X_t^\alpha=\left[b_{\mathrm{cons}}(X_t^\alpha)+\alpha b_{\mathrm{diss}}(X_t^\alpha)\right]\mathrm{d}t+\sqrt{2\alpha D(X_t^\alpha)}\,\mathrm{d}W_t,\qquad\alpha\downarrow0. \tag*{(145)} d X t α = [ b cons ( X t α ) + α b diss ( X t α ) ] d t + 2 α D ( X t α ) d W t , α ↓ 0. (145)
To justify the phrase fluctuation–dissipation partner, assume that a Θ \Theta Θ -invariant density π \pi π satisfies
b d i s s , i ( x ) = π ( x ) − 1 ∂ j ( D i j ( x ) π ( x ) ) , (146) b_{\mathrm{diss},i}(x)=\pi(x)^{-1}\partial_j\left(D_{ij}(x)\pi(x)\right), \tag*{(146)} b diss , i ( x ) = π ( x ) − 1 ∂ j ( D ij ( x ) π ( x ) ) , (146)
with the corresponding zero-flux or periodic boundary condition, and that ∇ ⋅ ( b c o n s π ) = 0 \nabla\cdot(b_{\mathrm{cons}}\pi)=0 ∇ ⋅ ( b cons π ) = 0 . The same factor then suppresses the dissipative response and its fluctuation partner. On every fixed finite time interval, under the usual tightness and coefficient-convergence hypotheses, the limit is X ˙ = b c o n s ( X ) \dot{X}=b_{\mathrm{cons}}(X) X ˙ = b cons ( X ) . It is Θ \Theta Θ -reversible if b c o n s b_{\mathrm{cons}} b cons satisfies (131) and the process is prepared in the Θ \Theta Θ -invariant stationary law. By contrast, the collective scaling (104) suppresses only the noise unless an additional scaling is imposed on the gradient drift, so reversibility cannot be inferred from that limit alone.
The fixed-time qualification cannot be dropped. Even if the stationary entropy-production rate is O ( α ) O(\alpha) O ( α ) , observation on a time scale T α = O ( α − 1 ) T_\alpha= O(\alpha^{-1}) T α = O ( α − 1 ) can accumulate an order-one forward–reverse relative entropy. Hence the limits α ↓ 0 \alpha\downarrow0 α ↓ 0 and T → ∞ T \to\infty T → ∞ need not commute.
This is the precise sense in which the framework unifies the two temporal descriptions. The common generator contains a conservative transport sector, a fluctuation–dissipation sector, and any external drive. Changing scale can alter which sector is visible, but the arrow of time at that scale is always decided by parity, probability current, boundary and initial data, and the forward–reverse path law. The apparent conflict between microscopic and macroscopic time is therefore removed by separating distinct mathematical questions, not by assigning reversibility to the disappearance of one coefficient.
Physical entropy production and a coarse-graining no-go result
The path-space calculation acquires a thermodynamic meaning only after the forward protocol, reversed protocol, reservoirs, and physical time parameter have been specified. If their path laws are mutually absolutely continuous, define the trajectory-wise total entropy production by
Δ s t o t [ ω ] : = k B log d P F d ( P R ∘ T T − 1 ) ( ω ) . (147) \Delta s_{\mathrm{tot}}[\omega] := k_{\mathrm{B}} \log\frac{\mathrm{d}P_{\mathrm{F}}}{\mathrm{d}(P_{\mathrm{R}} \circ T_{\mathrm{T}}^{-1})(\omega).} \tag*{(147)} Δ s tot [ ω ] := k B log d ( P R ∘ T T − 1 ) ( ω ) . d P F (147)
Normalization of the Radon–Nikodym derivative gives the integral fluctuation identity and its relative-entropy consequence [103], [47], and [144]:
E F exp ( − Δ s t o t k B ) = 1 , E F Δ s t o t = k B D K L ( P F ∥ P R ∘ T T − 1 ) ≥ 0. (148) \mathbb{E}_{\mathrm{F}}\exp\left(-\frac{\Delta s_{\mathrm{tot}}}{k_{\mathrm{B}}}\right)=1,\qquad\mathbb{E}_{\mathrm{F}}\Delta s_{\mathrm{tot}=k_{\mathrm{B}}D_{\mathrm{KL}}\left(P_{\mathrm{F}}\mathbin{\Vert}P_{\mathrm{R}}\circ T_{\mathrm{T}}^{-1}\right)\ge0.} \tag*{(148)} E F exp ( − k B Δ s tot ) = 1 , E F Δ s tot = k B D KL ( P F ∥ P R ∘ T T − 1 ) ≥ 0. (148)
The first equality is measure theoretic. Interpreting it as physical entropy production additionally requires local detailed balance and a declared heat, work, and reservoir bookkeeping convention. In particular, the auxiliary stochastic relaxation time used in a Euclidean sampler has not been proved equal to Lorentzian physical time.
The Crooks and Jarzynski work relations are a more specialized consequence, not synonyms for (148) . Suppose the system is coupled to an isothermal reservoir at the physical energy scale θ = k B T b a t h \theta= k_{\mathrm{B}}T_{\mathrm{bath}} θ = k B T bath , satisfies local detailed balance, and is driven by an energy E ( x ; λ t ) E(x;\lambda_t) E ( x ; λ t ) . The forward process starts in canonical equilibrium at λ 0 \lambda_0 λ 0 , and the reverse process starts in canonical equilibrium at λ T \lambda_T λ T with the reversed protocol and the correct parity of all external controls. In the absence of other work sources, use the work-on-system convention
W [ ω ] : = ∫ 0 T ∂ λ E ( X t ; λ t ) λ ˙ t d t , Δ F e q : = − θ log ( Z ( λ T ) Z ( λ 0 ) ) . (149) W[\omega] := \int_0^T \partial_\lambda E(X_t;\lambda_t)\dot{\lambda}_t\,\mathrm{d}t,\qquad\Delta F_{\mathrm{eq}} := -\theta\log\left(\frac{Z(\lambda_T)}{Z(\lambda_0)}\right). \tag*{(149)} W [ ω ] := ∫ 0 T ∂ λ E ( X t ; λ t ) λ ˙ t d t , Δ F eq := − θ log ( Z ( λ 0 ) Z ( λ T ) ) . (149)
This F e q F_{\mathrm{eq}} F eq is the canonical equilibrium free energy, not the relative-free-energy functional of an arbitrary evolving density. The first law Δ E = W − Q b a t h \Delta E = W - Q_{\mathrm{bath}} Δ E = W − Q bath , together with the equilibrium endpoint densities, gives the path identity
log d P F d ( P R ∘ T T − 1 ) ( ω ) = W [ ω ] − Δ F e q θ . (150) \log\frac{\mathrm{d}\mathbb{P}_{\mathrm{F}}}{\mathrm{d}(\mathbb{P}_{\mathrm{R}} \circ\mathcal{T}_{T}^{-1})(\omega)} = \frac{W[\omega]-\Delta F_{\mathrm{eq}}}{\theta}. \tag*{(150)} log d ( P R ∘ T T − 1 ) ( ω ) d P F = θ W [ ω ] − Δ F eq . (150)
Here the reservoir entropy change is Q b a t h / T b a t h Q_{\mathrm{bath}}/T_{\mathrm{bath}} Q bath / T bath . The endpoint system-entropy term cancels Δ E / θ \Delta E/\theta Δ E / θ , leaving the right-hand side. If forward and reversed work densities exist, marginalizing the path identity and using W R [ T ω ] = − W [ ω ] W_{\mathrm{R}}[\mathcal{T}\omega]=-W[\omega] W R [ T ω ] = − W [ ω ] gives
p F ( w ) p R ( − w ) = exp ( w − Δ F e q θ ) , E F e − W / θ = e − Δ F e q / θ . (151) \frac{p_{\mathrm{F}}(w)}{p_{\mathrm{R}}(-w)}=\exp\left(\frac{w-\Delta F_{\mathrm{eq}}}{\theta}\right), \qquad\mathbb{E}_{\mathrm{F}}e^{-W/\theta}=e^{-\Delta F_{\mathrm{eq}}/\theta}. \tag*{(151)} p R ( − w ) p F ( w ) = exp ( θ w − Δ F eq ) , E F e − W / θ = e − Δ F eq / θ . (151)
The second equality follows directly by substituting (150) in (148); it does not require work densities. Noncanonical initial laws, multiple reservoirs, or a mismatch between stochastic and physical time require additional boundary or entropy terms. These work identities therefore apply only after the stated physical interface has been established [103], [47], and [144].
There is also a useful no-go theorem for emergence by exact filtering. Let Π \Pi Π be a measurable path projection that intertwines reversal, Π ∘ T T = T T Q ∘ Π \Pi\circ\mathcal{T}_{T}=\mathcal{T}_{T}^{Q}\circ\Pi Π ∘ T T = T T Q ∘ Π . Data processing gives
D K L ( P F ∥ P R ∘ T T − 1 ) ≥ D K L ( Π # P F ∥ Π # P R ∘ ( T T Q ) − 1 ) . (152) D_{\mathrm{KL}}\left(\mathbb{P}_{\mathrm{F}} \,\middle\|\, \mathbb{P}_{\mathrm{R}}\circ\mathcal{T}_{T}^{-1}\right)\ge D_{\mathrm{KL}}\left(\Pi_{\#}\mathbb{P}_{\mathrm{F}} \,\middle\|\, \Pi_{\#}\mathbb{P}_{\mathrm{R}}\circ(\mathcal{T}_{T}^{Q})^{-1}\right). \tag*{(152)} D KL ( P F P R ∘ T T − 1 ) ≥ D KL ( Π # P F Π # P R ∘ ( T T Q ) − 1 ) . (152)
Consequently, if the complete stationary master path law is reversible, every exact reversal-compatible projection of it is also reversible. Coarse graining can hide entropy production; it cannot create fundamental positive path-space entropy production from an exactly reversible law. An irreversible macroscopic closure must therefore identify at least one additional source: a nonequilibrium low-entropy initial condition, external driving or reservoirs, unobserved odd variables, memory discarded by a Markov approximation, or a singular noncommutation of long-time and regulator limits. The forcing and dissipation assumed in the finite Navier–Stokes branch are such physical interface data; they do not follow from equilibrium Gibbs notation alone.
The same distinction survives quantization. For a closed state with unitary evolution,
S v N ( U ρ U † ) = S v N ( ρ ) . (153) S_{\mathrm{vN}}(U\rho U^{\dagger})=S_{\mathrm{vN}}(\rho). \tag*{(153)} S vN ( U ρ U † ) = S vN ( ρ ) . (153)
Growth of a reduced quantum entropy must come from entanglement with discarded degrees of freedom, an environment, measurement, coarse graining, or special boundary data. The framework therefore resolves the apparent conflict between reversible and irreversible descriptions by identifying their different path laws and information levels. It does not derive the cosmological low-entropy initial condition, often called the past hypothesis.
quantum theory, time reversal is represented antiunitarily, and the weak interaction permits fundamental CP and T violation while remaining compatible with CPT under its standard hypotheses. Fundamental T-odd couplings, violation of stochastic detailed balance, and thermodynamic entropy production are therefore different statements. When the microscopic dynamics is T-odd, the reversed experiment uses the conjugate time-reversed dynamics and reversed external protocol; it is not obtained by running the same coefficients backward. CKM or PMNS phases alone do not derive the second law or a low-entropy cosmological boundary condition.
4.11. Four complementary meanings of a time arrow. The distinction between reversible microscopic equations and irreversible macroscopic behavior becomes clearer when one specifies the object that is being reversed. An equation, a probability law, a Markov evolution, and a recorded history are different objects. The four descriptions below connect these objects without treating them as equivalent. We adapt the organization of the autonomous isothermal analysis in [100]; its information-theoretic arguments apply to the regulated process independently of a thermodynamic interpretation.
Fix a regulator and let P t P_t P t be its conservative Markov semigroup with invariant probability μ \mu μ . Write
ν t = ν 0 P t , f t = d ν t d μ , H μ ( t ) = D K L ( ν t ∥ μ ) . (154) \nu_t = \nu_0 P_t,\qquad f_t = \frac{d\nu_t}{d\mu},\qquad H_\mu(t) = D_{\mathrm{KL}}(\nu_t\Vert\mu). \tag*{(154)} ν t = ν 0 P t , f t = d μ d ν t , H μ ( t ) = D KL ( ν t ∥ μ ) . (154)
The entropy is dimensionless. If the reference is a physically calibrated Gibbs equilibrium, with energy scale θ = k B T b a t h \theta= k_{\mathrm{B}}T_{\mathrm{bath}} θ = k B T bath , the excess free energy is θ H μ \theta H_\mu θ H μ . Invariance alone does not make μ \mu μ a thermal equilibrium: a driven stationary law can also be invariant.
4.11.1. Generator structure and physical reversal. Let Θ \Theta Θ be the declared state-space parity involution and assume Θ # μ = μ \Theta_\#\mu= \mu Θ # μ = μ . Its action on observables is U Θ h = h ∘ Θ U_\Theta h = h \circ\Theta U Θ h = h ∘ Θ , so U Θ 2 = I U_\Theta^2 = I U Θ 2 = I . For fixed external parameters that are unchanged by physical reversal, generalized detailed balance takes the form
D Θ : = L † , μ − U Θ L U Θ = 0. (155) \mathcal{D}_\Theta:= \mathcal{L}^{\dagger,\mu} - U_\Theta\mathcal{L} U_\Theta= 0. \tag*{(155)} D Θ := L † , μ − U Θ L U Θ = 0. (155)
The equality is understood on a common invariant core and extended to the semigroups under their well-posedness hypotheses. External magnetic fields or other time-odd parameters must instead be reversed in the comparison generator. For the underdamped equilibrium decomposition L = A + S \mathcal{L} = A + S L = A + S , one can have
A † , μ = − A , U Θ A U Θ = − A , S † , μ = S , U Θ S U Θ = S . (156) A^{\dagger,\mu} = -A,\quad U_\Theta A U_\Theta= -A,\qquad S^{\dagger,\mu} = S,\quad U_\Theta S U_\Theta= S. \tag*{(156)} A † , μ = − A , U Θ A U Θ = − A , S † , μ = S , U Θ S U Θ = S . (156)
Then L † , μ = S − A = U Θ L U Θ \mathcal{L}^{\dagger,\mu} = S - A = U_\Theta\mathcal{L} U_\Theta L † , μ = S − A = U Θ L U Θ . Friction, noise, and Hamiltonian transport can therefore all be nonzero in a statistically reversible stationary process. Positivity of the dissipative Dirichlet form and violation of stationary physical reversal are not the same assertion.
Information loss and the absence of a Markov inverse
For every invariant-measure-preserving Markov kernel, data processing gives
D K L ( ν P t ∥ μ ) ≤ D K L ( ν ∥ μ ) . (157) D_{\mathrm{KL}}(\nu P_t\Vert\mu) \le D_{\mathrm{KL}}(\nu\Vert\mu). \tag*{(157)} D KL ( ν P t ∥ μ ) ≤ D KL ( ν ∥ μ ) . (157)
Suppose the inequality is strict for a specified ν \nu ν with finite relative entropy. No Markov kernel Q Q Q can simultaneously satisfy
( ν P t ) Q = ν , μ Q = μ . (158) (\nu P_t)Q=\nu,\qquad\mu Q=\mu. \tag*{(158)} ( ν P t ) Q = ν , μ Q = μ . (158)
Indeed, data processing under Q Q Q would imply D K L ( ν ∥ μ ) ≤ D K L ( ν P t ∥ μ ) D_{\mathrm{KL}}(\nu\Vert\mu) \le D_{\mathrm{KL}}(\nu P_t\Vert\mu) D KL ( ν ∥ μ ) ≤ D KL ( ν P t ∥ μ ) , contradicting strictness. This is an obstruction to positive probability recovery with the same reference law. It does not forbid constructing a reversed diffusion, solving a backward equation on a smaller domain, or introducing a recovery operation that uses additional information or changes the reference law. The physical reverse of a diffusion is not generally its inverse semigroup.
Ordering, convergence, and nonrecurrence of laws
Contraction orders the evolving distributions by H μ H_\mu H μ . To establish attraction to μ \mu μ , rather than mere nonincrease, suppose that a diffusion entropy chain rule and a coercive estimate give
d d t H μ ( t ) = − I μ ( f t ) , I μ ( f ) ≥ 2 λ H μ ( f μ ) , λ > 0. (159) \frac{d}{dt}H_\mu(t)=-\mathcal{I}_\mu(f_t),\qquad\mathcal{I}_\mu(f)\ge2\lambda H_\mu(f\mu),\qquad\lambda>0. \tag*{(159)} d t d H μ ( t ) = − I μ ( f t ) , I μ ( f ) ≥ 2 λ H μ ( f μ ) , λ > 0. (159)
Here H μ ( f μ ) = D K L ( f μ ∥ μ ) H_\mu(f\mu)=D_{\mathrm{KL}}(f\mu\Vert\mu) H μ ( f μ ) = D KL ( f μ ∥ μ ) , and λ \lambda λ has inverse-time units in the declared clock. For instance, the second inequality follows from an appropriately normalized logarithmic Sobolev inequality. Integration yields
H μ ( t ) ≤ e − 2 λ ( t − s ) H μ ( s ) , t ≥ s . (160) H_\mu(t)\le e^{-2\lambda(t-s)}H_\mu(s),\qquad t\ge s. \tag*{(160)} H μ ( t ) ≤ e − 2 λ ( t − s ) H μ ( s ) , t ≥ s . (160)
If ν t = ν s \nu_t=\nu_s ν t = ν s for t > s t>s t > s and H μ ( s ) < ∞ H_\mu(s)<\infty H μ ( s ) < ∞ , this inequality forces H μ ( s ) = 0 H_\mu(s)=0 H μ ( s ) = 0 and hence ν s = μ \nu_s=\mu ν s = μ . Thus a nonequilibrium probability law cannot exactly recur under this strict contraction.
This statement does not exclude return of an individual trajectory to a previously visited region. Poincare recurrence concerns a measure-preserving transformation on a finite invariant measure space, not equality of an evolving nonequilibrium ensemble with its earlier law. Its hypotheses do not require finite dimensionality. Nor does a Gibbs minimizer by itself prove (160). In kinetic models, noise acts only in momentum directions; quantitative convergence may require hypocoercivity rather than an elliptic estimate. Convergence of laws also does not imply that fluctuating sample paths settle at a single deterministic field configuration.
Distinguishability of paths and relaxation of histories.
For a finite window of duration T T T , retain the declared physical reverse experiment and its endpoint convention. Its law, pulled back by path reversal to the forward coordinates, is denoted by P ~ R , T \widetilde{\mathbb{P}}_{\mathrm{R},T} P R , T . Define
Σ T = D K L ( P F , T ∥ P ~ R , T ) . (161) \Sigma_T = D_{\mathrm{KL}}\left(\mathbb{P}_{\mathrm{F},T}\mathbin{\|}\widetilde{\mathbb{P}}_{\mathrm{R},T}\right). \tag*{(161)} Σ T = D KL ( P F , T ∥ P R , T ) . (161)
This measures the information distinguishing forward and reversed movies. At a stationary nonequilibrium law, H μ ( t ) = 0 H_\mu(t)=0 H μ ( t ) = 0 when the system starts from μ \mu μ , while Σ T / T \Sigma_T/T Σ T / T can be strictly positive. In particular, neither Σ T \Sigma_T Σ T nor its rate is generally the decrease of H μ H_\mu H μ . Physical entropy units require multiplication by k B k_B k B and the local-detailed-balance interface of § 4.10 .
A history law introduces a third quantity. For a moving window Y t ( s ) = X t + s Y_t(s)=X_{t+s} Y t ( s ) = X t + s , − h ≤ s ≤ 0 -h\le s\le0 − h ≤ s ≤ 0 , let η t \eta_t η t be its law and η s s \eta_{\mathrm{ss}} η ss a stationary history law. A history relaxation functional would be
H h i s t ( t ) = D K L ( η t ∥ η s s ) . (162) H_{\mathrm{hist}}(t)=D_{\mathrm{KL}}\left(\eta_t\mathbin{\|}\eta_{\mathrm{ss}}\right). \tag*{(162)} H hist ( t ) = D KL ( η t ∥ η ss ) . (162)
It must be analyzed with the actual shift-and-boundary-update dynamics of the window, or a valid augmented state when the projected process has memory. A symmetric sampler of a chosen path-space Gibbs density is not automatically that history process. A stationary circulating process has η t = η s s \eta_t=\eta_{\mathrm{ss}} η t = η ss but can still have positive Σ T / T \Sigma_T/T Σ T / T . History relaxation and history reversal therefore remain different questions.
Table 1. Four complementary descriptions of time direction. None is defined by the vanishing of a single drift or diffusion coefficient.
For the unified programme, these distinctions separate three uses of equilibrium: selecting a background, proving attraction of a regulated ensemble, and testing detailed balance of its stationary histories. The HJB value function selects an admissible feedback for a specified cost; it is not interchangeable with any of these entropy functionals. A sustained turbulent state belongs to the driven branch: stationary statistics need not eliminate injection, dissipation, or path asymmetry. The detailed proofs, endpoint bookkeeping, and exactly solvable comparisons are developed in the thermodynamics companion paper. This classification does not determine a cosmological low-entropy initial condition or prove global convergence of gravitational collapse.
A common fast–slow formulation
A single equation can contain both regimes through a scale parameter ε \varepsilon ε . The following display is schematic until one specifies domains, boundary conditions, and a finite-dimensional or regularized realization in which M Q M_Q M Q and M Z M_Z M Z are positive self-adjoint mobility operators with sufficiently regular square roots. If either mobility depends on the state, the corresponding reference-volume divergence correction from (80) is also required:
d Q t ε = − M Q ∇ Q Φ ( Q t ε , Z t ε ) d t + 2 θ M Q d B t , d Z t ε = − ε − 1 M Z ∇ Z C ( Q t ε , Z t ε ) d t + 2 θ ε − 1 M Z d W t . (163) \begin{aligned}
\mathrm{d}Q_t^\varepsilon&= -M_Q\nabla_Q\Phi(Q_t^\varepsilon,Z_t^\varepsilon)\,\mathrm{d}t+\sqrt{2\theta M_Q}\,\mathrm{d}B_t,\\
\mathrm{d}Z_t^\varepsilon&= -\varepsilon^{-1}M_Z\nabla_Z C(Q_t^\varepsilon,Z_t^\varepsilon)\,\mathrm{d}t+\sqrt{2\theta\varepsilon^{-1}M_Z}\,\mathrm{d}W_t. \tag*{(163)}
\end{aligned} d Q t ε d Z t ε = − M Q ∇ Q Φ ( Q t ε , Z t ε ) d t + 2 θ M Q d B t , = − ε − 1 M Z ∇ Z C ( Q t ε , Z t ε ) d t + 2 θ ε − 1 M Z d W t . (163)
For these equations to use one potential, impose ∇ Z C = ∇ Z Φ \nabla_Z C=\nabla_Z\Phi ∇ Z C = ∇ Z Φ , equivalently C ( Q , Z ) = Φ ( Q , Z ) + c ( Q ) C(Q,Z)=\Phi(Q,Z)+c(Q) C ( Q , Z ) = Φ ( Q , Z ) + c ( Q ) on each connected fast fiber. Otherwise the display describes two different potentials and need not have a joint reversible equilibrium. Under this compatibility condition, a finite conditional partition function gives
Z ( Q ) = ∫ e − Φ ( Q , z ) / θ ν Z ( d z ) , F s l o w ( Q ) = − θ log Z ( Q ) , ∇ Q F s l o w ( Q ) = ∫ ∇ Q Φ ( Q , z ) π ( d z ∣ Q ) . (164) \begin{aligned}
Z(Q) &= \int e^{-\Phi(Q,z)/\theta}\,\nu_Z(\mathrm{d}z),\qquad F_{\mathrm{slow}}(Q)=-\theta\log Z(Q),\\
\nabla_QF_{\mathrm{slow}}(Q) &= \int\nabla_Q\Phi(Q,z)\,\pi(\mathrm{d}z\mid Q). \tag*{(164)}
\end{aligned} Z ( Q ) ∇ Q F slow ( Q ) = ∫ e − Φ ( Q , z ) / θ ν Z ( d z ) , F slow ( Q ) = − θ log Z ( Q ) , = ∫ ∇ Q Φ ( Q , z ) π ( d z ∣ Q ) . (164)
Here ν Z \nu_Z ν Z is independent of Q Q Q , differentiation under the integral is assumed justified, and π ( d z ∣ Q ) = Z ( Q ) − 1 e − Φ ( Q , z ) / θ ν Z ( d z ) \pi(\mathrm{d}z\mid Q)=Z(Q)^{-1}e^{-\Phi(Q,z)/\theta}\nu_Z(\mathrm{d}z) π ( d z ∣ Q ) = Z ( Q ) − 1 e − Φ ( Q , z ) / θ ν Z ( d z ) . The last identity follows by differentiating Z ( Q ) Z(Q) Z ( Q ) and dividing by − Z ( Q ) / θ -Z(Q)/\theta − Z ( Q ) / θ . For constant slow mobility, averaging thus produces the drift − M Q ∇ Q F s l o w -M_Q\nabla_QF_{\mathrm{slow}} − M Q ∇ Q F slow ; its slow noise remains 2 θ M Q d B t \sqrt{2\theta M_Q}\,\mathrm{d}B_t 2 θ M Q d B t . Taking ε → 0 \varepsilon\to0 ε → 0 alone does not eliminate that noise. A deterministic macroscopic limit additionally uses the collective or small-noise scaling specified earlier. Convergence to the averaged equation requires uniform mixing of the fast dynamics, tightness, and a justified averaging theorem; time-scale separation alone does not supply these estimates.
Symmetry breaking and the emergence of four sectors
Residual symmetry
Let φ ∗ \varphi_* φ ∗ denote a chosen vacuum value of the fields that break the local symmetry. In a local trivialization its stabilizer in the finite-dimensional isotropy group H H H is
H φ ∗ : = { h ∈ H : h ⋅ φ ∗ = φ ∗ } . (165) H_{\varphi_*}:=\{h\in H:h\cdot\varphi_*=\varphi_*\}. \tag*{(165)} H φ ∗ := { h ∈ H : h ⋅ φ ∗ = φ ∗ } . (165)
The symbol φ ∗ \varphi_* φ ∗ is distinct from the real gauge-singlet scalar σ \sigma σ used in § 5.4 ; that singlet alone cannot break the internal gauge group. Replacing φ ∗ \varphi_* φ ∗ by another representative conjugates H φ ∗ H_{\varphi_*} H φ ∗ , so the residual subgroup is defined intrinsically up to conjugacy. Globally, a vacuum section of constant orbit type defines a reduction to this subgroup when such a reduction exists. The stabilizer of the whole section inside the gauge-transformation group G \mathcal{G} G is generally infinite dimensional; it must not be identified with the finite-dimensional fiber group H φ ∗ H_{\varphi_*} H φ ∗ . The low-energy gauge group is therefore not assumed to be fundamental. It is the stabilizer of a selected phase, provided the selection problem has a solution. In a gauge-invariant finite-volume measure, a local gauge-noninvariant expectation value is zero by Elitzur’s theorem [64]. Accordingly, φ ∗ \varphi_* φ ∗ and the displayed brackets below denote either a representative on a declared gauge-fixed slice or an orbit-level order parameter. The physical statements are the stabilizer of the orbit, gauge-invariant composites such as φ ∗ † φ ∗ \varphi_*^\dagger\varphi_* φ ∗ † φ ∗ , and the covariant-derivative mass matrix. A literal local gauge-symmetry breaking expectation value requires a separately specified gauge fixing and, where relevant, an infinite-volume limit.
If the selected phase admits a reductive complement, the corresponding Lie algebra decomposition is
h = h φ ∗ ⊕ h b r o k e n (166) \mathfrak{h} = \mathfrak{h}_{\varphi_*} \oplus\mathfrak{h}_{\mathrm{broken}} \tag*{(166)} h = h φ ∗ ⊕ h broken (166)
where h φ ∗ = Lie ( H φ ∗ ) \mathfrak{h}_{\varphi_*} = \operatorname{Lie}(H_{\varphi_*}) h φ ∗ = Lie ( H φ ∗ ) , and this decomposition is available only after the vacuum has been selected. The broken directions acquire masses or decouple only if the Hessian of the specified vacuum and kinetic terms has the required positive gap. The translational Cartan quotient g / h \mathfrak{g}/\mathfrak{h} g / h supplies the coframe; it is not a list of massive internal gauge bosons.
Candidate Standard-Model phase
A candidate residual internal symmetry is
G S M = S U ( 3 ) c × S U ( 2 ) L × U ( 1 ) Y . (167) G_{\mathrm{SM}} = SU(3)_{\mathrm{c}} \times SU(2)_{\mathrm{L}} \times U(1)_{\mathrm{Y}}. \tag*{(167)} G SM = S U ( 3 ) c × S U ( 2 ) L × U ( 1 ) Y . (167)
Equation (168) is a candidate output of the abstract selection problem and a prescribed input of the worked model in § 5.4 . A complete derivation would have to show that it solves the selection problem introduced below and that all competing admissible phases have larger master cost or fail consistency conditions.
After a scalar order parameter Φ H \Phi_H Φ H is selected in a representation that, at low energy, is equivalent to the usual electroweak doublet, the residual breaking is
S U ( 2 ) L × U ( 1 ) Y ⟶ U ( 1 ) e m . (168) SU(2)_{\mathrm{L}} \times U(1)_{\mathrm{Y}} \longrightarrow U(1)_{\mathrm{em}}. \tag*{(168)} S U ( 2 ) L × U ( 1 ) Y ⟶ U ( 1 ) em . (168)
With the conventional normalization,
Q = T 3 + Y . (169) Q = T_3 + Y. \tag*{(169)} Q = T 3 + Y . (169)
Here Y Y Y is the Standard Model hypercharge convention used in (235) and in the anomaly sums below. A convention using Y / 2 Y/2 Y /2 instead would require rescaling every displayed hypercharge and the associated U ( 1 ) U(1) U ( 1 ) coupling; no such rescaling is made in this paper. The representation and normalization must ultimately be derived from the master selection problem if the theory is to claim prediction rather than reconstruction.
Geometric and gauge recovery from one Euler system
Gravity as a low-energy projection
Before reducing the curvature blocks, write the connection component of the master Euler–Lagrange equation in constitutive form. Let ι h : h ↪ g \iota_h:\mathfrak{h}\hookrightarrow\mathfrak{g} ι h : h ↪ g be the inclusion of the isotropy algebra and let ι h ∗ : g ∗ → h ∗ \iota_h^*:\mathfrak{g}^*\to\mathfrak{h}^* ι h ∗ : g ∗ → h ∗ be its dual restriction. At fixed coframe an admissible connection variation has the form δ A = ι h ( δ a ) \delta A=\iota_h(\delta a) δ A = ι h ( δ a ) , with δ a ∈ Ω 1 ( M , ad P H ) \delta a\in\Omega^1(M,\operatorname{ad}P_H) δ a ∈ Ω 1 ( M , ad P H ) ; the translational component ℓ − 1 δ e \ell^{-1}\delta e ℓ − 1 δ e is not varied in this calculation. If L M \mathcal{L}_{\mathrm{M}} L M is the local master density, set the differential-form convention, at fixed coframe and fixed constitutive tensor,
δ a S M = ∫ M ⟨ P ∧ D A ι h ( δ a ) ⟩ + ∫ M ⟨ J e x t ∧ ι h ( δ a ) ⟩ , P ∈ Ω 2 ( M , ad ∗ P ) , J e x t ∈ Ω 3 ( M , ad ∗ P ) . (170) \delta_a S_{\mathrm{M}}=\int_M\langle\mathbb{P}\wedge\mathbb{D}_A\iota_h(\delta a)\rangle+\int_M\langle\mathbb{J}_{\mathrm{ext}}\wedge\iota_h(\delta a)\rangle,
\qquad
\mathbb{P}\in\Omega^2(M,\operatorname{ad}^*P),\quad
\mathbb{J}_{\mathrm{ext}}\in\Omega^3(M,\operatorname{ad}^*P). \tag*{(170)} δ a S M = ∫ M ⟨ P ∧ D A ι h ( δ a )⟩ + ∫ M ⟨ J ext ∧ ι h ( δ a )⟩ , P ∈ Ω 2 ( M , ad ∗ P ) , J ext ∈ Ω 3 ( M , ad ∗ P ) . (170)
The brackets in this formula pair the adjoint and coadjoint bundles; no identification between them is assumed. The boundary term is assumed to vanish. Since deg P = 2 \deg\mathbb{P}=2 deg P = 2 , covariant integration by parts gives P ∧ D A δ A = D A ( P ∧ δ A ) − D A P ∧ δ A \mathbb{P}\wedge\mathbb{D}_A\delta A=\mathbb{D}_A(\mathbb{P}\wedge\delta A)-\mathbb{D}_A\mathbb{P}\wedge\delta A P ∧ D A δ A = D A ( P ∧ δ A ) − D A P ∧ δ A . Stationarity therefore fixes the sign convention below. When L M \mathcal{L}_{\mathrm{M}} L M is written as a local four-form, the notation ∂ L M / ∂ F \partial\mathcal{L}_{\mathrm{M}}/\partial F ∂ L M / ∂ F means precisely this fiber derivative, including the Hodge and reference-density factors; it is not a derivative of a scalar with an untyped form degree. Thus
P : = ∂ L M ∂ F , ι h ∗ ( D P − J e x t ) = 0. (171) \mathbb{P}:=\frac{\partial\mathcal{L}_{\mathrm{M}}}{\partial F},
\qquad
\iota_h^*(\mathbb{D}\mathbb{P}-\mathbb{J}_{\mathrm{ext}})=0. \tag*{(171)} P := ∂ F ∂ L M , ι h ∗ ( DP − J ext ) = 0. (171)
Here D \mathbb{D} D acts on coadjoint-valued forms. The second relation is exactly the fixed-coframe connection equation: testing only against δ a \delta a δ a determines only the restriction to h ∗ \mathfrak{h}^* h ∗ . It does not assert that the translational component of D P − J e x t \mathbb{D}\mathbb{P}-\mathbb{J}_{\mathrm{ext}} DP − J ext vanishes. The coframe equation also varies the Hodge operator, the volume form, and every other dependence on e e e ; matter and order-parameter variations supply further equations. Let π ω \pi_\omega π ω and π i n t \pi_{\mathrm{int}} π int denote the projections from h ∗ \mathfrak{h}^* h ∗ onto the rotational and residual internal dual blocks of a selected gauge-fixed splitting. If the reduced covariant derivative preserves these blocks and the residual constitutive pairing has no off-diagonal term, equation (171) gives
π ω ι h ∗ ( D P ) = π ω ι h ∗ ( J e x t ) , π i n t ι h ∗ ( D P ) = π i n t ι h ∗ ( J e x t ) . (172) \pi_\omega\iota_h^*(\mathbb{D}\mathbb{P})=\pi_\omega\iota_h^*(\mathbb{J}_{\mathrm{ext}}),
\qquad
\pi_{\mathrm{int}}\iota_h^*(\mathbb{D}\mathbb{P})=\pi_{\mathrm{int}}\iota_h^*(\mathbb{J}_{\mathrm{ext}}). \tag*{(172)} π ω ι h ∗ ( DP ) = π ω ι h ∗ ( J ext ) , π int ι h ∗ ( DP ) = π int ι h ∗ ( J ext ) . (172)
These are two components of the same fixed-coframe variational equation, not two primitive actions. They must, however, be distinguished from the coframe equation. Let the full variational derivative of the master action, including all matter and order-parameter terms, be denoted by
E M : = ( E e , E ω , E A , E Φ , E σ , E Ψ , E χ , … ) = 0 , E e : = δ S M δ e , E ω : = δ S M δ ω , \mathcal{E}_{\mathrm{M}} := (\mathcal{E}_e,\mathcal{E}_\omega,\mathcal{E}_A,\mathcal{E}_\Phi,\mathcal{E}_\sigma,\mathcal{E}_\Psi,\mathcal{E}_\chi,\ldots)=0,
\qquad
\mathcal{E}_e := \frac{\delta S_{\mathrm{M}}}{\delta e},\quad
\mathcal{E}_\omega:= \frac{\delta S_{\mathrm{M}}}{\delta\omega}, E M := ( E e , E ω , E A , E Φ , E σ , E Ψ , E χ , … ) = 0 , E e := δ e δ S M , E ω := δ ω δ S M ,
E A : = δ S M δ A , E Φ : = δ S M δ Φ , E σ : = δ S M δ σ , E Ψ : = δ S M δ Ψ , \mathcal{E}_A := \frac{\delta S_{\mathrm{M}}}{\delta A},\quad
\mathcal{E}_\Phi:= \frac{\delta S_{\mathrm{M}}}{\delta\Phi},
\qquad
\mathcal{E}_\sigma:= \frac{\delta S_{\mathrm{M}}}{\delta\sigma},\quad
\mathcal{E}_\Psi:= \frac{\delta S_{\mathrm{M}}}{\delta\Psi}, E A := δ A δ S M , E Φ := δ Φ δ S M , E σ := δ σ δ S M , E Ψ := δ Ψ δ S M ,
E χ : = δ S M δ χ . (173) \mathcal{E}_\chi:= \frac{\delta S_{\mathrm{M}}}{\delta\chi}. \tag*{(173)} E χ := δ χ δ S M . (173)
Here A A A is the residual internal connection, and the omitted entries are any additional explicitly dynamical auxiliary-field equations. The reference information parameter is fixed in this formulation. The fixed-coframe connection component of (173) is E a : = ι h ∗ ( D P − J e x t ) = 0 \mathcal{E}_a:=\iota^*_h(\mathbb{D}P-J_{\mathrm{ext}})=0 E a := ι h ∗ ( D P − J ext ) = 0 . In a Cartan decomposition, its rotational part is the ω \omega ω equation, whereas the Einstein tensor comes from the independent coframe (or metric) equation E e = 0 \mathcal{E}_e=0 E e = 0 . This distinction is essential because the Hodge map and the volume form in the master pairing may themselves depend on e e e .
The precise sector-recovery statement used below is the following.
Theorem 9 (Controlled sector recovery from one unified Euler system).
Assume that, after gauge fixing or passage to the physical quotient, the selected phase has a light-sector splitting into a gravitational Cartan block and a residual internal block, with the fixed-coframe rotational and internal projections π ω \pi_\omega π ω and π i n t \pi_{\mathrm{int}} π int defined above. Assume that the constitutive tensor is block compatible, that the gravitational block contains an isolated Palatini term with a nonzero, spatially constant coefficient κ P \kappa_{\mathrm{P}} κ P , and that the coframe is non-degenerate. Here “isolated” means that, at the retained effective order, E ω \mathcal{E}_\omega E ω receives no unsuppressed curvature-square, torsion-square, or mixed contribution. The exact Palatini implication below uses zero connection remainder. More generally, if heavy connection modes have been integrated out, assume on specified normed spaces that the retained spinless equation has the form 0 = L e T + R ω 0=\mathcal{L}_e T+R_\omega 0 = L e T + R ω , where L e \mathcal{L}_e L e is invertible, ∥ L e − 1 ∥ ≤ C e \lVert\mathcal{L}_e^{-1}\rVert\leq C_e ∥ L e − 1 ∥ ≤ C e , and ∥ R ω ∥ ≤ δ ω \lVert R_\omega\rVert\leq\delta_\omega ∥ R ω ∥ ≤ δ ω . Assume that the matter spin current in the ω \omega ω equation vanishes (or is negligible at the matching scale), J ω s p i n = 0 J^{\mathrm{spin}}_\omega=0 J ω spin = 0 ; otherwise the same calculation gives the Einstein–Cartan branch with torsion sourced by spin. Assume also that the internal block has a positive A d Ad A d -invariant form κ ^ A B \widehat{\kappa}_{AB} κ A B compatible with the Hodge operator. The hat records that this form may depend on the selected scalar and coframe; on a constant scalar/metric branch it is denoted simply by κ A B \kappa_{AB} κ A B . Assume further that, after the stated reduction and continuation, the retained coframe-dependent Lorentzian action has the
effective basis displayed in (207) , with every other retained coframe-dependent term included in S n o n g r a v , L S_{\mathrm{nongrav,L}} S nongrav , L rather than silently discarded. This is the action induced after the retained connection equation has been solved, or after the indicated connection modes have been integrated out consistently; it is not obtained by restricting an unreduced independent-connection action before varying it. Assume moreover that every fixed information, reference, regulator, and matching datum entering the coframe variation is a diffeomorphism-covariant spurion and either belongs to the varied on-shell system or satisfies the associated Noether balance law. Then the single system (173) has the following three linked low-energy consequences, corresponding to the connection, coframe, and residual internal-connection components:
E ω = 0 in the Palatini truncation ⟹ T a = 0. (174) \mathcal{E}_{\omega}=0\ \text{in the Palatini truncation}\Longrightarrow T^{a}=0. \tag*{(174)} E ω = 0 in the Palatini truncation ⟹ T a = 0. (174)
If the controlled connection remainder just described is retained instead, the same equation and the inverse bound give
∥ T ∥ ≤ ∥ L e − 1 ∥ ∥ R ω ∥ ≤ C e δ ω . (175) \lVert T\rVert\le\lVert\mathcal{L}_{e}^{-1}\rVert\lVert R_{\omega}\rVert\le C_{e}\delta_{\omega}. \tag*{(175)} ∥ T ∥ ≤ ∥ L e − 1 ∥ ∥ R ω ∥ ≤ C e δ ω . (175)
After imposing the torsionless branch and the stated continuation, the coframe equation E e = 0 \mathcal{E}_{e}=0 E e = 0 gives
G μ ν + Λ L g μ ν + 16 π G e f f ∑ i α ˉ i H ˉ μ ν ( i ) = 8 π G e f f T μ ν L . (176) G_{\mu\nu}+\Lambda_{\mathrm{L}}g_{\mu\nu}+16\pi G_{\mathrm{eff}}\sum_{i}\bar{\alpha}^{i}\bar{H}^{(i)}_{\mu\nu}=8\pi G_{\mathrm{eff}}T^{\mathrm{L}}_{\mu\nu}. \tag*{(176)} G μν + Λ L g μν + 16 π G eff i ∑ α ˉ i H ˉ μν ( i ) = 8 π G eff T μν L . (176)
and
π i n t E a = 0 ⟺ D ( κ ^ A B ⋆ F B ) = − ⋆ J A , ⟺ D μ ( κ ^ A B F B μ ν ) = J A ν . (177) \pi_{\mathrm{int}}\mathcal{E}_{a}=0\Longleftrightarrow D\left(\widehat{\kappa}_{AB}\star F^{B}\right)=-\star J_{A},\Longleftrightarrow D_{\mu}\left(\widehat{\kappa}_{AB}F^{B\mu\nu}\right)=J_{A}^{\nu}. \tag*{(177)} π int E a = 0 ⟺ D ( κ A B ⋆ F B ) = − ⋆ J A , ⟺ D μ ( κ A B F B μν ) = J A ν . (177)
On a frozen constant-kinetic branch, canonical normalization reduces (177) to the textbook form (318) . The Noether hypothesis ensures that the stress tensor in (176) has no unaccounted external exchange term. If it is not satisfied, a fixed non-on-shell datum can spoil covariant conservation and the coframe equation must include the corresponding exchange contribution. The ordinary Einstein equation is the controlled regime in which the last higher-curvature term in (176) is negligible. Thus the two familiar equations are separate projections of one full variational system: Einstein comes from its coframe/metric component after the connection equation selects the torsionless branch, whereas Yang–Mills comes from the residual internal-connection component. The hypotheses specify the phase, constitutive structure, and retained scale; they do not assume either field equation. The two equations are obtained by varying the same master action, so the implication is conditional but not circular. In particular, a spatially varying coefficient multiplying the Palatini term is not covered by the torsionless implication. For the joint candidate below this coefficient is proportional to w ( σ ) w(\sigma) w ( σ ) , so a constant scalar branch or a controlled scalar-gradient approximation is required before applying the displayed Einstein reduction. The exact implication E ω = 0 ⇒ T = 0 \mathcal{E}_{\omega}=0 \Rightarrow T=0 E ω = 0 ⇒ T = 0 is therefore a statement about the retained constant-coefficient Palatini truncation; with a nonzero controlled connection remainder, the full theory gives the quantitative torsion estimate (176). It is not an identity for the unreduced SO(5) curvature-square witness, whose independent connection equation contains additional D ω ⋆ R D_{\omega}\star R D ω ⋆ R and torsion terms and therefore represents the more general higher-derivative theory before the controlled sector is taken.
Proof. The equality E a = 0 \mathcal{E}_{a}=0 E a = 0 is exactly (172). Block compatibility gives (173). On the isolated Palatini branch, the variation with respect to ω \omega ω is computed in (194); its algebraic inversion is given in (195)–(196), and proves (175). On the controlled-remainder branch, invertibility gives T = − L e − 1 R ω T=-\mathcal{L}_{e}^{-1}\mathcal{R}_{\omega} T = − L e − 1 R ω ; taking the declared norms proves (176). With T a = 0 T^{a}=0 T a = 0 , the independent coframe variation is the metric variation of the induced effective action. Equations (205)–(210), using the retained effective-action hypothesis, then give (177), including the stated Euclidean-to-Lorentzian continuation. Finally, P H A = κ ^ A B ⋆ F B \mathsf{P}_{H}^{A}=\hat{\kappa}_{AB}\star F^{B} P H A = κ ^ A B ⋆ F B by Hodge compatibility, so the internal component of E a = 0 \mathcal{E}_{a}=0 E a = 0 is the first equality in (178); the coordinate calculation in (314)–(318) proves the last equality. This also shows why a field-dependent or mixed constitutive tensor would leave the more general restricted constitutive equation ι ξ ∗ [ D ( ∂ L / ∂ F ) − J e x t ] = 0 \iota_{\xi}^{*}[D(\partial\mathcal{L}/\partial F)-J_{\mathrm{ext}}]=0 ι ξ ∗ [ D ( ∂ L / ∂ F ) − J ext ] = 0 instead. □
The word “controlled” records the domain of the theorem, not an unfinished step in its proof. Once its hypotheses hold, the displayed equations follow from the variational calculation and no classification of all global solutions or global regularity theorem for the full gauge–gravity evolution is needed. The light-sector splitting, constitutive form, continuation, and scale separation are sector and effective-matching hypotheses; non-degeneracy, admissibility of the variations, and the Noether balance are mathematical consistency hypotheses. Showing that a particular ultraviolet completion selects this domain uniquely would be a stronger result about model selection and predictivity, not a missing step in the conditional recovery theorem.
The next two subsections provide the detailed gravitational and internal calculations used in this theorem. They are projections of the same master Euler system, not independent sector actions.
Suppose the selected phase suppresses internal components at a length scale L L L , leaves a gravitational Cartan connection ( e , ω ) (e,\omega) ( e , ω ) light, and selects an invariant pairing/order-parameter insertion whose curvature cross-term is of Palatini type. Concretely, this means that the reduced density contains, for some coefficient κ \kappa κ fixed by the microscopic pairing,
⟨ F M e x t , F M e x t ⟩ h vol g ⊃ κ ϵ a b c d e a ∧ e b ∧ R c d . (178) \left\langle\mathbf{F}^{\mathrm{ext}}_{M},\mathbf{F}^{\mathrm{ext}}_{M}\right\rangle_{h}\,\operatorname{vol}_{g}\supset\kappa\,\epsilon_{abcd}e^{a}\wedge e^{b}\wedge R^{cd}. \tag*{(178)} ⟨ F M ext , F M ext ⟩ h vol g ⊃ κ ϵ ab c d e a ∧ e b ∧ R c d . (178)
The left-hand side is a scalar density written as a four-form by multiplication with vol g \operatorname{vol}_g vol g . Thus the displayed inclusion is an equality of four-forms; equivalently, after division by the nowhere-vanishing volume form on the coframe chart, it reads
⟨ F M e x t , F M e x t ⟩ h ⊃ κ ϵ a b c d e e a ∧ e b ∧ R c d vol g . (179) \left\langle\mathbf{F}_{\mathrm{M}}^{\mathrm{ext}},\mathbf{F}_{\mathrm{M}}^{\mathrm{ext}}\right\rangle_{\mathbf{h}} \supset\kappa\frac{\epsilon_{abcde}e^a \wedge e^b \wedge R^{cd}}{\operatorname{vol}_g}. \tag*{(179)} ⟨ F M ext , F M ext ⟩ h ⊃ κ vol g ϵ ab c d e e a ∧ e b ∧ R c d . (179)
The quotient in (180) is the Hodge-type block of the extended pairing, not an untyped identification of a scalar with a differential form. This is an additional algebraic assumption; it does not follow from a generic positive curvature-square norm. One sufficient mechanism for the term is a symmetry-breaking J J J -sector curvature
F J a b = R a b + λ e a ∧ e b (180) F_J^{ab}=R^{ab}+\lambda e^a\wedge e^b \tag*{(180)} F J ab = R ab + λ e a ∧ e b (180)
and an induced contraction proportional to ϵ a b c d \epsilon_{abcd} ϵ ab c d . The parameter λ \lambda λ has the inverse-length-squared dimension required by the chosen coframe normalization. In this display R a b R^{ab} R ab denotes the curvature of the Lorentz connection after any C P a P b J c d e a ∧ e b C_{P_aP_b}J_{cd}e^a\wedge e^b C P a P b J c d e a ∧ e b contribution has been separated; if that contribution is retained in R a b R^{ab} R ab , its coefficient must instead be absorbed into λ \lambda λ . Such a tensor is required to be invariant under the residual subgroup used in the effective theory; it need not be invariant under the full microscopic group before symmetry breaking. Its algebraic expansion is
ϵ a b c d F J a b ∧ F J c d = ϵ a b c d R a b ∧ R c d + 2 λ ϵ a b c d e e a ∧ e b ∧ R c d + λ 2 ϵ a b c d e e a ∧ e b ∧ e c ∧ e d . (181) \epsilon_{abcd}F_J^{ab}\wedge F_J^{cd}
=\epsilon_{abcd}R^{ab}\wedge R^{cd}
+2\lambda\epsilon_{abcde}e^a\wedge e^b\wedge R^{cd}
+\lambda^2\epsilon_{abcde}e^a\wedge e^b\wedge e^c\wedge e^d. \tag*{(181)} ϵ ab c d F J ab ∧ F J c d = ϵ ab c d R ab ∧ R c d + 2 λ ϵ ab c d e e a ∧ e b ∧ R c d + λ 2 ϵ ab c d e e a ∧ e b ∧ e c ∧ e d . (181)
The overall factors depend on the normalization of J a b J_{ab} J ab and are absorbed into κ \kappa κ . Thus the Palatini term is an output of the specified symmetry-breaking insertion, not a consequence of an arbitrary positive inner product. The ϵ \epsilon ϵ -contraction is a signed residual invariant; it is compatible with the positive Euclidean master norm only if it arises from a cross-block or order-parameter insertion whose full graded pairing remains positive on the chosen real slice. Positivity of that full pairing must be checked in the concrete model. Under the stated assumption, the finite-dimensional analogue makes the check explicit: if the relevant curvature/order-parameter block of h \mathbf{h} h has matrix ( A B B ∗ C ) \begin{pmatrix} A & B \\ B^* & C \end{pmatrix} ( A B ∗ B C ) with A > 0 A>0 A > 0 , then nonnegativity requires the Schur complement
C − B ∗ A − 1 B ⪰ 0. (182) C-B^*A^{-1}B\succeq0. \tag*{(182)} C − B ∗ A − 1 B ⪰ 0. (182)
For a positive definite, non-degenerate pairing in finite dimensions the inequality must be strict. In a Hilbert realization, a uniformly positive norm additionally requires a coercive lower bound on this complement and boundedness of the operators used in the factorization. Thus a signed cross term can occur without making the full Euclidean pairing indefinite, but only when (183) (or its operator analogue) is verified. The reduced curvature is then
F ⇝ ℓ − 1 T a P a + 1 2 ( R a b + λ e a ∧ e b ) J a b . (183) \mathbb{F} \leadsto\ell^{-1}T^{a}P_{a}+\frac{1}{2}\left(R^{ab}+\lambda e^{a}\wedge e^{b}\right)J_{ab}. \tag*{(183)} F ⇝ ℓ − 1 T a P a + 2 1 ( R ab + λ e a ∧ e b ) J ab . (183)
If the selected vacuum imposes the torsionless branch,
T a = 0 , (184) T^{a}=0, \tag*{(184)} T a = 0 , (184)
then, for a non-degenerate coframe, ω \omega ω is identified with the Levi–Civita spin connection of that coframe.
The linear Einstein term requires specified curvature components and a specified pairing. The Palatini-type pairing above is one mechanism; Section 5.3 gives another using a coframe-dependent positive Hodge norm. With the convention
ϵ a b c d e e a ∧ e b ∧ R c d = 2 R vol g (185) \epsilon_{abcde}e^{a}\wedge e^{b}\wedge R^{cd}=2R\,\operatorname{vol}_{g} \tag*{(185)} ϵ ab c d e e a ∧ e b ∧ R c d = 2 R vol g (185)
up to convention-dependent normalization. In an oriented orthonormal coframe, write R c d = 1 2 R e f c d e e ∧ e f R^{cd}=\frac{1}{2}R^{cd}_{\ \ ef}e^{e}\wedge e^{f} R c d = 2 1 R e f c d e e ∧ e f . Then
ϵ a b c d e e a ∧ e b ∧ R c d = 1 2 ϵ a b c d e R e f c d e a ∧ e b ∧ e e ∧ e f = 1 2 ϵ a b c d e ϵ a b e f R e f c d vol g = ( δ c e δ d f − δ c f δ d e ) R e f c d vol g = 2 R vol g , (186) \begin{aligned}
\epsilon_{abcde}e^{a}\wedge e^{b}\wedge R^{cd}
&=\frac{1}{2}\epsilon_{abcde}R^{cd}_{\ \ ef}e^{a}\wedge e^{b}\wedge e^{e}\wedge e^{f}\\
&=\frac{1}{2}\epsilon_{abcde}\epsilon^{abef}R^{cd}_{\ \ ef}\operatorname{vol}_{g}\\
&=\left(\delta^{e}_{c}\delta^{f}_{d}-\delta^{f}_{c}\delta^{e}_{d}\right)R^{cd}_{\ \ ef}\operatorname{vol}_{g}=2R\,\operatorname{vol}_{g}, \tag*{(186)}
\end{aligned} ϵ ab c d e e a ∧ e b ∧ R c d = 2 1 ϵ ab c d e R e f c d e a ∧ e b ∧ e e ∧ e f = 2 1 ϵ ab c d e ϵ ab e f R e f c d vol g = ( δ c e δ d f − δ c f δ d e ) R e f c d vol g = 2 R vol g , (186)
where ϵ a b c d ϵ a b e f = 2 ( δ c e δ d f − δ c f δ d e ) \epsilon_{abcd}\epsilon^{abef}=2(\delta^{e}_{c}\delta^{f}_{d}-\delta^{f}_{c}\delta^{e}_{d}) ϵ ab c d ϵ ab e f = 2 ( δ c e δ d f − δ c f δ d e ) and the last equality uses the scalar-curvature convention R : = R c d c d R:=R^{cd}_{\ \ cd} R := R c d c d and the antisymmetry of R e f c d R^{cd}_{\ \ ef} R e f c d in e , f e,f e , f together with the chosen curvature convention. Reversing the orientation or the sign convention for R c d R^{cd} R c d reverses the displayed sign. The low-energy gravitational action therefore has the form
S g r a v , e f f = ∫ M g [ Λ e f f − R 16 π G e f f + α 1 R 2 + α 2 R μ ν R μ ν + α 3 R μ ν ρ σ R μ ν ρ σ + ⋯ ] d 4 x . (187) S_{\mathrm{grav,eff}}=\int_{M}\sqrt{g}\left[\Lambda_{\mathrm{eff}}-\frac{R}{16\pi G_{\mathrm{eff}}}+\alpha_{1}R^{2}+\alpha_{2}R_{\mu\nu}R^{\mu\nu}+\alpha_{3}R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma}+\cdots\right]\mathrm{d}^{4}x. \tag*{(187)} S grav , eff = ∫ M g [ Λ eff − 16 π G eff R + α 1 R 2 + α 2 R μν R μν + α 3 R μν ρ σ R μν ρ σ + ⋯ ] d 4 x . (187)
The coefficients are not predicted by geometry alone; they depend on the precise master algebra, measure, regulator, and vacuum. The Einstein term is therefore a structural consequence conditional on those data, whereas a numerical prediction for G e f f G_{\mathrm{eff}} G eff is a separate problem. The geometric normalization is to be compared with the classical gravitational field equation only after the convention for R a b R^{ab} R ab has been fixed [63]. The boundedness and positivity assumptions imposed on the regulated master functional should not be confused with boundedness of this truncated Einstein–higher-curvature expansion; the latter is an effective description and can require additional stabilizing terms.
The logical origin of the gravitational equation can now be stated without introducing a second action. The gravitational projection of the connection equation in (172) supplies the rotational connection equation and, on the isolated Palatini branch, the torsionless condition. The Einstein tensor is supplied by the independent coframe equation in (173) ; it is not obtained by applying a connection projection alone. Under the block-projection assumptions, the torsionless Palatini branch, and the specified coframe variation, the exact conditional chain is
E ω = 0 ⟹ T a = 0 , E e = 0 ⟹ G μ ν + Λ L g μ ν + 16 π G e f f ∑ i α ˉ i H ˉ μ ν ( i ) = 8 π G e f f T μ ν L (188) \begin{aligned}
\mathcal{E}_{\omega} &= 0 \quad\Longrightarrow\quad T^{a}=0,\\
\mathcal{E}_{e} &= 0 \quad\Longrightarrow\quad G_{\mu\nu}+\Lambda_{L}g_{\mu\nu}+16\pi G_{\mathrm{eff}}\sum_i \bar{\alpha}_i\bar{\mathcal{H}}_{\mu\nu}^{(i)}\\
&=8\pi G_{\mathrm{eff}}T_{\mu\nu}^{L} \tag*{(188)}
\end{aligned} E ω E e = 0 ⟹ T a = 0 , = 0 ⟹ G μν + Λ L g μν + 16 π G eff i ∑ α ˉ i H ˉ μν ( i ) = 8 π G eff T μν L (188)
Both lines are understood after the stated continuation and matching. The first line is the ω \omega ω component of the unified Euler system and is proved in (193) –(195) . The second line is the coframe/metric component after the connection equation has selected the torsionless branch; the Euclidean-to-Lorentzian continuation and the normalization of the Palatini coefficient are included in the phrase “stated continuation and matching.” Thus (188) is a conditional projection of the single master variational system, not an independent Einstein action appended to it. In the controlled low-curvature regime in which the higher-curvature term is negligible, it reduces to the ordinary Einstein equation with cosmological constant.
Metric variation and controlled recovery of the Einstein equation
The occurrence of the scalar-curvature term in (187) is not by itself a derivation of Einstein’s field equation. The latter is an Euler–Lagrange equation and requires a variation with respect to the coframe, or equivalently with respect to the induced metric on a non-degenerate branch. We make this step explicit. Write
I 1 = R 2 , I 2 = R ρ σ R ρ σ , I 3 = R ρ σ λ τ R ρ σ λ τ . (189) I_1=R^2,\qquad I_2=R_{\rho\sigma}R^{\rho\sigma},\qquad I_3=R_{\rho\sigma\lambda\tau}R^{\rho\sigma\lambda\tau}. \tag*{(189)} I 1 = R 2 , I 2 = R ρ σ R ρ σ , I 3 = R ρ σ λ τ R ρ σ λ τ . (189)
For any curvature invariant I i I_i I i , define its metric variational tensor by
H μ ν ( i ) : = 1 g δ δ g μ ν ∫ M g I i d 4 x . (190) \mathcal{H}_{\mu\nu}^{(i)}:=\frac{1}{\sqrt{g}}\frac{\delta}{\delta g^{\mu\nu}}\int_M\sqrt{g}\,I_i\,\mathrm{d}^4x. \tag*{(190)} H μν ( i ) := g 1 δ g μν δ ∫ M g I i d 4 x . (190)
This definition includes the variation of the volume density. For example, with variations of compact support (or boundary conditions that remove the boundary term),
H μ ν ( 1 ) = 2 R R μ ν − 1 2 g μ ν R 2 + 2 ( g μ ν □ − ∇ μ ∇ ν ) R . (191) \mathcal{H}_{\mu\nu}^{(1)}=2RR_{\mu\nu}-\frac{1}{2}g_{\mu\nu}R^2+2\left(g_{\mu\nu}\Box-\nabla_\mu\nabla_\nu\right)R. \tag*{(191)} H μν ( 1 ) = 2 R R μν − 2 1 g μν R 2 + 2 ( g μν □ − ∇ μ ∇ ν ) R . (191)
For I 2 I_2 I 2 and I 3 I_3 I 3 , equation (190) is the unambiguous definition; their expanded expressions contain the corresponding curvature contractions and second covariant derivatives. This choice avoids hiding a sign convention for the Riemann tensor in a formula that is later continued to Lorentzian signature.
The torsionless statement also has a precise variational status. If the Palatini block is isolated with a nonzero constant coefficient κ P \kappa_P κ P as
S P = κ P 2 ∫ M ϵ a b c d e a ∧ e b ∧ R c d , (192) S_P = \frac{\kappa_P}{2}\int_M \epsilon_{abcd}e^a \wedge e^b \wedge R^{cd}, \tag*{(192)} S P = 2 κ P ∫ M ϵ ab c d e a ∧ e b ∧ R c d , (192)
then, using δ R c d = D δ ω c d \delta R^{cd}=D\delta\omega^{cd} δ R c d = D δ ω c d and discarding the boundary term,
δ ω S P = − κ P ∫ M ϵ a b c d T a ∧ e b ∧ δ ω c d . (193) \delta_\omega S_P=-\kappa_P\int_M \epsilon_{abcd}T^a \wedge e^b \wedge\delta\omega^{cd}. \tag*{(193)} δ ω S P = − κ P ∫ M ϵ ab c d T a ∧ e b ∧ δ ω c d . (193)
For a non-degenerate coframe, the connection equation implies ϵ a b c d T a ∧ e b = 0 \epsilon_{abcd}T^a \wedge e^b=0 ϵ ab c d T a ∧ e b = 0 . For completeness, the elementary algebraic inversion is as follows. Write T a = 1 2 T a m n e m ∧ e n T^a=\frac{1}{2}T^a{}_{mn}e^m\wedge e^n T a = 2 1 T a mn e m ∧ e n and set V n : = T a a n V_n:=T^a{}_{an} V n := T a an . Taking the Hodge dual of the three-form equation and using ⋆ ( e m ∧ e n ∧ e b ) = ϵ m n b s e s \star(e^m\wedge e^n\wedge e^b)=\epsilon^{mnb}{}_s e^s ⋆ ( e m ∧ e n ∧ e b ) = ϵ mnb s e s gives
0 = 1 2 ϵ a b c d T a m n ϵ m n b s e s = − ( T s c d + δ s d V c − δ c s V d ) e s . (194) 0=\frac{1}{2}\epsilon_{abcd}T^a{}_{mn}\epsilon^{mnb}{}_s e^s=-\left(T^s{}_{cd}+\delta^d_s V_c-\delta^s_c V_d\right)e^s. \tag*{(194)} 0 = 2 1 ϵ ab c d T a mn ϵ mnb s e s = − ( T s c d + δ s d V c − δ c s V d ) e s . (194)
Here the displayed contraction uses the Euclidean orthonormal frame before Lorentzian continuation. Reversing orientation changes both epsilon tensors in this contraction, leaving their product unchanged. The contraction used in (194) , with δ a c d m n s \delta^{mns}_{acd} δ a c d mn s denoting the generalized Kronecker delta, is ϵ a b c d ϵ m n b s = − δ a c d m n s \epsilon_{abcd}\epsilon^{mnb}{}_s=-\delta^{mns}_{acd} ϵ ab c d ϵ mnb s = − δ a c d mn s , together with δ a c d m n s T a m n = 2 ( T s c d + δ d s V c − δ c s V d ) \delta^{mns}_{acd}T^a{}_{mn}=2(T^s{}_{cd}+\delta^s_d V_c-\delta^s_c V_d) δ a c d mn s T a mn = 2 ( T s c d + δ d s V c − δ c s V d ) . Consequently,
T s c d = V d δ c s − V c δ d s . (195) T^s{}_{cd}=V_d\delta^s_c-V_c\delta^s_d. \tag*{(195)} T s c d = V d δ c s − V c δ d s . (195)
Taking the trace s = c s=c s = c and summing over c c c gives V d = 4 V d − V d = 3 V d V_d=4V_d-V_d=3V_d V d = 4 V d − V d = 3 V d , hence V d = 0 V_d=0 V d = 0 ; equation (195) then gives T s c d = 0 T^s{}_{cd}=0 T s c d = 0 . Thus T a = 0 T^a=0 T a = 0 on the non-degenerate torsionless branch. This statement is for the spinless or vacuum sector. If matter has a nonzero spin current, the connection equation is instead ϵ a b c d T a ∧ e b = J c d s p i n \epsilon_{abcd}T^a\wedge e^b=J^{\mathrm{spin}}_{cd} ϵ ab c d T a ∧ e b = J c d spin , and the appropriate low-energy theory is Einstein–Cartan with torsion sourced by spin. The ordinary Einstein branch therefore requires J s p i n = 0 J^{\mathrm{spin}}=0 J spin = 0 (or a controlled suppression) at the matching scale. The torsionless implication is also not automatic if the full master pairing contains independent torsion-square or mixed terms; in that case the additional terms must be included in the connection equation. Likewise, variation of the coframe gives
δ e S P = κ P ∫ M ϵ a b c d δ e a ∧ e b ∧ R c d , (196) \delta_e S_P=\kappa_P\int_M\epsilon_{abcd}\delta e^a\wedge e^b\wedge R^{cd}, \tag*{(196)} δ e S P = κ P ∫ M ϵ ab c d δ e a ∧ e b ∧ R c d , (196)
which must be combined with the cosmological, higher-curvature, and matter variations before it can be identified with an Einstein equation.
For the joint candidate, the common scalar generally makes the Palatini coefficient position dependent. The distinction can be derived without assuming a torsion ansatz. For a smooth positive function w w w , put
S P [ w , e , ω ] = κ P 2 ∫ M w ϵ a b c d e a ∧ e b ∧ R c d , E a : = w e a . (197) S_P[w,e,\omega]=\frac{\kappa_P}{2}\int_M w\,\epsilon_{abcde^a}\wedge e^b\wedge R^{cd},\qquad E^a:=\sqrt{w}\,e^a. \tag*{(197)} S P [ w , e , ω ] = 2 κ P ∫ M w ϵ ab c d e a ∧ e b ∧ R c d , E a := w e a . (197)
Holding e e e and w w w fixed while varying ω \omega ω , integration by parts gives
δ ω S P [ w , e , ω ] = − κ P 2 ∫ M ϵ a b c d D ω ( w e a ∧ e b ) ∧ δ ω c d = − κ P ∫ M ϵ a b c d ( w T a ∧ e b + 1 2 d w ∧ e a ∧ e b ) ∧ δ ω c d . (198) \delta_{\omega}S_P[w,e,\omega] = -\frac{\kappa_P}{2}\int_M \epsilon_{abcd}D_\omega(we^a\wedge e^b)\wedge\delta\omega^{cd}
= -\kappa_P\int_M \epsilon_{abcd}\left(wT^a\wedge e^b+\frac{1}{2}dw\wedge e^a\wedge e^b\right)\wedge\delta\omega^{cd}. \tag*{(198)} δ ω S P [ w , e , ω ] = − 2 κ P ∫ M ϵ ab c d D ω ( w e a ∧ e b ) ∧ δ ω c d = − κ P ∫ M ϵ ab c d ( w T a ∧ e b + 2 1 d w ∧ e a ∧ e b ) ∧ δ ω c d . (198)
Since w e a ∧ e b = E a ∧ E b we^a\wedge e^b=E^a\wedge E^b w e a ∧ e b = E a ∧ E b , the same algebraic inversion as in (194) –(195) , now applied to the non-degenerate coframe E E E , shows that the spinless connection equation is D ω E a = 0 D_\omega E^a=0 D ω E a = 0 . Expanding this identity yields
0 = D ω E a = w ( T a + 1 2 d log w ∧ e a ) , T a = 1 2 e a ∧ d log w . (199) 0=D_\omega E^a=\sqrt{w}\left(T^a+\frac{1}{2}d\log w\wedge e^a\right),\qquad T^a=\frac{1}{2}e^a\wedge d\log w. \tag*{(199)} 0 = D ω E a = w ( T a + 2 1 d log w ∧ e a ) , T a = 2 1 e a ∧ d log w . (199)
Thus ω \omega ω is the Levi–Civita spin connection of the conformally rescaled coframe E E E , but it is generally not the torsionless connection of the original coframe e e e . The reduction to the ordinary Einstein equation for the metric g e g_e g e therefore requires d w = 0 dw=0 d w = 0 , or a specified approximation that controls the scalar gradient and the remaining connection terms. Even if one imposes ω = ω ( e ) \omega=\omega(e) ω = ω ( e ) before variation, the metric variation already displays the missing derivative terms:
1 g δ δ g μ ν ∫ M g w R ( g ) d 4 x = w G μ ν + ( g μ ν □ − ∇ μ ∇ ν ) w . (200) \frac{1}{\sqrt{g}}\frac{\delta}{\delta g^{\mu\nu}}\int_M\sqrt{g}\,wR(g)\,d^4x=wG_{\mu\nu}+(g_{\mu\nu}\Box-\nabla_\mu\nabla_\nu)w. \tag*{(200)} g 1 δ g μν δ ∫ M g w R ( g ) d 4 x = w G μν + ( g μν □ − ∇ μ ∇ ν ) w . (200)
Here w w w is held fixed in the metric variation. Eliminating the independent connection from (197) also produces terms quadratic in d w dw d w ; it is not the same operation as imposing ω ( e ) \omega(e) ω ( e ) at the outset. Consequently, the constant-coefficient equations (205) and (209) describe a constant-scalar branch or a controlled effective limit, rather than every solution of the coupled weighted theory.
To display the remaining signs, first stay on the Euclidean slice used in (187) . In this subsection S m a t , E S_{\mathrm{mat},E} S mat , E denotes the total non-gravitational effective action, including the Yang–Mills, scalar, and matter sectors. Define the Euclidean stress tensor by
T μ ν E : = 2 g δ S m a t , E δ g μ ν . (201) T^E_{\mu\nu}:=\frac{2}{\sqrt{g}}\frac{\delta S_{\mathrm{mat},E}}{\delta g^{\mu\nu}}. \tag*{(201)} T μν E := g 2 δ g μν δ S mat , E . (201)
The elementary identities
δ g = − 1 2 g g μ ν δ g μ ν , δ R = R μ ν δ g μ ν + g μ ν □ δ g μ ν − ∇ μ ∇ ν δ g μ ν (202) \delta\sqrt{g}=-\frac{1}{2}\sqrt{g}g_{\mu\nu}\delta g^{\mu\nu},\qquad
\delta R=R_{\mu\nu}\delta g^{\mu\nu}+g_{\mu\nu}\Box\delta g^{\mu\nu}-\nabla_\mu\nabla_\nu\delta g^{\mu\nu} \tag*{(202)} δ g = − 2 1 g g μν δ g μν , δ R = R μν δ g μν + g μν □ δ g μν − ∇ μ ∇ ν δ g μν (202)
and covariant integration by parts give
δ ∫ M g R d 4 x = ∫ M g G μ ν δ g μ ν d 4 x , G μ ν : = R μ ν − 1 2 R g μ ν . (203) \delta\int_M\sqrt{g}R\,d^4x=\int_M\sqrt{g}G_{\mu\nu}\delta g^{\mu\nu}\,d^4x,\qquad
G_{\mu\nu}:=R_{\mu\nu}-\frac{1}{2}Rg_{\mu\nu}. \tag*{(203)} δ ∫ M g R d 4 x = ∫ M g G μν δ g μν d 4 x , G μν := R μν − 2 1 R g μν . (203)
Consequently, varying the complete Euclidean effective action, including the matter term, yields
0 = − 1 16 π G e f f G μ ν − 1 2 Λ e f f g μ ν + ∑ i α i H μ ν ( i ) + 1 2 T μ ν E (204) 0=-\frac{1}{16\pi G_{\mathrm{eff}}}G_{\mu\nu}-\frac{1}{2}\Lambda_{\mathrm{eff}}g_{\mu\nu}+\sum_i\alpha_iH^{(i)}_{\mu\nu}+\frac{1}{2}T^E_{\mu\nu} \tag*{(204)} 0 = − 16 π G eff 1 G μν − 2 1 Λ eff g μν + i ∑ α i H μν ( i ) + 2 1 T μν E (204)
G μ ν + 8 π G e f f Λ e f f g μ ν − 16 π G e f f ∑ i α i H μ ν ( i ) = 8 π G e f f T μ ν E . (205) G_{\mu\nu}+8\pi G_{\mathrm{eff}}\Lambda_{\mathrm{eff}}g_{\mu\nu}-16\pi G_{\mathrm{eff}}\sum_i\alpha_i H_{\mu\nu}^{(i)}
=8\pi G_{\mathrm{eff}}T_{\mu\nu}^{E}. \tag*{(205)} G μν + 8 π G eff Λ eff g μν − 16 π G eff i ∑ α i H μν ( i ) = 8 π G eff T μν E . (205)
The first displayed line is the direct variation; the following two lines are the same equation after multiplication by − 16 π G e f f -16\pi G_{\mathrm{eff}} − 16 π G eff . It is therefore the precise Euclidean statement associated with the sign convention in (187) . Here Λ e f f \Lambda_{\mathrm{eff}} Λ eff is a vacuum-energy density in the Euclidean action. The curvature cosmological constant has dimension two and is related, up to the sign fixed by the continuation, by Λ E = 8 π G e f f Λ e f f \Lambda_E=8\pi G_{\mathrm{eff}}\Lambda_{\mathrm{eff}} Λ E = 8 π G eff Λ eff ; it must not be identified with Λ e f f \Lambda_{\mathrm{eff}} Λ eff itself.
The usual Lorentzian Einstein equation requires an additional analytic continuation and a Lorentzian matter convention. To make the continuation explicit, choose the mostly-plus Lorentzian convention g L = − ( e L 0 ) 2 + ∑ j = 1 3 ( e L j ) 2 g_L=-(e_L^0)^2+\sum_{j=1}^{3}(e_L^j)^2 g L = − ( e L 0 ) 2 + ∑ j = 1 3 ( e L j ) 2 and the Wick map
e E 4 ⟼ i e L 0 , e E j ⟼ e L j ( j = 1 , 2 , 3 ) , e_E^4\longmapsto ie_L^0,\qquad e_E^j\longmapsto e_L^j\quad(j=1,2,3), e E 4 ⟼ i e L 0 , e E j ⟼ e L j ( j = 1 , 2 , 3 ) ,
g E ⟼ g L , g E d 4 x ⟼ i − g L d 4 x , S O ( 5 ) ⟼ S O ( 1 , 4 ) , R E ⟼ R L . (206) g_E\longmapsto g_L,\qquad\sqrt{g_E}\,\mathrm{d}^4x\longmapsto i\sqrt{-g_L}\,\mathrm{d}^4x,\qquad SO(5)\longmapsto SO(1,4),\qquad R_E\longmapsto R_L. \tag*{(206)} g E ⟼ g L , g E d 4 x ⟼ i − g L d 4 x , S O ( 5 ) ⟼ S O ( 1 , 4 ) , R E ⟼ R L . (206)
The volume multiplier in (206) includes an orientation and contour convention. We take the oriented Euclidean volume in the order e E 4 ∧ e E 1 ∧ e E 2 ∧ e E 3 e_E^4\wedge e_E^1\wedge e_E^2\wedge e_E^3 e E 4 ∧ e E 1 ∧ e E 2 ∧ e E 3 and the Lorentzian volume in the order e L 0 ∧ e L 1 ∧ e L 2 ∧ e L 3 e_L^0\wedge e_L^1\wedge e_L^2\wedge e_L^3 e L 0 ∧ e L 1 ∧ e L 2 ∧ e L 3 . The declared map then multiplies the volume by i i i . If instead the Euclidean wedge order is 1234, the associated orientation or contour Jacobian must also be reversed to retain this convention. All Levi–Civita symbols are defined relative to the chosen orientation; no additional sign is suppressed. The last line uses the curvature convention transported by the complexified coframe and spin connection; components with a time index acquire the corresponding factors of i i i . Writing W \mathcal{W} W for this continuation, the relation between the path-integral weights is fixed by S L = i W ( S E ) S_L=i\mathcal{W}(S_E) S L = i W ( S E ) , so that exp [ − W ( S E ) ] = exp ( i S L ) \exp[-\mathcal{W}(S_E)]=\exp(iS_L) exp [ − W ( S E )] = exp ( i S L ) . In particular, the Euclidean density Λ e f f − R E / ( 16 π G e f f ) \Lambda_{\mathrm{eff}}-R_E/(16\pi G_{\mathrm{eff}}) Λ eff − R E / ( 16 π G eff ) continues to the Lorentzian density R L / ( 16 π G e f f ) − Λ e f f R_L/(16\pi G_{\mathrm{eff}})-\Lambda_{\mathrm{eff}} R L / ( 16 π G eff ) − Λ eff . Thus this convention gives Λ L = 8 π G e f f Λ e f f \Lambda_L=8\pi G_{\mathrm{eff}}\Lambda_{\mathrm{eff}} Λ L = 8 π G eff Λ eff ; for unchanged real scalar curvature-square coefficients it gives α ˉ i = − α i \bar{\alpha}_i=-\alpha_i α ˉ i = − α i at this continuation step. Other real-form or signature conventions are equivalent after changing all associated signs consistently. The map is a continuation of the regulated fields and boundary data; a formal substitution alone does not establish Lorentzian hyperbolicity, causality, or a globally defined solution. In particular, the Euclidean S O ( 5 ) / S O ( 4 ) SO(5)/SO(4) S O ( 5 ) / S O ( 4 ) round-sphere vacuum is not itself a Lorentzian solution until this continuation and the relevant boundary conditions have been supplied. When the stated continuation exists, write the continued action as
S L = ∫ M − g [ R − 2 Λ L 16 π G e f f + ∑ i α ˉ i I i + L n o n g r a v , L ] d 4 x , (207) S_L=\int_M\sqrt{-g}\left[\frac{R-2\Lambda_L}{16\pi G_{\mathrm{eff}}}+\sum_i\bar{\alpha}_i I_i+\mathcal{L}_{\mathrm{nongrav},L}\right]\mathrm{d}^4x, \tag*{(207)} S L = ∫ M − g [ 16 π G eff R − 2 Λ L + i ∑ α ˉ i I i + L nongrav , L ] d 4 x , (207)
where L n o n g r a v , L \mathcal{L}_{\mathrm{nongrav},L} L nongrav , L contains the continued Yang–Mills, scalar, and matter terms, and where its action is S n o n g r a v , L : = ∫ M − g L n o n g r a v , L d 4 x S_{\mathrm{nongrav},L} := \int_M \sqrt{-g}\,\mathcal{L}_{\mathrm{nongrav},L}\,\mathrm{d}^4x S nongrav , L := ∫ M − g L nongrav , L d 4 x . Define
T μ ν L : = − 2 − g δ S n o n g r a v , L δ g μ ν , H ˉ μ ν ( i ) : = 1 − g δ δ g μ ν ∫ M − g I i d 4 x . (208) T^L_{\mu\nu} := -\frac{2}{\sqrt{-g}}\frac{\delta S_{\mathrm{nongrav},L}}{\delta g^{\mu\nu}}, \qquad\bar{H}^{(i)}_{\mu\nu} := \frac{1}{\sqrt{-g}}\frac{\delta}{\delta g^{\mu\nu}}\int_M \sqrt{-g}\,I_i\,\mathrm{d}^4x. \tag*{(208)} T μν L := − − g 2 δ g μν δ S nongrav , L , H ˉ μν ( i ) := − g 1 δ g μν δ ∫ M − g I i d 4 x . (208)
The Lorentzian Euler–Lagrange equation is then
G μ ν + Λ L g μ ν + 16 π G e f f ∑ i α ˉ i H ˉ μ ν ( i ) = 8 π G e f f T μ ν L . (209) G_{\mu\nu}+\Lambda_L g_{\mu\nu}+16\pi G_{\mathrm{eff}}\sum_i\bar{\alpha}_i\bar{H}^{(i)}_{\mu\nu}=8\pi G_{\mathrm{eff}}T^L_{\mu\nu}. \tag*{(209)} G μν + Λ L g μν + 16 π G eff i ∑ α ˉ i H ˉ μν ( i ) = 8 π G eff T μν L . (209)
The barred coefficients include the specified Wick sign and any separately declared renormalization or matching changes. Thus the ordinary Einstein equation is recovered, with an explicit error estimate, in a controlled low-curvature regime. A scale-based condition that remains meaningful near an exact vacuum is the following. Let L L L be a characteristic curvature length and suppose, for the displayed curvature invariants, that ∥ H ˉ ( i ) ∥ ≤ C i L − 4 \|\bar{H}^{(i)}\|\leq C_iL^{-4} ∥ H ˉ ( i ) ∥ ≤ C i L − 4 for fixed dimensionless constants C i C_i C i . Require
16 π G e f f ∑ i ∣ α ˉ i ∣ C i L − 2 ≤ ε , 0 < ε ≪ 1. (210) 16\pi G_{\mathrm{eff}}\sum_i|\bar{\alpha}_i|C_iL^{-2}\leq\varepsilon,\qquad0<\varepsilon\ll1. \tag*{(210)} 16 π G eff i ∑ ∣ α ˉ i ∣ C i L − 2 ≤ ε , 0 < ε ≪ 1. (210)
Define the residual of the ordinary Einstein equation by
R μ ν E : = G μ ν + Λ L g μ ν − 8 π G e f f T μ ν L . (211) \mathcal{R}^{E}_{\mu\nu}:=G_{\mu\nu}+\Lambda_Lg_{\mu\nu}-8\pi G_{\mathrm{eff}}T^L_{\mu\nu}. \tag*{(211)} R μν E := G μν + Λ L g μν − 8 π G eff T μν L . (211)
Equation (210), the triangle inequality, and (211) give the quantitative bound
∥ R E ∥ ≤ 16 π G e f f ∑ i ∣ α ˉ i ∣ ∥ H ˉ ( i ) ∥ ≤ ε L − 2 . (212) \|\mathcal{R}^{E}\|\leq16\pi G_{\mathrm{eff}}\sum_i|\bar{\alpha}_i|\|\bar{H}^{(i)}\|\leq\varepsilon L^{-2}. \tag*{(212)} ∥ R E ∥ ≤ 16 π G eff i ∑ ∣ α ˉ i ∣∥ H ˉ ( i ) ∥ ≤ ε L − 2 . (212)
Equivalently, the total higher-curvature contribution is bounded by an ε \varepsilon ε -fraction of a fixed leading curvature scale of order L − 2 L^{-2} L − 2 . This is a scale-separation condition, not a pointwise identity. If G + Λ L g = 0 G+\Lambda_Lg=0 G + Λ L g = 0 exactly, no ratio with that tensor is defined; an exact solution then requires ∑ i α ˉ i H ˉ ( i ) = 0 \sum_i\bar{\alpha}_i\bar{H}^{(i)}=0 ∑ i α ˉ i H ˉ ( i ) = 0 , or a constant part of this correction must be absorbed into the definition of Λ L \Lambda_L Λ L . The continuation, the torsionless coframe branch, and the estimate (211) define the domain and accuracy of the recovery statement. They are not claims of global regularity and need not hold for every solution of the unreduced equations. In particular, the fixed-background connection calculation in Appendix A does not replace this metric/coframe variation; conversely, proving the local variational identity does not require a global existence theory for all solutions.
For the displayed quadratic-curvature basis, a sufficient local version of the assumed Euler-tensor bound controls derivatives as well as curvature. Fix a positive reference tensor norm on a compact observation region and suppose
∥ R m ∥ ≤ c 0 L − 2 , ∥ ∇ R m ∥ ≤ c 1 L − 3 , ∥ ∇ 2 R m ∥ ≤ c 2 L − 4 . (213) \|\mathrm{Rm}\|\leq c_0L^{-2},\qquad\|\nabla\mathrm{Rm}\|\leq c_1L^{-3},\qquad\|\nabla^2\mathrm{Rm}\|\leq c_2L^{-4}. \tag*{(213)} ∥ Rm ∥ ≤ c 0 L − 2 , ∥∇ Rm ∥ ≤ c 1 L − 3 , ∥ ∇ 2 Rm ∥ ≤ c 2 L − 4 . (213)
Products of two curvatures and second derivatives of curvature are then bounded by constants times L − 4 L^{-4} L − 4 . These are the terms appearing in the quadratic-curvature Euler tensors, so the constants C i C_i C i above can be chosen from c 0 , c 1 , c 2 c_0,c_1,c_2 c 0 , c 1 , c 2 and the fixed contraction norms. An indefinite Lorentzian contraction is not used as a positive error norm. Small ∥ R m ∥ \lVert\mathrm{Rm}\rVert ∥ Rm ∥ alone does not control a rapidly varying ∇ 2 R m \nabla^2\mathrm{Rm} ∇ 2 Rm ; higher-derivative operators require corresponding additional derivative estimates.
The conclusion of (212) is an equation residual bound on the matched branch. A bound on the distance between an effective solution and an Einstein solution additionally requires common admissible initial or boundary data, a stability estimate in a declared norm, constraint propagation, and exclusion or treatment of additional higher-derivative modes. No such solution-distance bound follows from the residual estimate alone.
A positive Cartan-curvature realization
A concrete gravitational choice removes the need to leave the Einstein coefficient as an unspecified algebraic insertion. Take the compact Cartan pair ( S O ( 5 ) , S O ( 4 ) ) (SO(5),SO(4)) ( S O ( 5 ) , S O ( 4 )) , fix ℓ > 0 \ell>0 ℓ > 0 and g c > 0 g_c>0 g c > 0 , and use the defining real matrix representation. On an oriented orthonormal frame bundle the local connection matrix is
A c = ( ω a b ℓ − 1 e a − ℓ − 1 e b 0 ) , a , b ∈ { 1 , 2 , 3 , 4 } . (214) A_{\mathrm{c}}=\begin{pmatrix}
\omega^a{}_b & \ell^{-1}e^a\\
-\ell^{-1}e_b & 0
\end{pmatrix},\qquad a,b\in\{1,2,3,4\}. \tag*{(214)} A c = ( ω a b − ℓ − 1 e b ℓ − 1 e a 0 ) , a , b ∈ { 1 , 2 , 3 , 4 } . (214)
The matrix is skew symmetric, so its values belong to s o ( 5 ) \mathfrak{so}(5) so ( 5 ) . The inclusion of s o ( 4 ) \mathfrak{so}(4) so ( 4 ) is the upper-left block, and the four off-diagonal components are the soldering form. The ordinary transformation laws for ω \omega ω and e e e prove S O ( 4 ) SO(4) S O ( 4 ) -equivariance and reproduction of vertical generators. Non-degeneracy of the soldering form makes the map from the four horizontal directions an isomorphism onto s o ( 5 ) / s o ( 4 ) \mathfrak{so}(5)/\mathfrak{so}(4) so ( 5 ) / so ( 4 ) . Together with the six vertical directions this verifies the Cartan isomorphism axiom. The matrix commutator supplies all brackets and their Jacobi identities. This construction uses the frame bundle and requires no global trivialization of its coframe.
Matrix multiplication of forms gives
F c = ( R a b − ℓ − 2 e a ∧ e b ℓ − 1 T a − ℓ − 1 T b 0 ) , T a = d e a + ω a b ∧ e b . (215) F_{\mathrm{c}}=\begin{pmatrix}
R^a{}_b-\ell^{-2}e^a\wedge e_b & \ell^{-1}T^a\\
-\ell^{-1}T_b & 0
\end{pmatrix},\qquad T^a=de^a+\omega^a{}_b\wedge e^b. \tag*{(215)} F c = ( R a b − ℓ − 2 e a ∧ e b − ℓ − 1 T b ℓ − 1 T a 0 ) , T a = d e a + ω a b ∧ e b . (215)
Let ⋆ e \star_e ⋆ e be the spacetime Hodge operator of g e = δ a b e a ⊗ e b g_e=\delta_{ab}e^a\otimes e^b g e = δ ab e a ⊗ e b . With I , J ∈ { 1 , … , 5 } I,J\in\{1,\ldots,5\} I , J ∈ { 1 , … , 5 } , define a single curvature norm
S c [ e , ω ] = 1 2 g c 2 ∫ M ∑ I < J F c I J ∧ ⋆ e F c I J . (216) S_{\mathrm{c}}[e,\omega]=\frac{1}{2g_c^2}\int_M\sum_{I<J}F_{\mathrm{c}}^{IJ}\wedge\star_eF_{\mathrm{c}}^{IJ}. \tag*{(216)} S c [ e , ω ] = 2 g c 2 1 ∫ M I < J ∑ F c I J ∧ ⋆ e F c I J . (216)
The Lie-algebra pairing is the positive multiple of the negative trace form in the defining representation. Its norm, and hence the integrand in (216) , is nonnegative on the Euclidean slice. The Hodge operator is part of the coframe-dependent data and must vary when the coframe varies. The gauge transformations used here are those of the declared Cartan isotropy group S O ( 4 ) SO(4) S O ( 4 ) .
Theorem 10 (Exact coefficients of a Cartan norm). Write R a b = 1 2 R c d a b e c ∧ e d R^{ab} = \frac{1}{2}R^{ab}_{cd}e^c \wedge e^d R ab = 2 1 R c d ab e c ∧ e d , set R ( ω ) = R a b a b R(\omega)=R^{ab}_{ab} R ( ω ) = R ab ab , and define τ 2 = ∑ a ∣ T a ∣ g e 2 \tau^2=\sum_a |T^a|_{g_e}^2 τ 2 = ∑ a ∣ T a ∣ g e 2 . Then
S c = 1 2 g c 2 ∫ M [ 1 4 R c d a b R c d a b − R ( ω ) ℓ 2 + 6 ℓ 4 + 7 ℓ 2 ] vol g e . (217) S_c=\frac{1}{2g_c^2}\int_M\left[\frac{1}{4}R^{ab}_{cd}R^{ab}_{cd}-\frac{R(\omega)}{\ell^2}+\frac{6}{\ell^4}+\frac{7}{\ell^2}\right]\operatorname{vol}_{g_e}. \tag*{(217)} S c = 2 g c 2 1 ∫ M [ 4 1 R c d ab R c d ab − ℓ 2 R ( ω ) + ℓ 4 6 + ℓ 2 7 ] vol g e . (217) On the torsionless restriction ω = ω ( e ) \omega=\omega(e) ω = ω ( e ) , its coefficients in (188) are
G e f f = g c 2 ℓ 2 8 π , Λ e f f = 3 g c 2 ℓ 4 , α 1 = α 2 = 0 , α 3 = 1 8 g c 2 , (218) G_{\mathrm{eff}}=\frac{g_c^2\ell^2}{8\pi},\quad\Lambda_{\mathrm{eff}}=\frac{3}{g_c^2\ell^4},\quad\alpha_1=\alpha_2=0,\quad\alpha_3=\frac{1}{8g_c^2}, \tag*{(218)} G eff = 8 π g c 2 ℓ 2 , Λ eff = g c 2 ℓ 4 3 , α 1 = α 2 = 0 , α 3 = 8 g c 2 1 , (218) in the unreduced curvature-invariant basis used there. The associated Euclidean curvature cosmological constant obeys
Λ E = 3 ℓ 2 , 64 π G e f f Λ E α 3 = 3. (219) \Lambda_E=\frac{3}{\ell^2},\qquad64\pi G_{\mathrm{eff}}\Lambda_E\alpha_3=3. \tag*{(219)} Λ E = ℓ 2 3 , 64 π G eff Λ E α 3 = 3. (219)
Proof. The two-forms e a ∧ e b e^a\wedge e^b e a ∧ e b , a < b a<b a < b , form an orthonormal basis. There are six such forms. Antisymmetry in each curvature index pair gives
∑ a < b ∣ R a b ∣ 2 = 1 4 ∑ a , b , c , d ( R c d a b ) 2 , ∑ a < b ⟨ R a b , e a ∧ e b ⟩ = ∑ a < b R a b a b = 1 2 R ( ω ) , \sum_{a<b}|R^{ab}|^2=\frac{1}{4}\sum_{a,b,c,d}(R^{ab}_{cd})^2,\qquad\sum_{a<b}\langle R^{ab},e^a\wedge e^b\rangle=\sum_{a<b}R^{ab}_{ab}=\frac{1}{2}R(\omega), a < b ∑ ∣ R ab ∣ 2 = 4 1 a , b , c , d ∑ ( R c d ab ) 2 , a < b ∑ ⟨ R ab , e a ∧ e b ⟩ = a < b ∑ R ab ab = 2 1 R ( ω ) ,
∑ a < b ∣ e a ∧ e b ∣ 2 = 6. (220) \sum_{a<b}|e^a\wedge e^b|^2=6. \tag*{(220)} a < b ∑ ∣ e a ∧ e b ∣ 2 = 6. (220)
Expand each square ∣ R a b − ℓ − 2 e a ∧ e b ∣ 2 |R^{ab}-\ell^{-2}e^a\wedge e^b|^2 ∣ R ab − ℓ − 2 e a ∧ e b ∣ 2 . Its cross term is − 2 ℓ − 2 ⟨ R a b , e a ∧ e b ⟩ -2\ell^{-2}\langle R^{ab},e^a\wedge e^b\rangle − 2 ℓ − 2 ⟨ R ab , e a ∧ e b ⟩ , whose sum is − ℓ − 2 R ( ω ) -\ell^{-2}R(\omega) − ℓ − 2 R ( ω ) . The ( a , 5 ) (a,5) ( a , 5 ) components add ℓ − 2 τ 2 \ell^{-2}\tau^2 ℓ − 2 τ 2 . These four contributions prove (218). Under the torsionless restriction the curvature is Riemannian, and comparison of the coefficient of − R -R − R gives 1 / ( 16 π G e f f ) = 1 / ( 2 g c 2 ℓ 2 ) 1/(16\pi G_{\mathrm{eff}})=1/(2g_c^2\ell^2) 1/ ( 16 π G eff ) = 1/ ( 2 g c 2 ℓ 2 ) . The constant and curvature-square coefficients give the other entries of (219). Substitution into Λ E = 8 π G e f f Λ e f f \Lambda_E=8\pi G_{\mathrm{eff}}\Lambda_{\mathrm{eff}} Λ E = 8 π G eff Λ eff and then into 64 π G e f f Λ E α 3 64\pi G_{\mathrm{eff}}\Lambda_E\alpha_3 64 π G eff Λ E α 3 proves (220). □
The round four-sphere of radius ℓ \ell ℓ gives an exact Euclidean configuration with T a = 0 T^a=0 T a = 0 and R a b = ℓ − 2 e a ∧ e b R^{ab}=\ell^{-2}e^a\wedge e^b R ab = ℓ − 2 e a ∧ e b . Thus F c = 0 F_c=0 F c = 0 and S c = 0 S_c=0 S c = 0 . Every first variation vanishes: both the variation of the curvature and that of the Hodge operator occur multiplied by at least one curvature factor. This configuration is a minimum of the positive action and has R μ ν = 3 ℓ − 2 g μ ν R_{\mu\nu}=3\ell^{-2}g_{\mu\nu} R μν = 3 ℓ − 2 g μν . It supplies an explicit Einstein vacuum of the Euclidean example.
For general configurations, imposing ω = ω ( e ) \omega=\omega(e) ω = ω ( e ) before variation defines a restricted metric theory. It is not equivalent without further analysis to varying the independent connection in (217): the curvature-square and torsion-square terms contribute to that connection equation. The existence of the round vacuum proves neither uniqueness of the selected phase nor suppression of all torsional fluctuations.
The coefficient relation also exposes a scale restriction. In the metric equation, 16 π G e f f α 3 = ℓ 2 / 4 16\pi G_{\mathrm{eff}}\alpha_3 = \ell^2/4 16 π G eff α 3 = ℓ 2 /4 . For fields with curvature and derivatives characterized by a length L L L , the curvature-square correction has the relative derivative-expansion scale ℓ 2 / L 2 \ell^2/L^2 ℓ 2 / L 2 . The round vacuum itself has curvature scale ℓ − 2 \ell^{-2} ℓ − 2 ; its stationarity does not imply a small higher-derivative correction for arbitrary perturbations. An independently justified vacuum-energy mechanism would be needed to separate these two scales. Additional matter thresholds or independent finite counterterms can also change (220). In four dimensions the Euler density allows a change of curvature-invariant basis, so the relation is stated in the specified basis and is not by itself an invariant scattering prediction.
This gravitational example belongs to the established Cartan and gauge-geometric setting [145] and [116]. Its algebra, pairing, and coefficients are explicit inputs and consequences. The next construction places this Cartan block and the three Standard Model gauge factors in one specified master system and makes the remaining selection assumptions explicit.
A joint microscopic candidate with one order parameter
The phrase “one microscopic model” has a precise dimension constraint in Cartan geometry. Let
G S M : = S U ( 3 ) c × S U ( 2 ) L × U ( 1 ) Y , G ~ 0 : = S O ( 5 ) × G S M , G_{\mathrm{SM}} := SU(3)_c \times SU(2)_L \times U(1)_Y,\qquad\widetilde{G}_0 := SO(5) \times G_{\mathrm{SM}}, G SM := S U ( 3 ) c × S U ( 2 ) L × U ( 1 ) Y , G 0 := S O ( 5 ) × G SM ,
H 0 : = S O ( 4 ) × G S M . (221) H_0 := SO(4) \times G_{\mathrm{SM}}. \tag*{(221)} H 0 := S O ( 4 ) × G SM . (221)
Then
dim G ~ 0 = 10 + 8 + 3 + 1 = 22 , dim H 0 = 6 + 8 + 3 + 1 = 18 , \dim\widetilde{G}_0 = 10 + 8 + 3 + 1 = 22,\qquad\dim H_0 = 6 + 8 + 3 + 1 = 18, dim G 0 = 10 + 8 + 3 + 1 = 22 , dim H 0 = 6 + 8 + 3 + 1 = 18 ,
dim ( G ~ 0 / H 0 ) = 4. (222) \dim(\widetilde{G}_0/H_0) = 4. \tag*{(222)} dim ( G 0 / H 0 ) = 4. (222)
The quotient dimension is therefore the dimension of the spacetime manifold. This condition rules out simply choosing a larger group whose quotient has many extra directions and then silently identifying all of them with a four-dimensional coframe.
Let Γ 0 \Gamma_0 Γ 0 denote the full master-data package (55) obtained by supplementing ( G ~ 0 , H 0 ) (\widetilde{G}_0,H_0) ( G 0 , H 0 ) with the specified principal bundle, representations, invariant pairing, order-parameter map, reference measure, regularization, mobility, and information data. It is an explicit candidate witness for the coupled construction, not a selected minimizer of the selection problem in Section 6.1. Accordingly, M ~ Γ 0 \widetilde{\mathcal{M}}_{\Gamma_0} M Γ 0 below denotes the tangent mobility induced by that entire package, including its gauge fixing or quotient prescription.
The group G S M G_{\mathrm{SM}} G SM , the hypercharge normalization, and the one-generation representation displayed below are prescribed inputs in this candidate. They are included to exhibit one common, well-typed realization and are not outputs of the minimization problem in § 9.1 . Moreover, the Lie algebra is a direct product: the candidate is coupled through the shared scalar and mobility data, but it is not a proof of an algebraically indecomposable master algebra. The no-factorization test in Definition 2 remains an additional requirement.
On a principal H 0 H_0 H 0 -bundle, use the single Cartan-type master field
A 0 = ( ( ω a b ℓ − 1 e a − ℓ − 1 e b 0 ) , A ) , A = A A T A ∈ Ω 1 ( M , g S M ) . (223) \mathcal{A}_0=\left(\begin{pmatrix}\omega^a{}_b & \ell^{-1}e^a \\ -\ell^{-1}e_b & 0\end{pmatrix}, A\right),\qquad A=A^A T_A\in\Omega^1(M,\mathfrak{g}_{\mathrm{SM}}). \tag*{(223)} A 0 = ( ( ω a b − ℓ − 1 e b ℓ − 1 e a 0 ) , A ) , A = A A T A ∈ Ω 1 ( M , g SM ) . (223)
The first entry is an s o ( 5 ) \mathfrak{so}(5) so ( 5 ) -valued Cartan connection of type S O ( 5 ) / S O ( 4 ) SO(5)/SO(4) S O ( 5 ) / S O ( 4 ) ; its displayed matrix is the descended vector-representation expression. The Spin(4) lift is included so that spinor matter can be defined globally; if no spin lift exists, only the bosonic Cartan block is available. The second entry is the internal connection. Its curvature is the single ordered tuple
F 0 = ( R a b − ℓ − 2 e a ∧ e b , ℓ − 1 T a , F A ) , F A = d A A + 1 2 f B C A A B ∧ A C . (224) \mathcal{F}_0=\left(R^{ab}-\ell^{-2}e^a\wedge e^b,\ \ell^{-1}T^a,\ F^A\right),\qquad F^A=\mathrm{d}A^A+\frac{1}{2}f_{BC}{}^A A^B\wedge A^C. \tag*{(224)} F 0 = ( R ab − ℓ − 2 e a ∧ e b , ℓ − 1 T a , F A ) , F A = d A A + 2 1 f B C A A B ∧ A C . (224)
All three entries are components of one connection; they are not three independently postulated actions.
To make the coupling between the two geometric blocks explicit, introduce a dimensionless scalar coordinate σ \sigma σ , use
w ( σ ) : = exp ( − α σ σ ) > 0 , α σ ∈ R , (225) w(\sigma):=\exp(-\alpha_\sigma\sigma)>0,\qquad\alpha_\sigma\in\mathbb{R}, \tag*{(225)} w ( σ ) := exp ( − α σ σ ) > 0 , α σ ∈ R , (225)
and let Φ H \Phi_H Φ H be a scalar in the representation ( 1 , 2 ) 1 / 2 (1,2)_{1/2} ( 1 , 2 ) 1/2 of G S M G_{\mathrm{SM}} G SM . The scalar σ \sigma σ is a shared master coordinate: the same function w ( σ ) w(\sigma) w ( σ ) multiplies the Cartan and internal curvature blocks. With F j o i n t \mathcal{F}_{\mathrm{joint}} F joint denoting the tuple consisting of F 0 \mathcal{F}_0 F 0 , D Φ H D\Phi_H D Φ H , d σ \mathrm{d}\sigma d σ , U H \mathcal{U}_H U H , D Ψ D\Psi D Ψ , define the one-norm action by
S j o i n t : = 1 2 ∫ M ⟨ F j o i n t , h j o i n t ( σ ) F j o i n t ⟩ vol g e . (226) S_{\mathrm{joint}}:=\frac{1}{2}\int_M\langle\mathcal{F}_{\mathrm{joint}},h_{\mathrm{joint}}(\sigma)\mathcal{F}_{\mathrm{joint}}\rangle\,\operatorname{vol}_{g_{\mathrm{e}}}. \tag*{(226)} S joint := 2 1 ∫ M ⟨ F joint , h joint ( σ ) F joint ⟩ vol g e . (226)
The prefactor 1 / 2 1/2 1/2 is fixed by the master normalization. Let T T T embed the displayed joint tuple into the extended master tuple, including the reference-scale factors in (51). With T ∗ T^* T ∗ denoting the adjoint for the background fiber pairings, set
h j o i n t = 1 2 T ∗ h T . (227) h_{\mathrm{joint}}=\frac{1}{2}T^*hT. \tag*{(227)} h joint = 2 1 T ∗ h T . (227)
Thus 1 2 ⟨ U , h j o i n t U ⟩ = 1 4 ⟨ T U , h T U ⟩ \frac{1}{2}\langle U,h_{\mathrm{joint}}U\rangle=\frac{1}{4}\langle TU,hTU\rangle 2 1 ⟨ U , h joint U ⟩ = 4 1 ⟨ T U , h T U ⟩ . When the component scales coincide, this is one half of the restricted master pairing. Equation (227) is therefore the corresponding restriction of (52). The final matter term below is shorthand for the regularized fermionic block contraction contained in the same extended pairing; it is not an additional primitive action. The following displayed form fixes the relevant blocks and their normalization:
S j o i n t = 1 2 g c 2 ∫ M w ( σ ) [ ∑ a < b F c a b ∧ ⋆ e F c a b + ∑ a F c a 5 ∧ ⋆ e F c a 5 ] + 1 2 ∫ M w ( σ ) κ A B F A ∧ ⋆ e F B + 1 2 ∫ M [ ∣ D Φ H ∣ 2 + f σ 2 ∣ d σ ∣ 2 + ∣ U H ( Φ H , σ ) ∣ 2 ] vol g e + S m a t t e r [ Ψ ; A 0 , Φ H , σ , μ i n f o ] . (228) \mathcal{S}_{\mathrm{joint}} = \frac{1}{2g_c^2}\int_M w(\sigma)\left[\sum_{a<b}F_c^{ab}\wedge\star eF_c^{ab}+\sum_a F_c^{a5}\wedge\star eF_c^{a5}\right]
+\frac{1}{2}\int_M w(\sigma)\kappa_{AB}F^A\wedge\star eF^B+\frac{1}{2}\int_M\left[|D\Phi_H|^2+f_\sigma^2|d\sigma|^2+|\mathcal{U}_H(\Phi_H,\sigma)|^2\right]\operatorname{vol}_{g_e}
+\mathcal{S}_{\mathrm{matter}}[\Psi;A_0,\Phi_H,\sigma,\mu_{\mathrm{info}}]. \tag*{(228)} S joint = 2 g c 2 1 ∫ M w ( σ ) [ a < b ∑ F c ab ∧ ⋆ e F c ab + a ∑ F c a 5 ∧ ⋆ e F c a 5 ] + 2 1 ∫ M w ( σ ) κ A B F A ∧ ⋆ e F B + 2 1 ∫ M [ ∣ D Φ H ∣ 2 + f σ 2 ∣ d σ ∣ 2 + ∣ U H ( Φ H , σ ) ∣ 2 ] vol g e + S matter [ Ψ ; A 0 , Φ H , σ , μ info ] . (228)
Here F c a b = R a b − ℓ − 2 e a ∧ e b F_c^{ab}=R^{ab}-\ell^{-2}e^a\wedge e^b F c ab = R ab − ℓ − 2 e a ∧ e b and F c a 5 = ℓ − 1 T a F_c^{a5}=\ell^{-1}T^a F c a 5 = ℓ − 1 T a . The last line is included as a component of the extended norm after the fermionic regularization; it is not an independently added fourth action. The internal bilinear form κ A B \kappa_{AB} κ A B is positive on the Euclidean residual sector. The explicit choice
U H ( Φ H , σ ) = λ H ( Φ H † Φ H − v H 2 2 ) (229) \mathcal{U}_H(\Phi_H,\sigma)=\sqrt{\lambda_H}\left(\Phi_H^\dagger\Phi_H-\frac{v_H^2}{2}\right) \tag*{(229)} U H ( Φ H , σ ) = λ H ( Φ H † Φ H − 2 v H 2 ) (229)
with λ H > 0 \lambda_H>0 λ H > 0 produces a quartic component 1 2 ∣ U H ∣ 2 \frac{1}{2}|\mathcal{U}_H|^2 2 1 ∣ U H ∣ 2 bounded below. Here f σ > 0 f_\sigma>0 f σ > 0 is a reference mass scale for the dimensionless coordinate σ \sigma σ ; setting f σ = 1 f_\sigma=1 f σ = 1 is allowed only after declaring a fully nondimensionalized convention. The parameters λ H , v H \lambda_H,v_H λ H , v H are selected microscopic data at this stage.
The same σ \sigma σ prevents a decomposition into two independent functionals. For example, differentiation of (229) gives
∂ ∂ σ δ 2 S j o i n t δ F c δ F c = − α σ w ( σ ) 1 g c 2 , ∂ ∂ σ δ 2 S j o i n t δ F A δ F B = − α σ w ( σ ) κ A B . (230) \frac{\partial}{\partial\sigma}\frac{\delta^2\mathcal{S}_{\mathrm{joint}}}{\delta F_c\delta F_c}=-\alpha_\sigma w(\sigma)\frac{1}{g_c^2},\qquad\frac{\partial}{\partial\sigma}\frac{\delta^2\mathcal{S}_{\mathrm{joint}}}{\delta F^A\delta F^B}=-\alpha_\sigma w(\sigma)\kappa_{AB}. \tag*{(230)} ∂ σ ∂ δ F c δ F c δ 2 S joint = − α σ w ( σ ) g c 2 1 , ∂ σ ∂ δ F A δ F B δ 2 S joint = − α σ w ( σ ) κ A B . (230)
The displayed factors use the standard form-Hessian normalization in (229); fixed form-index contractions are understood. These derivatives would vanish for a sum of a gravitational functional, an internal functional, and a separate scalar functional. A field-dependent rescaling that removes w w w from one block transfers the coupling to the derivative terms, so it does not produce two independent stochastic systems. The candidate is therefore one coupled master system whenever α σ ≠ 0 \alpha_\sigma\ne0 α σ = 0 and the shared coordinate is retained. A full proof of irreducibility in a continuum realization still requires the no-factorization test in Definition 2 after gauge fixing and regularization.
The stochastic law uses one mobility on the complete coordinate vector. In the finite-dimensional affine realization of the candidate, let B Γ 0 \mathbf{B}_{\Gamma_0} B Γ 0 be the complete reversible Itô drift obtained from (61) by replacing the master action, the contravariant mobility, and the reference density with those of Γ 0 \Gamma_0 Γ 0 . With the field-space noise frame of (60), the law is
d X t = B Γ 0 ( X t ) d t + 2 θ M ^ Γ 0 , G 1 / 2 ( X t ) E X t d W t . (231) dX_t=\mathbf{B}_{\Gamma_0}(X_t)\,dt+\sqrt{2\theta}\,\widehat{M}_{\Gamma_0,G}^{1/2}(X_t)\mathcal{E}_{X_t}\,dW_t. \tag*{(231)} d X t = B Γ 0 ( X t ) d t + 2 θ M Γ 0 , G 1/2 ( X t ) E X t d W t . (231)
In these coordinates, if the reference measure is m ν , Γ 0 ( x ) d x m_{\nu,\Gamma_0}(x)\,\mathrm{d}x m ν , Γ 0 ( x ) d x , then
B Γ 0 i = − M Γ 0 i j ∂ j S j o i n t + θ m ν , Γ 0 − 1 ∂ j ( m ν , Γ 0 M Γ 0 i j ) . (232) \mathbb{B}_{\Gamma_0}^{i}=-M_{\Gamma_0}^{ij}\partial_j S_{\mathrm{joint}}+\theta m_{\nu,\Gamma_0}^{-1}\partial_j\left(m_{\nu,\Gamma_0}M_{\Gamma_0}^{ij}\right). \tag*{(232)} B Γ 0 i = − M Γ 0 ij ∂ j S joint + θ m ν , Γ 0 − 1 ∂ j ( m ν , Γ 0 M Γ 0 ij ) . (232)
The second term vanishes for constant mobility and a flat reference density; otherwise it is part of the drift that enters the path cost and the equilibrium equation. The composite noise coefficient has coordinate covariance 2 θ M Γ 0 2\theta M_{\Gamma_0} 2 θ M Γ 0 ; its Hilbert-space realization requires the admissibility conditions stated for (62) . The covariance is allowed to contain gravitational, internal, and scalar blocks. It must be symmetric and nonnegative; in a two-block regulator the condition is the Schur complement inequality, provided that M c M_c M c is strictly positive on the retained gravitational range:
M K − C ∗ M c − 1 C ⪰ 0. (233) M_K-C^*M_c^{-1}C\succeq0. \tag*{(233)} M K − C ∗ M c − 1 C ⪰ 0. (233)
If M c M_c M c is only nonnegative, the ordinary inverse in this display is replaced by the Moore–Penrose inverse and the corresponding range condition is imposed. Thus the action, the mobility, the Wiener process, and the path cost belong to the same specified model.
For completeness, the left-handed matter content of one anomaly-free generation may be chosen as
Q L = ( 3 , 2 ) 1 / 6 , u c = ( 3 ‾ , 1 ) − 2 / 3 , d c = ( 3 ‾ , 1 ) 1 / 3 , L L = ( 1 , 2 ) − 1 / 2 , e c = ( 1 , 1 ) 1 . (234) Q_L=(3,2)_{1/6},\quad u^c=(\overline{3},1)_{-2/3},\quad d^c=(\overline{3},1)_{1/3},\quad L_L=(1,2)_{-1/2},
\qquad e^c=(1,1)_1. \tag*{(234)} Q L = ( 3 , 2 ) 1/6 , u c = ( 3 , 1 ) − 2/3 , d c = ( 3 , 1 ) 1/3 , L L = ( 1 , 2 ) − 1/2 , e c = ( 1 , 1 ) 1 . (234)
with an optional neutral ν c = ( 1 , 1 ) 0 \nu^c=(1,1)_0 ν c = ( 1 , 1 ) 0 . The anomaly calculations use the left-handed convention. With T ( 3 ) = T ( 2 ) = 1 / 2 T(3)=T(2)=1/2 T ( 3 ) = T ( 2 ) = 1/2 ,
A S U ( 3 ) 3 = 2 − 1 − 1 = 0 , A S U ( 3 ) 2 U ( 1 ) = 2 ( 1 6 ) 1 2 − 3 ( 2 3 ) 1 2 + 3 ( 1 3 ) 1 2 = 0 , A S U ( 2 ) 2 U ( 1 ) = 3 ( 1 6 ) 1 2 − 1 ( 1 2 ) 1 2 = 0 , A g r a v 2 U ( 1 ) = 6 ( 1 6 ) + 3 ( − 2 3 ) + 3 ( 1 3 ) + 2 ( − 1 2 ) + 1 = 0 , A U ( 1 ) 3 = 6 ( 1 6 ) 3 + 3 ( − 2 3 ) 3 + 3 ( 1 3 ) 3 + 2 ( − 1 2 ) 3 + 1 = 0. (235) \begin{aligned}
\mathcal{A}_{SU(3)^3}&=2-1-1=0,\\
\mathcal{A}_{SU(3)^2U(1)}&=2\left(\frac{1}{6}\right)\frac{1}{2}-3\left(\frac{2}{3}\right)\frac{1}{2}+3\left(\frac{1}{3}\right)\frac{1}{2}=0,\\
\mathcal{A}_{SU(2)^2U(1)}&=3\left(\frac{1}{6}\right)\frac{1}{2}-1\left(\frac{1}{2}\right)\frac{1}{2}=0,\\
\mathcal{A}_{\mathrm{grav}^2U(1)}&=6\left(\frac{1}{6}\right)+3\left(-\frac{2}{3}\right)+3\left(\frac{1}{3}\right)+2\left(-\frac{1}{2}\right)+1=0,\\
\mathcal{A}_{U(1)^3}&=6\left(\frac{1}{6}\right)^3+3\left(-\frac{2}{3}\right)^3+3\left(\frac{1}{3}\right)^3+2\left(-\frac{1}{2}\right)^3+1=0. \tag*{(235)}
\end{aligned} A S U ( 3 ) 3 A S U ( 3 ) 2 U ( 1 ) A S U ( 2 ) 2 U ( 1 ) A grav 2 U ( 1 ) A U ( 1 ) 3 = 2 − 1 − 1 = 0 , = 2 ( 6 1 ) 2 1 − 3 ( 3 2 ) 2 1 + 3 ( 3 1 ) 2 1 = 0 , = 3 ( 6 1 ) 2 1 − 1 ( 2 1 ) 2 1 = 0 , = 6 ( 6 1 ) + 3 ( − 3 2 ) + 3 ( 3 1 ) + 2 ( − 2 1 ) + 1 = 0 , = 6 ( 6 1 ) 3 + 3 ( − 3 2 ) 3 + 3 ( 3 1 ) 3 + 2 ( − 2 1 ) 3 + 1 = 0. (235)
There are four S U ( 2 ) SU(2) S U ( 2 ) doublets after color multiplicity is counted, so the global Witten anomaly is absent. This verifies consistency of the displayed residual representation; it does not derive the representation from the master algebra.
Suppose a homogeneous value σ ∗ \sigma_* σ ∗ is selected by a specified stabilization or boundary condition, and write w ∗ = w ( σ ∗ ) w_*=w(\sigma_*) w ∗ = w ( σ ∗ ) . On this constant-scalar, torsionless branch, the explicit Cartan expansion and the internal kinetic term give, at tree level,
1 16 π G e f f = w ∗ 2 g c 2 ℓ 2 , g a − 2 = w ∗ κ a , α 3 = w ∗ 8 g c 2 . (236) \frac{1}{16\pi G_{\mathrm{eff}}}=\frac{w_*}{2g_c^2\ell^2},\qquad g_a^{-2}=w_*\kappa_a,\qquad\alpha_3=\frac{w_*}{8g_c^2}. \tag*{(236)} 16 π G eff 1 = 2 g c 2 ℓ 2 w ∗ , g a − 2 = w ∗ κ a , α 3 = 8 g c 2 w ∗ . (236)
These equalities identify coefficients in a restricted homogeneous background; they are not the field equations for an arbitrary varying σ \sigma σ . In particular, when d σ ≠ 0 \mathrm{d}\sigma\ne0 d σ = 0 , (199) supplies a nonzero connection response and the coframe equation contains scalar-derivative terms. Consequently,
G e f f g a − 2 = g c 2 ℓ 2 κ a 8 π , 64 π G e f f Λ E α 3 = 3 , Λ E = 3 ℓ 2 . (237) G_{\mathrm{eff}}g_a^{-2}=\frac{g_c^2\ell^2\kappa_a}{8\pi},\qquad64\pi G_{\mathrm{eff}}\Lambda_E\alpha_3=3,\qquad\Lambda_E=\frac{3}{\ell^2}. \tag*{(237)} G eff g a − 2 = 8 π g c 2 ℓ 2 κ a , 64 π G eff Λ E α 3 = 3 , Λ E = ℓ 2 3 . (237)
The first relation is independent of the shared order-parameter value σ ∗ \sigma_* σ ∗ at tree level. It becomes a physical prediction only if g c , ℓ , κ a g_c,\ell,\kappa_a g c , ℓ , κ a are fixed by the same selection problem and if loop, threshold, and basis corrections are computed. Writing the one-loop and threshold terms as Δ a \Delta_a Δ a and Δ G \Delta_G Δ G , the measured quantities obey instead
g a − 2 = w ∗ κ a + Δ a , G e f f − 1 = 8 π w ∗ g c 2 ℓ 2 + Δ G . (238) g_a^{-2}=w_*\kappa_a+\Delta_a,\qquad G_{\mathrm{eff}}^{-1}=\frac{8\pi w_*}{g_c^2\ell^2}+\Delta_G. \tag*{(238)} g a − 2 = w ∗ κ a + Δ a , G eff − 1 = g c 2 ℓ 2 8 π w ∗ + Δ G . (238)
These correction terms are part of the calculation required for a numerical prediction.
Section 14.39 completes a concrete orbital recovery test on the specified torsionless, scalar-decoupled stationary sector. It retains the curvature-square coefficient inherited from the Cartan norm, verifies its exact cancellation on the constructed Einstein exterior, and derives the test-body orbit from the same matter metric. The zero-vacuum-energy matching gives the standard periapsis advance; the unmodified Cartan constant gives an additional cosmological term. Thus the matching assumptions remain visible in the observable itself.
The internal group and representation in (221) and (234) are explicit inputs of this candidate witness. To claim that they are predicted, the minimization over a candidate class must select this Γ 0 \Gamma_0 Γ 0 uniquely up to gauge and field redefinitions. The candidate therefore proves existence of a single well-typed coupled realization; it does not by itself prove uniqueness, ultraviolet completeness, or a parameter-free Standard Model.
The scalar equation and confinement of its constant mode
The common positive weight couples the curvature blocks, but positivity alone does not confine the shared scalar. This point is relevant both to the classical vacuum and to the equilibrium construction. Work first on a smooth connected closed Riemannian four-manifold, keep the coframe fixed while varying σ \sigma σ , and consider the bosonic action with the auxiliary map (229) . Define the nonnegative curvature density
K ( e , ω , A ) : = 1 2 g c 2 ∑ I < J ∣ F c I J ∣ g e 2 + 1 2 κ A B ⟨ F A , F B ⟩ g e . (239) \mathcal{K}(e,\omega,A):=\frac{1}{2g_c^2}\sum_{I<J}\left|F_c^{IJ}\right|_{g_e}^2+\frac{1}{2}\kappa_{AB}\langle F^A,F^B\rangle_{g_e}. \tag*{(239)} K ( e , ω , A ) := 2 g c 2 1 I < J ∑ F c I J g e 2 + 2 1 κ A B ⟨ F A , F B ⟩ g e . (239)
Thus the part depending on σ \sigma σ is ∫ M [ w ( σ ) K + 1 2 f σ 2 ∣ d σ ∣ 2 ] v o l g e \int_M [w(\sigma)\mathcal{K}+\frac{1}{2}f_\sigma^2|d\sigma|^2]\,\mathrm{vol}_{g_e} ∫ M [ w ( σ ) K + 2 1 f σ 2 ∣ d σ ∣ 2 ] vol g e . For a real test function η \eta η , the two variations are
δ η ∫ M w ( σ ) K v o l g e = − α σ ∫ M η w ( σ ) K v o l g e , δ η ∫ M 1 2 f σ 2 ∣ d σ ∣ 2 v o l g e = − f σ 2 ∫ M η Δ g σ v o l g e . (240) \begin{aligned}
\delta_\eta\int_M w(\sigma)\mathcal{K}\,\mathrm{vol}_{g_e} &= -\alpha_\sigma\int_M\eta w(\sigma)\mathcal{K}\,\mathrm{vol}_{g_e},\\
\delta_\eta\int_M \frac{1}{2}f_\sigma^2|d\sigma|^2\,\mathrm{vol}_{g_e} &= -f_\sigma^2\int_M\eta\Delta_{g}\sigma\,\mathrm{vol}_{g_e}. \tag*{(240)}
\end{aligned} δ η ∫ M w ( σ ) K vol g e δ η ∫ M 2 1 f σ 2 ∣ d σ ∣ 2 vol g e = − α σ ∫ M η w ( σ ) K vol g e , = − f σ 2 ∫ M η Δ g σ vol g e . (240)
There is no derivative of (229) with respect to σ \sigma σ . Therefore the scalar Euler equation is
− f σ 2 Δ g σ − α σ w ( σ ) K = 0. (241) -f_\sigma^2\Delta_g\sigma-\alpha_\sigma w(\sigma)\mathcal{K}=0. \tag*{(241)} − f σ 2 Δ g σ − α σ w ( σ ) K = 0. (241)
Integration over M M M removes the Laplacian. If α σ ≠ 0 \alpha_\sigma\ne0 α σ = 0 , it gives ∫ M w ( σ ) K v o l g e = 0 \int_M w(\sigma)\mathcal{K}\,\mathrm{vol}_{g_e}=0 ∫ M w ( σ ) K vol g e = 0 . Since w > 0 w>0 w > 0 and K ≥ 0 \mathcal{K}\ge0 K ≥ 0 , smoothness implies K = 0 \mathcal{K}=0 K = 0 everywhere. Positive definiteness of the curvature pairings then gives F c = 0 F_c=0 F c = 0 and F = 0 F=0 F = 0 . Equation (241) reduces to Δ g σ = 0 \Delta_g\sigma=0 Δ g σ = 0 ; multiplication by σ \sigma σ and integration by parts give ∫ M ∣ d σ ∣ 2 v o l g e = 0 \int_M|d\sigma|^2\,\mathrm{vol}_{g_e}=0 ∫ M ∣ d σ ∣ 2 vol g e = 0 , so σ \sigma σ is constant. A nonzero scalar-dependent matter source, a boundary flux, or an additional scalar potential changes this argument and must appear explicitly in the scalar equation. A prescribed constant σ ∗ \sigma_\ast σ ∗ is consequently not an arbitrary solution with nonzero master-curvature density K \mathcal{K} K in the unstabilized closed bosonic model. This does not force the Riemann curvature of the coframe metric to vanish: F c = 0 F_c=0 F c = 0 permits R a b = ℓ − 2 e a ∧ e b R^a{}_b=\ell^{-2}e^a\wedge e^b R a b = ℓ − 2 e a ∧ e b , as on the round-sphere branch.
The same omission obstructs the simplest Gibbs construction. In a regulator that retains the constant scalar mode, write σ = s + σ ⊥ \sigma=s+\sigma_\perp σ = s + σ ⊥ , with ∫ M σ ⊥ v o l g e = 0 \int_M\sigma_\perp\,\mathrm{vol}_{g_e}=0 ∫ M σ ⊥ vol g e = 0 . At fixed values of the other regulated coordinates,
S j o i n t ( s ) = e − α σ s C + S 0 , C : = ∫ M e − α σ σ ⊥ K v o l g e ≥ 0 , (242) S_{\mathrm{joint}}(s)=e^{-\alpha_\sigma s}C+S_0,\qquad C:=\int_M e^{-\alpha_\sigma\sigma_\perp}\mathcal{K}\,\mathrm{vol}_{g_e}\ge0, \tag*{(242)} S joint ( s ) = e − α σ s C + S 0 , C := ∫ M e − α σ σ ⊥ K vol g e ≥ 0 , (242)
where S 0 S_0 S 0 is independent of s s s in this bosonic calculation. For α σ ≠ 0 \alpha_\sigma\ne0 α σ = 0 , along α σ s → + ∞ \alpha_\sigma s\to+\infty α σ s → + ∞ , the Gibbs integrand tends to exp ( − S 0 / θ ) > 0 \exp(-S_0/\theta)>0 exp ( − S 0 / θ ) > 0 . Hence its integral over the Lebesgue variable s s s diverges. For α σ = 0 \alpha_\sigma=0 α σ = 0 it is independent of s s s and diverges as well. With a product reference measure, Tonelli’s theorem extends this divergence to the full regulated partition function whenever the remaining finite-action coordinates have positive measure. This conclusion is specific to the stated unconfined reference measure; a prescribed confining reference density is additional model data.
An explicit bosonic modification within the same extended-norm construction is to enlarge its algebraic auxiliary fiber and replace the displayed map by
H H s t a b ( Φ H , σ ) : = ( λ H ( Φ H † Φ H − v H 2 / 2 ) m σ f σ ( σ − σ 0 ) ) , m σ > 0. (243) \mathcal{H}^{\mathrm{stab}}_H(\Phi_H,\sigma):=
\begin{pmatrix}
\sqrt{\lambda_H}(\Phi_H^\dagger\Phi_H-v_H^2/2)\\
m_\sigma f_\sigma(\sigma-\sigma_0)
\end{pmatrix},
\qquad m_\sigma>0. \tag*{(243)} H H stab ( Φ H , σ ) := ( λ H ( Φ H † Φ H − v H 2 /2 ) m σ f σ ( σ − σ 0 ) ) , m σ > 0. (243)
Both entries have mass dimension two; m σ m_\sigma m σ and f σ f_\sigma f σ have mass dimension one, and σ 0 \sigma_0 σ 0 is dimensionless. This declares new microscopic data rather than inferring a potential from the weight w w w . The same single squared norm now contains 1 2 m σ 2 f σ 2 ( σ − σ 0 ) 2 \frac{1}{2}m_{\sigma}^{2}f_{\sigma}^{2}(\sigma-\sigma_{0})^{2} 2 1 m σ 2 f σ 2 ( σ − σ 0 ) 2 , and the scalar equation becomes
− f σ 2 Δ g σ − α σ w ( σ ) K + m σ 2 f σ 2 ( σ − σ 0 ) = 0. (244) -f_{\sigma}^{2}\Delta_{g}\sigma-\alpha_{\sigma}w(\sigma)\mathcal{K}+m_{\sigma}^{2}f_{\sigma}^{2}(\sigma-\sigma_{0})=0. \tag*{(244)} − f σ 2 Δ g σ − α σ w ( σ ) K + m σ 2 f σ 2 ( σ − σ 0 ) = 0. (244)
This additional force can balance the curvature source. It does not prove existence of a solution of the coupled coframe and connection equations.
Theorem 11 (Fixed-background stabilized scalar). Let ( M , g ) (M,g) ( M , g ) be a smooth connected closed Riemannian manifold, let K ∈ C ∞ ( M ) \mathcal{K}\in C^{\infty}(M) K ∈ C ∞ ( M ) be nonnegative, and let f σ , m σ > 0 f_{\sigma},m_{\sigma}>0 f σ , m σ > 0 , α σ , σ 0 ∈ R \alpha_{\sigma},\sigma_{0}\in\mathbb{R} α σ , σ 0 ∈ R be fixed. With Δ g = div g grad g \Delta_{g}=\operatorname{div}_{g}\operatorname{grad}_{g} Δ g = div g grad g , equation (244) has exactly one smooth solution. It is the unique finite-energy minimizer of the fixed-background scalar action
J σ [ σ ] = ∫ M [ f σ 2 2 ∣ ∇ σ ∣ 2 + m σ 2 f σ 2 2 ( σ − σ 0 ) 2 + K e − α σ σ ] vol g . (245) J_{\sigma}[\sigma]=\int_{M}\left[\frac{f_{\sigma}^{2}}{2}|\nabla\sigma|^{2}+\frac{m_{\sigma}^{2}f_{\sigma}^{2}}{2}(\sigma-\sigma_{0})^{2}+\mathcal{K}e^{-\alpha_{\sigma}\sigma}\right]\operatorname{vol}_{g}. \tag*{(245)} J σ [ σ ] = ∫ M [ 2 f σ 2 ∣∇ σ ∣ 2 + 2 m σ 2 f σ 2 ( σ − σ 0 ) 2 + K e − α σ σ ] vol g . (245) For α σ = 0 \alpha_{\sigma}=0 α σ = 0 the solution is σ = σ 0 \sigma=\sigma_{0} σ = σ 0 . For α σ ≠ 0 \alpha_{\sigma}\ne0 α σ = 0 , put
u = α σ ( σ − σ 0 ) , a = α σ 2 e − α σ σ 0 f σ 2 K , ( − Δ g + m σ 2 ) u = a e − u . (246) u=\alpha_{\sigma}(\sigma-\sigma_{0}),\qquad a=\frac{\alpha_{\sigma}^{2}e^{-\alpha_{\sigma}\sigma_{0}}}{f_{\sigma}^{2}}\mathcal{K},\qquad(-\Delta_{g}+m_{\sigma}^{2})u=ae^{-u}. \tag*{(246)} u = α σ ( σ − σ 0 ) , a = f σ 2 α σ 2 e − α σ σ 0 K , ( − Δ g + m σ 2 ) u = a e − u . (246) Then
0 ≤ u ≤ b , b e b = ∥ a ∥ ∞ m σ 2 , b ≥ 0. (247) 0\le u\le b,\qquad be^{b}=\frac{\lVert a\rVert_{\infty}}{m_{\sigma}^{2}},\qquad b\ge0. \tag*{(247)} 0 ≤ u ≤ b , b e b = m σ 2 ∥ a ∥ ∞ , b ≥ 0. (247) Equivalently b = W 0 ( ∥ a ∥ ∞ / m σ 2 ) b=W_{0}(\lVert a\rVert_{\infty}/m_{\sigma}^{2}) b = W 0 (∥ a ∥ ∞ / m σ 2 ) , where W 0 W_{0} W 0 is the nonnegative inverse of z ↦ z e z z\mapsto ze^{z} z ↦ z e z . If K ≢ 0 \mathcal{K}\not\equiv0 K ≡ 0 , then u > 0 u>0 u > 0 . For α σ ≠ 0 \alpha_{\sigma}\ne0 α σ = 0 , the solution is constant if and only if K \mathcal{K} K is constant.
Proof. The case α σ = 0 \alpha_{\sigma}=0 α σ = 0 follows by testing ( − Δ g + m σ 2 ) ( σ − σ 0 ) = 0 (-\Delta_{g}+m_{\sigma}^{2})(\sigma-\sigma_{0})=0 ( − Δ g + m σ 2 ) ( σ − σ 0 ) = 0 with σ − σ 0 \sigma-\sigma_{0} σ − σ 0 . Suppose α σ ≠ 0 \alpha_{\sigma}\ne0 α σ = 0 . The substitution in (246) gives the stated equation. Define on H 1 ( M ) H^{1}(M) H 1 ( M ) the extended nonnegative functional
E [ u ] = ∫ M [ 1 2 ∣ ∇ u ∣ 2 + m σ 2 2 u 2 + a e − u ] vol g . (248) E[u]=\int_{M}\left[\frac{1}{2}|\nabla u|^{2}+\frac{m_{\sigma}^{2}}{2}u^{2}+ae^{-u}\right]\operatorname{vol}_{g}. \tag*{(248)} E [ u ] = ∫ M [ 2 1 ∣∇ u ∣ 2 + 2 m σ 2 u 2 + a e − u ] vol g . (248) It is finite at zero and coercive in H 1 ( M ) H^{1}(M) H 1 ( M ) . For ϕ x ( t ) = m σ 2 t 2 / 2 + a ( x ) e − t \phi_{x}(t)=m_{\sigma}^{2}t^{2}/2+a(x)e^{-t} ϕ x ( t ) = m σ 2 t 2 /2 + a ( x ) e − t , one has ϕ x ′ ( t ) < 0 \phi_{x}'(t)<0 ϕ x ′ ( t ) < 0 for t < 0 t<0 t < 0 , while for t > b t>b t > b ,
ϕ x ′ ( t ) = m σ 2 t − a ( x ) e − t ≥ m σ 2 b − ∥ a ∥ ∞ e − b = 0. (249) \phi_{x}'(t)=m_{\sigma}^{2}t-a(x)e^{-t}\ge m_{\sigma}^{2}b-\lVert a\rVert_{\infty}e^{-b}=0. \tag*{(249)} ϕ x ′ ( t ) = m σ 2 t − a ( x ) e − t ≥ m σ 2 b − ∥ a ∥ ∞ e − b = 0. (249) The truncation T ( t ) = min { b , max { 0 , t } } T(t)=\min\{b,\max\{0,t\}\} T ( t ) = min { b , max { 0 , t }} therefore decreases the potential pointwise and, being 1-Lipschitz, does not increase the gradient energy. Replace a minimizing sequence by its truncations. Weak compactness in H 1 H^{1} H 1 , compactness in L 2 L^{2} L 2 , and passage to an almost-everywhere convergent subsequence give a limit u u u with 0 ≤ u ≤ b 0\le u\le b 0 ≤ u ≤ b . Weak lower semicontinuity of the quadratic energy and Fatou’s lemma for the exponential term show that u u u minimizes E E E . Since truncation decreases E E E on its entire domain, this is an unrestricted minimizer.
For every smooth test function η \eta η , boundedness of u u u and η \eta η justifies differentiating E [ u + t η ] E[u+t\eta] E [ u + t η ] under the integral. Its first variation is
∫ M [ ⟨ ∇ u , ∇ η ⟩ + m σ 2 u η − a e − u η ] vol g = 0. (250) \int_M \left[\langle\nabla u,\nabla\eta\rangle+ m_\sigma^2u\eta-ae^{-u}\eta\right]\operatorname{vol}_g=0. \tag*{(250)} ∫ M [ ⟨ ∇ u , ∇ η ⟩ + m σ 2 u η − a e − u η ] vol g = 0. (250)
The right-hand side of the weak equation is bounded. Elliptic regularity first gives u ∈ W 2 , p ( M ) u\in W^{2,p}(M) u ∈ W 2 , p ( M ) for every finite p p p , and then smoothness by Sobolev embedding and bootstrapping. If u , v u,v u , v are two smooth solutions, subtract their equations and test with u − v u-v u − v to obtain
∫ M [ ∣ ∇ ( u − v ) ∣ 2 + m σ 2 ( u − v ) 2 ] vol g = ∫ M a ( e − u − e − v ) ( u − v ) vol g ≤ 0. (251) \int_M\left[|\nabla(u-v)|^2+m_\sigma^2(u-v)^2\right]\operatorname{vol}_g=\int_Ma(e^{-u}-e^{-v})(u-v)\operatorname{vol}_g\leq0. \tag*{(251)} ∫ M [ ∣∇ ( u − v ) ∣ 2 + m σ 2 ( u − v ) 2 ] vol g = ∫ M a ( e − u − e − v ) ( u − v ) vol g ≤ 0. (251)
Thus u = v u=v u = v . Strict convexity gives uniqueness of the energy minimizer, and J σ = ( f σ 2 / α σ 2 ) E J_\sigma=(f_\sigma^2/\alpha_\sigma^2)E J σ = ( f σ 2 / α σ 2 ) E under the substitution. The strong maximum principle gives u > 0 u>0 u > 0 if a ≢ 0 a\not\equiv0 a ≡ 0 . Finally a constant u u u satisfies a ( x ) = m σ 2 u e u a(x)=m_\sigma^2ue^u a ( x ) = m σ 2 u e u , so a a a must be constant. Conversely, constant a a a has the unique constant solution u = W 0 ( a / m σ 2 ) u=W_0(a/m_\sigma^2) u = W 0 ( a / m σ 2 ) . □
The fixed-background second variation at this solution is
δ 2 J σ [ η , η ] = ∫ M [ f σ 2 ∣ ∇ η ∣ 2 + ( m σ 2 f σ 2 + α σ 2 K e − α σ σ ∗ ) η 2 ] vol g ≥ m σ 2 f σ 2 ∥ η ∥ L 2 ( g ) 2 . (252) \begin{aligned}
\delta^2J_\sigma[\eta,\eta]=\int_M\left[f_\sigma^2|\nabla\eta|^2+\left(m_\sigma^2f_\sigma^2+\alpha_\sigma^2\mathcal{K}e^{-\alpha_\sigma\sigma_*}\right)\eta^2\right]\operatorname{vol}_g \\
&\geq m_\sigma^2f_\sigma^2\|\eta\|_{L^2(g)}^2. \tag*{(252)}
\end{aligned} δ 2 J σ [ η , η ] = ∫ M [ f σ 2 ∣∇ η ∣ 2 + ( m σ 2 f σ 2 + α σ 2 K e − α σ σ ∗ ) η 2 ] vol g ≥ m σ 2 f σ 2 ∥ η ∥ L 2 ( g ) 2 . (252)
Hence the canonically normalized scalar block has operator
H σ = − Δ g + m σ 2 + α σ 2 K e − α σ σ ∗ f σ 2 ⪰ m σ 2 I . (253) H_\sigma=-\Delta_g+m_\sigma^2+\frac{\alpha_\sigma^2\mathcal{K}e^{-\alpha_\sigma\sigma_*}}{f_\sigma^2}\succeq m_\sigma^2I. \tag*{(253)} H σ = − Δ g + m σ 2 + f σ 2 α σ 2 K e − α σ σ ∗ ⪰ m σ 2 I . (253)
For constant K \mathcal{K} K and α σ ≠ 0 \alpha_\sigma\neq0 α σ = 0 , the squared scalar mass in this block is m σ 2 ( 1 + u ∗ ) m_\sigma^2(1+u_*) m σ 2 ( 1 + u ∗ ) . For variable K \mathcal{K} K , the solution and the added multiplication potential are generally spatially variable; a prescribed constant σ ∗ \sigma_* σ ∗ is then not the exact scalar solution. These conclusions concern the scalar at fixed coframe and connection. They do not establish a solution of the coupled field equations or a lower bound for the full Hessian after mixing with the other fields.
For the constant mode, nonnegativity of the other bosonic norm components gives a direct integrability estimate. Let V g : = ∫ M vol g e > 0 V_g:=\int_M\operatorname{vol}_{g_e}>0 V g := ∫ M vol g e > 0 . The zero-mean property of σ ⊥ \sigma_\perp σ ⊥ removes the mixed term in the potential, and consequently
S j o i n t s t a b ≥ m σ 2 f σ 2 2 [ V g ( s − σ 0 ) 2 + ∥ σ ⊥ ∥ L 2 ( g e ) 2 ] , S_{\mathrm{joint}}^{\mathrm{stab}}\geq\frac{m_\sigma^2f_\sigma^2}{2}\left[V_g(s-\sigma_0)^2+\|\sigma_\perp\|_{L^2(g_e)}^2\right], S joint stab ≥ 2 m σ 2 f σ 2 [ V g ( s − σ 0 ) 2 + ∥ σ ⊥ ∥ L 2 ( g e ) 2 ] ,
∫ R e − S j o i n t s t a b / θ d s ≤ ( 2 π θ m σ 2 f σ 2 V g ) 1 / 2 exp [ − m σ 2 f σ 2 2 θ ∥ σ ⊥ ∥ L 2 ( g e ) 2 ] . (254) \int_{\mathbb{R}}e^{-S_{\mathrm{joint}}^{\mathrm{stab}}/\theta}\,ds\leq\left(\frac{2\pi\theta}{m_\sigma^2f_\sigma^2V_g}\right)^{1/2}\exp\left[-\frac{m_\sigma^2f_\sigma^2}{2\theta}\|\sigma_\perp\|_{L^2(g_e)}^2\right]. \tag*{(254)} ∫ R e − S joint stab / θ d s ≤ ( m σ 2 f σ 2 V g 2 π θ ) 1/2 exp [ − 2 θ m σ 2 f σ 2 ∥ σ ⊥ ∥ L 2 ( g e ) 2 ] . (254)
Thus the modification removes this particular scalar zero-mode divergence at fixed geometry. Controlling the gauge quotient, degenerate coframes, the remaining field coordinates, a fermion determinant, and the continuum limit still requires separate estimates. In particular, (255) is not a construction of a four-dimensional interacting field measure. The tree-level curvature identities (237) are unchanged at a prescribed constant scalar value. If that value differs from σ 0 \sigma_0 σ 0 , the new potential contributes to the vacuum energy; the cosmological relation (238) then refers only to the Cartan curvature contribution and must be corrected for this added energy.
Inertia, gravitation, and the absence of ordinary material shielding.
The controlled recovery theorem identifies the low-energy gravitational field equation under its stated matching hypotheses. We now distinguish its physical interpretation from the stronger, still conditional claim that a microscopic equilibrium principle has derived the origin of the selected geometry. Let e ∗ a μ e_*^{a}{}_{\mu} e ∗ a μ be a non-degenerate Lorentzian coframe on the selected low-energy branch, and define
g μ ν ∗ : = η a b e ∗ a μ e ∗ b ν . (255) g^*_{\mu\nu} := \eta_{ab}e_*^{a}{}_{\mu}e_*^{b}{}_{\nu}. \tag*{(255)} g μν ∗ := η ab e ∗ a μ e ∗ b ν . (255)
The asterisk records the branch under discussion. Equation (255) is a coframe representation of its metric; by itself it does not prove that a global equilibrium problem has selected e ∗ e_* e ∗ from all admissible coframes.
Universal coupling and free motion.
Assume that ordinary low-energy matter fields ψ A \psi^A ψ A have the covariant action
S m [ g ∗ , ψ ] = ∫ M L m ( g ∗ , ψ , ∇ ( g ∗ ) ψ ) − g ∗ d 4 x , (256) S_{\mathrm{m}}[g^*,\psi] = \int_M \mathcal{L}_{\mathrm{m}}(g^*,\psi,\nabla^{(g^*)}\psi)\sqrt{-g^*}\,\mathrm{d}^4x, \tag*{(256)} S m [ g ∗ , ψ ] = ∫ M L m ( g ∗ , ψ , ∇ ( g ∗ ) ψ ) − g ∗ d 4 x , (256)
with no fixed, non-covariant background that exchanges energy–momentum with the matter sector. Equivalently, one may use S m [ e ∗ , ω ∗ , ψ ] S_{\mathrm{m}}[e_*,\omega_*,\psi] S m [ e ∗ , ω ∗ , ψ ] on the torsionless branch, on which ω ∗ \omega_* ω ∗ is the Levi–Civita spin connection of e ∗ e_* e ∗ . Fix the stress-tensor convention by
δ S m = 1 2 ∫ M − g ∗ T μ ν δ g μ ν ∗ d 4 x + ∫ M E A δ ψ A d 4 x . (257) \delta S_{\mathrm{m}} = \frac{1}{2}\int_M \sqrt{-g^*}\,T^{\mu\nu}\,\delta g^*_{\mu\nu}\,\mathrm{d}^4x + \int_M E_A\,\delta\psi^A\,\mathrm{d}^4x. \tag*{(257)} δ S m = 2 1 ∫ M − g ∗ T μν δ g μν ∗ d 4 x + ∫ M E A δ ψ A d 4 x . (257)
For a compactly supported infinitesimal diffeomorphism generated by ζ μ \zeta^\mu ζ μ , covariance gives δ ζ g μ ν ∗ = 2 ∇ ( μ ( g ∗ ) ζ ν ) \delta_\zeta g^*_{\mu\nu}=2\nabla^{(g^*)}_{(\mu}\zeta_{\nu)} δ ζ g μν ∗ = 2 ∇ ( μ ( g ∗ ) ζ ν ) and δ ζ ψ A = L ζ ψ A \delta_\zeta\psi^A=\mathcal{L}_\zeta\psi^A δ ζ ψ A = L ζ ψ A . Thus, when the matter equations E A = 0 E_A=0 E A = 0 hold, integration by parts in (258) gives
∇ μ ( g ∗ ) T μ ν = 0. (258) \nabla^{(g^*)}_{\mu}T^{\mu\nu}=0. \tag*{(258)} ∇ μ ( g ∗ ) T μν = 0. (258)
The qualification concerning fixed backgrounds is necessary: a non-dynamical datum that is neither varied on shell nor transformed as a diffeomorphism-covariant spurion contributes an exchange term to (259). This is precisely the Noether-balance condition appearing in Theorem 9.
For pressureless, spinless matter, set
T μ ν = ρ u μ u ν , J μ = ρ u μ , ∇ μ ( g ∗ ) J μ = 0 , (259) T^{\mu\nu}=\rho u^\mu u^\nu,\qquad J^\mu=\rho u^\mu,\qquad\nabla^{(g^*)}_{\mu}J^\mu=0, \tag*{(259)} T μν = ρ u μ u ν , J μ = ρ u μ , ∇ μ ( g ∗ ) J μ = 0 , (259)
where u μ u μ = − 1 u^\mu u_\mu=-1 u μ u μ = − 1 and ρ > 0 \rho>0 ρ > 0 on the support of the dust. Equations (259) and (260) yield
0 = ∇ μ ( g ∗ ) ( ρ u μ u ν ) = u ν ∇ μ ( g ∗ ) ( ρ u μ ) + ρ u μ ∇ μ ( g ∗ ) u ν = ρ u μ ∇ μ ( g ∗ ) u ν . (260) 0=\nabla_\mu^{(g^*)}(\rho u^\mu u^\nu)=u^\nu\nabla_\mu^{(g^*)}(\rho u^\mu)+\rho u^\mu\nabla_\mu^{(g^*)}u^\nu=\rho u^\mu\nabla_\mu^{(g^*)}u^\nu. \tag*{(260)} 0 = ∇ μ ( g ∗ ) ( ρ u μ u ν ) = u ν ∇ μ ( g ∗ ) ( ρ u μ ) + ρ u μ ∇ μ ( g ∗ ) u ν = ρ u μ ∇ μ ( g ∗ ) u ν . (260)
Hence a freely falling dust element obeys
u μ ∇ μ ( g ∗ ) u ν = 0. (261) u^\mu\nabla_\mu^{(g^*)}u^\nu=0. \tag*{(261)} u μ ∇ μ ( g ∗ ) u ν = 0. (261)
This is the standard test-body limit of universally coupled matter [63] and [158]. Spinning, extended, pressured, or non-minimally coupled bodies can acquire additional multipole, spin-curvature, pressure, or direct-curvature forces, so (262) is not a trajectory equation for every possible material model.
In this precise low-energy sense, inertia and gravitation are represented by one geometry. Inertia is free propagation relative to the connection of g ∗ g^* g ∗ , whereas the invariant tidal content of gravitation is curvature of the same connection. For neighboring geodesics with tangent u μ u^\mu u μ and separation ξ μ \xi^\mu ξ μ ,
D 2 ξ μ D τ 2 = − R μ α ν β [ g ∗ ] u α ξ ν u β . (262) \frac{D^2\xi^\mu}{D\tau^2}=-R^\mu{}_{\alpha\nu\beta}[g^*]u^\alpha\xi^\nu u^\beta. \tag*{(262)} D τ 2 D 2 ξ μ = − R μ α ν β [ g ∗ ] u α ξ ν u β . (262)
Thus vanishing coordinate acceleration in a local inertial frame does not remove the tidal observable in (263). On the controlled low-curvature branch, this curvature is governed by (177), which arises from the coframe component of the master Euler system.
Spectral mass, inertial propagation, and gravitational response
The preceding dust argument starts from a stress tensor and therefore does not yet identify the origin of the mass that appears in that tensor. The effective-action formulation supplies the missing interface. Let ϕ \phi ϕ denote collectively the physical matter and order-parameter coordinates after gauge fixing, removal of constraints, and passage to the physical quotient. At a selected stationary background ( e ∗ , ϕ ∗ ) (e_*,\phi_*) ( e ∗ , ϕ ∗ ) , write η : = ϕ − ϕ ∗ \eta:=\phi-\phi_* η := ϕ − ϕ ∗ . The renormalized effective action has the local expansion
δ Γ e f f δ ϕ ∣ ( e ∗ , ϕ ∗ ) = 0 , Γ e f f [ e ∗ , ϕ ∗ + η ] = Γ ∗ + 1 2 ⟨ η , K ∗ η ⟩ + O ( ∥ η ∥ 3 ) , (263) \left.\frac{\delta\Gamma_{\mathrm{eff}}}{\delta\phi}\right|_{(e_*,\phi_*)}=0,\qquad\Gamma_{\mathrm{eff}}[e_*,\phi_*+\eta]=\Gamma_*+\frac{1}{2}\langle\eta,\mathcal{K}_*\eta\rangle+O(\lVert\eta\rVert^3), \tag*{(263)} δ ϕ δ Γ eff ( e ∗ , ϕ ∗ ) = 0 , Γ eff [ e ∗ , ϕ ∗ + η ] = Γ ∗ + 2 1 ⟨ η , K ∗ η ⟩ + O (∥ η ∥ 3 ) , (263)
K ∗ : = δ 2 Γ e f f p h y s δ ϕ δ ϕ ∣ ( e ∗ , ϕ ∗ ) . \mathcal{K}_*:=\left.\frac{\delta^2\Gamma_{\mathrm{eff}}^{\mathrm{phys}}}{\delta\phi\,\delta\phi}\right|_{(e_*,\phi_*)}. K ∗ := δ ϕ δ ϕ δ 2 Γ eff phys ( e ∗ , ϕ ∗ ) .
The superscript phys is essential. Before the gauge and constraint quotients, K ∗ \mathcal{K}_* K ∗ contains gauge zero modes, and its raw eigenvalues are not particle masses. Even after this reduction, K ∗ \mathcal{K}_* K ∗ is a spacetime two-point operator: its eigenvalues generally include kinetic momentum and boundary spectra. For a bosonic second-order channel in a local derivative expansion, separate the positive kinetic form K ∗ k i n K_*^{\mathrm{kin}} K ∗ kin from the zero-momentum potential Hessian V ∗ \mathcal{V}_* V ∗ . Curvature masses are then defined by
V ∗ v j = m H , j 2 K ∗ k i n v j , K ∗ k i n ≻ 0 , V ∗ : = ∂ 2 V e f f ∂ ϕ ∂ ϕ ∣ ϕ ∗ p h y s . (264) \mathcal{V}_*v_j=m_{H,j}^2K_*^{\mathrm{kin}}v_j,\qquad K_*^{\mathrm{kin}}\succ0,\qquad\mathcal{V}_*:=\left.\frac{\partial^2V_{\mathrm{eff}}}{\partial\phi\,\partial\phi}\right|_{\phi_*}^{\mathrm{phys}}. \tag*{(264)} V ∗ v j = m H , j 2 K ∗ kin v j , K ∗ kin ≻ 0 , V ∗ := ∂ ϕ ∂ ϕ ∂ 2 V eff ϕ ∗ phys . (264)
The positive eigenvalues of this kinetic-normalized potential pencil give the bosonic curvature mass squared, not the eigenvalues of the full spacetime Hessian. Fermionic first-order channels instead use the singular values of the matched Dirac mass map, or the appropriate Majorana diagonalization, before quantum pole corrections. On a translation-invariant Lorentzian branch, a stable physical mass is instead a pole of the gauge-invariant or gauge-reduced two-point function. With the mostly-plus convention used here, put s : = − p L 2 s:=-p_L^2 s := − p L 2 . For a possibly mixed channel j j j , define
P j ( s ; q ∗ , μ ) : = det [ Γ j , p h y s ( 2 ) ( p L ; q ∗ , μ ) ] ∣ − p L 2 = s , P j ( M j , p o l e 2 ; q ∗ , μ ) = 0. (265) \mathcal{P}_j(s;\mathbf{q}_*,\mu):=\left.\det\left[\Gamma_{j,\mathrm{phys}}^{(2)}(p_L;\mathbf{q}_*,\mu)\right]\right|_{-p_L^2=s},\qquad\mathcal{P}_j(M_{j,\mathrm{pole}}^2;\mathbf{q}_*,\mu)=0. \tag*{(265)} P j ( s ; q ∗ , μ ) := det [ Γ j , phys ( 2 ) ( p L ; q ∗ , μ ) ] − p L 2 = s , P j ( M j , pole 2 ; q ∗ , μ ) = 0. (265)
Here q ∗ \mathbf{q}_* q ∗ denotes the selected vacuum coordinates and renormalized couplings. A simple, isolated real zero with positive residue defines a stable one-particle mass. An unstable excitation has a complex pole on the appropriate continued sheet; a confined channel must be defined through a gauge-invariant composite correlator. On a generic curved background there need not be a global momentum p L p_L p L , and the corresponding invariant statement uses spectral thresholds or the long-distance decay of covariant two-point functions. These qualifications prevent a stochastic relaxation rate, a gauge-dependent Hessian eigenvalue, or an arbitrary inverse length from being renamed a physical mass.
The connection between the pole in (266) and inertial motion can be seen without postulating a separate inertial force. Consider, for clarity, a minimally coupled scalar representative. In this paragraph m = M p o l e m=M_{\mathrm{pole}} m = M pole is the matched physical mass. Its short-wavelength equation is
( ℏ 2 □ g ∗ − m 2 − ℏ 2 ξ R [ g ∗ ] ) ψ = 0. (266) \left(\hbar^2\Box_{g_*}-m^2-\hbar^2\xi R[g_*]\right)\psi=0. \tag*{(266)} ( ℏ 2 □ g ∗ − m 2 − ℏ 2 ξ R [ g ∗ ] ) ψ = 0. (266)
For ψ = a exp ( i S / ℏ ) \psi=a\exp(iS/\hbar) ψ = a exp ( i S /ℏ ) , the order-ℏ 0 \hbar^0 ℏ 0 term is
g ∗ μ ν ∂ μ S ∂ ν S + m 2 = 0. (267) g_*^{\mu\nu}\partial_\mu S\,\partial_\nu S+m^2=0. \tag*{(267)} g ∗ μν ∂ μ S ∂ ν S + m 2 = 0. (267)
The curvature coupling and amplitude transport enter at subsequent eikonal orders. Put p μ : = ∂ μ S p_\mu:=\partial_\mu S p μ := ∂ μ S . Since covariant derivatives commute on a scalar phase,
p ν ∇ ν ( g ∗ ) p μ = p ν ∇ μ ( g ∗ ) p ν = 1 2 ∇ μ ( g ∗ ) ( p ν p ν ) = 0. (268) p^\nu\nabla_\nu^{(g_*)}p_\mu=p^\nu\nabla_\mu^{(g_*)}p_\nu=\frac{1}{2}\nabla_\mu^{(g_*)}(p^\nu p_\nu)=0. \tag*{(268)} p ν ∇ ν ( g ∗ ) p μ = p ν ∇ μ ( g ∗ ) p ν = 2 1 ∇ μ ( g ∗ ) ( p ν p ν ) = 0. (268)
For m > 0 m>0 m > 0 , u μ : = p μ / m u^\mu:=p^\mu/m u μ := p μ / m therefore obeys u ν ∇ ν ( g ∗ ) u μ = 0 u^\nu\nabla_\nu^{(g_*)}u^\mu=0 u ν ∇ ν ( g ∗ ) u μ = 0 . This eikonal reduction assumes a stable narrow excitation and background variation on scales large compared with its Compton wavelength. The analogous principal-symbol calculation gives the same timelike characteristic for minimally coupled spinor and vector wave packets after their physical polarizations have been selected. Here the geodesic statement concerns a freely propagating, neutral mode with constant matched mass. For a charged Abelian mode with constant charge q e l q_{\mathrm{el}} q el , define the gauge-invariant eikonal momentum by k μ = ∂ μ S − q e l A μ k_\mu= \partial_\mu S - q_{\mathrm{el}} A_\mu k μ = ∂ μ S − q el A μ , with F μ ν = ∂ μ A ν − ∂ ν A μ F_{\mu\nu} = \partial_\mu A_\nu- \partial_\nu A_\mu F μν = ∂ μ A ν − ∂ ν A μ . If a background scalar makes the effective mass position dependent, the leading eikonal relation and its differentiated characteristic are
k μ k μ = − m ( x ) 2 , k ν ∇ ν k μ = 1 2 ∂ μ ( k ν k ν ) + q e l F μ ν k ν = q e l F μ ν k ν − 1 2 ∂ μ m ( x ) 2 . (269) k^\mu k_\mu= -m(x)^2,\qquad k^\nu\nabla_\nu k_\mu= \frac{1}{2}\partial_\mu\left(k^\nu k_\nu\right) + q_{\mathrm{el}}F_{\mu\nu}k^\nu
= q_{\mathrm{el}}F_{\mu\nu}k^\nu- \frac{1}{2}\partial_\mu m(x)^2. \tag*{(269)} k μ k μ = − m ( x ) 2 , k ν ∇ ν k μ = 2 1 ∂ μ ( k ν k ν ) + q el F μν k ν = q el F μν k ν − 2 1 ∂ μ m ( x ) 2 . (269)
The first equality in the second line uses ∇ ν k μ − ∇ μ k ν = q e l F μ ν \nabla_\nu k_\mu- \nabla_\mu k_\nu= q_{\mathrm{el}}F_{\mu\nu} ∇ ν k μ − ∇ μ k ν = q el F μν . Thus universal gravitational coupling coexists with a Lorentz force or scalar force; a pure geodesic follows when those terms vanish or are negligible in the declared approximation. Thus inertia is the characteristic propagation of a spectral excitation of the selected vacuum; it is not an extra force appended to the geometric field equation.
The same conclusion, including the gravitational source, follows from the localized worldline limit. For a stable, structureless excitation, let
S p p [ z ; g ∗ ] = − m ∫ d s = − m ∫ − g μ ν ∗ ( z ) z ˙ μ z ˙ ν d λ . (270) S_{\mathrm{pp}}[z;g^*] = -m\int ds = -m\int\sqrt{-g^*_{\mu\nu}(z)\dot z^\mu\dot z^\nu}\,d\lambda. \tag*{(270)} S pp [ z ; g ∗ ] = − m ∫ d s = − m ∫ − g μν ∗ ( z ) z ˙ μ z ˙ ν d λ . (270)
This worldline action must be obtained by matching the isolated matter pole and the stress-tensor Ward identity; merely postulating it would build the mass equivalence into the assumption. Once that matching has been made, endpoint-fixed variation of z z z , followed by the proper-time choice of parameter, gives
d 2 z μ d τ 2 + Γ α β μ [ g ∗ ] d z α d τ d z β d τ = 0. (271) \frac{d^2 z^\mu}{d\tau^2}+\Gamma^\mu_{\alpha\beta}[g^*]\frac{dz^\alpha}{d\tau}\frac{dz^\beta}{d\tau}=0. \tag*{(271)} d τ 2 d 2 z μ + Γ α β μ [ g ∗ ] d τ d z α d τ d z β = 0. (271)
Metric variation of the same term gives
δ g S p p = m 2 ∫ u μ u ν δ g μ ν ∗ ( z ( τ ) ) d τ = 1 2 ∫ M − g ∗ T p p μ ν ( x ) δ g μ ν ∗ ( x ) d 4 x , (272) \delta_g S_{\mathrm{pp}}=\frac{m}{2}\int u^\mu u^\nu\delta g^*_{\mu\nu}(z(\tau))\,d\tau=\frac{1}{2}\int_M\sqrt{-g^*}\,T_{\mathrm{pp}}^{\mu\nu}(x)\delta g^*_{\mu\nu}(x)\,d^4x, \tag*{(272)} δ g S pp = 2 m ∫ u μ u ν δ g μν ∗ ( z ( τ )) d τ = 2 1 ∫ M − g ∗ T pp μν ( x ) δ g μν ∗ ( x ) d 4 x , (272)
where
T p p μ ν ( x ) = m − g ∗ ( x ) ∫ u μ u ν δ ( 4 ) ( x − z ( τ ) ) d τ . (273) T_{\mathrm{pp}}^{\mu\nu}(x)=\frac{m}{\sqrt{-g^*(x)}}\int u^\mu u^\nu\delta^{(4)}(x-z(\tau))\,d\tau. \tag*{(273)} T pp μν ( x ) = − g ∗ ( x ) m ∫ u μ u ν δ ( 4 ) ( x − z ( τ )) d τ . (273)
Equations (271) and (273) are not two independent assignments: they are the worldline and coframe variations of one covariant functional. At the quantum level, the corresponding non-circular normalization is the zero-momentum stress-tensor form factor. For a stable normalized spin-zero one-particle state, write
⟨ p ′ ∣ T μ ν ( 0 ) ∣ p ⟩ = 2 P μ P ν A ( q 2 ) + ( q μ q ν − g μ ν ∗ q 2 ) C ( q 2 ) , P : = p + p ′ 2 , q : = p ′ − p . (274) \langle p'|T_{\mu\nu}(0)|p\rangle= 2P_\mu P_\nu A(q^2)+(q_\mu q_\nu-g^*_{\mu\nu}q^2)C(q^2),\qquad P:=\frac{p+p'}{2},\qquad q:=p'-p. \tag*{(274)} ⟨ p ′ ∣ T μν ( 0 ) ∣ p ⟩ = 2 P μ P ν A ( q 2 ) + ( q μ q ν − g μν ∗ q 2 ) C ( q 2 ) , P := 2 p + p ′ , q := p ′ − p . (274)
For a spinning state, additional spin-dependent tensors occur, but its spin-averaged forward normalization has the same Ward-identity content. An anomaly-free diffeomorphism Ward identity and canonical state normalization give A ( 0 ) = 1 A(0)=1 A ( 0 ) = 1 . This condition matches the pole four-momentum to the long-wavelength tensor gravitational charge, including loop corrections. It is not available for a confined colored state, an infraparticle without an isolated pole, or a theory with an anomalous Ward identity.
For clarity, the normalization does not involve division by a distributional state overlap. Use ⟨ p ′ ∣ p ⟩ = 2 E p ( 2 π ) 3 δ ( 3 ) ( p ′ − p ) \langle p'|p\rangle=2E_p(2\pi)^3\delta^{(3)}(\mathbf{p}'-\mathbf{p}) ⟨ p ′ ∣ p ⟩ = 2 E p ( 2 π ) 3 δ ( 3 ) ( p ′ − p ) and the translation charge P ^ μ = ∫ T 0 μ ( 0 , x ) d 3 x \widehat{P}^{\mu}=\int T^{0\mu}(0,\mathbf{x})\,\mathrm{d}^3x P μ = ∫ T 0 μ ( 0 , x ) d 3 x . Spatial translation covariance of the matrix element gives
⟨ p ′ ∣ P ^ μ ∣ p ⟩ = ( 2 π ) 3 δ ( 3 ) ( p ′ − p ) ⟨ p ′ ∣ T 0 μ ( 0 ) ∣ p ⟩ = 2 E p p μ ( 2 π ) 3 δ ( 3 ) ( p ′ − p ) . (275) \langle p'|\widehat{P}^{\mu}|p\rangle=(2\pi)^3\delta^{(3)}(\mathbf{p}'-\mathbf{p})\langle p'|T^{0\mu}(0)|p\rangle=2E_p p^{\mu}(2\pi)^3\delta^{(3)}(\mathbf{p}'-\mathbf{p}). \tag*{(275)} ⟨ p ′ ∣ P μ ∣ p ⟩ = ( 2 π ) 3 δ ( 3 ) ( p ′ − p ) ⟨ p ′ ∣ T 0 μ ( 0 ) ∣ p ⟩ = 2 E p p μ ( 2 π ) 3 δ ( 3 ) ( p ′ − p ) . (275)
The forward limit of the form-factor decomposition is ⟨ p ∣ T 0 μ ( 0 ) ∣ p ⟩ = 2 E p p μ A ( 0 ) \langle p|T^{0\mu}(0)|p\rangle=2E_p p^{\mu}A(0) ⟨ p ∣ T 0 μ ( 0 ) ∣ p ⟩ = 2 E p p μ A ( 0 ) . Equality of the coefficients of the common delta distribution therefore gives A ( 0 ) = 1 A(0)=1 A ( 0 ) = 1 , or equivalently the same result follows after smearing with normalizable wave packets. This uses the total conserved stress tensor and its translation generator. A partial-sector stress tensor, a gauge-dependent excitation, or an independently normalized extra scalar force need not have this forward gravitational interpretation.
This identity becomes especially transparent in the Newtonian limit. Write g 00 ∗ = − ( 1 + 2 Φ N ) g^{*}_{00}=-(1+2\Phi_N) g 00 ∗ = − ( 1 + 2 Φ N ) , take g 0 i ∗ = 0 g^{*}_{0i}=0 g 0 i ∗ = 0 , and retain first order in Φ N \Phi_N Φ N and second order in ∣ v ∣ |\mathbf{v}| ∣ v ∣ . Then
d s d t = 1 + Φ N − 1 2 ∣ v ∣ 2 + O ( Φ N 2 , Φ N ∣ v ∣ 2 , ∣ v ∣ 4 ) , (276) \frac{\mathrm{d}s}{\mathrm{d}t}=1+\Phi_N-\frac{1}{2}|\mathbf{v}|^2+O(\Phi_N^2,\Phi_N|\mathbf{v}|^2,|\mathbf{v}|^4), \tag*{(276)} d t d s = 1 + Φ N − 2 1 ∣ v ∣ 2 + O ( Φ N 2 , Φ N ∣ v ∣ 2 , ∣ v ∣ 4 ) , (276)
so, after removal of the constant rest term,
L N R = 1 2 m ∣ v ∣ 2 − m Φ N , m x ¨ = − m ∇ Φ N . (277) L_{\mathrm{NR}}=\frac{1}{2}m|\mathbf{v}|^2-m\Phi_N,\qquad m\ddot{\mathbf{x}}=-m\nabla\Phi_N. \tag*{(277)} L NR = 2 1 m ∣ v ∣ 2 − m Φ N , m x ¨ = − m ∇ Φ N . (277)
Moreover, T p p 00 = m δ ( 3 ) ( x − z ) T^{00}_{\mathrm{pp}}=m\delta^{(3)}(\mathbf{x}-\mathbf{z}) T pp 00 = m δ ( 3 ) ( x − z ) to this order, and ∇ 2 Φ N = 4 π G e f f T 00 \nabla^2\Phi_N=4\pi G_{\mathrm{eff}}T^{00} ∇ 2 Φ N = 4 π G eff T 00 after neglecting the cosmological and higher-curvature remainders. The same m m m therefore enters the active source. Inertial, passive gravitational, and active monopole mass are one coefficient in the same generally covariant effective action, rather than two or three masses whose equality is imposed after the fact.
This statement is restricted to the tensor branch with universal minimal coupling. The shared-scalar sector in Section 11.11 can carry composition-dependent scalar charges and then produces a fifth force and an Eötvös signal. Nonminimal curvature terms, spin, finite size, radiation reaction, or strong self-gravity also require a corrected effective action. They can invalidate the elementary point-particle reduction without invalidating the diffeomorphism Ward identity in its proper domain.
The resulting interpretation is more specific than the phrase mass resists acceleration . Mass is the pole or spectral stiffness of a physical excitation around the selected equilibrium; inertia is the covariant phase propagation of that excitation; gravitational charge is the coframe response of the same effective term; and observable gravity is the tidal curvature relating neighboring characteristics. The extension sought by the present framework is not a replacement of this general-relativistic low-energy mechanism. It is a derivation, from one selected microscopic package, of the vacuum, its physical Hessian, the coframe, and their universal coupling.
What would determine numerical inertial masses?
The preceding argument identifies what mass means, but it does not by itself calculate any observed mass. A non-circular calculation requires the selected data q ∗ \mathbf{q}_\ast q ∗ to be fixed independently of the masses that will test the theory. In a declared renormalization scheme, the closed system to be solved is schematically
∇ q S s e l ( q ∗ ) = 0 , C A ( q ∗ ) = 0 , P j ( M j 2 ; q ∗ , μ ) = 0 , μ d M j d μ = 0. (278) \nabla_{\mathbf{q}}\mathcal{S}_{\mathrm{sel}}(\mathbf{q}_\ast)=0,\qquad\mathcal{C}_A(\mathbf{q}_\ast)=0,\qquad\mathcal{P}_j(M_j^2;\mathbf{q}_\ast,\mu)=0,\qquad\mu\frac{\mathrm{d}M_j}{\mathrm{d}\mu}=0. \tag*{(278)} ∇ q S sel ( q ∗ ) = 0 , C A ( q ∗ ) = 0 , P j ( M j 2 ; q ∗ , μ ) = 0 , μ d μ d M j = 0. (278)
Here C A \mathcal{C}_A C A contains the gauge, anomaly, vacuum, and matching constraints, and the last identity states renormalization-group invariance of the physical pole after all threshold terms have been included. At a simple pole, an uncertainty or a change in selected microscopic data propagates as
δ M j 2 = − ∂ q P j ( M j 2 ; q ∗ , μ ) ⋅ δ q ∗ ∂ s P j ( M j 2 ; q ∗ , μ ) . (279) \delta M_j^2=-\frac{\partial_{\mathbf{q}}\mathcal{P}_j(M_j^2;\mathbf{q}_\ast,\mu)\cdot\delta\mathbf{q}_\ast}{\partial_s\mathcal{P}_j(M_j^2;\mathbf{q}_\ast,\mu)}. \tag*{(279)} δ M j 2 = − ∂ s P j ( M j 2 ; q ∗ , μ ) ∂ q P j ( M j 2 ; q ∗ , μ ) ⋅ δ q ∗ . (279)
If the microscopic construction selects only dimensionless coordinates, then one can at most obtain
M i M j = m i ( q ∗ ) m j ( q ∗ ) . (280) \frac{M_i}{M_j}=\frac{m_i(\mathbf{q}_\ast)}{m_j(\mathbf{q}_\ast)}. \tag*{(280)} M j M i = m j ( q ∗ ) m i ( q ∗ ) . (280)
Absolute masses additionally require a selected scale μ ∗ \mu_\ast μ ∗ , for example through a rigorously controlled dimensional-transmutation mechanism,
Λ d y n = μ exp [ − ∫ g ( μ ) d g β ( g ) ] , (281) \Lambda_{\mathrm{dyn}}=\mu\exp\left[-\int^{g(\mu)}\frac{\mathrm{d}g}{\beta(g)}\right], \tag*{(281)} Λ dyn = μ exp [ − ∫ g ( μ ) β ( g ) d g ] , (281)
with a declared renormalization boundary condition. One then has M j = μ ∗ m j ( q ∗ ) M_j=\mu_\ast m_j(\mathbf{q}_\ast) M j = μ ∗ m j ( q ∗ ) . Fitting q ∗ \mathbf{q}_\ast q ∗ , μ ∗ \mu_\ast μ ∗ , or Yukawa matrices to the target masses and reproducing them through (278) is matching, not prediction. The present paper supplies the equations and some conditional mass relations, but it does not yet solve this closed system for all observed masses.
A precise Machian response criterion.
The statement that inertia is influenced by the rest of the universe has several inequivalent meanings. Universal geodesic motion proves none of them by itself. A formula-level version can, however, be stated in the present selection framework. Let φ \varphi φ collect the physical non-coframe variables, introduce physical sources through Γ J = Γ e f f − ⟨ J e , e ⟩ − ⟨ J φ , φ ⟩ \Gamma_J = \Gamma_{\mathrm{eff}} - \langle J_e,e\rangle- \langle J_\varphi,\varphi\rangle Γ J = Γ eff − ⟨ J e , e ⟩ − ⟨ J φ , φ ⟩ , and evaluate the reduced Hessian at a selected solution. On the gauge quotient it has the block form
K ∗ r e d = ( K e e K e φ K φ e K φ φ ) . (282) \mathcal{K}^{\mathrm{red}}_* =
\begin{pmatrix}
K_{ee} & K_{e\varphi} \\
K_{\varphi e} & K_{\varphi\varphi}
\end{pmatrix}. \tag*{(282)} K ∗ red = ( K ee K φ e K e φ K φφ ) . (282)
Linearizing the two stationarity equations gives
\begin{pmatrix}
K_{ee} & K_{e\varphi} \\
K_{\varphi e} & K_{\varphi\varphi}
\end{pmatrix}
\begin{pmatrix}
\delta e_* \\
\delta\varphi_*
\end{pmatrix}
=
\begin{pmatrix}
\delta J_e \\
\delta J_\varphi
\end{pmatrix}. \tag*{(283)}\end{pmatrix}
Assume that K φ φ K_{\varphi\varphi} K φφ is invertible on the physical subspace and define the coframe Schur complement
K M a c h : = K e e − K e φ K φ φ − 1 K φ e . (284) K_{\mathrm{Mach}} := K_{ee} - K_{e\varphi}K_{\varphi\varphi}^{-1}K_{\varphi e}. \tag*{(284)} K Mach := K ee − K e φ K φφ − 1 K φ e . (284)
If the boundary conditions and global selection rule also make K M a c h K_{\mathrm{Mach}} K Mach invertible, elimination of δ φ ∗ \delta\varphi_* δ φ ∗ yields
δ e ∗ = K M a c h − 1 ( δ J e − K e φ K φ φ − 1 δ J φ ) . (285) \delta e_* = K_{\mathrm{Mach}}^{-1}\left(\delta J_e - K_{e\varphi}K_{\varphi\varphi}^{-1}\delta J_\varphi\right). \tag*{(285)} δ e ∗ = K Mach − 1 ( δ J e − K e φ K φφ − 1 δ J φ ) . (285)
The integral kernel of the second term,
χ e φ ( x , y ) : = − [ K M a c h − 1 K e φ K φ φ − 1 ] ( x , y ) , (286) \chi_{e\varphi}(x,y) := -\left[K_{\mathrm{Mach}}^{-1}K_{e\varphi}K_{\varphi\varphi}^{-1}\right](x,y), \tag*{(286)} χ e φ ( x , y ) := − [ K Mach − 1 K e φ K φφ − 1 ] ( x , y ) , (286)
is a quantitative Mach susceptibility: it measures how a physical change of matter or order-parameter data at y y y changes the selected local coframe at x x x . It is nonlocal because it contains Green operators, not because the theory permits superluminal signalling. Causal response requires the corresponding retarded Lorentzian kernel; the Euclidean inverse is used only for equilibrium selection.
Equation (285) distinguishes three levels of claim. First, the off-diagonal block K e φ ≠ 0 K_{e\varphi} \ne0 K e φ = 0 gives a weak relational statement: matter and inertial geometry respond to common data. Second, a strong Machian closure requires the physical kernel of K M a c h K_{\mathrm{Mach}} K Mach to vanish after boundary conditions, so that there is no freely specifiable homogeneous coframe perturbation independent of the selected global state. Third, a complete origin theorem must derive those boundary conditions and the selected state rather than insert them. One may state the strong condition without coordinate ambiguity as
{ X : E M [ X ] = 0 , O m a t [ X ] = D , B r e l [ X ] = 0 } Diff ( M ) ⋉ G = { [ X ∗ ( D ) ] } . (287) \frac{\left\{X : E_M[X] = 0,\ O_{\mathrm{mat}}[X] = D,\ B_{\mathrm{rel}}[X] = 0\right\}}{\operatorname{Diff}(M) \ltimes\mathcal{G}} = \left\{\left[X_*(D)\right]\right\}. \tag*{(287)} Diff ( M ) ⋉ G { X : E M [ X ] = 0 , O mat [ X ] = D , B rel [ X ] = 0 } = { [ X ∗ ( D ) ] } . (287)
Here D D D is the complete declared relational matter data and B r e l B_{\mathrm{rel}} B rel is a physically fixed boundary condition. Vacuum Einstein solutions and freely propagating gravitational waves show that ordinary general relativity does not satisfy this strong requirement in an unqualified form. The current master theory provides the response operator and a route by which global equilibrium data could impose the stronger closure, but it has not proved (288) for the physical branch. This is the precise boundary between a formula-derived Machian mechanism and a verbal appeal to Mach’s principle.
Quantum geometry and the vacuum-curvature balance
Classical stochastic fluctuations of e e e are not, by themselves, quantum gravity. A quantum interpretation would require a physical finite-regulator Euclidean measure, schematically
d μ a , N p h y s ( Φ ) = 1 Z a , N exp [ − S E a , N ( Φ ) ℏ ] d ν a , N p h y s ( Φ ) , (288) \mathrm{d}\mu^{\mathrm{phys}}_{a,N}(\Phi)=\frac{1}{Z_{a,N}}\exp\left[-\frac{S_{E}^{a,N}(\Phi)}{\hbar}\right]\mathrm{d}\nu^{\mathrm{phys}}_{a,N}(\Phi), \tag*{(288)} d μ a , N phys ( Φ ) = Z a , N 1 exp [ − ℏ S E a , N ( Φ ) ] d ν a , N phys ( Φ ) , (288)
where the quotient, gauge fixing, determinants, and physical observables are defined consistently. One would then have to prove the regulator-independent Schwinger limits (857), reflection positivity, the remaining Osterwalder–Schrader axioms, and the gravitational constraint or relational-observable conditions stated in the quantum-completion section. Identifying the classical noise scale θ \theta θ with ℏ \hbar ℏ is not a derivation of any of these properties.
For a direct Lorentzian construction, the Gauss, local Lorentz, diffeomorphism, and Hamiltonian constraints, collectively C ^ A \widehat{C}_{A} C A , would have to satisfy
C ^ A Ψ p h y s = 0 , [ C ^ A , C ^ B ] = i ℏ f ^ A B C C ^ C , Q B R S T 2 = 0 , (289) \widehat{C}_{A}\Psi_{\mathrm{phys}}=0,\qquad[\widehat{C}_{A},\widehat{C}_{B}]=i\hbar\widehat{f}_{AB}{}^{C}\widehat{C}_{C},\qquad Q_{\mathrm{BRST}}^{2}=0, \tag*{(289)} C A Ψ phys = 0 , [ C A , C B ] = i ℏ f A B C C C , Q BRST 2 = 0 , (289)
with the field-dependent structure functions, domains, regularization, and possible boundary charges treated explicitly. An anomaly in this algebra, an indefinite physical inner product, or the absence of relational observables prevents the construction from being a quantum theory of geometry.
If such a quantum completion exists, the in-in effective action gives two related response equations. Its first coframe variation yields the semiclassical fixed point
G μ ν + Λ r e n g μ ν + H μ ν r e n = 8 π G r e n ⟨ T ^ μ ν ⟩ ω ∗ , g r e n , (290) G_{\mu\nu}+\Lambda_{\mathrm{ren}}g_{\mu\nu}+\mathcal{H}^{\mathrm{ren}}_{\mu\nu}=8\pi G_{\mathrm{ren}}\langle\widehat{T}_{\mu\nu}\rangle^{\mathrm{ren}}_{\omega_{*},g}, \tag*{(290)} G μν + Λ ren g μν + H μν ren = 8 π G ren ⟨ T μν ⟩ ω ∗ , g ren , (290)
while its second physical variation determines the connected coframe two-point function through
∫ Γ e e , p h y s ( 2 ) ( x , z ) G e e p h y s ( z , y ) d z = P p h y s ( x , y ) . (291) \int\Gamma^{(2)}_{ee,\mathrm{phys}}(x,z)G^{\mathrm{phys}}_{ee}(z,y)\,\mathrm{d}z=P_{\mathrm{phys}}(x,y). \tag*{(291)} ∫ Γ ee , phys ( 2 ) ( x , z ) G ee phys ( z , y ) d z = P phys ( x , y ) . (291)
Here P p h y s P_{\mathrm{phys}} P phys projects away gauge directions and enforces the constraints. Real-time response must also be causal:
δ ⟨ T ^ μ ν ( x ) ⟩ ω ∗ , g r e n δ g ρ σ ( y ) = 0 when y ∉ J − ( x ) . (292) \frac{\delta\langle\widehat{T}_{\mu\nu}(x)\rangle^{\mathrm{ren}}_{\omega_{*},g}}{\delta g_{\rho\sigma}(y)}=0\qquad\text{when }y\notin J^{-}(x). \tag*{(292)} δ g ρ σ ( y ) δ ⟨ T μν ( x ) ⟩ ω ∗ , g ren = 0 when y ∈ / J − ( x ) . (292)
This condition is naturally formulated by the Schwinger–Keldysh effective action and does not follow from an in-out or Euclidean Hessian alone. Formula (292) gives the proposed extension of the classical interpretation: the mean coframe is the first-response variable selected by the effective action, whereas quantum geometry is encoded by the physical inverse Hessian and higher connected coframe correlations. These are completion conditions, not a construction of a positive graviton Hilbert space in the present manuscript.
The same effective action separates a possible explanation of the cosmological constant from an unsupported cancellation claim. On a constant vacuum branch write
Γ v a c [ g , φ ] = ∫ M − g [ R − 2 Λ L ( q , μ ) 16 π G ( q , μ ) − V e f f ( φ ; q , μ ) + L h c ] d 4 x . (293) \Gamma_{\mathrm{vac}}[g,\varphi] = \int_M \sqrt{-g}\left[\frac{R-2\Lambda_L(\mathbf{q},\mu)}{16\pi G(\mathbf{q},\mu)}-V_{\mathrm{eff}}(\varphi;\mathbf{q},\mu)+\mathcal{L}_{\mathrm{hc}}\right]\mathrm{d}^{4}x. \tag*{(293)} Γ vac [ g , φ ] = ∫ M − g [ 16 π G ( q , μ ) R − 2 Λ L ( q , μ ) − V eff ( φ ; q , μ ) + L hc ] d 4 x . (293)
Let φ ∗ \varphi_\ast φ ∗ satisfy the renormalized vacuum equations. On a maximally symmetric four-dimensional branch with R μ ν = 3 H ∗ 2 g μ ν R_{\mu\nu}=3H_\ast^2g_{\mu\nu} R μν = 3 H ∗ 2 g μν , define Q h c ( H ∗ 2 ) g μ ν Q_{\mathrm{hc}}(H_\ast^2)g_{\mu\nu} Q hc ( H ∗ 2 ) g μν to be the complete higher-curvature and nonlocal vacuum contribution after multiplication by 8 π G 8\pi G 8 π G . The coframe equation reduces to the scalar balance
3 H ∗ 2 = Λ L ( q ∗ , μ ) + 8 π G ( q ∗ , μ ) V e f f ( φ ∗ ; q ∗ , μ ) + Q h c ( H ∗ 2 ; q ∗ , μ ) . (294) 3H_\ast^2=\Lambda_L(\mathbf{q}_\ast,\mu)+8\pi G(\mathbf{q}_\ast,\mu)V_{\mathrm{eff}}(\varphi_\ast;\mathbf{q}_\ast,\mu)+Q_{\mathrm{hc}}(H_\ast^2;\mathbf{q}_\ast,\mu). \tag*{(294)} 3 H ∗ 2 = Λ L ( q ∗ , μ ) + 8 π G ( q ∗ , μ ) V eff ( φ ∗ ; q ∗ , μ ) + Q hc ( H ∗ 2 ; q ∗ , μ ) . (294)
Thus the observed curvature is a renormalized balance among the geometric Cartan term, the selected vacuum energy, and the remaining quantum and higher-curvature response. It is not equal to any one bare zero-point sum. When all beta functions and counterterms are included consistently, the physical root must satisfy
μ d d μ [ Λ L + 8 π G V e f f + Q h c ] ( φ ∗ , q ∗ , H ∗ ) = 0. (295) \mu\frac{\mathrm{d}}{\mathrm{d}\mu}\left[\Lambda_L+8\pi G V_{\mathrm{eff}}+Q_{\mathrm{hc}}\right]_{(\varphi_\ast,\mathbf{q}_\ast,H_\ast)}=0. \tag*{(295)} μ d μ d [ Λ L + 8 π G V eff + Q hc ] ( φ ∗ , q ∗ , H ∗ ) = 0. (295)
Renormalization-scale invariance removes an unphysical scale dependence; it does not explain why the resulting root is small. For a field of mass M i M_i M i , a typical one-loop threshold contribution has the form
Δ V i = ( − 1 ) F i n i M i 4 64 π 2 ( log M i 2 μ 2 − c i ) , (296) \Delta V_i=\frac{(-1)^{F_i}n_iM_i^4}{64\pi^2}\left(\log\frac{M_i^2}{\mu^2}-c_i\right), \tag*{(296)} Δ V i = 64 π 2 ( − 1 ) F i n i M i 4 ( log μ 2 M i 2 − c i ) , (296)
so every threshold must be included in the matching and stability test. Ordinary equilibrium minimization alone cannot solve the additive vacuum-energy problem. To see this directly, fix the coframe while selecting an internal vacuum coordinate q q q , and let
V C ( q ) : = V 0 ( q ) + C , ∂ q V 0 ( q ∗ ) = 0 , ∂ q 2 V 0 ( q ∗ ) > 0. (297) V_C(q):=V_0(q)+C,\qquad\partial_qV_0(q_\ast)=0,\qquad\partial_q^2V_0(q_\ast)>0. \tag*{(297)} V C ( q ) := V 0 ( q ) + C , ∂ q V 0 ( q ∗ ) = 0 , ∂ q 2 V 0 ( q ∗ ) > 0. (297)
Then
d q ∗ d C = 0 , d d C V C ( q ∗ ) = 1 , ∂ ∂ C ( Λ L + 8 π G V C ( q ∗ ) + Q h c ) ∣ q ∗ , G , Λ L , Q h c = 8 π G . (298) \begin{aligned}
\frac{\mathrm{d}q_\ast}{\mathrm{d}C}&=0,\qquad\frac{\mathrm{d}}{\mathrm{d}C}V_C(q_\ast)=1,\\
\left.\frac{\partial}{\partial C}\left(\Lambda_L+8\pi G V_C(q_\ast)+Q_{\mathrm{hc}}\right)\right|_{q_\ast,G,\Lambda_L,Q_{\mathrm{hc}}}&=8\pi G. \tag*{(298)}
\end{aligned} d C d q ∗ ∂ C ∂ ( Λ L + 8 π G V C ( q ∗ ) + Q hc ) q ∗ , G , Λ L , Q hc = 0 , d C d V C ( q ∗ ) = 1 , = 8 π G . (298)
The first equality follows because an additive constant does not enter the internal stationarity equation; the second is the envelope identity; and the third is the direct change of the curvature source with its other inputs held fixed. It is not the full implicit response when the higher-curvature term depends on the root. To retain that feedback, put y ( C ) : = H ∗ 2 ( C ) y(C) := H_*^2(C) y ( C ) := H ∗ 2 ( C ) and hold G G G , Λ L \Lambda_L Λ L , and V 0 ( q ∗ ) V_0(q_*) V 0 ( q ∗ ) fixed while solving
3 y ( C ) = Λ L + 8 π G ( V 0 ( q ∗ ) + C ) + Q h c ( y ( C ) ) . (299) 3y(C) = \Lambda_L + 8\pi G\left(V_0(q_*) + C\right) + \mathcal{Q}_{\mathrm{hc}}(y(C)). \tag*{(299)} 3 y ( C ) = Λ L + 8 π G ( V 0 ( q ∗ ) + C ) + Q hc ( y ( C )) . (299)
At a differentiable simple root, implicit differentiation gives
d y d C = 8 π G 3 − Q h c ′ ( y ) , 3 − Q h c ′ ( y ) ≠ 0. (300) \frac{\mathrm{d}y}{\mathrm{d}C} = \frac{8\pi G}{3-\mathcal{Q}'_{\mathrm{hc}}(y)}, \qquad3-\mathcal{Q}'_{\mathrm{hc}}(y) \ne0. \tag*{(300)} d C d y = 3 − Q hc ′ ( y ) 8 π G , 3 − Q hc ′ ( y ) = 0. (300)
In the ordinary Einstein limit the derivative is 8 π G / 3 8\pi G/3 8 π G /3 ; for the curvature quantity 3 y 3y 3 y it is 8 π G 8\pi G 8 π G . A vanishing denominator invalidates the simple-root argument rather than proving insensitivity. If the selected matter vacuum, couplings, or counterterms also vary, their chain-rule terms must be included. Thus higher-curvature feedback changes the response but does not automatically cancel the vacuum-energy sensitivity. A global volume constraint, a symmetry, a sequestering variable, or another dynamical zero-mode could modify this response, but it must be part of the master data and must satisfy the radiative-stability test
d d C [ Λ L ( q ∗ ( C ) ) + 8 π G ( q ∗ ( C ) ) V C ( ϕ ∗ ( C ) , q ∗ ( C ) ) + Q h c ( H ∗ 2 ( C ) , q ∗ ( C ) ) ] ≈ 0 (301) \frac{\mathrm{d}}{\mathrm{d}C}\left[\Lambda_L(q_*(C)) + 8\pi G(q_*(C))V_C(\phi_*(C),q_*(C)) + \mathcal{Q}_{\mathrm{hc}}(H_*^2(C),q_*(C))\right] \approx0 \tag*{(301)} d C d [ Λ L ( q ∗ ( C )) + 8 π G ( q ∗ ( C )) V C ( ϕ ∗ ( C ) , q ∗ ( C )) + Q hc ( H ∗ 2 ( C ) , q ∗ ( C )) ] ≈ 0 (301)
over the declared range of threshold corrections. The present local master action does not yet prove this criterion or calculate the observed value of H ∗ H_* H ∗ . Its explanatory content is a common response architecture and a sharp obstruction test: inertia, gravitation, quantum coframe correlations, and vacuum curvature are successive first- and second-variation data of one selected effective functional, while numerical mass and cosmological-constant predictions require the additional selection and radiative-stability equations displayed above.
5.5.6. Relation to the stochastic control cost. The functional (65) measures the quadratic deviation of an admissible absolutely continuous skeleton from a specified uncontrolled drift. Therefore
J T [ x ] = 0 ⟺ x ˙ t = B ( x t ) for almost every t , (302) \mathcal{J}_T[x] = 0 \Longleftrightarrow\dot{x}_t = \mathbb{B}(x_t) \quad\text{for almost every } t, \tag*{(302)} J T [ x ] = 0 ⟺ x ˙ t = B ( x t ) for almost every t , (302)
on the range of the mobility and under the hypotheses of § 3.9 . If a test-body reduction of that drift is independently derived in the form
B t b ( x , v ) = ( v μ , − Γ α β μ [ g ∗ ] v α v β ) , (303) \mathbb{B}_{\mathrm{tb}}(x,v) = \left(v^\mu,-\Gamma^\mu_{\alpha\beta}[g_*]v^\alpha v^\beta\right), \tag*{(303)} B tb ( x , v ) = ( v μ , − Γ α β μ [ g ∗ ] v α v β ) , (303)
then (302) gives the geodesic equation (261) . The order of this implication matters. Specifying (303) in advance and then calling its paths “zero-cost” does not derive the origin of inertial motion; it merely assigns zero control expenditure to a preselected drift.
A microscopic origin result would require the same independently calibrated selection data to determine the coframe, universal matter coupling, and reduced drift. Schematically, the required stronger conclusion is
Γ ∗ ∈ arg min Γ ∈ G a d m S s e l ( Γ ) ⟹ ( e ∗ , ω ∗ , S m [ e ∗ , ω ∗ , ψ ] , B t b ) . (304) \Gamma_* \in\operatorname*{arg\,min}_{\Gamma\in\mathcal{G}_{\mathrm{adm}}} \mathcal{S}_{\mathrm{sel}}(\Gamma) \quad\Longrightarrow\quad\left(e_*,\omega_*,S_{\mathrm{m}}[e_*,\omega_*,\psi],\mathbb{B}_{\mathrm{tb}}\right). \tag*{(304)} Γ ∗ ∈ Γ ∈ G adm arg min S sel ( Γ ) ⟹ ( e ∗ , ω ∗ , S m [ e ∗ , ω ∗ , ψ ] , B tb ) . (304)
The right-hand side includes the universal coupling in (256) and the derivation of (303) . Equation (304) is a mathematically definite target for a concrete realization; it is not implied solely by the notation (255) .
Ordinary material response and shielding
Let a source, a passive material layer, and a probe have stress tensors T μ ν s r c T^{\mathrm{src}}_{\mu\nu} T μν src , T μ ν m a t T^{\mathrm{mat}}_{\mu\nu} T μν mat , and T μ ν p r T^{\mathrm{pr}}_{\mu\nu} T μν pr , respectively. The controlled metric equation then has the decomposed form
G μ ν [ g ∗ ] + Λ L g μ ν ∗ + 16 π G e f f ∑ i α ˉ i H ˉ μ ν ( i ) [ g ∗ ] = 8 π G e f f ( T μ ν s r c + T μ ν m a t + T μ ν p r ) . (305) G_{\mu\nu}[g^*] + \Lambda_{\mathrm{L}}g^*_{\mu\nu} + 16\pi G_{\mathrm{eff}}\sum_i \bar{\alpha}_i\bar{H}^{(i)}_{\mu\nu}[g^*] = 8\pi G_{\mathrm{eff}}\left(T^{\mathrm{src}}_{\mu\nu}+T^{\mathrm{mat}}_{\mu\nu}+T^{\mathrm{pr}}_{\mu\nu}\right). \tag*{(305)} G μν [ g ∗ ] + Λ L g μν ∗ + 16 π G eff i ∑ α ˉ i H ˉ μν ( i ) [ g ∗ ] = 8 π G eff ( T μν src + T μν mat + T μν pr ) . (305)
The material layer can change the total stress tensor and hence the solution g ∗ g^* g ∗ , but it does not remove the probe from the universal coupling (256) . The present ordinary-matter branch therefore contains no material parameter that changes (261) into free motion with respect to a second, shielded metric. This is the structural reason that it has no electromagnetic Faraday-cage analogue for externally imposed gravitational response.
This assertion is intentionally narrower than a no-go theorem for every gravitational configuration. Material distributions can modify curvature; a symmetric shell can make the Newtonian field generated by that shell vanish in its interior; selected source configurations can cancel the acceleration of a particular probe; and gravitational radiation can be scattered, absorbed, or phase shifted. Such effects alter the solution of (305) ; they do not universally decouple interior probes from g ∗ g^* g ∗ . A general no-shielding theorem would require an operational definition of shielding, a class of material stress tensors, boundary conditions, and suitable energy and regularity hypotheses. Neither universal coupling alone nor the absence of a simple negative gravitational charge supplies that theorem.
At low energy, this interpretation is intended to recover, rather than replace, the geometric content of general relativity. General relativity takes a metric and its universal matter coupling as fundamental low-energy structure. The additional programme here is to derive, from common selection data, why the effective coframe, its field equation, and universal coupling occur together. Until one concrete microscopic realization proves the selection statement (304) and derives the drift (303) , that origin claim remains a well-specified conditional objective rather than a consequence of the master cost alone.
Yang–Mills sectors and electromagnetism
For the residual internal block, the same master equation has P H A = κ ^ A B ⋆ F B \mathbb{P}_{H}^{A}=\widehat{\kappa}_{AB}\star F^{B} P H A = κ A B ⋆ F B , where κ ^ A B \widehat{\kappa}_{AB} κ A B is the internal constitutive form induced by the single master pairing. It may contain the shared order-parameter weight; for the joint candidate in (228) , κ ^ A B ( σ ) = w ( σ ) κ A B \widehat{\kappa}_{AB}(\sigma)=w(\sigma)\kappa_{AB} κ A B ( σ ) = w ( σ ) κ A B . The second projection in (172) is then
π i n t ι h ∗ ( D P − J e x t ) = 0 ⟺ D ( κ ^ A B ⋆ F B ) = − ⋆ J A . (306) \pi_{\mathrm{int}}\iota_{h}^{*}(\mathbb{D}\mathbb{P}-\mathbb{J}_{\mathrm{ext}})=0 \quad\Longleftrightarrow\quad D\left(\widehat{\kappa}_{AB}\star F^{B}\right)=-\star J_{A}. \tag*{(306)} π int ι h ∗ ( DP − J ext ) = 0 ⟺ D ( κ A B ⋆ F B ) = − ⋆ J A . (306)
In local coordinates, after fixing the Hodge convention and lowering the form index, this is precisely D μ ( κ ^ A B F B μ ν ) = J A ν D_{\mu}(\widehat{\kappa}_{AB}F^{B\mu\nu})=J_{A}^{\nu} D μ ( κ A B F B μν ) = J A ν , which is derived by direct variation below. A scalar-dependent but Hodge-compatible coefficient is already included inside the derivative in this equation. If the constitutive tensor has off-diagonal blocks or is not Hodge-compatible, one must instead retain the appropriate internal component of ι h ∗ D ( ∂ L / ∂ F ) \iota_{h}^{*}D(\partial\mathcal{L}/\partial F) ι h ∗ D ( ∂ L / ∂ F ) ; the displayed Yang–Mills form is the rigorously recovered residual equation on the stated Hodge-compatible branch. The displayed Hodge equation is Euclidean, because positivity of the master pairing and the stochastic construction are formulated on the Euclidean slice. If the Wick map (206) extends analytically to the internal connection and its source, write
⋆ E F E ⟼ ς ⋆ ⋆ L F L , J E ⟼ ς J J L , ς ⋆ , ς J ∈ { ± 1 , ± i } . (307) \star_{E}F_{E}\longmapsto\varsigma_{\star}\star_{L}F_{L},\qquad J_{E}\longmapsto\varsigma_{J}J_{L},\qquad\varsigma_{\star},\varsigma_{J}\in\{\pm1,\pm i\}. \tag*{(307)} ⋆ E F E ⟼ ς ⋆ ⋆ L F L , J E ⟼ ς J J L , ς ⋆ , ς J ∈ { ± 1 , ± i } . (307)
The continuation must fix these factors together with the Lorentzian action and real source normalization. An arbitrary imaginary factor cannot be absorbed into a coupling while retaining a real, physically admissible theory. With the real convention D L , μ ( κ ^ L , A B F L B μ ν ) = J L , A ν D_{L,\mu}(\widehat{\kappa}_{L,AB}F_{L}^{B\mu\nu})=J_{L,A}^{\nu} D L , μ ( κ L , A B F L B μν ) = J L , A ν , the corresponding differential-form equation is D L ( κ ^ L , A B ⋆ L F L B ) = − ⋆ L J L , A D_{L}(\widehat{\kappa}_{L,AB}\star_{L}F_{L}^{B})=-\star_{L}J_{L,A} D L ( κ L , A B ⋆ L F L B ) = − ⋆ L J L , A . Thus, under the specified continuation and real source convention, the physical Lorentzian Yang–Mills equation is recovered from the same projected Euler system. Establishing global hyperbolicity, global existence, or regularity for every solution is a separate analytic problem and is not a prerequisite for this local variational identity.
On a phase in which the internal connection survives as a light connection on a residual compact group G H G_{H} G H , the master curvature contains
F = d A + A ∧ A . (308) F=dA+A\wedge A. \tag*{(308)} F = d A + A ∧ A . (308)
In components, the wedge-product convention is ( A ∧ A ) A = 1 2 f B C A A B ∧ A C (A\wedge A)^{A}=\frac{1}{2}f_{BC}^{\ \ A}A^{B}\wedge A^{C} ( A ∧ A ) A = 2 1 f B C A A B ∧ A C , which is equivalent to (311) below. Let κ ^ A B \widehat{\kappa}_{AB} κ A B denote the internal bilinear form induced by the reduced master pairing after the residual symmetry and field normalization have been fixed. It is not an additional microscopic pairing. The corresponding quadratic curvature invariant is
S Y M , e f f = 1 2 ∫ M κ ^ A B F A ∧ ⋆ F B = 1 4 ∫ M g κ ^ A B F μ ν A F B μ ν d 4 x . (309) S_{\mathrm{YM,eff}}=\frac{1}{2}\int_{M}\widehat{\kappa}_{AB}F^{A}\wedge\star F^{B}=\frac{1}{4}\int_{M}\sqrt{g}\,\widehat{\kappa}_{AB}F^{A}_{\mu\nu}F^{B\mu\nu}\,d^{4}x. \tag*{(309)} S YM , eff = 2 1 ∫ M κ A B F A ∧ ⋆ F B = 4 1 ∫ M g κ A B F μν A F B μν d 4 x . (309)
On a branch on which the scalar and coframe are frozen, κ ^ A B = κ A B \widehat{\kappa}_{AB}=\kappa_{AB} κ A B = κ A B is constant; only on that branch does the following variation omit derivatives of the constitutive coefficient. After canonical normalization and symmetry breaking, this may take the familiar form
S Y M , e f f = ∫ M g [ 1 4 g 3 2 G μ ν a G a μ ν + 1 4 g 2 2 W μ ν i W i μ ν + 1 4 g 1 2 B μ ν B μ ν ] d 4 x . (310) S_{\mathrm{YM,eff}}=\int_M\sqrt{g}\left[\frac{1}{4g_3^2}G^a_{\mu\nu}G^{a\mu\nu}+\frac{1}{4g_2^2}W^i_{\mu\nu}W^{i\mu\nu}+\frac{1}{4g_1^2}B_{\mu\nu}B^{\mu\nu}\right]\,\mathrm{d}^4x. \tag*{(310)} S YM , eff = ∫ M g [ 4 g 3 2 1 G μν a G a μν + 4 g 2 2 1 W μν i W i μν + 4 g 1 2 1 B μν B μν ] d 4 x . (310)
Electromagnetism then appears after electroweak breaking as the unbroken U ( 1 ) e m U(1)_{\mathrm{em}} U ( 1 ) em connection. The photon is not a fourth microscopic connection in this construction; it is the massless residual gauge mode of the broken electroweak phase. The conventional gauge-theoretic and electroweak structures are recalled in [167], [81], [160], and [143], while the standard symmetry-breaking mechanism is represented by [95] and [65].
Internal-connection variation
We now verify the corresponding gauge equation rather than identifying the Yang–Mills term at the level of the action only. Choose a basis { T A } \{T_A\} { T A } of the residual Lie algebra with [ T B , T C ] = f A B C T A [T_B,T_C]=f^{A}{}_{BC}T_A [ T B , T C ] = f A B C T A , and write
F = F A T A , F μ ν A = 2 ∂ [ μ A ν ] A + f A B C A μ B A ν C . (311) F=F^A T_A,\qquad F^A_{\mu\nu}=2\partial_{[\mu}A^A_{\nu]}+f^A{}_{BC}A^B_\mu A^C_\nu. \tag*{(311)} F = F A T A , F μν A = 2 ∂ [ μ A ν ] A + f A B C A μ B A ν C . (311)
The invariant form κ ^ A B \widehat{\kappa}_{AB} κ A B is assumed here to be positive definite on the Euclidean residual sector and Ad ( G H ) \operatorname{Ad}(G_H) Ad ( G H ) -invariant. It is constant in the following subcalculation, as on the frozen scalar/coframe branch; for a dynamical κ ^ \widehat{\kappa} κ , the same integration by parts keeps it inside the covariant derivative. These assumptions are precisely what permit the following covariant integration by parts. The curvature variation is
δ F μ ν A = 2 ( D [ μ δ A ν ] ) A , ( D μ X ) A ≡ ∂ μ X A + f A B C A μ B X C . (312) \delta F^A_{\mu\nu}=2(D_{[\mu}\delta A_{\nu]})^A,\qquad(D_\mu X)^A\equiv\partial_\mu X^A+f^A{}_{BC}A^B_\mu X^C. \tag*{(312)} δ F μν A = 2 ( D [ μ δ A ν ] ) A , ( D μ X ) A ≡ ∂ μ X A + f A B C A μ B X C . (312)
Varying (309) and using the antisymmetry of F A μ ν F^{A\mu\nu} F A μν gives
δ S Y M , e f f = 1 2 ∫ M g κ ^ A B F A μ ν δ F μ ν B d 4 x = ∫ M g κ ^ A B F A μ ν ( D μ δ A ν ) B d 4 x . (313) \begin{aligned}
\delta S_{\mathrm{YM,eff}}=\frac{1}{2}\int_M\sqrt{g}\,\widehat{\kappa}_{AB}F^{A\mu\nu}\delta F^B_{\mu\nu}\,\mathrm{d}^4x \\
&=\int_M\sqrt{g}\,\widehat{\kappa}_{AB}F^{A\mu\nu}(D_\mu\delta A_\nu)^B\,\mathrm{d}^4x. \tag*{(313)}
\end{aligned} δ S YM , eff = 2 1 ∫ M g κ A B F A μν δ F μν B d 4 x = ∫ M g κ A B F A μν ( D μ δ A ν ) B d 4 x . (313)
Assume that the boundary term vanishes. Gauge-covariant integration by parts, together with the invariance of κ \kappa κ , then yields
δ S Y M , e f f = − ∫ M g D μ ( κ ^ A B F B μ ν ) δ A ν A d 4 x . (314) \delta S_{\mathrm{YM,eff}}=-\int_M\sqrt{g}\,D_\mu\left(\widehat{\kappa}_{AB}F^{B\mu\nu}\right)\delta A^A_\nu\,\mathrm{d}^4x. \tag*{(314)} δ S YM , eff = − ∫ M g D μ ( κ A B F B μν ) δ A ν A d 4 x . (314)
Define the connection current of the remaining low-energy matter and order-parameter action by
δ S s r c = ∫ M δ A A ∧ ⋆ J A = ∫ M g J A ν δ A ν A d 4 x , J A ν : = 1 g δ S s r c δ A ν A . (315) \delta S_{\mathrm{src}}=\int_M\delta A^A\wedge\star J_A=\int_M\sqrt{g}\,J^\nu_A\delta A^A_\nu\,\mathrm{d}^4x,\qquad J^\nu_A:=\frac{1}{\sqrt{g}}\frac{\delta S_{\mathrm{src}}}{\delta A^A_\nu}. \tag*{(315)} δ S src = ∫ M δ A A ∧ ⋆ J A = ∫ M g J A ν δ A ν A d 4 x , J A ν := g 1 δ A ν A δ S src . (315)
Here J A = J A , μ d x μ J_A=J_{A,\mu}\mathrm{d}x^\mu J A = J A , μ d x μ , and the order of the forms fixes the sign: ⋆ J A ∧ δ A A = − δ A A ∧ ⋆ J A \star J_A\wedge\delta A^A=-\delta A^A\wedge\star J_A ⋆ J A ∧ δ A A = − δ A A ∧ ⋆ J A . Thus the three-form source in (170) is J e x t , A = − ⋆ J A J_{\mathrm{ext},A}=-\star J_A J ext , A = − ⋆ J A . Equivalently,
D ( κ ^ A B ⋆ F B ) = − ⋆ [ g ν λ D μ ( κ ^ A B F B μ ν ) d x λ ] . (316) D\left(\widehat{\kappa}_{AB}\star F^B\right)=-\star\left[g_{\nu\lambda}D_\mu\left(\widehat{\kappa}_{AB}F^{B\mu\nu}\right)\mathrm{d}x^\lambda\right]. \tag*{(316)} D ( κ A B ⋆ F B ) = − ⋆ [ g ν λ D μ ( κ A B F B μν ) d x λ ] . (316)
In this identity D μ D_\mu D μ contains both the gauge connection and the Levi–Civita connection on spacetime indices. The latter is essential for integration against g d 4 x \sqrt{g}\,\mathrm{d}^4x g d 4 x . The stationarity condition is exactly
D μ ( κ ^ A B F B μ ν ) = J A ν . (317) D_\mu\left(\widehat{\kappa}_{AB}F^{B\mu\nu}\right)=J_A^\nu. \tag*{(317)} D μ ( κ A B F B μν ) = J A ν . (317)
In a basis with κ A B = δ A B \kappa_{AB}=\delta_{AB} κ A B = δ A B , this becomes
D μ F A μ ν = J A ν . (318) D_\mu F^{A\mu\nu}=J^{A\nu}. \tag*{(318)} D μ F A μν = J A ν . (318)
For a canonically normalized simple factor on the frozen branch with κ A B = δ A B / g a 2 \kappa_{AB}=\delta_{AB}/g_a^2 κ A B = δ A B / g a 2 , the same equation reads D μ F A μ ν = g a 2 J A ν D_\mu F^{A\mu\nu}=g_a^2J^{A\nu} D μ F A μν = g a 2 J A ν ; absorbing g a 2 g_a^2 g a 2 into the definition of the source gives the displayed canonical convention. If κ ^ \widehat{\kappa} κ depends on a scalar or coframe, the derivative in (317) must be expanded and the corresponding variation contributes to the scalar and metric equations. The covariant conservation law can be checked directly from the equation of motion. Set K A μ ν : = κ ^ A B F B μ ν K_A^{\mu\nu}:=\widehat{\kappa}_{AB}F^{B\mu\nu} K A μν := κ A B F B μν . Antisymmetry and an exchange of dummy indices give
D ν J A ν = D ν D μ K A μ ν = 1 2 [ D ν , D μ ] K A μ ν = 0. (319) D_\nu J_A^\nu=D_\nu D_\mu K_A^{\mu\nu}=\frac{1}{2}[D_\nu,D_\mu]K_A^{\mu\nu}=0. \tag*{(319)} D ν J A ν = D ν D μ K A μν = 2 1 [ D ν , D μ ] K A μν = 0. (319)
The spacetime-curvature contractions vanish because the Ricci tensor is symmetric and K A μ ν K_A^{\mu\nu} K A μν is antisymmetric. The gauge-curvature contraction vanishes by the symmetric Ad-invariant pairing: in a simple factor it is a scalar multiple of [ F ν μ , F μ ν ] = 0 [F_{\nu\mu},F^{\mu\nu}]=0 [ F ν μ , F μν ] = 0 . Abelian factors contribute no commutator. Gauge invariance of the source action yields the same identity on its matter equations. This is compatible with, but does not follow merely from, the differential Bianchi identity D [ μ F ν ρ ] = 0 D_{[\mu}F_{\nu\rho]}=0 D [ μ F ν ρ ] = 0 . Equations (317) –(318) are the internal low-energy projection of the fixed-coframe unified connection equation ι k ∗ ( D P − J e x t ) = 0 \iota^*_k(\mathbb{D}P-J_{\mathrm{ext}})=0 ι k ∗ ( D P − J ext ) = 0 in (171) , after the residual pairing has become block-diagonal and the background metric is fixed. In the present Hodge-compatible case, its internal constitutive momentum is P H A = κ ^ A B ⋆ F B P_H^A=\widehat{\kappa}_{AB}\star F^B P H A = κ A B ⋆ F B , so the projected equation is exactly D ( κ ^ A B ⋆ F B ) = − ⋆ J A D(\widehat{\kappa}_{AB}\star F^B)=-\star J_A D ( κ A B ⋆ F B ) = − ⋆ J A . A mixed or non-Hodge-compatible master pairing instead gives the restricted constitutive equation ι k ∗ [ D ( ∂ L / ∂ F ) − J e x t ] = 0 \iota^*_k[\mathbb{D}(\partial\mathcal{L}/\partial F)-J_{\mathrm{ext}}]=0 ι k ∗ [ D ( ∂ L / ∂ F ) − J ext ] = 0 . Therefore the displayed Yang–Mills equation is the exact internal projection under the stated constitutive hypotheses. Its conditional scope makes it a controlled recovery theorem, not an additional microscopic action.
From the common action to a controlled Einstein–Yang–Mills comparison
The preceding variations identify a common origin for the gravitational and internal equations. To compare their solutions with familiar physics, one must also specify which coefficients are held fixed, which additional terms are small, and which stability estimate turns an equation error into a field error. These questions connect the variational construction to the physical matching problem. They also distinguish a numerical approximation of the mother equations from a low-energy reduction of those equations.
The geometric source of the Einstein term. The coframe supplies a length-valued one-form that relates local frame components to spacetime distances. Its occurrence in the Cartan connection therefore carries more information than an independent internal gauge connection. In the geometric branch of (217), the curvature has components
F c a b = R a b − ℓ − 2 e a ∧ e b , F c a 5 = ℓ − 1 T a . (320) F_{\mathrm{c}}^{ab}=R^{ab}-\ell^{-2}e^a\wedge e^b,\qquad F_{\mathrm{c}}^{a5}=\ell^{-1}T^a. \tag*{(320)} F c ab = R ab − ℓ − 2 e a ∧ e b , F c a 5 = ℓ − 1 T a . (320)
For an orthonormal Euclidean coframe in four dimensions, the six forms e a ∧ e b e^a\wedge e^b e a ∧ e b , a < b a<b a < b , have unit squared norm. With the curvature convention of (1097),
∑ a < b ∣ R a b ∣ 2 = 1 4 ∣ R m ∣ 2 , ∑ a < b ⟨ R a b , e a ∧ e b ⟩ = 1 2 S . (321) \sum_{a<b}\lvert R^{ab}\rvert^2=\frac{1}{4}\lvert\mathrm{Rm}\rvert^2,\qquad\sum_{a<b}\langle R^{ab},e^a\wedge e^b\rangle=\frac{1}{2}S. \tag*{(321)} a < b ∑ ∣ R ab ∣ 2 = 4 1 ∣ Rm ∣ 2 , a < b ∑ ⟨ R ab , e a ∧ e b ⟩ = 2 1 S . (321)
Expanding the square gives the geometric contribution
S c , E = w ∗ 2 g c 2 ∫ M ( 1 4 ∣ R m ∣ 2 − ℓ − 2 S + 6 ℓ − 4 + ℓ − 2 ∑ a ∣ T a ∣ 2 ) vol g . (322) S_{\mathrm{c,E}}=\frac{w_*}{2g_{\mathrm{c}}^2}\int_M\left(\frac{1}{4}\lvert\mathrm{Rm}\rvert^2-\ell^{-2}S+6\ell^{-4}+\ell^{-2}\sum_a\lvert T^a\rvert^2\right)\operatorname{vol}_g. \tag*{(322)} S c , E = 2 g c 2 w ∗ ∫ M ( 4 1 ∣ Rm ∣ 2 − ℓ − 2 S + 6 ℓ − 4 + ℓ − 2 a ∑ ∣ T a ∣ 2 ) vol g . (322)
Thus the Einstein term comes from the cross contraction in the Cartan curvature norm. The continuation and consistent reduction stated in Section 14.39 yield M P 2 = w ∗ / ( g c 2 ℓ 2 ) M_{\mathrm{P}}^2=w_*/(g_{\mathrm{c}}^2\ell^2) M P 2 = w ∗ / ( g c 2 ℓ 2 ) and the unmodified vacuum coefficient Λ = 3 ℓ − 2 \Lambda=3\ell^{-2} Λ = 3 ℓ − 2 . Neither a generic connection norm nor an arbitrary additional scalar supplies this conclusion. In particular, changing the vacuum coefficient is additional matching data. The exact Einstein critical points under independent connection and coframe variation are proved in Proposition 159.
The generalized curvature in (52) is a typed tuple. Its matter and potential entries need not be differential forms of the same degree. The common pairing makes their contraction well defined; it does not identify the tuple with the square of an unspecified superconnection. Likewise, the coupling of sectors is a property of the action, reference law and admissible reductions together. A mixed Hessian can display a coupling in chosen coordinates, but is not an invariant criterion for the factorization in Definition 2. For example, an interaction x 2 y 2 x^2y^2 x 2 y 2 has zero mixed Hessian at the origin while coupling the variables away from it.
Two coefficient errors in the internal equation. Write H = κ ^ H=\widehat{\kappa} H = κ for the symmetric invariant kinetic pairing in the preceding calculation. Invariance permits independent constants on distinct compact simple ideals; abelian factors can have kinetic mixing. The comparison pairing H 0 H_0 H 0 is fixed, parallel and invertible. Both act from the adjoint bundle to its dual. All statements below use one gauge and one source convention, namely (315) . On a fixed branch define the equation defect r r r and the canonical comparison residual R A \mathcal{R}_A R A by
D μ ( H F μ ν ) = J ν + r ν , R A ν = D μ F μ ν − H 0 − 1 J ν . (323) D_\mu(HF^{\mu\nu}) = J^\nu+ r^\nu,\qquad\mathcal{R}_A^\nu= D_\mu F^{\mu\nu} - H_0^{-1}J^\nu. \tag*{(323)} D μ ( H F μν ) = J ν + r ν , R A ν = D μ F μν − H 0 − 1 J ν . (323)
Here a defect can represent a retained physical term or a numerical equation error; its origin must be specified before taking a limit.
Proposition 12 (The coefficient-normalized gauge residual). Suppose the terms in (323) are defined and D H 0 = 0 DH_0=0 D H 0 = 0 . Then
H 0 R A ν = r ν − ( D μ H ) F μ ν − ( H − H 0 ) D μ F μ ν . (324) H_0\mathcal{R}_A^\nu= r^\nu- (D_\mu H)F^{\mu\nu} - (H-H_0)D_\mu F^{\mu\nu}. \tag*{(324)} H 0 R A ν = r ν − ( D μ H ) F μν − ( H − H 0 ) D μ F μν . (324) For 1 ≤ p ≤ ∞ 1\leq p\leq\infty 1 ≤ p ≤ ∞ , use a positive auxiliary bundle norm and the induced norms for the displayed contractions. If their right-hand side is in L p L^p L p , then
∥ R A ∥ L p ≤ ∥ H 0 − 1 ∥ ( ∥ r ∥ L p + ∥ D H ∥ ∞ ∥ F ∥ L p + ∥ H − H 0 ∥ ∞ ∥ D μ F μ ∥ L p ) . (325) \|\mathcal{R}_A\|_{L^p}\leq\|H_0^{-1}\|\left(\|r\|_{L^p}+\|DH\|_\infty\|F\|_{L^p}+\|H-H_0\|_\infty\|D_\mu F^\mu\|_{L^p}\right). \tag*{(325)} ∥ R A ∥ L p ≤ ∥ H 0 − 1 ∥ ( ∥ r ∥ L p + ∥ D H ∥ ∞ ∥ F ∥ L p + ∥ H − H 0 ∥ ∞ ∥ D μ F μ ∥ L p ) . (325) If H H H is pointwise symmetric and A d Ad A d -invariant, the same equation also implies the distributional compatibility condition
D ν J ν = − D ν r ν . (326) D_\nu J^\nu=-D_\nu r^\nu. \tag*{(326)} D ν J ν = − D ν r ν . (326)
Proof. The product rule gives D μ ( H F μ ν ) = H D μ F μ ν + ( D μ H ) F μ ν D_\mu(HF^{\mu\nu})=HD_\mu F^{\mu\nu}+(D_\mu H)F^{\mu\nu} D μ ( H F μν ) = H D μ F μν + ( D μ H ) F μν . Subtract J J J , split H = H 0 + ( H − H 0 ) H=H_0+(H-H_0) H = H 0 + ( H − H 0 ) , and move both coefficient terms to the other side. This proves (324) . Apply H 0 − 1 H_0^{-1} H 0 − 1 , the triangle inequality and the multiplication bound ∥ B u ∥ L p ≤ ∥ B ∥ ∞ ∥ u ∥ L p \|Bu\|_{L^p}\leq\|B\|_\infty\|u\|_{L^p} ∥ B u ∥ L p ≤ ∥ B ∥ ∞ ∥ u ∥ L p . The contraction norm incorporates the sum over the spacetime index, so no unspecified component factor is suppressed. Finally, antisymmetry gives D ν D μ ( H F μ ν ) = 1 2 [ D ν , D μ ] ( H F μ ν ) = 0 D_\nu D_\mu(HF^{\mu\nu})=\frac{1}{2}[D_\nu,D_\mu](HF^{\mu\nu})=0 D ν D μ ( H F μν ) = 2 1 [ D ν , D μ ] ( H F μν ) = 0 . The Ricci contractions and the internal commutator vanish for the reasons given after (319) . Taking the divergence of (323) proves (326) . □
Small derivatives of H H H alone are insufficient for recovery with a prescribed source normalization. A nonmatching constant coefficient has zero derivative and still changes H 0 − 1 J H_0^{-1}J H 0 − 1 J . Conversely, a variable coefficient can give a canonical equation on a special solution if the two coefficient terms cancel. The proposition gives a sufficient bound, not a necessary characterization of every such solution. If a charged background replaces the invariant pairing, its own variation and the corresponding Noether identity must be included; the last conclusion does not extend by simply dropping that equation.
Variations, boundary data and physical units
The preceding local statements use compactly supported variations on the interior of M M M , or a closed Euclidean manifold. For a nonempty boundary the gauge surface variation contains ∫ ∂ M δ A A ∧ H A B ⋆ F B \int_{\partial M}\delta A^A\wedge H_{AB}\star F^B ∫ ∂ M δ A A ∧ H A B ⋆ F B , with the orientation inherited from Stokes’ theorem. Fixing the tangential connection removes this term. Prescribing flux instead requires a different boundary variational problem. There is no universal boundary functional ⟨ F , A ⟩ \langle F,A\rangle ⟨ F , A ⟩ that simultaneously supplies the gauge, coframe and independent-connection conditions. For the metric Einstein–Hilbert action, fixing the induced metric uses the corresponding Gibbons–Hawking–York term on a non-null boundary; higher-curvature terms require their own boundary data. These choices are unnecessary for the compact-support statements proved here.
All comparisons must also respect units. In four dimensions with ℏ = c = 1 \hbar= c = 1 ℏ = c = 1 , the component quantities A μ A_\mu A μ , F μ ν F_{\mu\nu} F μν , J ν J^\nu J ν have dimensions L − 1 L^{-1} L − 1 , L − 2 L^{-2} L − 2 , L − 3 L^{-3} L − 3 , while H H H is dimensionless. Every term in (324) consequently has dimension L − 3 L^{-3} L − 3 . The Einstein residual G μ ν + Λ g μ ν − 8 π G T μ ν G_{\mu\nu} + \Lambda g_{\mu\nu} - 8\pi G T_{\mu\nu} G μν + Λ g μν − 8 π G T μν has dimension L − 2 L^{-2} L − 2 . Choosing a reference length L 0 L_0 L 0 makes these residuals dimensionless by multiplying them by L 0 3 L_0^3 L 0 3 and L 0 2 L_0^2 L 0 2 , respectively, before defining a combined error. Their unscaled numerical values cannot be added as if they were the same observable.
Four distinct steps in recovery. Equation consistency asks whether a given configuration has a small Einstein–Yang–Mills residual. Compactness asks whether a family of configurations has a convergent subsequence. Stability asks whether a small residual implies proximity to a selected solution. Physical evolution asks whether the proximity persists on a time interval. Each step has assumptions not supplied by the preceding one.
For example, fix a compact smooth four-dimensional manifold, a finite bundle atlas, and compatible diffeomorphism and internal gauge choices. Suppose e j e_j e j are bounded in H s H^s H s , A j A_j A j , ω j \omega_j ω j are bounded in H s − 1 H^{s-1} H s − 1 , and s > 4 s > 4 s > 4 . Also impose a uniform lower bound on the smallest singular value of e j e_j e j , measured using fixed positive reference metrics. Sobolev compactness then gives a subsequence with e j → e e_j \to e e j → e in C 2 C^2 C 2 and A j , ω j → A , ω A_j,\omega_j \to A,\omega A j , ω j → A , ω in C 1 C^1 C 1 . This assertion does not provide bounds or gauges for an arbitrary sequence of mother-theory solutions. In particular, on these regularity assumptions D μ F μ ν D_\mu F^{\mu\nu} D μ F μν is naturally in H s − 3 H^{s-3} H s − 3 , not H s − 1 H^{s-1} H s − 1 .
If, in addition, H j → H 0 H_j \to H_0 H j → H 0 in C 1 C^1 C 1 , J j → J J_j \to J J j → J in L 1 L^1 L 1 , and r j → 0 r_j \to0 r j → 0 in distributions, testing (323) against a smooth one-form and integrating by parts passes the equation to H 0 D μ F μ ν = J ν H_0D_\mu F^{\mu\nu} = J^\nu H 0 D μ F μν = J ν . The integrated terms involve H j F j H_jF_j H j F j and first derivatives of the test form, with the convergent metric and connection coefficients. They therefore converge under the stated assumptions without demanding pointwise convergence of second derivatives of A j A_j A j . If instead r j → r ∞ ≠ 0 r_j \to r_\infty\ne0 r j → r ∞ = 0 , this source remains in the limiting equation. Applying the same argument to a quantitatively matched coframe equation requires convergence of the stress tensor and of its actual geometric remainder. The Einstein tensor itself is continuous under the above C 2 C^2 C 2 convergence. These conclusions do not imply C ∞ C^\infty C ∞ convergence or removal of all physical scales.
For local stationary stability, Proposition 157 requires an invertible gauge-fixed linearization with a bounded inverse, together with bounds on the nonlinear and remainder terms. Gauge modes and physical zero modes must first be treated. Under its hypotheses, an equation defect bounded by δ \delta δ gives a field displacement at most 2 K δ 2K\delta 2 K δ , where K K K bounds the inverse. The dependence on K K K is indispensable: the scalar equation ε 2 z − ε = 0 \varepsilon^2 z-\varepsilon=0 ε 2 z − ε = 0 has residual ε \varepsilon ε at zero but solution z = ε − 1 z=\varepsilon^{-1} z = ε − 1 .
For Lorentzian evolution, an elliptic inverse cannot replace an energy estimate. Suppose a specified hyperbolic reduction and its propagated constraints yield, on [ 0 , T ] [0,T] [ 0 , T ] , an energy E ( t ) ≥ 0 E(t)\ge0 E ( t ) ≥ 0 for the difference of two solutions satisfying
E ′ ( t ) ≤ C ( t ) E ( t ) + f ( t ) 2 , C ∈ L 1 ( 0 , T ) , f ∈ L 2 ( 0 , T ) . (327) E'(t) \le C(t)E(t)+f(t)^2,\qquad C\in L^1(0,T),\qquad f\in L^2(0,T). \tag*{(327)} E ′ ( t ) ≤ C ( t ) E ( t ) + f ( t ) 2 , C ∈ L 1 ( 0 , T ) , f ∈ L 2 ( 0 , T ) . (327)
Multiplication by exp ( − ∫ 0 t C ) \exp(-\int_0^t C) exp ( − ∫ 0 t C ) and integration give
E ( t ) ≤ e ∫ 0 t C ( u ) d u E ( 0 ) + ∫ 0 t e ∫ v t C ( u ) d u f ( v ) 2 d v . (328) E(t)\le e^{\int_0^t C(u)\,\mathrm{d}u}E(0)+\int_0^t e^{\int_v^t C(u)\,\mathrm{d}u}f(v)^2\,\mathrm{d}v. \tag*{(328)} E ( t ) ≤ e ∫ 0 t C ( u ) d u E ( 0 ) + ∫ 0 t e ∫ v t C ( u ) d u f ( v ) 2 d v . (328)
Here f f f must be obtained from the actual forcing and matching defect of the reduced equations, and E E E must control the chosen solution norm. The calculation does not establish the required hyperbolic reduction for an unrestricted higher-derivative branch. It specifies the estimate that a finite-time recovery argument must justify, including its initial data and constraint equations.
Finite examples that separate the errors. The following calculations use dimensionless synthetic fields. They isolate individual implications of the recovery argument; they are not observations or a nonlinear simulation of the entire mother theory. On the periodic interval [ 0 , 2 π ) [0,2\pi) [ 0 , 2 π ) , prescribe a transverse abelian potential and its kinetic coefficient, with ∣ α ∣ < 1 |\alpha|<1 ∣ α ∣ < 1 , by
A y = sin x , F x y = cos x , H = 1 + α cos x , J y = − sin x − α sin ( 2 x ) . (329) A_y=\sin x,\qquad F_{xy}=\cos x,\qquad H=1+\alpha\cos x,\qquad J_y=-\sin x-\alpha\sin(2x). \tag*{(329)} A y = sin x , F x y = cos x , H = 1 + α cos x , J y = − sin x − α sin ( 2 x ) . (329)
Then ∂ x ( H F x y ) = J y \partial_x(HF_{xy})=J_y ∂ x ( H F x y ) = J y , whereas ∂ x F x y − J y = α sin ( 2 x ) \partial_xF_{xy}-J_y=\alpha\sin(2x) ∂ x F x y − J y = α sin ( 2 x ) . Each of the two coefficient terms in (325) contributes half this canonical residual. With h = 2 π / N h=2\pi/N h = 2 π / N , the centered difference of the exact flux gives the exact discrete root-mean-square error
e h 2 = 1 2 [ ( 1 − sinc h ) 2 + α 2 ( 1 − sinc ( 2 h ) ) 2 ] , sinc z = sin z z . (330) e_h^2=\frac{1}{2}\left[(1-\operatorname{sinc}h)^2+\alpha^2(1-\operatorname{sinc}(2h))^2\right],\qquad\operatorname{sinc}z=\frac{\sin z}{z}. \tag*{(330)} e h 2 = 2 1 [ ( 1 − sinc h ) 2 + α 2 ( 1 − sinc ( 2 h ) ) 2 ] , sinc z = z sin z . (330)
The formula holds for the uniform meshes below, by discrete orthogonality of the two resolved sine modes. Its leading order is e h = h 2 ( 1 + 16 α 2 ) / 72 + O ( h 4 ) e_h=h^2\sqrt{(1+16\alpha^2)/72}+O(h^4) e h = h 2 ( 1 + 16 α 2 ) /72 + O ( h 4 ) . The canonical continuum residual instead has root-mean-square norm ∣ α ∣ / 2 |\alpha|/\sqrt{2} ∣ α ∣/ 2 . Thus a convergent discretization can converge accurately to a constitutive equation different from the canonical one.
A second comparison uses the actual quadratic gravitational multiplier of (1775). For Fourier modes k = 1 , … , 12 k=1,\ldots,12 k = 1 , … , 12 and a unit source in each mode, compare
h 0 ( k ) = k − 2 , h ℓ ( k ) = [ k 2 ( 1 + ℓ 2 k 2 ) ] − 1 , h ℓ , h ( k ) = [ q h ( k ) ( 1 + ℓ 2 q h ( k ) ) ] − 1 . (331) h_0(k)=k^{-2},\qquad h_\ell(k)=\left[k^2(1+\ell^2k^2)\right]^{-1},\qquad h_{\ell,h}(k)=\left[q_h(k)(1+\ell^2q_h(k))\right]^{-1}. \tag*{(331)} h 0 ( k ) = k − 2 , h ℓ ( k ) = [ k 2 ( 1 + ℓ 2 k 2 ) ] − 1 , h ℓ , h ( k ) = [ q h ( k ) ( 1 + ℓ 2 q h ( k )) ] − 1 . (331)
Table 2. Synthetic periodic gauge calculation at α = 0.2 \alpha= 0.2 α = 0.2 . The second column tests the constitutive equation; the third compares with the constant-coefficient equation using the same source.
where q h ( k ) = 4 h − 2 sin 2 ( k h / 2 ) q_h(k) = 4h^{-2}\sin^2(kh/2) q h ( k ) = 4 h − 2 sin 2 ( k h /2 ) is the centered Laplacian symbol. At ℓ = 0.1 \ell= 0.1 ℓ = 0.1 , increasing N N N from 64 to 1024 reduces ∥ h ℓ , h − h ℓ ∥ 2 / ∥ h ℓ ∥ 2 \|h_{\ell,h} - h_\ell\|_2/\|h_\ell\|_2 ∥ h ℓ , h − h ℓ ∥ 2 /∥ h ℓ ∥ 2 from 2.45772 ⋅ 10 − 3 2.45772 \cdot10^{-3} 2.45772 ⋅ 1 0 − 3 to 9.35572 ⋅ 10 − 6 9.35572 \cdot10^{-6} 9.35572 ⋅ 1 0 − 6 . The separate continuum correction ∥ h ℓ − h 0 ∥ 2 / ∥ h 0 ∥ 2 = 0.02435798 \|h_\ell- h_0\|_2/\|h_0\|_2 = 0.02435798 ∥ h ℓ − h 0 ∥ 2 /∥ h 0 ∥ 2 = 0.02435798 is unchanged. The sharp band bound (1776) is a worst-source bound; a source concentrated in low modes need not saturate it.
The role of stability can be seen without an infinite-dimensional existence assumption. For 0 < ε < 1 0 < \varepsilon< 1 0 < ε < 1 , consider
V ε ( x , y ) = 1 2 ( x 2 + y 2 ) + ε x y + ε x . (332) V_\varepsilon(x,y) = \frac{1}{2}(x^2+y^2) + \varepsilon xy + \varepsilon x. \tag*{(332)} V ε ( x , y ) = 2 1 ( x 2 + y 2 ) + ε x y + ε x . (332)
Its Hessian has eigenvalues 1 ± ε 1 \pm\varepsilon 1 ± ε , so it has the unique global minimizer
( x ε , y ε ) = ( − ε 1 − ε 2 , ε 2 1 − ε 2 ) , ∥ ( x ε , y ε ) ∥ 2 ≤ ε 1 − ε . (333) (x_\varepsilon,y_\varepsilon) = \left(-\frac{\varepsilon}{1-\varepsilon^2},\frac{\varepsilon^2}{1-\varepsilon^2}\right), \qquad\|(x_\varepsilon,y_\varepsilon)\|_2 \le\frac{\varepsilon}{1-\varepsilon}. \tag*{(333)} ( x ε , y ε ) = ( − 1 − ε 2 ε , 1 − ε 2 ε 2 ) , ∥ ( x ε , y ε ) ∥ 2 ≤ 1 − ε ε . (333)
Indeed, solving the two linear stationarity equations gives the displayed point, and the inverse Hessian has norm ( 1 − ε ) − 1 (1-\varepsilon)^{-1} ( 1 − ε ) − 1 . This finite example demonstrates the mechanism of a residual-to-distance bound, rather than verifying the Einstein linearization hypotheses.
Finally, a periodic 4 × 4 4 \times4 4 × 4 S U ( 2 ) SU(2) S U ( 2 ) lattice tests the internal covariance. Let T a = i σ a / 2 T_a = i\sigma_a/2 T a = i σ a /2 , with σ a \sigma_a σ a the Pauli matrices, and use link variables U x μ U_{x\mu} U xμ . The Wilson functional and a local gauge transformation are
S W = ∑ p ( 1 − 1 2 Re tr U p ) , U x μ ↦ G x U x μ G x + μ † . (334) S_W = \sum_p \left(1-\frac{1}{2}\operatorname{Re}\operatorname{tr} U_p\right), \qquad U_{x\mu} \mapsto G_x U_{x\mu}G_{x+\mu}^{\dagger}. \tag*{(334)} S W = p ∑ ( 1 − 2 1 Re tr U p ) , U xμ ↦ G x U xμ G x + μ † . (334)
For reproducibility, the link exponent vectors are drawn as 0.3 ξ 0.3\xi 0.3 ξ and the gauge exponent vectors as η \eta η , with independent standard Gaussian components and generator seed 20261028. The force f x μ , a = ∂ t S W ( e t T a U x μ ) ∣ t = 0 f_{x\mu,a} = \left.\partial_t S_W(e^{tT_a}U_{x\mu})\right|_{t=0} f xμ , a = ∂ t S W ( e t T a U xμ ) t = 0 is computed by differentiating each plaquette product. It transforms by f x μ , a ′ = R a b ( G x ) f x μ , b f'_{x\mu,a} = R_{ab}(G_x)f_{x\mu,b} f xμ , a ′ = R ab ( G x ) f xμ , b , where R a b ( G ) = − 2 Re tr ( T a G T b G † ) R_{ab}(G) = -2\operatorname{Re}\operatorname{tr}(T_aGT_bG^\dagger) R ab ( G ) = − 2 Re tr ( T a G T b G † ) . In double precision, the action difference is 8.89 ⋅ 10 − 16 8.89 \cdot10^{-16} 8.89 ⋅ 1 0 − 16 or less, and the largest component of the analytic force-covariance defect is 3.06 · 10 − 16 10^{-16} 1 0 − 16 or less. Central differences provide a separate check of the derivative. For a single varied link, S W ( t ) = a + b cos ( t / 2 ) + c sin ( t / 2 ) SW(t) = a + b\cos(t/2) + c\sin(t/2) S W ( t ) = a + b cos ( t /2 ) + c sin ( t /2 ) , so the central derivative equals the exact force multiplied by sinc ( t / 2 ) \operatorname{sinc}(t/2) sinc ( t /2 ) . This supplies its truncation error without choosing an arbitrary acceptance threshold. At very small steps, floating-point subtraction eventually dominates that error.
Figure 3. Four synthetic comparisons. (a) A variable-coefficient gauge equation is resolved with second-order accuracy while its canonical mismatch persists. (b) Mesh error and the fixed physical curvature correction have different limits. (c) A coercive finite stationary model obeys its explicit displacement bound. (d) Independent central differences of the lattice action agree with its analytic derivative until roundoff competes with truncation. These calculations test the stated finite mechanisms, not a full nonlinear or quantum continuum completion.
These comparisons make the next field-theoretic task precise. The common action supplies the coefficients and the exact projected equations. A selected branch must then supply uniform regularity, constraint propagation and stability, with the physical correction kept separate from the discretization error. The matter equations considered next determine the sources in that comparison and their exchange with the geometric and internal fields.
Matter, order parameter, and information.
Matter coupling.
The master Dirac-type operator is written schematically as
D A = γ μ D μ + Y ( Φ H , χ ) , (335) \mathcal{D}_{A} = \gamma^{\mu}D_{\mu} + \mathcal{Y}(\Phi_{H},\chi), \tag*{(335)} D A = γ μ D μ + Y ( Φ H , χ ) , (335)
where Y \mathcal{Y} Y is a master Yukawa-type endomorphism. The following notation denotes the matter component of the low-energy expansion of (52) , not an additional primitive action:
S m a t = ∫ M g Ψ ‾ D A Ψ d 4 x . (336) S_{\mathrm{mat}} = \int_{M} \sqrt{g}\,\overline{\Psi}\mathcal{D}_{A}\Psi\, \mathrm{d}^{4}x. \tag*{(336)} S mat = ∫ M g Ψ D A Ψ d 4 x . (336)
The displayed first-order Grassmann expression is a conditional effective identification. The extended norm in (51) contains the quadratic component ⟨ D Ψ , D Ψ ⟩ \langle D\Psi,D\Psi\rangle ⟨ D Ψ , D Ψ ⟩ ; obtaining a first-order Dirac action requires an additional fermionic construction, such as a supertrace, a spectral determinant, or the elimination of auxiliary variables. No such construction is assumed in the abstract framework, so the Dirac term is not claimed to follow from the norm alone. For a complete physical theory, the representations of Ψ \Psi Ψ , chirality, reality conditions, and anomaly cancellation must all follow from the master representation problem. The conventional normalization and chiral-field bookkeeping can be compared with [128] and [162].
More explicitly, whenever the selected Euclidean geometry admits a spin structure, the matter field has the form
Ψ ∈ Γ ( S g [ X ] ⊗ E R ) , E R : = P H × R V R , (337) \Psi\in\Gamma\left(S_{g}[X]\otimes E_{\mathcal{R}}\right), \qquad E_{\mathcal{R}} := P_{H}\times_{\mathcal{R}} V_{\mathcal{R}}, \tag*{(337)} Ψ ∈ Γ ( S g [ X ] ⊗ E R ) , E R := P H × R V R , (337)
where S g [ X ] S_{g}[X] S g [ X ] is the spinor bundle of the induced metric and R \mathcal{R} R is a representation supplied by the master representation problem. In the worked candidate, the familiar fields Q L , u R , d R , L L , e R Q_{L},u_{R},d_{R},L_{L},e_{R} Q L , u R , d R , L L , e R are prescribed representation components. In a predictive realization, the selection problem must instead determine why these components, and any additional fields, are retained. Equation (337) is conditional on the existence of a spin lift and on the extension of the relevant H H H -representation; it does not by itself prove the observed chiral spectrum.
Information as a state coordinate.
The information variable χ \chi χ is defined as a coordinate on the state manifold rather than as an independently added force. If its state space carries a linear representation ρ χ \rho_{\chi} ρ χ of the master group, a minimal covariant coupling is
D χ = d χ + ρ χ ( A ) χ . (338) D\chi= \mathrm{d}\chi+ \rho_{\chi}(A)\chi. \tag*{(338)} D χ = d χ + ρ χ ( A ) χ . (338)
The model may then admit an information–geometry interaction through the same master metric,
∥ D χ ∥ G 2 . (339) \lVert D\chi\rVert_G^2 . \tag*{(339)} ∥ D χ ∥ G 2 . (339)
The physical status of χ \chi χ remains open. It may ultimately be eliminated as an auxiliary variable, identified with an information-theoretic order parameter, or promoted to a microscopic field. The present formulation does not assume which interpretation is correct. Related geometric treatments of matter and internal data in a unified spectral setting include [45], [44], [10], [155], [83], [46], [28], and [30].
Spectral and heat-kernel reduction
The classical variations identify the field equations on a specified branch. Fluctuations around that branch contribute further local operators. The heat kernel organizes those operators by dimension; the determinant also tests whether the chosen background admits a stable Gaussian expansion. Positivity of a principal symbol and stability of the entire quadratic operator are different requirements.
One operator, one spectral object
Suppose the quadratic fluctuation operator around a selected background X ∗ X_\ast X ∗ is a Laplace-type operator
L = − ( g μ ν ∇ μ ∇ ν + E ) . (340) L = -\left(g^{\mu\nu}\nabla_\mu\nabla_\nu+ E\right). \tag*{(340)} L = − ( g μν ∇ μ ∇ ν + E ) . (340)
Here g g g is the metric induced by the selected coframe as in (37) ; on a preliminary regulator it means the auxiliary g ˉ \bar{g} g ˉ . Thus the heat kernel is a low-energy/effective expansion around a selected master geometry. It can generate the local invariants R R R , F 2 F^2 F 2 , and matter endomorphism terms once the operator and the selected background are specified, but it does not by itself construct a metric, a Cartan bundle, or a residual gauge group from no prior geometric data. For the determinant calculation, assume that the Euclidean manifold is smooth and closed and that L L L has a self-adjoint elliptic realization. After removing its finite-dimensional kernel, assume in addition that the remaining spectrum is strictly positive. Let P ker L P_{\ker L} P k e r L denote the orthogonal projection onto the kernel, and write
Tr ′ ( e − t L ) : = Tr ( e − t L ) − dim ker L = ∑ λ j > 0 e − t λ j . (341) \operatorname{Tr}'\left(e^{-tL}\right) := \operatorname{Tr}\left(e^{-tL}\right) - \dim\ker L = \sum_{\lambda_j>0} e^{-t\lambda_j}. \tag*{(341)} Tr ′ ( e − t L ) := Tr ( e − t L ) − dim ker L = λ j > 0 ∑ e − t λ j . (341)
The equality with the last sum uses the absence of negative eigenvalues. On a compact manifold the positive spectrum has a lower bound λ 1 > 0 \lambda_1 > 0 λ 1 > 0 . For t ≥ 1 t \ge1 t ≥ 1 ,
Tr ′ ( e − t L ) ≤ e − ( t − 1 ) λ 1 Tr ′ ( e − L ) , (342) \operatorname{Tr}'\left(e^{-tL}\right) \le e^{-(t-1)\lambda_1}\operatorname{Tr}'\left(e^{-L}\right), \tag*{(342)} Tr ′ ( e − t L ) ≤ e − ( t − 1 ) λ 1 Tr ′ ( e − L ) , (342)
so the large-t t t part of the proper-time integral converges. The Laplace-type principal symbol alone does not imply this positivity. For example, L = − Δ g − m 2 L = -\Delta_g - m^2 L = − Δ g − m 2 , with m > 0 m > 0 m > 0 , has a negative constant mode on a closed connected manifold; that mode contributes e t m 2 e^{tm^2} e t m 2 . Such a background requires a separate treatment of unstable directions and cannot be integrated as a positive Gaussian equilibrium.
The factor of the fluctuation scale can be fixed without a formal infinite-dimensional integral. Retain N N N positive modes, write their coordinates as y j y_j y j , and let L N L_N L N be the restricted Hessian. Gaussian integration gives
∫ R N exp ( − y T L N y 2 θ ) d y = ( 2 π θ ) N / 2 ( det L N ) − 1 / 2 , Δ S 1 -loop , N = θ 2 ∑ j = 1 N log ( λ j μ r e f 2 ) + C N . (343) \begin{aligned}
\int_{\mathbb{R}^N} \exp\left(-\frac{y^{\mathsf{T}}L_Ny}{2\theta}\right)\,dy &= (2\pi\theta)^{N/2}(\det L_N)^{-1/2},\\
\Delta S_{1\text{-loop},N} &= \frac{\theta}{2}\sum_{j=1}^{N}\log\left(\frac{\lambda_j}{\mu_{\mathrm{ref}}^2}\right)+C_N. \tag*{(343)}
\end{aligned} ∫ R N exp ( − 2 θ y T L N y ) d y Δ S 1 -loop , N = ( 2 π θ ) N /2 ( det L N ) − 1/2 , = 2 θ j = 1 ∑ N log ( μ ref 2 λ j ) + C N . (343)
Here μ r e f > 0 \mu_{\mathrm{ref}}>0 μ ref > 0 is a fixed mass unit and C N C_N C N records the chosen finite-dimensional reference normalization. The effective action is defined by the weight exp ( − S e f f / θ ) \exp(-S_{\mathrm{eff}}/\theta) exp ( − S eff / θ ) ; therefore its one-loop term contains θ \theta θ . Dividing the entire effective action by θ \theta θ instead gives the frequently used dimensionless 1 2 log det \frac{1}{2}\log\det 2 1 log det expression. These conventions agree with the explicit heavy-field integration in Section 11.10.
For positive λ \lambda λ and μ r e f 2 \mu_{\mathrm{ref}}^2 μ ref 2 , the identity
log λ μ r e f 2 = − ∫ 0 ∞ e − t λ − e − t μ r e f 2 t d t (344) \log\frac{\lambda}{\mu_{\mathrm{ref}}^2}
=
-\int_{0}^{\infty}\frac{e^{-t\lambda}-e^{-t\mu_{\mathrm{ref}}^2}}{t}\,dt \tag*{(344)} log μ ref 2 λ = − ∫ 0 ∞ t e − t λ − e − t μ ref 2 d t (344)
follows by differentiation with respect to λ \lambda λ and evaluation at λ = μ r e f 2 \lambda=\mu_{\mathrm{ref}}^2 λ = μ ref 2 . The numerator is O ( t ) O(t) O ( t ) at zero and decays at infinity. Summing this identity over the retained modes explains both the sign and the subtraction in a proper-time regulator.
In the continuum, the reference and scale must remain visible. If a positive reference operator L r e f L_{\mathrm{ref}} L ref is chosen with the same zero-mode convention and the relative heat trace has the required trace-class and small-time properties, define
log det ε ( L , L r e f ) : = − ∫ ε ∞ d t t { Tr ′ ( e − t L ) − Tr ′ ( e − t L r e f ) } , Δ S 1 -loop ( ε ) : = θ 2 log det ε ( L , L r e f ) . (345) \begin{aligned}
\log\det_{\varepsilon}(L,L_{\mathrm{ref}}) &:= -\int_{\varepsilon}^{\infty}\frac{dt}{t}\left\{\operatorname{Tr}'(e^{-tL})-\operatorname{Tr}'(e^{-tL_{\mathrm{ref}}})\right\},\\
\Delta S_{1\text{-loop}}^{(\varepsilon)} &:= \frac{\theta}{2}\log\det_{\varepsilon}(L,L_{\mathrm{ref}}). \tag*{(345)}
\end{aligned} log ε det ( L , L ref ) Δ S 1 -loop ( ε ) := − ∫ ε ∞ t d t { Tr ′ ( e − t L ) − Tr ′ ( e − t L ref ) } , := 2 θ log ε det ( L , L ref ) . (345)
This relative definition reduces in N N N dimensions to the sum of the Frullani identities after taking L r e f = μ r e f 2 I N L_{\mathrm{ref}}=\mu_{\mathrm{ref}}^2 I_N L ref = μ ref 2 I N . An alternative absolute definition uses spectral zeta continuation:
ζ L / μ r e f 2 ( s ) : = Tr ′ [ ( L / μ r e f 2 ) − s ] = μ r e f 2 s ζ L ( s ) , log det ζ ( L / μ r e f 2 ) : = − ζ L / μ r e f 2 ′ ( 0 ) = − ζ L ′ ( 0 ) − 2 ζ L ( 0 ) log μ r e f , (346) \begin{aligned}
\zeta_{L/\mu_{\mathrm{ref}}^2}(s) &:= \operatorname{Tr}'\left[(L/\mu_{\mathrm{ref}}^2)^{-s}\right]=\mu_{\mathrm{ref}}^{2s}\zeta_L(s),\\
\log\det_{\zeta}(L/\mu_{\mathrm{ref}}^2) &:= -\zeta'_{L/\mu_{\mathrm{ref}}^2}(0)=-\zeta'_L(0)-2\zeta_L(0)\log\mu_{\mathrm{ref}}, \tag*{(346)}
\end{aligned} ζ L / μ ref 2 ( s ) log ζ det ( L / μ ref 2 ) := Tr ′ [ ( L / μ ref 2 ) − s ] = μ ref 2 s ζ L ( s ) , := − ζ L / μ ref 2 ′ ( 0 ) = − ζ L ′ ( 0 ) − 2 ζ L ( 0 ) log μ ref , (346)
provided ζ L \zeta_L ζ L is regular at zero. Thus the normalization-scale dependence is explicit rather than hidden in an unspecified additive constant. If the relative small-time cancellations do not hold, removing the ultraviolet cutoff requires the local counterterms determined by the heat expansion. Their background dependence cannot in general be absorbed into a numerical constant. Finite scheme-dependent local terms must be fixed by declared renormalization conditions. Zero modes require collective coordinates or a quotient construction in addition to their removal from the determinant. Fermionic and ghost determinants have their own signs and multiplicities, fixed by the specified quadratic field content. The unprimed local heat trace has the asymptotic expansion [156]
Tr ( e − t L ) ∼ 1 ( 4 π t ) 2 ∑ k = 0 ∞ t k ∫ M tr ( a 2 k ( x , L ) ) vol g (347) \operatorname{Tr}(e^{-tL}) \sim\frac{1}{(4\pi t)^2}\sum_{k=0}^{\infty} t^k \int_M \operatorname{tr}(a_{2k}(x,L))\,\operatorname{vol}_g \tag*{(347)} Tr ( e − t L ) ∼ ( 4 π t ) 2 1 k = 0 ∑ ∞ t k ∫ M tr ( a 2 k ( x , L )) vol g (347)
The coefficients contain curvature, bundle curvature, and endomorphism invariants. Consequently, a single master fluctuation operator can generate gravitational and gauge invariants in the same expansion. The sign convention and integrated heat coefficients used here are summarized in Appendix D. For example, in a standard Laplace-type convention,
a 0 = Id , (348) a_0 = \operatorname{Id}, \tag*{(348)} a 0 = Id , (348)
a 2 = E + 1 6 R Id , (349) a_2 = E + \frac{1}{6}R\operatorname{Id}, \tag*{(349)} a 2 = E + 6 1 R Id , (349)
a 4 = 1 360 [ 60 ∇ μ ∇ μ E + 60 R E + 180 E 2 + 30 Ω μ ν Ω μ ν + ( 12 □ R + 5 R 2 − 2 R μ ν R μ ν + 2 R μ ν ρ σ R μ ν ρ σ ) Id ] , (350) a_4 = \frac{1}{360}\left[60\nabla^\mu\nabla_\mu E + 60RE + 180E^2 + 30\Omega_{\mu\nu}\Omega^{\mu\nu} + \left(12\Box R + 5R^2 - 2R_{\mu\nu}R^{\mu\nu} + 2R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma}\right)\operatorname{Id}\right], \tag*{(350)} a 4 = 360 1 [ 60 ∇ μ ∇ μ E + 60 R E + 180 E 2 + 30 Ω μν Ω μν + ( 12 □ R + 5 R 2 − 2 R μν R μν + 2 R μν ρ σ R μν ρ σ ) Id ] , (350)
where Ω μ ν = [ ∇ μ , ∇ ν ] \Omega_{\mu\nu} = [\nabla_\mu,\nabla_\nu] Ω μν = [ ∇ μ , ∇ ν ] . On a closed manifold the displayed ∇ 2 E \nabla^2E ∇ 2 E and □ R \Box R □ R terms integrate to zero, but they are part of the local coefficient. The precise signs and multiplicities depend on the convention for L L L and the bundle connection; the convention must be fixed before numerical matching. Standard references for these coefficients and their geometric applications are [78], [59], [17], and [158].
Effective action
After renormalization and symmetry reduction, one obtains schematically
S e f f = ∫ M g [ Λ e f f − R 16 π G e f f + ∑ a 1 4 g a 2 F μ ν a F a μ ν + ∣ D Φ H ∣ 2 + V e f f ( Φ H ) + Ψ ‾ ̸ D Ψ + S h i g h e r ] d 4 x . (351) S_{\mathrm{eff}} = \int_M \sqrt{g}\left[\Lambda_{\mathrm{eff}} - \frac{R}{16\pi G_{\mathrm{eff}}} + \sum_a \frac{1}{4g_a^2}F^a_{\mu\nu}F^{a\mu\nu} + |D\Phi_H|^2 + V_{\mathrm{eff}}(\Phi_H) + \overline{\Psi}\mathbin{\not\!D}\Psi+ S_{\mathrm{higher}}\right]\,d^4x. \tag*{(351)} S eff = ∫ M g [ Λ eff − 16 π G eff R + a ∑ 4 g a 2 1 F μν a F a μν + ∣ D Φ H ∣ 2 + V eff ( Φ H ) + Ψ D Ψ + S higher ] d 4 x . (351)
Equation (352) is a target effective structure. It is not a numerical prediction until the complete microscopic operator, field content, measure, regularization, and renormalization conditions are specified.
Group selection and the conditional (331) phase
The group-selection problem
The preceding reductions begin with a fixed candidate. Selecting among candidates adds a comparison problem: the equilibrium free energy must be defined with compatible units, normalization, and admissibility rules. Existence of an equilibrium within each candidate does not imply existence or uniqueness of a minimizer across candidates. The next results specify the additional compactness conditions and exhibit an obstruction when positive curvature coefficients remain unrestricted.
Why the gauge group cannot be inserted
A circular construction would begin with
g i n t = s u ( 3 ) ⊕ s u ( 2 ) ⊕ u ( 1 ) (352) \mathfrak{g}_{\mathrm{int}}=\mathfrak{su}(3)\oplus\mathfrak{su}(2)\oplus\mathfrak{u}(1) \tag*{(352)} g int = su ( 3 ) ⊕ su ( 2 ) ⊕ u ( 1 ) (352)
and then recover the Standard Model action from it. That is a reconstruction, not a derivation of the group. The construction first defines a type-admissible candidate class G c a n d \mathscr{G}_{\mathrm{cand}} G cand of master-data packages Γ \Gamma Γ of the form (55). The notation A \mathscr{A} A below is retained only as a short name for this class: A : = G c a n d \mathscr{A}:=\mathscr{G}_{\mathrm{cand}} A := G cand . Membership means that the candidate has a well-defined configuration and reference-measure class, an invariant pairing, admissible representations, and a positive Euclidean regularization. The anomaly, stability, locality, and stochastic well-posedness requirements are imposed in the consistency-admissible subclass G a d m \mathscr{G}_{\mathrm{adm}} G adm defined below; they are not inferred from the desired residual group.
For each candidate Γ ∈ A \Gamma\in\mathscr{A} Γ ∈ A , use the full master-data package (55); the notation Γ = ( g , h , R , B ) \Gamma=(\mathfrak{g},\mathfrak{h},\mathcal{R},\mathcal{B}) Γ = ( g , h , R , B ) in the next display is only a projection onto the algebra, pairing, representation, and vacuum components. All omitted components of the package, including the configuration space, reference measure, mobility, regularization, and information data, remain part of the candidate label. Fix a normalized reference measure ν Γ \nu_{\Gamma} ν Γ , a dimensionless normalization scale, and an explicitly specified candidate configuration space X Γ \mathcal{X}_{\Gamma} X Γ . The admissible probability class P Γ \mathcal{P}_{\Gamma} P Γ consists of normalized densities p p p on X Γ \mathcal{X}_{\Gamma} X Γ , not of pairs containing a second copy of the random configuration. Here B \mathcal{B} B denotes vacuum or order-parameter data and is distinct from the stochastic diffusion coefficient. If a candidate contains an externally fixed background β \beta β , it is included in the label Γ \Gamma Γ ; if it is to be optimized, one must instead specify a separate class B Γ \mathcal{B}_{\Gamma} B Γ and take an explicitly stated additional infimum over β ∈ B Γ \beta\in\mathcal{B}_{\Gamma} β ∈ B Γ . These choices, as well as the consistency class below, must be fixed before the low-energy target is inspected; otherwise the selection rule can encode the desired answer by construction. Cross-candidate comparison also requires a common dimensionless unit convention, a reference-measure normalization rule fixed for the entire candidate class, a common fluctuation scale θ \theta θ , and a common additive zero for the renormalized action and free energy. If an embedding or quotient map is used to compare configuration spaces, its treatment of the reference measure must be specified explicitly. Such a map is not required merely to evaluate the real-valued scores on different spaces. For the finite compact-group examples below, the common rule is product Haar probability on every compact group factor and the same declared scalar reference law. This rule is part of the comparison problem; it is not inferred from equilibrium of an individual candidate. Candidate-dependent shifts, rescalings, or regulator-dependent constants are not allowed. A prescribed common reference configuration, when available, can fix the additive action zero by subtraction of its action value. That subtraction does not fix the reference-measure normalization, which must still be supplied separately. Finally, assume that every P Γ \mathcal{P}_{\Gamma} P Γ is nonempty and that the stated coercivity/compactness hypotheses make the infimum finite and attained whenever C c o n s ( Γ ) < ∞ \mathcal{C}_{\mathrm{cons}}(\Gamma)<\infty C cons ( Γ ) < ∞ . For a candidate Γ \Gamma Γ , let F Γ [ p ] \mathcal{F}_{\Gamma}[p] F Γ [ p ] denote the free-energy functional of Section 4.1 restricted to X Γ \mathcal{X}_{\Gamma} X Γ , ν Γ \nu_{\Gamma} ν Γ , and the candidate master action. Define the selection functional
S s e l ( Γ ) : = inf p ∈ P Γ { F Γ [ p ] + C c o n s ( Γ ) } , (353) \mathcal{S}_{\mathrm{sel}}(\Gamma):=\inf_{p\in\mathcal{P}_{\Gamma}}\left\{\mathcal{F}_{\Gamma}[p]+\mathcal{C}_{\mathrm{cons}}(\Gamma)\right\}, \tag*{(353)} S sel ( Γ ) := p ∈ P Γ inf { F Γ [ p ] + C cons ( Γ ) } , (353)
where C c o n s \mathcal{C}_{\mathrm{cons}} C cons is not a phenomenological fit penalty but the indicator of a previously stated consistency class encoding normalization, anomaly, positivity, stability, locality, stochastic well-posedness, and any explicitly declared renormalization or effective-field-theory consistency constraints. It contains no adjustable group-dependent weights and does not refer to the observed group label. In the strict version, inconsistency is assigned infinite cost:
C c o n s ( Γ ) = + ∞ (354) \mathcal{C}_{\mathrm{cons}}(\Gamma)=+\infty \tag*{(354)} C cons ( Γ ) = + ∞ (354)
whenever the candidate violates a required consistency condition. The selected phase is
Γ ∗ ∈ arg min Γ ∈ G a d m S s e l ( Γ ) . (355) \Gamma_{\ast}\in\underset{\Gamma\in\mathcal{G}_{\mathrm{adm}}}{\arg\min}\,\mathcal{S}_{\mathrm{sel}}(\Gamma). \tag*{(355)} Γ ∗ ∈ Γ ∈ G adm arg min S sel ( Γ ) . (355)
The Standard Model group is predicted only if the minimizer is unique up to the declared notion of isomorphism and its residual group is the observed one. This statement has three logically distinct levels:
(S1) Existence : the admissible infimum is attained, so that
arg min Γ ∈ G a d m S s e l ( Γ ) ≠ ∅ . (356) \underset{\Gamma\in\mathcal{G}_{\mathrm{adm}}}{\arg\min}\,\mathcal{S}_{\mathrm{sel}}(\Gamma)\ne\varnothing. \tag*{(356)} Γ ∈ G adm arg min S sel ( Γ ) = ∅ . (356)
(S2) Uniqueness : the minimizing equivalence class is unique (or a finite list of classes is obtained together with a stated residual degeneracy).
(S3) Identification : the residual stabilizer of that class is S U ( 3 ) c × S U ( 2 ) L × U ( 1 ) Y SU(3)_{c}\times SU(2)_{L}\times U(1)_{Y} S U ( 3 ) c × S U ( 2 ) L × U ( 1 ) Y , and the subsequent electroweak stabilizer is U ( 1 ) e m U(1)_{\mathrm{em}} U ( 1 ) em .
Only the first two levels are variational existence or compactness questions; the third additionally requires the representation and symmetry-breaking calculation. None of the three levels is silently inferred from the notation G S M G_{\mathrm{SM}} G SM .
Anomaly cancellation as a constraint
For anomaly bookkeeping, let R L \mathcal{R}_{L} R L and R R \mathcal{R}_{R} R R denote the original left- and right-handed Weyl representations, respectively. A right-handed contribution enters with the opposite chirality sign. Equivalently, one may convert every right-handed field to a left-handed field in R R ∗ \mathcal{R}_{R}^{*} R R ∗ and add the corresponding conjugate trace. We use the equivalent two-chirality notation below, with T R T_{R} T R acting in the original right-handed representation. The perturbative cubic gauge-anomaly tensor is schematically
A a b c : = tr R L ( T a { T b , T c } ) − tr R R ( T a { T b , T c } ) = 0 , (357) \mathcal{A}^{abc}:=\operatorname{tr}_{\mathcal{R}_{L}}\left(T^{a}\{T^{b},T^{c}\}\right)-\operatorname{tr}_{\mathcal{R}_{R}}\left(T^{a}\{T^{b},T^{c}\}\right)=0, \tag*{(357)} A ab c := tr R L ( T a { T b , T c } ) − tr R R ( T a { T b , T c } ) = 0 , (357)
where the indices range over all gauge generators. The mixed gauge–gravitational condition is a separate linear trace constraint for each abelian generator,
A g r a v a : = tr R L ( T a ) − tr R R ( T a ) = 0. (358) \mathcal{A}_{\mathrm{grav}}^{a}:=\operatorname{tr}_{\mathcal{R}_{L}}(T^{a})-\operatorname{tr}_{\mathcal{R}_{R}}(T^{a})=0. \tag*{(358)} A grav a := tr R L ( T a ) − tr R R ( T a ) = 0. (358)
Thus the cubic equation covers U ( 1 ) 3 U(1)^{3} U ( 1 ) 3 and mixed abelian–nonabelian gauge anomalies, whereas the second equation covers the gauge–gravitational anomaly. Independent global conditions, such as the mod–2 S U ( 2 ) SU(2) S U ( 2 ) anomaly when applicable, must be checked separately. All of these tests must be carried out in a concrete representation. A candidate master phase that fails any required condition cannot define a consistent chiral quantum gauge theory in the usual framework.
Accordingly, in a realization in which these anomalies are part of the quantum consistency requirements, the admissible class can be written more explicitly as
G a d m : = { Γ ∈ G c a n d : A a b c ( Γ ) = 0 , A g r a v a ( Γ ) = 0 , A g l o b a l ( Γ ) = 0 , C r e g ( Γ ) < ∞ } . (359) \mathcal{G}_{\mathrm{adm}}:=\left\{\Gamma\in\mathcal{G}_{\mathrm{cand}}:\ \mathcal{A}^{abc}(\Gamma)=0,\quad\mathcal{A}_{\mathrm{grav}}^{a}(\Gamma)=0,\quad\mathcal{A}_{\mathrm{global}}(\Gamma)=0,\quad\mathcal{C}_{\mathrm{reg}}(\Gamma)<\infty\right\}. \tag*{(359)} G adm := { Γ ∈ G cand : A ab c ( Γ ) = 0 , A grav a ( Γ ) = 0 , A global ( Γ ) = 0 , C reg ( Γ ) < ∞ } . (359)
where A g l o b a l \mathcal{A}_{\mathrm{global}} A global denotes the collection of required global tests (for example, the mod–2 S U ( 2 ) SU(2) S U ( 2 ) condition when applicable) and C r e g \mathcal{C}_{\mathrm{reg}} C reg collects the previously stated positivity, stability, locality, compactness/coercivity, stochastic well-posedness, and the declared renormalization or effective-field-theory consistency criterion. Equation (360) is a definition of the candidate class, not an assertion that any particular master algebra has yet been found in it. If a proposed realization does not contain a chiral quantum sector, the corresponding anomaly conditions must be removed explicitly rather than silently declared satisfied.
Coercive selection and uniqueness modulo equivalence
The preceding definition specifies what may be compared, but it does not by itself prove that a comparison has a minimizer or that the minimizer is unique. Those are separate compactness questions. We make the relevant notion of sameness explicit before stating a selection result. Two master packages Γ \Gamma Γ and Γ ′ \Gamma' Γ ′ are called package-equivalent , written Γ ∼ Γ ′ \Gamma\sim\Gamma' Γ ∼ Γ ′ , if there is a bundle and field redefinition that consists of a Lie-group isomorphism u : G ~ Γ → G ~ Γ ′ u:\widetilde{G}_{\Gamma} \to\widetilde{G}_{\Gamma'} u : G Γ → G Γ ′ on the master groups, an isomorphism of the associated bundles and representations, and a configuration-space bijection T : X Γ → X Γ ′ \mathcal{T}:\mathcal{X}_{\Gamma} \to\mathcal{X}_{\Gamma'} T : X Γ → X Γ ′ , such that the following data are preserved (with v , w v,w v , w tangent vectors):
F Γ ′ ( T x ) = u ∗ F Γ ( x ) , u ∗ h Γ ′ = h Γ , T ∗ ν Γ ′ = ν Γ , S Γ ′ [ T x ] = S Γ [ x ] , F_{\Gamma'}(\mathcal{T}x)=u_*F_{\Gamma}(x), \qquad u^*h_{\Gamma'}=h_{\Gamma}, \qquad\mathcal{T}^*\nu_{\Gamma'}=\nu_{\Gamma}, \qquad S_{\Gamma'}[\mathcal{T}x]=S_{\Gamma}[x], F Γ ′ ( T x ) = u ∗ F Γ ( x ) , u ∗ h Γ ′ = h Γ , T ∗ ν Γ ′ = ν Γ , S Γ ′ [ T x ] = S Γ [ x ] ,
G Γ ′ ( T ∗ v , T ∗ w ) = G Γ ( v , w ) , M Γ ′ ( T x ) ∘ T − ∗ = T ∗ ∘ M Γ ( x ) , (360) G_{\Gamma'}(\mathcal{T}_*v,\mathcal{T}_*w)=G_{\Gamma}(v,w), \qquad M_{\Gamma'}(\mathcal{T}x)\circ\mathcal{T}^{-*}=\mathcal{T}_*\circ M_{\Gamma}(x), \tag*{(360)} G Γ ′ ( T ∗ v , T ∗ w ) = G Γ ( v , w ) , M Γ ′ ( T x ) ∘ T −∗ = T ∗ ∘ M Γ ( x ) , (360)
and the order-parameter and information maps intertwine under the same identifications. Here T − ∗ \mathcal{T}^{-*} T −∗ denotes the cotangent pushforward, the dual inverse of T ∗ \mathcal{T}_* T ∗ , so that both sides of the mobility identity have type T x ∗ X Γ → T T x X Γ ′ T_x^*\mathcal{X}_{\Gamma}\to T_{\mathcal{T}x}\mathcal{X}_{\Gamma'} T x ∗ X Γ → T T x X Γ ′ . The equality of measures in (360) is understood after the common dimensionless normalization fixed above. Thus a smooth invertible change of coordinates, an admissible gauge transformation, or a representation relabeling does not create a new candidate when it satisfies all these requirements. In particular the map T \mathcal{T} T is required to be a diffeomorphism on the regularized configuration manifolds, to transport P Γ \mathcal{P}_{\Gamma} P Γ bijectively onto P Γ ′ \mathcal{P}_{\Gamma'} P Γ ′ , and to preserve the consistency indicator and regularization domains. Preservation of the displayed action and measure alone would not guarantee preservation of a separately restricted admissible density class. Conversely, a change of a coupling, a change of the reference measure, a different regularization, or a different residual stabilizer is not declared equivalent unless an isomorphism satisfying (360) is supplied. We write Q : = G a d m / ∼ \mathcal{Q}:=\mathcal{G}_{\mathrm{adm}}/\sim Q := G adm / ∼ for the resulting set of equivalence classes and [ Γ ] [\Gamma] [ Γ ] for the class of Γ \Gamma Γ .
The package-equivalence requirement is also what makes the selection functional independent of the representative. Indeed, if p ∈ P Γ p\in\mathcal{P}_{\Gamma} p ∈ P Γ , pushforward by T \mathcal{T} T gives a density T # p ∈ P Γ ′ \mathcal{T}_{\#}p\in\mathcal{P}_{\Gamma'} T # p ∈ P Γ ′ . Equation (360) , together with the common entropy reference, then gives
F Γ ′ [ T # p ] = F Γ [ p ] , S s e l ( Γ ′ ) = S s e l ( Γ ) . (361) F_{\Gamma'}[\mathcal{T}_{\#}p]=F_{\Gamma}[p], \qquad\mathcal{S}_{\mathrm{sel}}(\Gamma')=\mathcal{S}_{\mathrm{sel}}(\Gamma). \tag*{(361)} F Γ ′ [ T # p ] = F Γ [ p ] , S sel ( Γ ′ ) = S sel ( Γ ) . (361)
Consequently S s e l \mathcal{S}_{\mathrm{sel}} S sel descends to a well-defined extended-real functional S ‾ s e l \overline{\mathcal{S}}_{\mathrm{sel}} S sel on Q \mathcal{Q} Q .
For an infinite candidate class, a precise coercivity assumption is the following. Equip Q \mathcal{Q} Q with the topology induced by the chosen topology on the master data and assume that it is Hausdorff. We call the selection functional coercive on the quotient if, for every finite R R R ,
K R : = { q ∈ Q : S ‾ s e l ( q ) ≤ R } is compact , (362) K_R:=\{q\in\mathcal{Q}:\overline{\mathcal{S}}_{\mathrm{sel}}(q)\le R\}\ \text{is compact}, \tag*{(362)} K R := { q ∈ Q : S sel ( q ) ≤ R } is compact , (362)
and if F ‾ s e l \overline{\mathcal F}_{\mathrm{sel}} F sel is lower semicontinuous. In a metrizable parameterization, (362) means that every sequence of bounded selection cost has a convergent subsequence whose limit remains in the corresponding sublevel. In particular, any sequence escaping every compact subset has cost tending to infinity. This condition is a property to be proved for a concrete class; it is not a consequence of merely listing the possible groups.
Theorem 13 (Coercive existence and uniqueness modulo package equivalence). Assume the following conditions.
(E1) The consistency class G a d m \mathcal G_{\mathrm{adm}} G adm and the indicator C c o n s \mathcal C_{\mathrm{cons}} C cons are fixed before the observed low-energy group is inspected. They use only intrinsic algebraic, analytic, stochastic, locality, positivity, anomaly, and renormalization conditions; in particular, they do not contain the statement that a specified residual group is preferred.
(E2) For every Γ ∈ G a d m \Gamma\in\mathcal G_{\mathrm{adm}} Γ ∈ G adm with C c o n s ( Γ ) < ∞ \mathcal C_{\mathrm{cons}}(\Gamma) < \infty C cons ( Γ ) < ∞ , the class P Γ \mathcal P_{\Gamma} P Γ is nonempty and sequentially compact in the topology used for probability measures, and F Γ \mathcal F_{\Gamma} F Γ is proper and lower semicontinuous on that class. The value inf p ∈ P Γ F Γ [ p ] \inf_{p\in\mathcal P_{\Gamma}} \mathcal F_{\Gamma}[p] inf p ∈ P Γ F Γ [ p ] is finite.
(E3) The common units, reference-measure normalization, fluctuation scale, and additive zero of the renormalized free energy have been fixed as in the preceding subsection, and package-equivalent candidates satisfy (361) .
(E4) The quotient Q \mathcal Q Q is Hausdorff and F ‾ s e l \overline{\mathcal F}_{\mathrm{sel}} F sel is proper, lower semicontinuous, and coercive in the sense of (362) ; in particular, it is bounded below and finite for at least one admissible class.
Then the following conclusions hold.
(i) For every admissible Γ \Gamma Γ with finite consistency cost, there exists p Γ ∈ P Γ p_{\Gamma} \in\mathcal P_{\Gamma} p Γ ∈ P Γ satisfying F Γ [ p Γ ] = inf p ∈ P Γ F Γ [ p ] \mathcal F_{\Gamma}[p_{\Gamma}] = \inf_{p\in\mathcal P_{\Gamma}} \mathcal F_{\Gamma}[p] F Γ [ p Γ ] = inf p ∈ P Γ F Γ [ p ] .
(ii) The quotient minimizer set
argmin q ∈ Q F ‾ s e l ( q ) (363) \underset{q\in\mathcal Q}{\operatorname{argmin}}\ \overline{\mathcal F}_{\mathrm{sel}}(q) \tag*{(363)} q ∈ Q argmin F sel ( q ) (363) is nonempty and compact.
(iii) The microscopic package is uniquely selected modulo the declared equivalence precisely when this minimizer set consists of one class. If q ∗ q_{\ast} q ∗ denotes that class and the quotient is not a singleton, a sufficient and directly checkable condition is the strict gap
Δ s e l : = inf q ∈ Q , q ≠ q ∗ ( F ‾ s e l ( q ) − F ‾ s e l ( q ∗ ) ) > 0. (364) \Delta_{\mathrm{sel}} := \inf_{\substack{q\in\mathcal Q,\,q\ne q_{\ast}}} \left(\overline{\mathcal F}_{\mathrm{sel}}(q)-\overline{\mathcal F}_{\mathrm{sel}}(q_{\ast})\right) > 0. \tag*{(364)} Δ sel := q ∈ Q , q = q ∗ inf ( F sel ( q ) − F sel ( q ∗ ) ) > 0. (364) A zero gap leaves uniqueness undecided by this criterion: distinct minimizing classes give a degeneracy, but a unique minimizer can also have zero gap. For example, s ( q ) = q 2 s(q) = q^{2} s ( q ) = q 2 on R \mathbb{R} R has the unique minimizer q = 0 q = 0 q = 0 , while inf q ≠ 0 s ( q ) = 0 \inf_{q\ne0}s(q) = 0 inf q = 0 s ( q ) = 0 .
The theorem gives no identification of a particular physical group. Such an identification is an additional statement that the residual stabilizer map of the selected class has the claimed value.
Proof. Fix Γ \Gamma Γ with finite consistency cost and let ( p n ) n ≥ 1 ⊂ P Γ (p_n)_{n\ge1} \subset\mathcal{P}_{\Gamma} ( p n ) n ≥ 1 ⊂ P Γ be a minimizing sequence for F Γ \mathcal{F}_{\Gamma} F Γ . By (E2), a subsequence converges in the prescribed weak topology to some p Γ ∈ P Γ p_{\Gamma} \in\mathcal{P}_{\Gamma} p Γ ∈ P Γ . Lower semicontinuity gives
F Γ [ p Γ ] ≤ lim inf n → ∞ F Γ [ p n ] = inf p ∈ P Γ F Γ [ p ] , (365) \mathcal{F}_{\Gamma}[p_{\Gamma}] \le\liminf_{n\to\infty} \mathcal{F}_{\Gamma}[p_n] = \inf_{p\in\mathcal{P}_{\Gamma}} \mathcal{F}_{\Gamma}[p], \tag*{(365)} F Γ [ p Γ ] ≤ n → ∞ lim inf F Γ [ p n ] = p ∈ P Γ inf F Γ [ p ] , (365)
while the reverse inequality follows from the definition of the infimum. This proves (i), and hence the inner infimum in (354) is an attained value.
Let s ∗ s_{\ast} s ∗ be the infimum of S s e l \mathscr{S}_{\mathrm{sel}} S sel on Q \mathcal{Q} Q , and choose a minimizing sequence ( q n ) (q_n) ( q n ) . Since at least one finite-cost class exists, s ∗ < + ∞ s_{\ast}<+\infty s ∗ < + ∞ ; after discarding finitely many terms, the sequence lies in the sublevel set K s ∗ + 1 \mathcal{K}_{s_{\ast}+1} K s ∗ + 1 . Coercivity makes this set compact, so a subnet (or a subsequence in the metrizable case) converges to some q ∗ ∈ K s ∗ + 1 q_{\ast}\in\mathcal{K}_{s_{\ast}+1} q ∗ ∈ K s ∗ + 1 . Lower semicontinuity yields
S s e l ( q ∗ ) ≤ lim inf S s e l ( q n ) = s ∗ . (366) \mathscr{S}_{\mathrm{sel}}(q_{\ast}) \le\liminf\mathscr{S}_{\mathrm{sel}}(q_n) = s_{\ast}. \tag*{(366)} S sel ( q ∗ ) ≤ lim inf S sel ( q n ) = s ∗ . (366)
The definition of s ∗ s_{\ast} s ∗ gives the opposite inequality, so equality holds. The minimizer set is a closed subset of the compact sublevel set and is therefore compact. By definition, it consists of one element exactly when the selected package is unique modulo ∼ \sim ∼ . If (365) holds, every other class has strictly larger cost, so the minimizer set is the singleton { q ∗ } \{q_{\ast}\} { q ∗ } . Conversely, a singleton minimizer does not in general imply a positive gap on a non-discrete quotient; therefore the gap is stated as a sufficient, not a necessary, criterion in that case. No step of the argument uses the name of an observed residual group.
The finite comparison used in an explicit regularization can be stated without any topological assumption. It is useful because it separates the existence of a candidate list from the physical identification of the selected phase.
Theorem 14 (Finite candidate selection). Let N ≥ 1 N \ge1 N ≥ 1 and let Q f i n = { q 1 , … , q N } ⊂ Q \mathcal{Q}_{\mathrm{fin}} = \{q_1,\ldots,q_N\} \subset\mathcal{Q} Q fin = { q 1 , … , q N } ⊂ Q be a finite list of pairwise inequivalent, target-blind admissible package classes, and suppose that each class has a finite real, attained inner free-energy minimum. Put
s i : = S s e l ( q i ) , s min : = min 1 ≤ i ≤ N s i , I ∗ : = { i : s i = s min } . (367) s_i := \mathscr{S}_{\mathrm{sel}}(q_i), \qquad s_{\min} := \min_{1\le i\le N} s_i, \qquad I_{\ast} := \{i : s_i = s_{\min}\}. \tag*{(367)} s i := S sel ( q i ) , s m i n := 1 ≤ i ≤ N min s i , I ∗ := { i : s i = s m i n } . (367) Then the selected classes in the finite list are exactly { q i : i ∈ I ∗ } \{q_i : i \in I_{\ast}\} { q i : i ∈ I ∗ } . In particular, the selection is unique modulo package equivalence if and only if ∣ I ∗ ∣ = 1 |I_{\ast}| = 1 ∣ I ∗ ∣ = 1 . When N > 1 N > 1 N > 1 and ∣ I ∗ ∣ = 1 |I_{\ast}| = 1 ∣ I ∗ ∣ = 1 , the strict finite gap
Δ f i n : = min i ≠ i ∗ ( s i − s i ∗ ) > 0 (368) \Delta_{\mathrm{fin}} := \min_{i\ne i_{\ast}}(s_i-s_{i_{\ast}}) > 0 \tag*{(368)} Δ fin := i = i ∗ min ( s i − s i ∗ ) > 0 (368)
is an explicit certificate of uniqueness within the listed class. If ∣ I ∗ ∣ > 1 |I_*| > 1 ∣ I ∗ ∣ > 1 , the tied classes are not distinguishable by the stated selection functional and no unique physical group may be claimed from this finite comparison.
Proof. The set of real numbers { s 1 , … , s N } \{s_1,\ldots,s_N\} { s 1 , … , s N } is finite, so its minimum s min s_{\min} s m i n exists. A class q i q_i q i belongs to the finite-list argmin precisely when s i = s min s_i=s_{\min} s i = s m i n , which is the definition of I ∗ I_* I ∗ . If ∣ I ∗ ∣ = 1 |I_*|=1 ∣ I ∗ ∣ = 1 , every other value exceeds s min s_{\min} s m i n ; the minimum of the finitely many positive differences is therefore positive and equals Δ f i n \Delta_{\mathrm{fin}} Δ fin . If ∣ I ∗ ∣ > 1 |I_*|>1 ∣ I ∗ ∣ > 1 , at least two inequivalent classes have exactly the same cost, so the selection rule leaves a genuine degeneracy. The argument is purely a comparison of the pre-specified target-blind costs and does not use the desired low-energy label. □
Thus, in a concrete model, uniqueness is a theorem only after one supplies the candidate topology or a finite exhaustive list, proves the inner free-energy compactness, fixes the common normalization, and verifies either the singleton quotient argmin or the strict gap in (364) or (368) . The residual group is then computed from the stabilizer of the selected vacuum; it is not inserted into the selection cost. The explicit candidate Γ 0 \Gamma_0 Γ 0 of § 5.4 can therefore serve as one member of such a list, but it is an existence witness until all competing classes have been included and compared.
An obstruction to unrestricted coupling and group selection
The compactness and separation assumptions in the selection theorems are substantive restrictions on the candidate class. A common positive norm and a unique equilibrium for each candidate do not supply them. First, the absolute selection score must be distinguished from the normalized equilibrium law. If the admissible density class contains the Gibbs minimizer, its score is − θ log Z -\theta\log Z − θ log Z . Replacing an action and reference measure by S + A S+A S + A and c ν c\nu c ν , respectively, gives
Z ′ = c e − A / θ Z , − θ log Z ′ = − θ log Z + A − θ log c . (369) Z' = c e^{-A/\theta} Z,\qquad-\theta\log Z'=-\theta\log Z+A-\theta\log c. \tag*{(369)} Z ′ = c e − A / θ Z , − θ log Z ′ = − θ log Z + A − θ log c . (369)
The normalized Gibbs probability and reversible generator are unchanged. This proves the necessity of the previously imposed common additive zero and reference normalization. Such candidate-dependent changes are excluded from the selection problem stated here. The following obstruction persists even after excluding them.
Theorem 15 (Monotonicity obstruction for a free positive coefficient). Fix a reference measure ν \nu ν , a scale θ > 0 \theta>0 θ > 0 , and measurable functions S 0 S_0 S 0 , V V V , with V V V finite and nonnegative almost everywhere. Assume
0 < Z 0 : = ∫ e − S 0 / θ d ν < ∞ , ∫ { V > 0 } e − S 0 / θ d ν > 0. (370) 0<Z_0:=\int e^{-S_0/\theta}\,d\nu<\infty,\qquad\int_{\{V>0\}}e^{-S_0/\theta}\,d\nu>0. \tag*{(370)} 0 < Z 0 := ∫ e − S 0 / θ d ν < ∞ , ∫ { V > 0 } e − S 0 / θ d ν > 0. (370)
For κ > 0 \kappa> 0 κ > 0 , let S κ = S 0 + κ V S_\kappa= S_0 + \kappa V S κ = S 0 + κV and suppose each package has zero consistency indicator and an admissible density class containing its Gibbs density. Use the relative free-energy extension just defined; alternatively, require the Gibbs energy moment to be finite and restrict the class to densities with separately integrable energy and entropy. Then its selection score is
s ( κ ) = − θ log Z κ , Z κ = ∫ e − ( S 0 + κ V ) / θ d ν , (371) s(\kappa) = -\theta\log Z_\kappa,\qquad Z_\kappa= \int e^{-(S_0+\kappa V)/\theta}\,\mathrm{d}\nu, \tag*{(371)} s ( κ ) = − θ log Z κ , Z κ = ∫ e − ( S 0 + κV ) / θ d ν , (371)
and s s s is continuous and strictly increasing on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . Moreover,
s ′ ( κ ) = ∫ V d π κ > 0 , lim κ ↓ 0 s ( κ ) = − θ log Z 0 . (372) s'(\kappa) = \int V\,\mathrm{d}\pi_\kappa> 0,\qquad\lim_{\kappa\downarrow0}s(\kappa) = -\theta\log Z_0. \tag*{(372)} s ′ ( κ ) = ∫ V d π κ > 0 , κ ↓ 0 lim s ( κ ) = − θ log Z 0 . (372)
Consequently the candidate family with every κ > 0 \kappa> 0 κ > 0 has no minimizer. If zero curvature coefficient is inadmissible, its infimum is approached only at an excluded endpoint. If instead the permitted interval is [ κ min , κ max ] [\kappa_{\min},\kappa_{\max}] [ κ m i n , κ m a x ] with κ min > 0 \kappa_{\min} > 0 κ m i n > 0 , minimization selects the prescribed boundary κ min \kappa_{\min} κ m i n .
Proof. The Gibbs identity gives (372). If 0 < κ 1 < κ 2 0 < \kappa_1 < \kappa_2 0 < κ 1 < κ 2 , the integrand defining Z κ 2 Z_{\kappa_2} Z κ 2 is at most that defining Z κ 1 Z_{\kappa_1} Z κ 1 and is strictly smaller on a set of positive weighted measure. Hence 0 < Z κ 2 < Z κ 1 ≤ Z 0 0 < Z_{\kappa_2} < Z_{\kappa_1} \le Z_0 0 < Z κ 2 < Z κ 1 ≤ Z 0 . Dominated convergence proves continuity and the limit at zero. For κ \kappa κ in a compact subinterval of ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , the bound v e − κ v / θ ≤ θ / ( e κ ) v e^{-\kappa v/\theta} \le\theta/(e\kappa) v e − κ v / θ ≤ θ / ( e κ ) for v ≥ 0 v \ge0 v ≥ 0 supplies an integrable dominator after multiplication by e − S 0 / θ e^{-S_0/\theta} e − S 0 / θ . Differentiation under the integral gives
Z κ ′ = − θ − 1 ∫ V e − S κ / θ d ν , s ′ ( κ ) = − θ Z κ ′ / Z κ , (373) Z'_\kappa= -\theta^{-1}\int V e^{-S_\kappa/\theta}\,\mathrm{d}\nu,\qquad s'(\kappa) = -\theta Z'_\kappa/Z_\kappa, \tag*{(373)} Z κ ′ = − θ − 1 ∫ V e − S κ / θ d ν , s ′ ( κ ) = − θ Z κ ′ / Z κ , (373)
which proves (373). Every positive κ \kappa κ is strictly improved by κ / 2 \kappa/2 κ /2 , proving nonattainment on the open family. Strict monotonicity also proves the endpoint assertion on a closed permitted interval. □ \square □
For the bosonic joint action, an independently variable internal coefficient has precisely this form, with
V a = 1 2 ∫ M e − α σ σ ∣ F a ∣ 2 vol g ≥ 0. (374) V_a = \frac{1}{2}\int_M e^{-\alpha\sigma\sigma\lvert F_a\rvert^2}\,\operatorname{vol}_g \ge0. \tag*{(374)} V a = 2 1 ∫ M e − α σ σ ∣ F a ∣ 2 vol g ≥ 0. (374)
This application fixes the regulator, reference measure, and all remaining data while varying that coefficient. If all positive values are retained, the sequence κ j ↓ 0 \kappa_j \downarrow0 κ j ↓ 0 has bounded selection score but no limit in the positive-coefficient class. The coercivity hypothesis therefore fails in its ordinary parameter topology. A restriction excluding this sequence may be physically appropriate, but must be separately supplied.
There are also inequivalent gauge groups with the same scalar confinement and finite-regulator equilibrium mechanism. For each integer n ≥ 2 n \ge2 n ≥ 2 , set
G ~ n = S O ( 5 ) × S U ( n ) , H n = S O ( 4 ) × S U ( n ) . (375) \widetilde{G}_n = SO(5) \times SU(n),\qquad H_n = SO(4) \times SU(n). \tag*{(375)} G n = S O ( 5 ) × S U ( n ) , H n = S O ( 4 ) × S U ( n ) . (375)
The quotient dimension is four for every n n n . Take a common fixed Cartan background and a bosonic matter choice. No chiral fermionic anomaly is introduced by this choice; a chiral spectrum is not being derived.
An explicit compact gauge regulator has four links U 1 , … , U 4 ∈ S U ( n ) U_1,\ldots,U_4 \in SU(n) U 1 , … , U 4 ∈ S U ( n ) with product Haar probability d U dU d U and
P ( U ) = U 1 U 2 U 3 − 1 U 4 − 1 , C n ( U ) = 1 2 ∥ I − P ( U ) ∥ F 2 = n − Re tr P ( U ) . (376) P(U) = U_1U_2U_3^{-1}U_4^{-1}, \qquad C_n(U) = \frac{1}{2}\lVert I-P(U)\rVert_F^2 = n-\operatorname{Re}\operatorname{tr}P(U). \tag*{(376)} P ( U ) = U 1 U 2 U 3 − 1 U 4 − 1 , C n ( U ) = 2 1 ∥ I − P ( U ) ∥ F 2 = n − Re tr P ( U ) . (376)
It is a smooth nonnegative gauge-invariant squared curvature coordinate. Conditional on three links, the plaquette is Haar distributed. Since a point has Haar measure zero in S U ( n ) SU(n) S U ( n ) , C n > 0 C_n>0 C n > 0 almost everywhere. This is a declared finite plaquette regulator, not an assertion of a continuum limit for the dynamical coframe.
Fix the same dimensionless constants v , m , f , θ > 0 v,m,f,\theta>0 v , m , f , θ > 0 , α σ ≠ 0 \alpha_\sigma\ne0 α σ = 0 , s 0 s_0 s 0 , and C c ≥ 0 C_c\ge0 C c ≥ 0 for every group. Use the explicit common normalization rule
ν n ( d U d s ) = d U e − s 2 / ( 2 v ) 2 π v d s , ν n ( 1 ) = 1 , (377) \nu_n(dU\,ds)=dU\,\frac{e^{-s^2/(2v)}}{\sqrt{2\pi v}}\,ds,\qquad\nu_n(1)=1, \tag*{(377)} ν n ( d U d s ) = d U 2 π v e − s 2 / ( 2 v ) d s , ν n ( 1 ) = 1 , (377)
and the same additive action zero in
S n , κ ( U , s ) = e − α σ s [ C c + κ C n ( U ) ] + m 2 f 2 2 ( s − s 0 ) 2 . (378) S_{n,\kappa}(U,s)=e^{-\alpha_\sigma s}\left[C_c+\kappa C_n(U)\right]+\frac{m^2f^2}{2}(s-s_0)^2. \tag*{(378)} S n , κ ( U , s ) = e − α σ s [ C c + κ C n ( U ) ] + 2 m 2 f 2 ( s − s 0 ) 2 . (378)
The compact factors always have Haar mass one, and the scalar reference law is identical; no group-dependent measure multiplier is used. This is an application of the common normalization prescription. It does not assert that the Haar laws for different groups are pushforwards under their natural inclusions. For example, the inclusion of S U ( 2 ) 4 SU(2)^4 S U ( 2 ) 4 into S U ( 3 ) 4 SU(3)^4 S U ( 3 ) 4 has a lower-dimensional image, of zero ambient Haar measure. Its pushed-forward probability is therefore singular relative to the Haar probability on S U ( 3 ) 4 SU(3)^4 S U ( 3 ) 4 . Each action is a single positive extended norm with the same exponential weight and quadratic stabilizer. Its partition function lies in ( 0 , 1 ] (0,1] ( 0 , 1 ] . The corresponding reversible scalar drift includes the reference term − θ s / v -\theta s/v − θ s / v in addition to − ∂ s S n , κ -\partial_sS_{n,\kappa} − ∂ s S n , κ . For the Lyapunov function V 0 ( s ) = 1 + s 2 V_0(s)=1+s^2 V 0 ( s ) = 1 + s 2 , put a = m 2 f 2 a=m^2f^2 a = m 2 f 2 and use C n ≤ 2 n C_n\le2n C n ≤ 2 n and z e − z ≤ e − 1 ze^{-z}\le e^{-1} z e − z ≤ e − 1 . Direct differentiation gives
L V 0 ≤ 2 θ + a s 0 2 + 2 ( C c + 2 κ n ) e − ( a + 2 θ / v ) s 2 . (379) \mathcal{L}V_0\le2\theta+as_0^2+\frac{2(C_c+2\kappa n)}{e}-(a+2\theta/v)s^2. \tag*{(379)} L V 0 ≤ 2 θ + a s 0 2 + e 2 ( C c + 2 κn ) − ( a + 2 θ / v ) s 2 . (379)
The finite-regulator proof in Theorem 19 therefore applies, with the reference drift included. Compactness of the links, ellipticity, and connectedness are unchanged. Hence each candidate has global regulated dynamics and a unique Gibbs equilibrium.
Write s n ( κ ) = − θ log Z n , κ s_n(\kappa)=-\theta\log Z_{n,\kappa} s n ( κ ) = − θ log Z n , κ . At κ = 0 \kappa=0 κ = 0 the normalized link integral is one, so s n ( 0 ) s_n(0) s n ( 0 ) is the same number s ∗ s_* s ∗ for every n n n . The preceding theorem gives strict increase and continuity. Since C n > 0 C_n > 0 C n > 0 almost everywhere, dominated convergence also gives
Z n , κ ⟶ 0 ( κ → ∞ ) , s n ( ( 0 , ∞ ) ) = ( s ∗ , ∞ ) . (380) Z_{n,\kappa} \longrightarrow0 \quad(\kappa\to\infty), \qquad s_n((0,\infty)) = (s_*,\infty). \tag*{(380)} Z n , κ ⟶ 0 ( κ → ∞ ) , s n (( 0 , ∞ )) = ( s ∗ , ∞ ) . (380)
Thus, for any declared score r > s ∗ r > s_* r > s ∗ , each group has exactly one positive κ n \kappa_n κ n with s n ( κ n ) = r s_n(\kappa_n) = r s n ( κ n ) = r . Two such packages for S U ( 2 ) SU(2) S U ( 2 ) and S U ( 3 ) SU(3) S U ( 3 ) have equal selection score but cannot be package-equivalent: their master groups have different dimensions. In the declared finite bosonic regulator class, assume that both packages satisfy the same previously specified consistency conditions, so that both indicators are zero. A class containing just these two packages then has two minimizers; the class retaining all their positive coefficients has no minimizer. Choosing other allowed coefficients can give either ordering. These choices demonstrate underdetermination; they are not a prescription for tuning a desired group into the minimum.
The examples establish an obstruction within the explicit algebraic, bosonic, and regulated class. A completed chiral, continuum, or Lorentzian consistency condition could exclude some of them, but an actual exclusion argument is required. Finally, even at a prescribed w ∗ > 0 w_* > 0 w ∗ > 0 , the tree relations g a − 2 = w ∗ κ a g_a^{-2} = w_*\kappa_a g a − 2 = w ∗ κ a reproduce any positive gauge-coupling triple by choosing κ a = g a − 2 / w ∗ \kappa_a = g_a^{-2}/w_* κ a = g a − 2 / w ∗ . Numerical coupling prediction therefore requires independent constraints fixing those normalizations; the common exponential weight alone does not fix them.
A quantitative one-plaquette check
A single U ( 1 ) U(1) U ( 1 ) plaquette makes the obstruction quantitatively explicit. Use a plaquette angle u ∈ [ − π , π ) u \in[-\pi,\pi) u ∈ [ − π , π ) with Haar measure d u / ( 2 π ) \mathrm{d}u/(2\pi) d u / ( 2 π ) , one scalar s ∈ R s \in\mathbb{R} s ∈ R with Lebesgue reference, and the dimensionless data
S κ ( u , s ) = 1 2 s 2 + κ e − α σ s ( 1 − cos u ) , θ = 1 , α σ = 0.6 , κ > 0. (381) S_\kappa(u,s) = \frac{1}{2}s^2 + \kappa e^{-\alpha_\sigma s}(1-\cos u), \qquad\theta= 1,\quad\alpha_\sigma= 0.6,\quad\kappa> 0. \tag*{(381)} S κ ( u , s ) = 2 1 s 2 + κ e − α σ s ( 1 − cos u ) , θ = 1 , α σ = 0.6 , κ > 0. (381)
The scalar action is confining, the curvature term is nonnegative, and the finite-regulator proof in Theorem 19 applies. Angular integration gives
I 0 ( a ) : = 1 2 π ∫ − π π e a cos u d u , I 1 ( a ) : = I 0 ′ ( a ) , R ( κ ) : = Z κ Z 0 = E s ∼ N ( 0 , 1 ) [ e − a I 0 ( a ) ] , a = κ e − α σ s , Z 0 = 2 π , F ( κ ) − F ( 0 ) = − log R ( κ ) . (382) \begin{aligned}
I_0(a) &:= \frac{1}{2\pi}\int_{-\pi}^{\pi} e^{a\cos u}\,\mathrm{d}u, \qquad I_1(a) := I_0'(a),\\
R(\kappa) &:= \frac{Z_\kappa}{Z_0} = \mathbb{E}_{s\sim N(0,1)}\left[e^{-a}I_0(a)\right], \qquad a = \kappa e^{-\alpha_\sigma s},\\
Z_0 &= \sqrt{2\pi}, \qquad F(\kappa)-F(0) = -\log R(\kappa). \tag*{(382)}
\end{aligned} I 0 ( a ) R ( κ ) Z 0 := 2 π 1 ∫ − π π e a c o s u d u , I 1 ( a ) := I 0 ′ ( a ) , := Z 0 Z κ = E s ∼ N ( 0 , 1 ) [ e − a I 0 ( a ) ] , a = κ e − α σ s , = 2 π , F ( κ ) − F ( 0 ) = − log R ( κ ) . (382)
The scaled Bessel function e − a I 0 ( a ) e^{-a}I_0(a) e − a I 0 ( a ) is bounded by one for a ≥ 0 a \ge0 a ≥ 0 , so this formula avoids a large-exponential cancellation in numerical integration. Differentiation, with the domination established in the selection argument, yields
F ′ ( κ ) = E [ e − α σ s e − a { I 0 ( a ) − I 1 ( a ) } ] R ( κ ) > 0. (383) F'(\kappa) = \frac{\mathbb{E}\left[e^{-\alpha_\sigma s}e^{-a}\{I_0(a)-I_1(a)\}\right]}{R(\kappa)} > 0. \tag*{(383)} F ′ ( κ ) = R ( κ ) E [ e − α σ s e − a { I 0 ( a ) − I 1 ( a )} ] > 0. (383)
The numerator is the Gaussian expectation of the nonnegative angular integral e − α σ s ( 1 − cos u ) e − a ( 1 − cos u ) e^{-\alpha_{\sigma}s}(1-\cos u)e^{-a(1-\cos u)} e − α σ s ( 1 − cos u ) e − a ( 1 − c o s u ) . It is strictly positive because the angle is not concentrated at zero.
The scalar drift is − s + α σ κ e − α σ s ( 1 − cos u ) -s+\alpha_{\sigma}\kappa e^{-\alpha_{\sigma}s}(1-\cos u) − s + α σ κ e − α σ s ( 1 − cos u ) . All its terms are integrable in equilibrium by the Gaussian domination and the derivative estimate. Either taking stationary expectations of the scalar equation or integrating the scalar Gibbs density by parts therefore gives the independent balance identity
E π κ s = α σ κ F ′ ( κ ) . (384) \mathbb{E}_{\pi_\kappa}s=\alpha_{\sigma}\kappa F'(\kappa). \tag*{(384)} E π κ s = α σ κ F ′ ( κ ) . (384)
This is a relation for a nonlinear coupled scalar–gauge equilibrium, not a small-fluctuation Gaussian identity.
Figure 4 evaluates (383)–(385) at 241 logarithmically spaced values of κ \kappa κ in [ 10 − 3 , 10 2 ] [10^{-3},10^{2}] [ 1 0 − 3 , 1 0 2 ] . Gauss–Hermite orders 256 and 384 differ by at most 9.13 × 10 − 16 9.13\times10^{-16} 9.13 × 1 0 − 16 over all reported quantities. The maximum stationary-balance residual is 4.56 × 10 − 15 4.56\times10^{-15} 4.56 × 1 0 − 15 . A fourth-order centered finite difference with relative step 2 × 10 − 3 2\times10^{-3} 2 × 1 0 − 3 differs from (383) by at most 5.39 × 10 − 11 5.39\times10^{-11} 5.39 × 1 0 − 11 . As an independent check, adaptive quadrature on s ∈ [ − 12 , 12 ] s\in[-12,12] s ∈ [ − 12 , 12 ] at κ = 10 − 3 , 10 − 2 , 10 − 1 , 1 , 10 , 10 2 \kappa=10^{-3},10^{-2},10^{-1},1,10,10^{2} κ = 1 0 − 3 , 1 0 − 2 , 1 0 − 1 , 1 , 10 , 1 0 2 differs from the order-384 Gauss–Hermite values by at most 3.72 × 10 − 15 3.72\times10^{-15} 3.72 × 1 0 − 15 ; the omitted standard normal tail probability is below 4 × 10 − 33 4\times10^{-33} 4 × 1 0 − 33 . These comparisons are numerical consistency checks, not certified global quadrature error bounds. The exact integral inequalities prove monotonicity independently of the plot.
Figure 4. Deterministic integrals for the scalar–U ( 1 ) U(1) U ( 1 ) plaquette model. Left: the strictly increasing free-energy difference, whose infimum corresponds to the excluded zero coefficient. Right: the directly integrated scalar mean and the independent stationary-balance formula. The curves represent the specified mathematical model and contain no observational data.
A conditional 331 family and flavor extension
The selection results above distinguish a consistency condition from a selection theorem. We now give a concrete chiral candidate class in which that distinction can be followed through anomaly cancellation, symmetry breaking, flavor, and threshold matching. The candidate package Γ 331 \Gamma_{331} Γ 331 contains the group, representations, scalar potential, flavor maps, regulator, mobility, reference measure, and matching conditions. It is admitted to the anomaly part of (359) only after all of those data have been specified. Neither the anomaly identities below nor the equilibrium construction bypasses the nonattainment result of Theorem 15 or the equal-score examples in (375) . Accordingly, Γ 331 \Gamma_{331} Γ 331 is an additional conditional realization, rather than a uniquely selected phase of the master theory.
Conditional anomaly and family-count mechanism
Let
G ~ 331 : = S U ( 3 ) c × S U ( 3 ) L × U ( 1 ) x , Q = T 3 − T 8 3 + X 331 . (385) \widetilde{G}_{331}:=\mathrm{SU}(3)_{\mathrm{c}}\times\mathrm{SU}(3)_{\mathrm{L}}\times\mathrm{U}(1)_{x},\qquad Q=T_{3}-\frac{T_{8}}{\sqrt{3}}+X_{331}. \tag*{(385)} G 331 := SU ( 3 ) c × SU ( 3 ) L × U ( 1 ) x , Q = T 3 − 3 T 8 + X 331 . (385)
For the global statements below, the compact U ( 1 ) \mathrm{U}(1) U ( 1 ) factor is normalized by an integral character charge x ∈ Z x\in\mathbb{Z} x ∈ Z . We use the phenomenological generator X 331 = x / 3 X_{331}=x/3 X 331 = x /3 , and the corresponding coupling normalization is fixed by
g x x = g X 331 , equivalently g x = g X 3 . (386) g_x x=gX_{331},\qquad\text{equivalently}\qquad g_x=\frac{gX}{3}. \tag*{(386)} g x x = g X 331 , equivalently g x = 3 g X . (386)
Thus the fractional labels in the following display are conventional X 331 X_{331} X 331 -charges, while the compact-group characters are the integral charges x = 3 X 331 x = 3X_{331} x = 3 X 331 . This convention leaves the vanishing of all local anomaly coefficients unchanged and will be used again in the global quotient test below. The following matter class is a 331 realization with right-handed neutrinos. Early 331 constructions were introduced in [70] and [130], and the stated right-handed-neutrino realization is described in [69]. We state its complete covering-group charged chiral content used in the anomaly count because the family-number conclusion depends on these representations, on the absence of additional active colored fields, and on the chosen normalization of X 331 X_{331} X 331 . The displayed subscripts denote X 331 X_{331} X 331 -charges. For i = 1 , … , N g i = 1,\ldots,N_g i = 1 , … , N g , r = 1 , … , k r = 1,\ldots,k r = 1 , … , k , and α = 1 , … , N g − k \alpha= 1,\ldots,N_g-k α = 1 , … , N g − k , take
L i L ∼ ( 1 , 3 ) − 1 / 3 , E i R ∼ ( 1 , 1 ) − 1 , Q L r ∼ ( 3 , 3 ) 1 / 3 Q L α ∼ ( 3 , 3 ‾ ) 0 , L_{iL} \sim(1,3)_{-1/3},\ E_{iR} \sim(1,1)_{-1},\ Q^r_L \sim(3,3)_{1/3}\quad Q^\alpha_L \sim(3,\overline{3})_0, L i L ∼ ( 1 , 3 ) − 1/3 , E i R ∼ ( 1 , 1 ) − 1 , Q L r ∼ ( 3 , 3 ) 1/3 Q L α ∼ ( 3 , 3 ) 0 ,
U i R ∼ ( 3 , 1 ) 2 / 3 , D i R ∼ ( 3 , 1 ) − 1 / 3 , T r R ∼ ( 3 , 1 ) 2 / 3 , B α R ∼ ( 3 , 1 ) − 1 / 3 . (387) U_{iR} \sim(3,1)_{2/3},\ D_{iR} \sim(3,1)_{-1/3},\ T_{rR} \sim(3,1)_{2/3},\ B_{\alpha R} \sim(3,1)_{-1/3}. \tag*{(387)} U i R ∼ ( 3 , 1 ) 2/3 , D i R ∼ ( 3 , 1 ) − 1/3 , T r R ∼ ( 3 , 1 ) 2/3 , B α R ∼ ( 3 , 1 ) − 1/3 . (387)
The declared representations have a common central kernel. With ω 3 = e 2 π i / 3 \omega_3 = e^{2\pi i/3} ω 3 = e 2 π i /3 , the faithful continuous gauge group is
K 331 = { ( ω 3 m 1 3 , ω 3 m 1 3 , ω 3 m ) : m = 0 , 1 , 2 } , K_{331} = \{(\omega_3^m 1_3,\omega_3^m 1_3,\omega_3^m) : m = 0,1,2\}, K 331 = {( ω 3 m 1 3 , ω 3 m 1 3 , ω 3 m ) : m = 0 , 1 , 2 } ,
G 331 = G ~ 331 / K 331 ≃ { ( U , V ) ∈ U ( 3 ) 2 : det U = det V } . (388) G_{331} = \widetilde{G}_{331}/K_{331} \simeq\{(U,V) \in U(3)^2 : \det U = \det V\}. \tag*{(388)} G 331 = G 331 / K 331 ≃ {( U , V ) ∈ U ( 3 ) 2 : det U = det V } . (388)
To verify that the kernel is complete, an element ( C , L , z ) (C,L,z) ( C , L , z ) acting trivially on E R , L L , U R E_R,L_L,U_R E R , L L , U R must satisfy z 3 = 1 z^3=1 z 3 = 1 , L = z 1 3 L=z\mathbf{1}_3 L = z 1 3 , and C = z − 2 1 3 = z 1 3 C=z^{-2}\mathbf{1}_3=z\mathbf{1}_3 C = z − 2 1 3 = z 1 3 . Each remaining multiplet has trialities a c a_c a c , b L b_L b L and character x x x with a c + b L + x = 0 a_c+b_L+x=0 a c + b L + x = 0 (mod 3), so these three elements act trivially on all of them. The unitary-pair isomorphism is induced by ( C , L , z ) ↦ ( z − 1 C , z − 1 L ) (C,L,z)\mapsto(z^{-1}C,z^{-1}L) ( C , L , z ) ↦ ( z − 1 C , z − 1 L ) . A cube root of the common determinant proves surjectivity, and its kernel is exactly K 331 K_{331} K 331 . Local Lie-algebra computations below are unchanged, but principal bundles and the global anomaly problem refer to this quotient rather than its cover.
The lepton triplet has electric charges ( 0 , − 1 , 0 ) (0,-1,0) ( 0 , − 1 , 0 ) ; we therefore write
L i L = ( ν i L , e i L , n i L c ) T . (389) L_{iL}=(\nu_{iL},e_{iL},n^c_{iL})^T. \tag*{(389)} L i L = ( ν i L , e i L , n i L c ) T . (389)
The embedded neutral field n i L c n^c_{iL} n i L c is distinct from the gauge-singlet field N i R N_{iR} N i R introduced below for the even type-I seesaw. In the declared matching action, n i L c n^c_{iL} n i L c instead obtains its odd-sector Dirac partner through S R S_R S R . The gauge-singlet fermions introduced below have X 331 = 0 X_{331}=0 X 331 = 0 and do not alter these gauge or mixed gauge–gravitational anomaly coefficients. The symbols in (388) are local to this candidate and are distinct from the prescribed Standard-Model fields in the earlier joint realization. In anomaly sums every right-handed field is replaced by its charge-conjugate left-handed field. With T ( 3 ) = 1 / 2 T(3)=1/2 T ( 3 ) = 1/2 , the potentially nonzero local anomaly coefficients are
A S U ( 3 ) c 3 331 = 3 N g − ( 2 N g + k + N g − k ) = 0 , A S U ( 3 ) c 2 X 331 331 = 1 2 [ k − N g 3 − 2 k 3 + N g − k 3 ] = 0 , A S U ( 3 ) L 3 331 = N g + 3 ( 2 k − N g ) = 6 k − 2 N g , A S U ( 3 ) L 2 X 331 331 = 1 2 [ N g 3 + k ] = 3 k − N g 6 , A X 331 3 331 = N g [ 3 ( − 1 3 ) 3 + 1 ] + 3 [ − 4 k 9 − 2 ( N g − k ) 9 ] = 2 ( N g − 3 k ) 9 , A g r a v 2 X 331 331 = 0. (390) \begin{aligned}
\mathcal{A}^{331}_{\mathrm{SU}(3)_c^3} &= 3N_g-(2N_g+k+N_g-k)=0,\\
\mathcal{A}^{331}_{\mathrm{SU}(3)_c^2X_{331}} &= \frac{1}{2}\left[k-\frac{N_g}{3}-\frac{2k}{3}+\frac{N_g-k}{3}\right]=0,\\
\mathcal{A}^{331}_{\mathrm{SU}(3)_L^3} &= N_g+3(2k-N_g)=6k-2N_g,\\
\mathcal{A}^{331}_{\mathrm{SU}(3)_L^2X_{331}} &= \frac{1}{2}\left[\frac{N_g}{3}+k\right]=\frac{3k-N_g}{6},\\
\mathcal{A}^{331}_{X_{331}^3} &= N_g\left[3\left(-\frac{1}{3}\right)^3+1\right]+3\left[-\frac{4k}{9}-\frac{2(N_g-k)}{9}\right]=\frac{2(N_g-3k)}{9},\\
\mathcal{A}^{331}_{\mathrm{grav}^2X_{331}} &=0. \tag*{(390)}
\end{aligned} A SU ( 3 ) c 3 331 A SU ( 3 ) c 2 X 331 331 A SU ( 3 ) L 3 331 A SU ( 3 ) L 2 X 331 331 A X 331 3 331 A grav 2 X 331 331 = 3 N g − ( 2 N g + k + N g − k ) = 0 , = 2 1 [ k − 3 N g − 3 2 k + 3 N g − k ] = 0 , = N g + 3 ( 2 k − N g ) = 6 k − 2 N g , = 2 1 [ 3 N g + k ] = 6 3 k − N g , = N g [ 3 ( − 3 1 ) 3 + 1 ] + 3 [ − 9 4 k − 9 2 ( N g − k ) ] = 9 2 ( N g − 3 k ) , = 0. (390)
For example, the first line counts the three S U ( 3 ) L \mathrm{SU}(3)_L SU ( 3 ) L components of each left-handed colored multiplet and then subtracts the charge-conjugated right-handed color fundamentals. The last coefficient vanishes separately in each lepton and quark family. All anomalies with one generator of a simple factor and generators of independent simple factors vanish by tracelessness. Thus the local anomaly conditions in this specified class reduce exactly to
N g = 3 k . (391) N_g=3k. \tag*{(391)} N g = 3 k . (391)
Theorem 16 (Conditional 331 three-family statement). Assume (387) , color-singlet scalar fields, no additional colored degrees of freedom in the energy interval in which the one-loop color criterion is imposed, and one-loop color asymptotic freedom. Then local anomaly cancellation together with that asymptotic-freedom condition implies N g = 3 N_g = 3 N g = 3 and k = 1 k = 1 k = 1 .
Proof. Equation (391) gives N g = 3 k N_g = 3k N g = 3 k . Since N g > 0 N_g > 0 N g > 0 , k k k is a positive integer. Each family contains two ordinary and one exotic Dirac color fundamental above the exotic thresholds. Hence there are 3 N g 3N_g 3 N g Dirac color fundamentals. In the convention μ ∂ μ g c = − b 0 , c ( 331 ) g c 3 / ( 16 π 2 ) + ⋯ \mu\partial_\mu g_c = -b_{0,c}^{(331)} g_c^3/(16\pi^2) + \cdots μ ∂ μ g c = − b 0 , c ( 331 ) g c 3 / ( 16 π 2 ) + ⋯ ,
b 0 , c ( 331 ) = 11 − 2 3 ( 3 N g ) = 11 − 2 N g . (392) b_{0,c}^{(331)} = 11 - \frac{2}{3}(3N_g) = 11 - 2N_g. \tag*{(392)} b 0 , c ( 331 ) = 11 − 3 2 ( 3 N g ) = 11 − 2 N g . (392) Color asymptotic freedom gives N g < 11 / 2 N_g < 11/2 N g < 11/2 . The only positive multiple of three below 11 / 2 11/2 11/2 is N g = 3 N_g = 3 N g = 3 ; substitution into (391) gives k = 1 k = 1 k = 1 . □
Ordinary-spin global consistency. The quotient admits two universal rank-three unitary bundles with the same determinant line. Write their common first Chern class as d d d . The classifying-space fibration has fiber B S U ( 3 ) c × B S U ( 3 ) L BSU(3)_c \times BSU(3)_L B S U ( 3 ) c × B S U ( 3 ) L over B U ( 1 ) BU(1) B U ( 1 ) . All integral fiber Chern monomials lift to the universal bundles on the total space. Leray–Hirsch [92, Theorem 4D.1] therefore gives
H ∗ ( B G 331 ; Z ) = Z [ d , c 2 c , c 2 L , c 3 c , c 3 L ] , ∣ d ∣ = 2 , ∣ c 2 j ∣ = 4 , ∣ c 3 j ∣ = 6. (393) H^*(BG_{331};\mathbb{Z}) = \mathbb{Z}[d,c_{2c},c_{2L},c_{3c},c_{3L}], \qquad|d| = 2,\quad|c_{2j}| = 4,\quad|c_{3j}| = 6. \tag*{(393)} H ∗ ( B G 331 ; Z ) = Z [ d , c 2 c , c 2 L , c 3 c , c 3 L ] , ∣ d ∣ = 2 , ∣ c 2 j ∣ = 4 , ∣ c 3 j ∣ = 6. (393)
The actual lifted generators remove multiplicative extension ambiguity; finite-type universal coefficients give free even-degree homology and zero odd-degree homology. In the ordinary spin-bordism spectral sequence, the only total-degree-five term is E 4 , 1 2 = ( Z / 2 ) 3 E^2_{4,1} = (\mathbb{Z}/2)^3 E 4 , 1 2 = ( Z /2 ) 3 . The incoming and outgoing differentials are the homology duals of S q 2 \mathrm{Sq}^2 Sq 2 , including mod-two reduction for the incoming integral group [55]. In the degree-four basis ( d 2 , c 2 c , c 2 L ) (d^2,c_{2c},c_{2L}) ( d 2 , c 2 c , c 2 L ) and the degree-six basis ( d 3 , d c 2 c , d c 2 L , c 3 c , c 3 L ) (d^3,dc_{2c},dc_{2L},c_{3c},c_{3L}) ( d 3 , d c 2 c , d c 2 L , c 3 c , c 3 L ) , they are
A = ( 1 0 0 ) , B = ( 0 0 0 0 0 0 1 0 1 0 0 0 1 0 1 ) , ker A = im B . (394) A = \begin{pmatrix} 1 & 0 & 0 \end{pmatrix}, \qquad B = \begin{pmatrix} 0 & 0 & 0 & 0 & 0 \\ 0 & 1 & 0 & 1 & 0 \\ 0 & 0 & 1 & 0 & 1 \end{pmatrix}, \qquad\ker A = \operatorname{im} B. \tag*{(394)} A = ( 1 0 0 ) , B = 0 0 0 0 1 0 0 0 1 0 1 0 0 0 1 , ker A = im B . (394)
Indeed, S q 2 d = d 2 \mathrm{Sq}^2 d = d^2 Sq 2 d = d 2 , S q 2 ( d 2 ) = 0 \mathrm{Sq}^2(d^2) = 0 Sq 2 ( d 2 ) = 0 , and S q 2 c 2 j = d c 2 j + c 3 j \mathrm{Sq}^2 c_{2j} = dc_{2j} + c_{3j} Sq 2 c 2 j = d c 2 j + c 3 j . These are operations on the actual quotient-bundle classes, not on hypothetical generators of its cover. No other degree-five entry or extension survives, and hence
Ω 5 S p i n ( B G 331 ) = 0. (395) \Omega^{\mathrm{Spin}}_5(BG_{331}) = 0. \tag*{(395)} Ω 5 Spin ( B G 331 ) = 0. (395)
This excludes a residual continuous-group fermion anomaly on ordinary spin spacetime after the local anomaly polynomial vanishes. Non-liftable quotient bundles are included. Paper II supplies the degree-by-degree integral and differential proofs. A spin–gauge identification, a gauged discrete parity, and ultraviolet line-operator choices are different questions and are not resolved by (395) .
The three-family theorem is conditional in two independent ways. Asymptotic freedom is an additional dynamical requirement rather than a universal quantum consistency condition, and colored messengers or other threshold states change (392) . For the minimal three-family content with three complex S U ( 3 ) L \mathrm{SU}(3)_{L} SU ( 3 ) L triplet scalars, the remaining one-loop coefficients are
b 0 , L ( 331 ) = 13 2 , b X 331 ( 331 ) = − 26 3 , β X ( g X ) = + ϵ l o o p 26 g X 3 3 ( 16 π 2 ) + O ( ϵ l o o p 2 g X 5 ) . (396) b^{(331)}_{0,L}=\frac{13}{2},\qquad b^{(331)}_{X_{331}}=-\frac{26}{3},\qquad\beta_{X}(g_{X})=+\epsilon_{\mathrm{loop}}\frac{26g_{X}^{3}}{3(16\pi^{2})}+O(\epsilon_{\mathrm{loop}}^{2}g_{X}^{5}). \tag*{(396)} b 0 , L ( 331 ) = 2 13 , b X 331 ( 331 ) = − 3 26 , β X ( g X ) = + ϵ loop 3 ( 16 π 2 ) 26 g X 3 + O ( ϵ loop 2 g X 5 ) . (396)
Here and below the one-loop convention is μ ∂ μ g = − ϵ l o o p b 0 g 3 / ( 16 π 2 ) + ⋯ \mu\,\partial_{\mu}g=-\epsilon_{\mathrm{loop}}b_{0}g^{3}/(16\pi^{2})+\cdots μ ∂ μ g = − ϵ loop b 0 g 3 / ( 16 π 2 ) + ⋯ . The sign of b X 331 ( 331 ) b^{(331)}_{X_{331}} b X 331 ( 331 ) is therefore consistent with the manuscript-wide convention (450) . The explicitly displayed positive β X \beta_{X} β X signals abelian ultraviolet growth and therefore requires a cutoff or a further ultraviolet completion. The numerical magnitude of b X 331 ( 331 ) b^{(331)}_{X_{331}} b X 331 ( 331 ) is tied to the generator normalization in (386) ; its sign and the matched products g X X 331 g_{X}X_{331} g X X 331 are unchanged by a consistent rescaling. The family mechanism itself is established 331 physics, not a new derivation of the observed family number from the master action.
The flavor and spectrum construction below the generalized-color and parent-group obstruction analyses specializes to the branch N g = 3 N_{g}=3 N g = 3 , k = 1 k=1 k = 1 selected in Theorem 16 . There is therefore one gauge-triplet quark multiplet and two gauge-antitriplet quark multiplets. Whenever a 3 × 3 3\times3 3 × 3 ordinary-flavor matrix is used in that construction, its left-handed rows are ordered as
( Q 1 L , Q 2 L , Q 3 L ) : = ( Q 1 L t , Q 1 L a , Q 2 L a ) . (397) (Q_{1L},Q_{2L},Q_{3L}):=(Q^{t}_{1L},Q^{a}_{1L},Q^{a}_{2L}). \tag*{(397)} ( Q 1 L , Q 2 L , Q 3 L ) := ( Q 1 L t , Q 1 L a , Q 2 L a ) . (397)
This is a gauge-basis bookkeeping convention; it is not an identification of the special gauge triplet with a physical mass generation, which is defined only after the Yukawa matrices are diagonalized.
Why anomaly cancellation does not select three colors
The preceding theorem must not be read as an independent derivation of S U ( 3 ) c \mathrm{SU}(3)_{c} SU ( 3 ) c . To see the limitation directly, replace color by S U ( N c ) \mathrm{SU}(N_{c}) SU ( N c ) , retain S U ( 3 ) L \mathrm{SU}(3)_{L} SU ( 3 ) L , and allow the ordinary quark-doublet hypercharge to be an unknown Y q Y_{q} Y q . The triplet and antitriplet charges are then
X t = Y q + 1 6 , X a = Y q − 1 6 , (398) X_{t}=Y_{q}+\frac{1}{6},\qquad X_{a}=Y_{q}-\frac{1}{6}, \tag*{(398)} X t = Y q + 6 1 , X a = Y q − 6 1 , (398)
whereas the corresponding right-handed ordinary and exotic charges are
X U = Y q + 1 2 , X D = Y q − 1 2 . (399) X_{U}=Y_{q}+\frac{1}{2},\qquad X_{D}=Y_{q}-\frac{1}{2}. \tag*{(399)} X U = Y q + 2 1 , X D = Y q − 2 1 . (399)
The two independent left-handed anomaly equations are
A S U ( 3 ) L 3 331 = N g + N c ( 2 k − N g ) = 0 , \mathcal{A}^{331}_{\mathrm{SU}(3)^3_L}=N_g+N_c(2k-N_g)=0, A SU ( 3 ) L 3 331 = N g + N c ( 2 k − N g ) = 0 ,
A S U ( 3 ) L 2 X 331 331 = − N g 6 + N c 2 [ N g Y q + 2 k − N g 6 ] = 0. (400) \mathcal{A}^{331}_{\mathrm{SU}(3)^2_LX_{331}}=-\frac{N_g}{6}+\frac{N_c}{2}\left[N_gY_q+\frac{2k-N_g}{6}\right]=0. \tag*{(400)} A SU ( 3 ) L 2 X 331 331 = − 6 N g + 2 N c [ N g Y q + 6 2 k − N g ] = 0. (400)
Substitution of the first equation into the second gives Y q = 1 / ( 2 N c ) Y_q=1/(2N_c) Y q = 1/ ( 2 N c ) . At that value, direct substitution in the cubic abelian sum gives
A S U ( 3 ) L 2 X 331 331 = A S U ( 3 ) L 3 331 12 , A X 331 3 331 = A S U ( 3 ) L 3 331 9 , (401) \mathcal{A}^{331}_{\mathrm{SU}(3)^2_LX_{331}}=\frac{\mathcal{A}^{331}_{\mathrm{SU}(3)^3_L}}{12},\qquad\mathcal{A}^{331}_{X^3_{331}}=\frac{\mathcal{A}^{331}_{\mathrm{SU}(3)^3_L}}{9}, \tag*{(401)} A SU ( 3 ) L 2 X 331 331 = 12 A SU ( 3 ) L 3 331 , A X 331 3 331 = 9 A SU ( 3 ) L 3 331 , (401)
and the color and mixed gravitational anomalies vanish. Consequently the remaining anomaly and color-asymptotic-freedom conditions are
2 N c k = ( N c − 1 ) N g , b 0 , c ( 331 ) = 11 N c 3 − 2 N g > 0. (402) 2N_ck=(N_c-1)N_g,\qquad b^{(331)}_{0,c}=\frac{11N_c}{3}-2N_g>0. \tag*{(402)} 2 N c k = ( N c − 1 ) N g , b 0 , c ( 331 ) = 3 11 N c − 2 N g > 0. (402)
For odd N c N_c N c , coprimality of N c N_c N c and ( N c − 1 ) / 2 (N_c-1)/2 ( N c − 1 ) /2 makes N g N_g N g a multiple of N c N_c N c ; the inequality in (402) then leaves
N g = N c , k = N c − 1 2 . (403) N_g=N_c,\qquad k=\frac{N_c-1}{2}. \tag*{(403)} N g = N c , k = 2 N c − 1 . (403)
For even N c N_c N c , integrality makes N g N_g N g a multiple of 2 N c 2N_c 2 N c , which violates the same inequality. Thus every odd N c ≥ 3 N_c\ge3 N c ≥ 3 survives these two tests. With the three scalar triplets used below, the SU(3)L _L L coefficient is
b 0 , L ( 331 ) = 11 − N g ( N c + 1 ) 3 − 1 2 . (404) b^{(331)}_{0,L}=11-\frac{N_g(N_c+1)}{3}-\frac{1}{2}. \tag*{(404)} b 0 , L ( 331 ) = 11 − 3 N g ( N c + 1 ) − 2 1 . (404)
The explicit alternative ( N c , N g , k ) = ( 5 , 5 , 2 ) (N_c,N_g,k)=(5,5,2) ( N c , N g , k ) = ( 5 , 5 , 2 ) has
b 0 , c ( 331 ) = 25 3 , b 0 , L ( 331 ) = 1 2 , (405) b^{(331)}_{0,c}=\frac{25}{3},\qquad b^{(331)}_{0,L}=\frac{1}{2}, \tag*{(405)} b 0 , c ( 331 ) = 3 25 , b 0 , L ( 331 ) = 2 1 , (405)
so both nonabelian factors are asymptotically free. Its ordinary quark charges are 3 / 5 3/5 3/5 and − 2 / 5 -2/5 − 2/5 . If the observed charges 2 / 3 2/3 2/3 and − 1 / 3 -1/3 − 1/3 are instead imposed, then Y q = 1 / 6 = 1 / ( 2 N c ) Y_q=1/6=1/(2N_c) Y q = 1/6 = 1/ ( 2 N c ) forces N c = 3 N_c=3 N c = 3 ; those observed charges are then empirical inputs. The anomaly mechanism therefore correlates the number of families and colors in this class but does not select three colors by itself.
An illustrative SU(6) splitting, global orientation, and chiral obstruction
It is useful to distinguish a calculable residual Lie algebra from a full global embedding of the matter representations. Fix an SU(6) parent, a traceless Hermitian adjoint Σ 6 \Sigma_6 Σ 6 , and the positive potential
U 6 ( Σ 6 ) = λ 6 tr [ ( Σ 6 2 − v 6 2 6 1 6 ) 2 ] , λ 6 , v 6 > 0. (406) \mathcal{U}_6(\Sigma_6)=\lambda_6\operatorname{tr}\left[\left(\Sigma_6^2-\frac{v_6^2}{6}\mathbf{1}_6\right)^2\right],\qquad\lambda_6,v_6>0. \tag*{(406)} U 6 ( Σ 6 ) = λ 6 tr [ ( Σ 6 2 − 6 v 6 2 1 6 ) 2 ] , λ 6 , v 6 > 0. (406)
The potential in (406) is a sum of squares. After diagonalizing Σ 6 \Sigma_6 Σ 6 , every zero of this sum has eigenvalues ± v 6 / 6 \pm v_6/\sqrt{6} ± v 6 / 6 . If r r r eigenvalues are positive, tracelessness gives ( 2 r − 6 ) v 6 / 6 = 0 (2r-6)v_6/\sqrt{6}=0 ( 2 r − 6 ) v 6 / 6 = 0 , hence r = 3 r=3 r = 3 . All zero-cost vacua therefore lie in the one SU(6) orbit of
Σ 6 , ∗ = v 6 6 diag ( 1 3 , − 1 3 ) . (407) \Sigma_{6,*}=\frac{v_6}{\sqrt{6}}\operatorname{diag}(\mathbf{1}_3,-\mathbf{1}_3). \tag*{(407)} Σ 6 , ∗ = 6 v 6 diag ( 1 3 , − 1 3 ) . (407)
Its stabilizer is S(U(3) × \times × U(3)), with Lie algebra
s u ( 3 ) ⊕ s u ( 3 ) ⊕ u ( 1 ) . (408) \mathfrak{su}(3)\oplus\mathfrak{su}(3)\oplus\mathfrak{u}(1). \tag*{(408)} su ( 3 ) ⊕ su ( 3 ) ⊕ u ( 1 ) . (408)
This is a conditional algebraic splitting into two SU(3) factors. The Lie ranks of SU(6) and SU(3) are five and two, respectively; the calculation does not distinguish which factor would be color.
The global quotient is essential. The homomorphism
( U , V , z ) ⟼ diag ( z U , z − 1 V ) (409) (U,V,z)\longmapsto\operatorname{diag}(zU,z^{-1}V) \tag*{(409)} ( U , V , z ) ⟼ diag ( z U , z − 1 V ) (409)
from SU(3) × \times × SU(3) × \times × U(1) has kernel
{ ( ω 3 − m 1 3 , ω 3 m 1 3 , ω 3 m ) : m = 0 , 1 , 2 } ≅ Z 3 , ω 3 = e 2 π i / 3 . (410) \left\{(\omega_3^{-m}\mathbf{1}_3,\omega_3^m\mathbf{1}_3,\omega_3^m):m=0,1,2\right\}\cong\mathbb{Z}_3,\qquad\omega_3=e^{2\pi i/3}. \tag*{(410)} { ( ω 3 − m 1 3 , ω 3 m 1 3 , ω 3 m ) : m = 0 , 1 , 2 } ≅ Z 3 , ω 3 = e 2 π i /3 . (410)
A representation with the two trialities ( a , b ) (a,b) ( a , b ) and integral U(1) charge n X n_X n X descends through this quotient only if
− a + b + n X = 0 ( m o d 3 ) . (411) -a+b+n_X=0\pmod{3}. \tag*{(411)} − a + b + n X = 0 ( mod 3 ) . (411)
In the normalization of (387), the charge appearing here is n X = x = 3 X 331 n_X=x=3X_{331} n X = x = 3 X 331 . The condition is consequently a global representation condition and not another local anomaly equation. The proposed quark antitriplet ( 3 , 3 ‾ ) 0 (\mathbf{3},\overline{\mathbf{3}})_0 ( 3 , 3 ) 0 has ( a , b , n X ) = ( 1 , − 1 , 0 ) (a,b,n_X)=(1,-1,0) ( a , b , n X ) = ( 1 , − 1 , 0 ) and violates (412). This excludes the direct identification of the parent blocks with physical color and the left group, not every global identification. Composing the color factor with complex conjugation instead gives
ψ ( C , L , z ) = diag ( z C ‾ , z − 1 L ) , ker ψ = K 331 . (412) \psi(C,L,z)=\operatorname{diag}(z\overline{C},z^{-1}L),\qquad\ker\psi=K_{331}. \tag*{(412)} ψ ( C , L , z ) = diag ( z C , z − 1 L ) , ker ψ = K 331 . (412)
The first block has determinant z 3 z^3 z 3 and the second z − 3 z^{-3} z − 3 , so the image lies in S(U(3) × \times × U(3)). Conversely a cube root of the first determinant supplies a preimage. The induced global-group isomorphism is therefore correct, and the descent condition becomes a C + b L + x = 0 ( m o d 3 ) a_C+b_L+x=0\pmod{3} a C + b L + x = 0 ( mod 3 ) , exactly the condition in (389).
Global compatibility nevertheless does not supply a chiral ultraviolet completion. The following obstruction is independent of the direct orientation failure and agrees with the full calculation in Paper II. Let P = e i π / 6 ( 0 1 3 1 3 0 ) P=e^{i\pi/6}\begin{pmatrix}\mathbf{0}&\mathbf{1}_3\\\mathbf{1}_3&\mathbf{0}\end{pmatrix} P = e iπ /6 ( 0 1 3 1 3 0 ) . The block exchange has determinant − 1 -1 − 1 , so det P = 1 \det P=1 det P = 1 , and
P ψ ( C , L , z ) P − 1 = ψ ( L ‾ , C ‾ , z − 1 ) . (413) P\psi(C,L,z)P^{-1}=\psi(\overline{L},\overline{C},z^{-1}). \tag*{(413)} P ψ ( C , L , z ) P − 1 = ψ ( L , C , z − 1 ) . (413)
Thus the normalizer acts on residual irreducible labels by τ ( R C , R L ) x = ( R L ‾ , R C ‾ ) − x \tau(R_C,R_L)_x=(\overline{R_L},\overline{R_C})_{-x} τ ( R C , R L ) x = ( R L , R C ) − x . For every finite-dimensional parent representation, its action of P P P intertwines the restricted representation with this pullback. Uniqueness of compact-group decomposition forces n ( r ) = n ( τ r ) n(r)=n(\tau r) n ( r ) = n ( τ r ) . Since τ \tau τ commutes with conjugation, the chiral index I ( r ) = n ( r ) − n ( r ˉ ) I(r)=n(r)-n(\bar r) I ( r ) = n ( r ) − n ( r ˉ ) obeys
I ( τ r ) = I ( r ) . (414) I(\tau r)=I(r). \tag*{(414)} I ( τ r ) = I ( r ) . (414)
In the following chiral-index counts the subscripts denote the integer character x = 3 X 331 x=3X_{331} x = 3 X 331 , rather than the fractional labels in (388). In the three-family left-handed spectrum, r = ( 1 , 3 ) − 1 r=(\mathbf{1},\mathbf{3})_{-1} r = ( 1 , 3 ) − 1 has index 3, whereas τ r = ( 3 ‾ , 1 ) + 1 \tau r=(\overline{\mathbf{3}},\mathbf{1})_{+1} τ r = ( 3 , 1 ) + 1 has index 5 from the three ordinary and two exotic down-type singlet conjugates. Already 3 ≠ 5 3\ne5 3 = 5 contradicts (415). Independently, ( 1 , 1 ) + 3 (\mathbf{1},\mathbf{1})_{+3} ( 1 , 1 ) + 3 has index 3 and is sent to its conjugate, which would require that index to vanish. Additional vectorlike pairs change neither index. Thus no finite collection of complete four-dimensional SU(6) multiplets yields this target spectrum through the stated adjoint breaking. A completion must change the parent, breaking pattern, spectrum, or restriction mechanism, and must still verify its own anomaly and unwanted-state conditions.
A 331 auxiliary-map vacuum, parity, and flavor sector
We next give a scalar realization in the G 331 G_{331} G 331 candidate itself. It is to be understood as a candidate square-norm component of the auxiliary map entering the master scalar potential U H \mathcal{U}_{H} U H , rather than as an unrelated fundamental action. At this point U 331 \mathcal{U}_{331} U 331 is an effective scalar block; it becomes a member of G a d m \mathcal{G}_{\mathrm{adm}} G adm only after an equivariant auxiliary map, regulator, mobility, reference measure, and continuation and matching data have been fixed. Use three distinct fields
Φ η , Φ Ξ ∼ ( 1 , 3 ) − 1 / 3 , Φ ρ ∼ ( 1 , 3 ) 2 / 3 . (415) \Phi_{\eta},\Phi_{\Xi}\sim(\mathbf{1},\mathbf{3})_{-1/3},\qquad\Phi_{\rho}\sim(\mathbf{1},\mathbf{3})_{2/3}. \tag*{(415)} Φ η , Φ Ξ ∼ ( 1 , 3 ) − 1/3 , Φ ρ ∼ ( 1 , 3 ) 2/3 . (415)
Take v η , v ρ , v Ξ , Λ D > 0 v_{\eta},v_{\rho},v_{\Xi},\Lambda_{D}>0 v η , v ρ , v Ξ , Λ D > 0 . An explicit positive component that imposes the desired orbit is
U 331 = ∑ a ∈ { η , ρ , Ξ } λ a 331 ( Φ a † Φ a − v a 2 2 ) 2 + ∑ a , b ∈ { η , ρ , Ξ } a < b ζ a b ∣ Φ a † Φ b ∣ 2 + ζ D Λ D 2 ∣ det ( Φ η , Φ ρ , Φ Ξ ) − v η v ρ v Ξ 2 2 ∣ 2 , λ a 331 , ζ a b , ζ D > 0. (416) \mathcal{U}_{331}
=
\sum_{a\in\{\eta,\rho,\Xi\}}\lambda^{331}_{a}\left(\Phi_a^\dagger\Phi_a-\frac{v_a^2}{2}\right)^2
+
\sum_{\substack{a,b\in\{\eta,\rho,\Xi\}\\a<b}}\zeta_{ab}\lvert\Phi_a^\dagger\Phi_b\rvert^2
+
\frac{\zeta_D}{\Lambda_D^2}\left\lvert\det(\Phi_{\eta},\Phi_{\rho},\Phi_{\Xi})-\frac{v_{\eta}v_{\rho}v_{\Xi}}{2\sqrt{2}}\right\rvert^2,
\qquad\lambda^{331}_a,\zeta_{ab},\zeta_D>0. \tag*{(416)} U 331 = a ∈ { η , ρ , Ξ } ∑ λ a 331 ( Φ a † Φ a − 2 v a 2 ) 2 + a , b ∈ { η , ρ , Ξ } a < b ∑ ζ ab ∣ Φ a † Φ b ∣ 2 + Λ D 2 ζ D det ( Φ η , Φ ρ , Φ Ξ ) − 2 2 v η v ρ v Ξ 2 , λ a 331 , ζ ab , ζ D > 0. (416)
The determinant has total X 331 X_{331} X 331 -charge zero. Every summand in (417) is nonnegative and all vanish at
⟨ Φ η ⟩ = v η 2 e 1 , ⟨ Φ ρ ⟩ = v ρ 2 e 2 , ⟨ Φ Ξ ⟩ = v Ξ 2 e 3 . (417) \langle\Phi_{\eta}\rangle=\frac{v_{\eta}}{\sqrt{2}}\mathbf{e}_1,\qquad
\langle\Phi_{\rho}\rangle=\frac{v_{\rho}}{\sqrt{2}}\mathbf{e}_2,\qquad
\langle\Phi_{\Xi}\rangle=\frac{v_{\Xi}}{\sqrt{2}}\mathbf{e}_3. \tag*{(417)} ⟨ Φ η ⟩ = 2 v η e 1 , ⟨ Φ ρ ⟩ = 2 v ρ e 2 , ⟨ Φ Ξ ⟩ = 2 v Ξ e 3 . (417)
Conversely, a zero of the potential fixes the three lengths, their pairwise orthogonality, and the determinant phase. Hence the normalized column matrix
U Φ : = 2 ( Φ η v η , Φ ρ v ρ , Φ Ξ v Ξ ) (418) U_{\Phi}:=\sqrt{2}\left(\frac{\Phi_{\eta}}{v_{\eta}},\frac{\Phi_{\rho}}{v_{\rho}},\frac{\Phi_{\Xi}}{v_{\Xi}}\right) \tag*{(418)} U Φ := 2 ( v η Φ η , v ρ Φ ρ , v Ξ Φ Ξ ) (418)
obeys U Φ † U Φ = 1 3 U_{\Phi}^{\dagger}U_{\Phi}=\mathbf{1}_3 U Φ † U Φ = 1 3 and det U Φ = 1 \det U_{\Phi}=1 det U Φ = 1 , and so belongs to S U ( 3 ) L \mathrm{SU}(3)_L SU ( 3 ) L . Every zero-cost vacuum is consequently gauge-equivalent to (417) . The final term in (416) contains a dimension-six effective interaction and requires a cutoff or an auxiliary-field completion. The hierarchy v Ξ ≫ v η , v ρ v_{\Xi}\gg v_{\eta},v_{\rho} v Ξ ≫ v η , v ρ and its radiative stability remain additional input.
The e 3 e_3 e 3 expectation value is invariant under the upper-block S U ( 2 ) L \mathrm{SU}(2)_L SU ( 2 ) L and under Y = − T 8 / 3 + X 331 Y=-T_8/\sqrt{3}+X_{331} Y = − T 8 / 3 + X 331 , whereas the three aligned expectations preserve precisely Q = T 3 + Y Q=T_3+Y Q = T 3 + Y . Thus, under the stated hierarchy, the alignment realizes the breaking chain at the Lie-algebra level, with finite central quotients suppressed in this display,
G 331 ⟶ S U ( 3 ) c × S U ( 2 ) L × U ( 1 ) Y ⟶ S U ( 3 ) c × U ( 1 ) e m . (419) G_{331}\longrightarrow\mathrm{SU}(3)_c\times\mathrm{SU}(2)_L\times\mathrm{U}(1)_Y\longrightarrow\mathrm{SU}(3)_c\times\mathrm{U}(1)_{\mathrm{em}}. \tag*{(419)} G 331 ⟶ SU ( 3 ) c × SU ( 2 ) L × U ( 1 ) Y ⟶ SU ( 3 ) c × U ( 1 ) em . (419)
The three displayed ⟨ Φ ⟩ \langle\Phi\rangle ⟨ Φ ⟩ are representatives of the selected orbit on the gauge-fixed realization used for this calculation. The gauge-invariant content of the breaking chain is the orbit stabilizer and the mass spectrum of ( D μ Φ ) † D μ Φ (D_{\mu}\Phi)^{\dagger}D_{\mu}\Phi ( D μ Φ ) † D μ Φ ; it is not an assertion of a gauge-noninvariant local order parameter in a finite-volume unfixed measure. At this order, write
v E W 2 : = v η 2 + v ρ 2 , M W 2 = g 3 L 2 v E W 2 4 (420) v_{\mathrm{EW}}^{2}:=v_{\eta}^{2}+v_{\rho}^{2},\qquad M_{W}^{2}=\frac{g_{3L}^{2}v_{\mathrm{EW}}^{2}}{4} \tag*{(420)} v EW 2 := v η 2 + v ρ 2 , M W 2 = 4 g 3 L 2 v EW 2 (420)
at tree level after the electroweak stage, with the coupling evaluated at the chosen common matching scale.
An ordinary–exotic separation can be imposed rather than assumed. Give Φ Ξ , U R , D R , E R \Phi_{\Xi},U_R,D_R,E_R Φ Ξ , U R , D R , E R intrinsic parity + 1 +1 + 1 , and give Φ η , Φ ρ , Q L t , Q L a , L L , T R , B R \Phi_{\eta},\Phi_{\rho},Q^{t}_L,Q^{a}_L,L_L,T_R,B_R Φ η , Φ ρ , Q L t , Q L a , L L , T R , B R intrinsic parity − 1 -1 − 1 . The product of this parity with diag ( − 1 , − 1 , 1 ) ∈ S U ( 3 ) L \operatorname{diag}(-1,-1,1)\in\mathrm{SU}(3)_L diag ( − 1 , − 1 , 1 ) ∈ SU ( 3 ) L defines a residual parity Π s p \Pi_{\mathrm{sp}} Π sp that preserves (417) . The ordinary doublet components are Π s p \Pi_{\mathrm{sp}} Π sp -even and the exotic third components are Π s p \Pi_{\mathrm{sp}} Π sp -odd. In particular, the parity-even ordinary Yukawa operators can be written as
− L Y , o r d = Q ‾ r L t Φ η ( Y t u ) r i U i R + Q ‾ r L t Φ ρ ( Y t d ) r i D i R + Q ‾ α L a Φ ρ ∗ ( Y a u ) α i U i R + Q ‾ α L a Φ η ∗ ( Y a d ) α i D i R + h . c . , (421) -\mathcal{L}_{Y,\mathrm{ord}}=\overline{Q}^{t}_{rL}\Phi_{\eta}(Y^{u}_{t})_{ri}U_{iR}+\overline{Q}^{t}_{rL}\Phi_{\rho}(Y^{d}_{t})_{ri}D_{iR}
+\overline{Q}^{a}_{\alpha L}\Phi_{\rho}^{*}(Y^{u}_{a})_{\alpha i}U_{iR}+\overline{Q}^{a}_{\alpha L}\Phi_{\eta}^{*}(Y^{d}_{a})_{\alpha i}D_{iR}+\mathrm{h.c.}, \tag*{(421)} − L Y , ord = Q r L t Φ η ( Y t u ) r i U i R + Q r L t Φ ρ ( Y t d ) r i D i R + Q α L a Φ ρ ∗ ( Y a u ) α i U i R + Q α L a Φ η ∗ ( Y a d ) α i D i R + h.c. , (421)
where repeated family indices are summed. The exotic operators
Q ‾ r L t Φ Ξ ( Y T ) r s T s R and Q ‾ α L a Φ Ξ ∗ ( Y B ) α β B β R (422) \overline{Q}^{t}_{rL}\Phi_{\Xi}(Y_T)_{rs}T_{sR}\qquad\text{and}\qquad\overline{Q}^{a}_{\alpha L}\Phi_{\Xi}^{*}(Y_B)_{\alpha\beta}B_{\beta R} \tag*{(422)} Q r L t Φ Ξ ( Y T ) r s T s R and Q α L a Φ Ξ ∗ ( Y B ) α β B β R (422)
are also parity even, whereas ordinary–exotic mass mixing is forbidden. Full rank of both Y T Y_T Y T and Y B Y_B Y B is necessary and sufficient for every exotic quark to have a nonzero tree-level mass. On the k = 1 k=1 k = 1 branch, Y T ≠ 0 Y_T\ne0 Y T = 0 is equivalent to full rank. The exotic masses are uniformly of order v Ξ v_{\Xi} v Ξ only when the corresponding singular values are bounded above and below by positive constants independent of v E W / v Ξ v_{\mathrm{EW}}/v_{\Xi} v EW / v Ξ and of the flavor expansion parameter. A quantum-consistent realization of Π s p \Pi_{\mathrm{sp}} Π sp and an acceptable fate for its lightest odd state are still required; in particular, a stable colored lightest state is not viable.
Introduce a G 331 G_{331} G 331 -singlet, Π s p \Pi_{\mathrm{sp}} Π sp -even flavor spurion φ F N ∼ ( 1 , 1 ) 0 \varphi_{\mathrm{FN}} \sim(1,1)_0 φ FN ∼ ( 1 , 1 ) 0 of flavor charge − 1 -1 − 1 , and fix its phase convention by
ε F N : = ⟨ φ F N ⟩ Λ F , 0 < ε F N ≪ 1. (423) \varepsilon_{\mathrm{FN}} := \frac{\langle\varphi_{\mathrm{FN}}\rangle}{\Lambda_F}, \qquad0 < \varepsilon_{\mathrm{FN}} \ll1. \tag*{(423)} ε FN := Λ F ⟨ φ FN ⟩ , 0 < ε FN ≪ 1. (423)
After choosing an ordering of the ordinary flavor rows in (422), assign
F ( Q i L ) = − q i , F ( U j R ) = u j , F ( D j R ) = d j , (424) F(Q_{iL})=-q_i,\qquad F(U_{jR})=u_j,\qquad F(D_{jR})=d_j, \tag*{(424)} F ( Q i L ) = − q i , F ( U j R ) = u j , F ( D j R ) = d j , (424)
and give the scalar triplets zero flavor charge. The effective operators then carry powers q i + u j q_i+u_j q i + u j and q i + d j q_i+d_j q i + d j . This is the Froggatt–Nielsen mechanism [74]. For the illustrative choice
q = ( 3 , 2 , 0 ) , u = ( 5 , 2 , 0 ) , d = ( 4 , 3 , 3 ) , (425) q=(3,2,0),\qquad u=(5,2,0),\qquad d=(4,3,3), \tag*{(425)} q = ( 3 , 2 , 0 ) , u = ( 5 , 2 , 0 ) , d = ( 4 , 3 , 3 ) , (425)
the Yukawa matrices have the factorized form
Y u = D q C u D u , Y d = D q C d D d , D f = diag ( ε F N f 1 , ε F N f 2 , ε F N f 3 ) . (426) \mathbf{Y}_u=\mathbf{D}_q\mathbf{C}_u\mathbf{D}_u,\qquad
\mathbf{Y}_d=\mathbf{D}_q\mathbf{C}_d\mathbf{D}_d,\qquad
\mathbf{D}_f=\operatorname{diag}\left(\varepsilon_{\mathrm{FN}}^{f_1},\varepsilon_{\mathrm{FN}}^{f_2},\varepsilon_{\mathrm{FN}}^{f_3}\right). \tag*{(426)} Y u = D q C u D u , Y d = D q C d D d , D f = diag ( ε FN f 1 , ε FN f 2 , ε FN f 3 ) . (426)
The row-dependent electroweak vacuum factors in (422) are included in C u , C d \mathbf{C}_u,\mathbf{C}_d C u , C d ; the statements below require them to introduce no additional powers of ε F N \varepsilon_{\mathrm{FN}} ε FN . Assume that the relevant minors of C u , C d \mathbf{C}_u,\mathbf{C}_d C u , C d remain nonzero and of order one as ε F N → 0 \varepsilon_{\mathrm{FN}}\to0 ε FN → 0 .
For a matrix Y \mathbf{Y} Y with descending singular values σ 1 ( Y ) ≥ σ 2 ( Y ) ≥ σ 3 ( Y ) \sigma_1(\mathbf{Y})\geq\sigma_2(\mathbf{Y})\geq\sigma_3(\mathbf{Y}) σ 1 ( Y ) ≥ σ 2 ( Y ) ≥ σ 3 ( Y ) , the exterior powers satisfy
∥ ⋀ r Y ∥ o p = ∏ j = 1 r σ j ( Y ) . (427) \left\|\bigwedge\nolimits^r \mathbf{Y}\right\|_{\mathrm{op}}=\prod_{j=1}^{r}\sigma_j(\mathbf{Y}). \tag*{(427)} ⋀ r Y op = j = 1 ∏ r σ j ( Y ) . (427)
The generic-minor assumptions and (427) give
( ∥ Y u ∥ o p , ∥ ⋀ 2 Y u ∥ o p , ∣ det Y u ∣ ) ≍ ( 1 , ε F N 4 , ε F N 12 ) , \left(\|\mathbf{Y}_u\|_{\mathrm{op}},\|\bigwedge\nolimits^2\mathbf{Y}_u\|_{\mathrm{op}},|\det\mathbf{Y}_u|\right)\asymp\left(1,\varepsilon_{\mathrm{FN}}^4,\varepsilon_{\mathrm{FN}}^{12}\right), ( ∥ Y u ∥ op , ∥ ⋀ 2 Y u ∥ op , ∣ det Y u ∣ ) ≍ ( 1 , ε FN 4 , ε FN 12 ) ,
( ∥ Y d ∥ o p , ∥ ⋀ 2 Y d ∥ o p , ∣ det Y d ∣ ) ≍ ( ε F N 3 , ε F N 8 , ε F N 15 ) . (428) \left(\|\mathbf{Y}_d\|_{\mathrm{op}},\|\bigwedge\nolimits^2\mathbf{Y}_d\|_{\mathrm{op}},|\det\mathbf{Y}_d|\right)\asymp\left(\varepsilon_{\mathrm{FN}}^3,\varepsilon_{\mathrm{FN}}^8,\varepsilon_{\mathrm{FN}}^{15}\right). \tag*{(428)} ( ∥ Y d ∥ op , ∥ ⋀ 2 Y d ∥ op , ∣ det Y d ∣ ) ≍ ( ε FN 3 , ε FN 8 , ε FN 15 ) . (428)
Dividing the second and third products in (429) by their preceding products yields
m u : m c : m t ∼ ε F N 8 : ε F N 4 : 1 , m d : m s : m b ∼ ε F N 7 : ε F N 5 : ε F N 3 . (429) m_u:m_c:m_t\sim\varepsilon_{\mathrm{FN}}^8:\varepsilon_{\mathrm{FN}}^4:1,\qquad
m_d:m_s:m_b\sim\varepsilon_{\mathrm{FN}}^7:\varepsilon_{\mathrm{FN}}^5:\varepsilon_{\mathrm{FN}}^3. \tag*{(429)} m u : m c : m t ∼ ε FN 8 : ε FN 4 : 1 , m d : m s : m b ∼ ε FN 7 : ε FN 5 : ε FN 3 . (429)
Absent cancellations between the two left rotations, the same charge differences give the scaling estimates
∣ V u s ∣ ∼ ε F N , ∣ V c b ∣ ∼ ε F N 2 , ∣ V u b ∣ ∼ ε F N 3 . (430) |V_{us}|\sim\varepsilon_{\mathrm{FN}},\qquad
|V_{cb}|\sim\varepsilon_{\mathrm{FN}}^2,\qquad
|V_{ub}|\sim\varepsilon_{\mathrm{FN}}^3. \tag*{(430)} ∣ V u s ∣ ∼ ε FN , ∣ V c b ∣ ∼ ε FN 2 , ∣ V u b ∣ ∼ ε FN 3 . (430)
These are power laws, not numerical predictions: the coefficients, phases, charges in (426), and ε F N \varepsilon_{\mathrm{FN}} ε FN remain independent microscopic data. A gauged flavor symmetry also requires its own anomaly analysis, while a global continuous version requires a mechanism for the associated would-be Goldstone mode. The spurion presentation is an effective description and contains no messenger fields in the spectrum used in Theorem 16 . If the asymptotic-freedom premise of that theorem is imposed through a scale at which colored flavor messengers are active, their Dynkin indices and thresholds must be included in (392) ; the family-count proof applies only on a range with no additional active colored states.
Neutral fermions, alignment-dependent FCNCs, and heavy masses
The neutral third component of L L L_L L L is not automatically made heavy by ⟨ Φ Ξ ⟩ \langle\Phi_\Xi\rangle ⟨ Φ Ξ ⟩ . To give a fully declared tree-level candidate, introduce three Π s p \Pi_{\mathrm{sp}} Π sp -even gauge singlets N R ∼ ( 1 , 1 ) 0 N_R \sim(1,1)_0 N R ∼ ( 1 , 1 ) 0 and three Π s p \Pi_{\mathrm{sp}} Π sp -odd gauge singlets S R ∼ ( 1 , 1 ) 0 S_R \sim(1,1)_0 S R ∼ ( 1 , 1 ) 0 . The corresponding neutral-sector Lagrangian is
− L Y , ν = L ‾ Φ Y ν N R + L ‾ Φ Y S S R + 1 2 N R c ‾ M N N R + h . c . (431) -\mathcal{L}_{Y,\nu}=\overline{L}_\Phi Y_\nu N_R+\overline{L}_\Phi Y_S S_R+\frac{1}{2}\overline{N_R^c}M_NN_R+\mathrm{h.c.} \tag*{(431)} − L Y , ν = L Φ Y ν N R + L Φ Y S S R + 2 1 N R c M N N R + h.c. (431)
Here Y S Y_S Y S is a 3 × 3 3 \times3 3 × 3 matrix. To assert that every odd neutral third component is heavy, assume constants 0 < c S ≤ C S < ∞ 0<c_S\le C_S<\infty 0 < c S ≤ C S < ∞ independent of v E W / v Ξ v_{\mathrm{EW}}/v_\Xi v EW / v Ξ and of ε F N \varepsilon_{\mathrm{FN}} ε FN such that
c S ≤ σ min ( Y S ) ≤ σ max ( Y S ) ≤ C S . (432) c_S\le\sigma_{\min}(Y_S)\le\sigma_{\max}(Y_S)\le C_S. \tag*{(432)} c S ≤ σ m i n ( Y S ) ≤ σ m a x ( Y S ) ≤ C S . (432)
For the matching action used here, the independently allowed odd-singlet Majorana block M S M_S M S is set to zero and independent dimension-five neutrino operators are excluded at the matching scale. These are explicit action assumptions, not consequences of the gauge group or parity. On this M S = 0 M_S=0 M S = 0 branch, the odd neutral components and S R S_R S R then form three Dirac fermions with mass singular values
σ i ( m S ) = v Ξ 2 σ i ( Y S ) , m S = v Ξ 2 Y S , i = 1 , 2 , 3 , (433) \sigma_i(m_S)=\frac{v_\Xi}{\sqrt{2}}\sigma_i(Y_S),\qquad m_S=\frac{v_\Xi}{\sqrt{2}}Y_S,\qquad i=1,2,3, \tag*{(433)} σ i ( m S ) = 2 v Ξ σ i ( Y S ) , m S = 2 v Ξ Y S , i = 1 , 2 , 3 , (433)
and are uniformly of order v Ξ v_\Xi v Ξ . Without (432) , the displayed gauge-invariant operator alone establishes neither full rank nor a common heavy scale. The charged-lepton operator and its mass matrix are
− L Y ℓ = L ‾ Φ Y e E R + h . c . , M e = v ρ 2 Y e . (434) -\mathcal{L}^{\ell}_Y=\overline{L}_\Phi Y_e E_R+\mathrm{h.c.},\qquad M_e=\frac{v_\rho}{\sqrt{2}}Y_e. \tag*{(434)} − L Y ℓ = L Φ Y e E R + h.c. , M e = 2 v ρ Y e . (434)
If M S M_S M S is retained, the complete odd mass matrix must be diagonalized; a parametrically large block can generate light odd states despite full-rank Y S Y_S Y S . For the even sector, assume M N = M N T M_N=M_N^T M N = M N T , det M N ≠ 0 \det M_N\ne0 det M N = 0 , and ∥ m ν D M N − 1 ∥ ≪ 1 \lVert m_{\nu D}M_N^{-1}\rVert\ll1 ∥ m ν D M N − 1 ∥ ≪ 1 . The leading type-I seesaw expression is
m ν = − m ν D M N − 1 m ν D T , m ν D = v η 2 Y ν . (435) m_\nu=-m_{\nu D}M_N^{-1}m_{\nu D}^T,\qquad m_{\nu D}=\frac{v_\eta}{\sqrt{2}}Y_\nu. \tag*{(435)} m ν = − m ν D M N − 1 m ν D T , m ν D = 2 v η Y ν . (435)
Let D ℓ \mathbf{D}_\ell D ℓ and D n \mathbf{D}_n D n be invertible diagonal flavor-power matrices. In particular, D ℓ T = D ℓ \mathbf{D}_\ell^T=\mathbf{D}_\ell D ℓ T = D ℓ and D n T = D n \mathbf{D}_n^T=\mathbf{D}_n D n T = D n . If
Y ν = D ℓ C D D n , M N = M 0 D n C N D n , Y_\nu=\mathbf{D}_\ell C_D\mathbf{D}_n,\qquad M_N=M_0\mathbf{D}_nC_N\mathbf{D}_n, Y ν = D ℓ C D D n , M N = M 0 D n C N D n ,
C N = C N T , det C N ≠ 0 , M 0 > 0 , (436) C_N=C_N^T,\qquad\det C_N\ne0,\qquad M_0>0, \tag*{(436)} C N = C N T , det C N = 0 , M 0 > 0 , (436)
then M N M_N M N is symmetric and invertible, and the inverse is
M N − 1 = 1 M 0 D n − 1 C N − 1 D n − 1 . (437) M_N^{-1} = \frac{1}{M_0}D_n^{-1}C_N^{-1}D_n^{-1}. \tag*{(437)} M N − 1 = M 0 1 D n − 1 C N − 1 D n − 1 . (437)
Using the transpose of the first line in (436) in (435) , the adjacent D n D_n D n and D n − 1 D_n^{-1} D n − 1 factors cancel, and therefore
m ν = − v η 2 2 M 0 D ℓ C D C N − 1 C D T D ℓ . (438) m_\nu= -\frac{v_\eta^2}{2M_0}D_\ell C_D C_N^{-1} C_D^T D_\ell. \tag*{(438)} m ν = − 2 M 0 v η 2 D ℓ C D C N − 1 C D T D ℓ . (438)
Equal lepton-doublet charges can allow large mixing, but do not determine its angles.
The remaining numerical freedom can be proved without a fit. For any positive diagonal masses and any unitary V C K M V_{\mathrm{CKM}} V CKM , choose
M u = D u m a s s , M d = V C K M D d m a s s . (439) M_u = D_u^{\mathrm{mass}}, \qquad M_d = V_{\mathrm{CKM}}D_d^{\mathrm{mass}}. \tag*{(439)} M u = D u mass , M d = V CKM D d mass . (439)
The left diagonalizations give the prescribed V C K M V_{\mathrm{CKM}} V CKM . Similarly, for any positive diagonal D ν D_\nu D ν , any unitary U P M N S U_{\mathrm{PMNS}} U PMNS , and M N = M 1 3 M_N = M\mathbf{1}_3 M N = M 1 3 with M M M sufficiently large for the seesaw expansion, the choice
m ν D = i M U P M N S ∗ D ν (440) m_{\nu D} = i\sqrt{M}\,U_{\mathrm{PMNS}}^*\sqrt{D_\nu} \tag*{(440)} m ν D = i M U PMNS ∗ D ν (440)
gives, at the leading seesaw order stated in (435) , m ν = U P M N S ∗ D ν U P M N S † m_\nu= U_{\mathrm{PMNS}}^*D_\nu U_{\mathrm{PMNS}}^\dagger m ν = U PMNS ∗ D ν U PMNS † through (435) . With diagonal charged-lepton masses, U P M N S T m ν U P M N S = D ν U_{\mathrm{PMNS}}^T m_\nu U_{\mathrm{PMNS}} = D_\nu U PMNS T m ν U PMNS = D ν , so U P M N S U_{\mathrm{PMNS}} U PMNS is the Takagi matrix. Gauge invariance and the displayed flavor power counting therefore do not fix numerical CKM or PMNS data without a further selection principle.
Anomaly cancellation distinguishes one quark representation, but it does not identify that representation with a measured generation before the Yukawa matrices are diagonalized. Let P f a m = e f a m e f a m † P_{\mathrm{fam}} = e_{\mathrm{fam}}e_{\mathrm{fam}}^\dagger P fam = e fam e fam † project onto the special gauge triplet. In a quark mass basis, a nonuniversal neutral interaction has the form
G q = c 0 1 + Δ c U q L † P f a m U q L , ( G q ) i j = Δ c ( U q L ) f a m , i ( U q L ) f a m , j ∗ ( i ≠ j ) . (441) G_q = c_0\mathbf{1} + \Delta c\,U_{qL}^\dagger P_{\mathrm{fam}}U_{qL}, \qquad(G_q)_{ij} = \Delta c\,(U_{qL})_{\mathrm{fam},i}(U_{qL})_{\mathrm{fam},j}^* \quad(i \ne j). \tag*{(441)} G q = c 0 1 + Δ c U q L † P fam U q L , ( G q ) ij = Δ c ( U q L ) fam , i ( U q L ) fam , j ∗ ( i = j ) . (441)
The CKM matrix alone is insufficient to determine these couplings. Indeed, for every unitary W W W , the pair U u L = W U_{uL} = W U uL = W , U d L = W V C K M U_{dL} = W V_{\mathrm{CKM}} U d L = W V CKM has the same U u L † U d L = V C K M U_{uL}^\dagger U_{dL} = V_{\mathrm{CKM}} U uL † U d L = V CKM , but in general produces different projectors in (441) . Neutral meson mixing consequently probes flavor alignment and the heavy neutral boson scale jointly.
For canonically normalized gauge fields,
D μ = ∂ μ − i g 3 L T a A μ a − i g X X 331 B μ , (442) D_\mu= \partial_\mu- ig_{3L}T_a A_\mu^a - ig_X X_{331}B_\mu, \tag*{(442)} D μ = ∂ μ − i g 3 L T a A μ a − i g X X 331 B μ , (442)
the leading v Ξ v_\Xi v Ξ -scale masses follow directly from ∣ D μ ⟨ Φ Ξ ⟩ ∣ 2 \lvert D_\mu\langle\Phi_\Xi\rangle\rvert^2 ∣ D μ ⟨ Φ Ξ ⟩ ∣ 2 . The off-diagonal vectors have
M o f f 2 = g 3 L 2 v Ξ 2 4 . (443) M_{\mathrm{off}}^2 = \frac{g_{3L}^2v_\Xi^2}{4}. \tag*{(443)} M off 2 = 4 g 3 L 2 v Ξ 2 . (443)
Since the neutral third component has T 8 = − 1 / 3 T_8=-1/\sqrt{3} T 8 = − 1/ 3 and X 331 = − 1 / 3 X_{331}=-1/3 X 331 = − 1/3 , the neutral mass matrix is
M 0 , 331 2 = v Ξ 2 ( g 3 L 2 / 3 g 3 L g X / ( 3 3 ) g 3 L g X / ( 3 3 ) g X 2 / 9 ) . (444) M^2_{0,331}=v_\Xi^2\begin{pmatrix} g_{3L}^2/3 & g_{3L}g_X/(3\sqrt{3}) \\ g_{3L}g_X/(3\sqrt{3}) & g_X^2/9 \end{pmatrix}. \tag*{(444)} M 0 , 331 2 = v Ξ 2 ( g 3 L 2 /3 g 3 L g X / ( 3 3 ) g 3 L g X / ( 3 3 ) g X 2 /9 ) . (444)
Its determinant is zero, and its nonzero eigenvalue gives
M Z 331 ′ 2 = v Ξ 2 ( g 3 L 2 3 + g X 2 9 ) , M Z 331 ′ 2 M o f f 2 = 4 3 + 4 g X 2 9 g 3 L 2 , 1 g Y 2 = 1 3 g 3 L 2 + 1 g X 2 . (445) M^2_{Z'_{331}}=v_\Xi^2\left(\frac{g_{3L}^2}{3}+\frac{g_X^2}{9}\right),\qquad\frac{M^2_{Z'_{331}}}{M^2_{\mathrm{off}}}=\frac{4}{3}+\frac{4g_X^2}{9g_{3L}^2},\qquad\frac{1}{g_Y^2}=\frac{1}{3g_{3L}^2}+\frac{1}{g_X^2}. \tag*{(445)} M Z 331 ′ 2 = v Ξ 2 ( 3 g 3 L 2 + 9 g X 2 ) , M off 2 M Z 331 ′ 2 = 3 4 + 9 g 3 L 2 4 g X 2 , g Y 2 1 = 3 g 3 L 2 1 + g X 2 1 . (445)
The final identity matches the canonically normalized massless combination to Y = − T 8 / 3 + X 331 Y=-T_8/\sqrt{3}+X_{331} Y = − T 8 / 3 + X 331 , so the unbroken charge convention agrees with Q = T 3 + Y Q=T_3+Y Q = T 3 + Y used for the Standard-Model phase. It requires all couplings to be evaluated at a common matching scale. Positivity of the abelian kinetic normalization requires
1 g X 2 = 1 g Y 2 − 1 3 g 3 L 2 > 0 , sin 2 θ W < 3 4 , tan θ W = g Y g 3 L . (446) \frac{1}{g_X^2}=\frac{1}{g_Y^2}-\frac{1}{3g_{3L}^2}>0,\qquad\sin^2\theta_W<\frac{3}{4},\qquad\tan\theta_W=\frac{g_Y}{g_{3L}}. \tag*{(446)} g X 2 1 = g Y 2 1 − 3 g 3 L 2 1 > 0 , sin 2 θ W < 4 3 , tan θ W = g 3 L g Y . (446)
This is a consistency condition of the declared 331 embedding, not a prediction of v Ξ v_\Xi v Ξ . Electroweak corrections are relatively O ( v E W 2 / v Ξ 2 ) O(v_{\mathrm{EW}}^2/v_\Xi^2) O ( v EW 2 / v Ξ 2 ) , and loop and threshold corrections require a separate calculation. Exotic-quark masses are the singular values of v Ξ Y T , B / 2 v_\Xi Y_{T,B}/\sqrt{2} v Ξ Y T , B / 2 . For the isolated radial potential λ Ξ ( Φ Ξ † Φ Ξ − v Ξ 2 / 2 ) 2 \lambda_\Xi(\Phi_\Xi^\dagger\Phi_\Xi-v_\Xi^2/2)^2 λ Ξ ( Φ Ξ † Φ Ξ − v Ξ 2 /2 ) 2 , one obtains m h Ξ 2 = 2 λ Ξ v Ξ 2 m_{h_\Xi}^2=2\lambda_\Xi v_\Xi^2 m h Ξ 2 = 2 λ Ξ v Ξ 2 ; the complete multi-scalar theory instead requires the full canonically normalized Hessian.
6.12. What remains to be selected. At fixed dimensionless couplings, the replacement v Ξ ↦ a v Ξ v_\Xi\mapsto av_\Xi v Ξ ↦ a v Ξ rescales all leading heavy masses by a a a . The common master action does not eliminate this scaling freedom. The physical masses to be compared with experiment are propagator poles after continuation and matching; stochastic relaxation eigenvalues are not automatically those poles. A predictive completion must therefore supply a global-group-compatible chiral parent, a selected flavor map and vacuum scale, threshold matching, and a target-independent comparison proving uniqueness of the candidate class. Before those further constructions, the anomaly identities, vacuum orbit, flavor power laws, flavor-changing-neutral-current structure, and leading mass ratio above are conditional consequences of the explicitly stated data. This intermediate-scale 331 block can obey the existing conditional Yang–Mills projection once its kinetic and continuation hypotheses are verified; it does not add a separate derivation of the Einstein equation, whose status remains that of the Cartan branch analyzed earlier.
Well-posedness, conditional recovery, and parameter matching
7.1. Conditional emergence theorem. The following theorem states precisely what the framework can establish once a concrete master algebra and measure have been supplied. Because the hypotheses explicitly include the residual group and the effective-sign and stability conditions, the result is a conditional implication; it is not an existence proof for a master algebra or an independent derivation of the observed group.
Theorem 17 (Conditional effective-sector implication). Assume the following hypotheses.
(H1) The master configuration space is well-defined and the master action S M S_{\mathrm{M}} S M is bounded below on the admissible class.
(H2) A specified realization of the stochastic law (62) admits a global weak solution, and its invariant probability measure is unique in the selected phase. If this probability is identified with a Gibbs law for state-dependent mobility or a nonconstant reference density, the realization uses the complete reversible drift (80) , or its intrinsic counterpart (81) , rather than the uncorrected gradient drift alone.
(H3) The group-selection problem (355) has a unique admissible minimizer Γ ∗ \Gamma_\ast Γ ∗ , whose residual internal symmetry is S U ( 3 ) c × S U ( 2 ) L × U ( 1 ) Y \mathrm{SU}(3)_{\mathrm{c}} \times\mathrm{SU}(2)_{\mathrm{L}} \times\mathrm{U}(1)_{\mathrm{Y}} SU ( 3 ) c × SU ( 2 ) L × U ( 1 ) Y .
(H4) The selected matter representation is anomaly-free and contains the required low-energy chiral representations. If the claim includes propagating Dirac fermions, the microscopic or regularized action must additionally specify a non-degenerate first-order Dirac kinetic operator on those representations; this is an extra hypothesis and does not follow from the quadratic norm alone.
(H5) The fluctuation operator about the selected phase is of Laplace type after gauge fixing, and the heat-kernel expansion is valid in the relevant low-energy regime.
(H6) The Lorentzian continuation exists and satisfies the required positivity, causality, and unitarity conditions.
(H7) After gauge fixing or passage to the quotient by gauge orbits, the selected phase admits a reductive decomposition with a non-degenerate physical potential Hessian on the broken directions. Its positive kinetic pencil defines the tree-level curvature-mass scales. A positive mobility-weighted relaxation spectrum, when available, is a separate stochastic property and is not identified with a physical pole mass. The residual gauge symmetry protects the unbroken directions from an explicit mass term at that order. In addition, the quadratic form induced by ∣ D Φ H ∣ 2 |D\Phi_{\mathrm{H}}|^2 ∣ D Φ H ∣ 2 on the broken gauge generators has full rank on g b r o k e n \mathfrak{g}_{\mathrm{broken}} g broken after the vacuum is inserted; this is the condition that actually gives masses to all claimed broken vector modes.
(H8) The scalar/order-parameter representation contains the required electroweak doublet with its stated hypercharge, and its renormalized potential is bounded below and has a stable, non-degenerate vacuum implementing $\mathrm{SU}(2)_{\mathrm{L}} \times
U ( 1 ) Y → U ( 1 ) e m U(1)_{\mathrm{Y}} \to U(1)_{\mathrm{em}} U ( 1 ) Y → U ( 1 ) em . The regularization and matching prescription gives nonzero, finite coefficients for the scalar-curvature, gauge-curvature, scalar, and matter terms at the claimed effective order; moreover G e f f > 0 G_{\mathrm{eff}} > 0 G eff > 0 , the gauge kinetic matrix is positive on the Euclidean slice, and the scalar kinetic and potential terms have the signs required for stability. If a propagating fermionic matter action is claimed, the regularized matter block is additionally assumed to contain a non-degenerate first-order Dirac operator compatible with the selected representation. Without this extra hypothesis the conclusion about matter is only the selected representation content, not a derived Dirac kinetic term.
Then the low-energy effective action contains, to the corresponding order in the heat-kernel expansion, an Einstein term with positive Newton coefficient, Yang–Mills fields of the residual gauge group with positive kinetic form, and a stable Higgs-type symmetry-breaking sector. It also contains the matter terms associated with those selected representations for which a kinetic operator has been explicitly included. Electromagnetism appears as the unbroken U ( 1 ) e m U(1)_{\mathrm{em}} U ( 1 ) em subgroup after electroweak symmetry breaking. The theorem does not by itself derive a first-order Dirac kinetic term; that conclusion is conditional on the additional Dirac clause in (H8).
Proof. By (H3), the selected residual group is the stated product. By (H7), the reductive decomposition and the full-rank broken-vector quadratic form give mass terms to the claimed broken vector modes, while residual gauge symmetry forbids an explicit local mass term for the unbroken connections at that order. The latter statement is not a claim that every residual sector has a massless physical excitation: a confining sector may have a physical mass gap without such a local mass term. By (H4), the selected representations are consistent with the residual gauge symmetry. A Dirac kinetic term is available only when the additional first-order hypothesis in (H8) is satisfied. By (H5) and (H8), the one-loop determinant has a valid heat-kernel expansion and its renormalized coefficients include the scalar-curvature, gauge-curvature, scalar, and any explicitly specified matter structures displayed in (351) . By (H6), the Euclidean effective action can be continued to a Lorentzian effective theory without violating the stated physical consistency requirements. The final electromagnetic subgroup follows from the residual stabilizer of the electroweak order parameter. The theorem still does not determine numerical coupling values without an explicit matching calculation.
If the selected propagating matter has a non-negligible spin current, the connection component of Theorem 9 lies on its Einstein–Cartan branch. The torsionless Einstein reduction additionally requires the spin-current suppression stated there. When algebraic torsion can be integrated out with an invertible local field equation, the effective action instead acquires the corresponding local spin-current interactions; the Dirac-operator hypothesis alone does not remove them.
The hypotheses identify the mathematical conditions that must be verified in a concrete realization. In particular, (H3) and (H8) are not proved by the abstract master-action definitions: they supply the group-selection and renormalized matching information needed to turn the implication into a physical derivation. The theorem therefore records the logical dependence of the effective-sector statement without claiming that the Standard Model group or its numerical couplings have already been selected. All of these hypotheses are mathematical premises of the implication, but those selecting the residual group, vacuum, kinetic signs, scale separation, and Lorentzian continuation specify the physical effective branch rather than merely technical regularity.
A non-circular definition of prediction
A parameter is predicted by the framework only if it is not inserted into the microscopic model as an independent free datum and is uniquely determined by the selection, normalization, and matching procedures. Let λ \lambda λ denote a low-energy parameter. We call λ \lambda λ a structural prediction if
λ = Λ ( Γ ∗ ) (447) \lambda= \Lambda(\Gamma_{*}) \tag*{(447)} λ = Λ ( Γ ∗ ) (447)
for a uniquely selected microscopic object Γ ∗ \Gamma_{*} Γ ∗ , where Λ \Lambda Λ is fixed by the theory and not by fitting the measured value of λ \lambda λ .
In particular, the following hierarchy should be distinguished:
Structural consequence: follows algebraically from the master definitions.
Conditional consequence: follows once explicitly stated hypotheses are assumed.
Matching result: follows after fixing renormalization and normalization conditions.
Empirical input: supplied by experiment and not predicted.
This distinction specifies which observables can be compared with data after matching without using those same observations as inputs.
Renormalized parameter matching and identifiability
The preceding definition of prediction becomes operational only after the microscopic data have been mapped to renormalized observables. We make this map explicit in order to keep three logically different operations separate: the evaluation of the selected microscopic action, the integration of modes between two scales, and the choice of quantities that are calibrated from experiment. Let Q Γ ⊂ R d Q_{\Gamma} \subset\mathbb{R}^{d} Q Γ ⊂ R d be a local coordinate space for the finite collection of microscopic parameters that remains after the algebraic constraints, gauge quotient, and vacuum equations for a candidate Γ \Gamma Γ have been imposed. Typical coordinates, when they are independent in a particular realization, include w ∗ w_\ast w ∗ , q c q_c q c , ℓ \ell ℓ , entries of the invariant pairing κ \kappa κ , and heavy masses and coupling vectors. The notation Q Γ \mathcal{Q}_{\Gamma} Q Γ does not assume that these quantities are free: a successful selection problem may reduce it to a single point.
Fix a matching scale μ m a t c h \mu_{\mathrm{match}} μ match , a renormalization prescription R \mathcal{R} R , and reference units for dimensionful observables. The renormalized matching map is
M μ , R ( Γ , q ) = O ( μ ) : = ( G e f f − 1 ( μ ) g 1 − 2 ( μ ) g 2 − 2 ( μ ) g 3 − 2 ( μ ) λ H ( μ ) m H 2 ( μ ) ⋮ ) , q ∈ Q Γ . (448) \mathcal{M}_{\mu,\mathcal{R}}(\Gamma,\mathbf{q})=\mathbf{O}(\mu):=
\begin{pmatrix}
G_{\mathrm{eff}}^{-1}(\mu)\\
g_1^{-2}(\mu)\\
g_2^{-2}(\mu)\\
g_3^{-2}(\mu)\\
\lambda_H(\mu)\\
m_H^2(\mu)\\
\vdots
\end{pmatrix},
\qquad\mathbf{q}\in\mathcal{Q}_{\Gamma}. \tag*{(448)} M μ , R ( Γ , q ) = O ( μ ) := G eff − 1 ( μ ) g 1 − 2 ( μ ) g 2 − 2 ( μ ) g 3 − 2 ( μ ) λ H ( μ ) m H 2 ( μ ) ⋮ , q ∈ Q Γ . (448)
Here the target space is the finite product of the corresponding observable spaces, with dimensions and units fixed by the chosen reference convention. The ellipsis denotes only observables for which the selected realization supplies a renormalized operator and a finite matching prescription; it does not silently add an independently postulated sector. The map is defined only after the regulator, field normalization, subtraction scheme, and residual operator basis have been specified. A change of scheme changes the coordinates of O \mathbf{O} O and the finite terms below, but not a scheme-consistent observable relation.
For the joint candidate, the tree-level component of M \mathcal{M} M is the previously derived relation
g a − 2 ( μ m a t c h ) = w ∗ κ a , G e f f − 1 ( μ m a t c h ) = 8 π w ∗ g c 2 ℓ 2 , Λ E = 3 ℓ 2 . (449) g_a^{-2}(\mu_{\mathrm{match}})=w_\ast\kappa_a,\qquad
G_{\mathrm{eff}}^{-1}(\mu_{\mathrm{match}})=\frac{8\pi w_\ast}{g_c^2\ell^2},\qquad
\Lambda_E=\frac{3}{\ell^2}. \tag*{(449)} g a − 2 ( μ match ) = w ∗ κ a , G eff − 1 ( μ match ) = g c 2 ℓ 2 8 π w ∗ , Λ E = ℓ 2 3 . (449)
The first two identities are the components of (237); the last one fixes the curvature convention used in the Cartan witness. They are relations among selected microscopic coordinates, not numerical values, until the selection problem fixes those coordinates.
To state the perturbative part without assigning uncomputed constants, introduce a dimensionless loop-counting parameter ϵ l o o p \epsilon_{\mathrm{loop}} ϵ loop , set to one after the perturbative order has been identified. This parameter is distinct from the fixed fluctuation scale θ \theta θ . Choose the convention
β a ( g a ) : = μ d g a d μ = − ϵ l o o p b a g a 3 16 π 2 + O ( ϵ l o o p 2 g a 5 ) . (450) \beta_a(g_a):=\mu\frac{\mathrm{d}g_a}{\mathrm{d}\mu}
=-\epsilon_{\mathrm{loop}}\frac{b_a g_a^3}{16\pi^2}
+O\left(\epsilon_{\mathrm{loop}}^2 g_a^5\right). \tag*{(450)} β a ( g a ) := μ d μ d g a = − ϵ loop 16 π 2 b a g a 3 + O ( ϵ loop 2 g a 5 ) . (450)
The chain rule gives d ( g a − 2 ) / d log μ = − 2 g a − 3 β a ( g a ) \mathrm{d}(g_a^{-2})/\mathrm{d}\log\mu=-2g_a^{-3}\beta_a(g_a) d ( g a − 2 ) / d log μ = − 2 g a − 3 β a ( g a ) . Integrating between μ m a t c h \mu_{\mathrm{match}} μ match and μ \mu μ , and adding the finite threshold and subtraction contributions, gives
g a − 2 ( μ ) = w ∗ κ a + ϵ l o o p b a 8 π 2 log ( μ μ m a t c h ) + Δ a t h ( μ m a t c h ; Γ , q , R ) + Δ a f i n ( μ m a t c h ; Γ , q , R ) + O ( ϵ l o o p 2 ) + O ( E 2 M H 2 ) , G e f f − 1 ( μ ) = 8 π w ∗ g c 2 ℓ 2 + Δ G l o o p ( μ , μ m a t c h ; Γ , q , R ) + Δ G t h ( μ m a t c h ; Γ , q , R ) + M r e f 2 O ( ϵ l o o p ) . (451) \begin{aligned}
g_a^{-2}(\mu) &= w_*\kappa_a+\epsilon_{\mathrm{loop}}\frac{b_a}{8\pi^2}\log\left(\frac{\mu}{\mu_{\mathrm{match}}}\right) \\
&\quad+\Delta_a^{\mathrm{th}}(\mu_{\mathrm{match}};\Gamma,\mathbf{q},\mathcal{R})+\Delta_a^{\mathrm{fin}}(\mu_{\mathrm{match}};\Gamma,\mathbf{q},\mathcal{R}) \\
&\quad+O(\epsilon_{\mathrm{loop}}^2)+O\left(\frac{E^2}{M_H^2}\right),\qquad
G_{\mathrm{eff}}^{-1}(\mu)=\frac{8\pi w_*}{g_c^2\ell^2}+\Delta_G^{\mathrm{loop}}(\mu,\mu_{\mathrm{match}};\Gamma,\mathbf{q},\mathcal{R})\\
&\quad+\Delta_G^{\mathrm{th}}(\mu_{\mathrm{match}};\Gamma,\mathbf{q},\mathcal{R})+M_{\mathrm{ref}}^2O(\epsilon_{\mathrm{loop}}). \tag*{(451)}
\end{aligned} g a − 2 ( μ ) = w ∗ κ a + ϵ loop 8 π 2 b a log ( μ match μ ) + Δ a th ( μ match ; Γ , q , R ) + Δ a fin ( μ match ; Γ , q , R ) + O ( ϵ loop 2 ) + O ( M H 2 E 2 ) , G eff − 1 ( μ ) = g c 2 ℓ 2 8 π w ∗ + Δ G loop ( μ , μ match ; Γ , q , R ) + Δ G th ( μ match ; Γ , q , R ) + M ref 2 O ( ϵ loop ) . (451)
Here M r e f > 0 M_{\mathrm{ref}}>0 M ref > 0 is a fixed matching mass, introduced to state the dimension-two gravitational remainder in dimensionless units. The Δ \Delta Δ -terms include their declared loop orders and any retained tree-level threshold. The order symbols concern a specified perturbative truncation on a fixed range of scales; they are not nonperturbative error bounds unless such bounds are proved separately. The coefficient b a b_a b a is determined by the selected light spectrum and representations. The scale E E E is the largest external momentum in the effective calculation and M H M_H M H is the smallest integrated-out mass in the retained expansion. The displayed remainder is therefore a controlled statement only when E / M H E/M_H E / M H is small and the loop expansion is valid. A different beta-function convention changes the sign assigned to b a b_a b a and the logarithm together, leaving the integrated equation unchanged. No numerical value of b a b_a b a , a threshold, or a coupling is asserted here.
The heavy-mode calculation in Section 11.10 supplies an explicit component of the threshold term. For one scalar and the operator basis used there,
Δ C i j ( ϕ ) = c i c j m ϕ 2 m ∗ 2 , C e f f = C d i r e c t + Δ C ( ϕ ) + C l o o p , (452) \Delta C_{ij}^{(\phi)}=\frac{c_i c_j}{m_\phi^2m_*^2},\qquad C_{\mathrm{eff}}=C_{\mathrm{direct}}+\Delta C^{(\phi)}+C_{\mathrm{loop}}, \tag*{(452)} Δ C ij ( ϕ ) = m ϕ 2 m ∗ 2 c i c j , C eff = C direct + Δ C ( ϕ ) + C loop , (452)
where the second identity is understood in one fixed nonredundant basis and renormalization scheme. (452) is the rank-one contribution of the specified exchange; it does not replace the determinant, direct terms, other heavy fields, or finite counterterms. Those terms are part of Δ t h \Delta^{\mathrm{th}} Δ th or Δ f i n \Delta^{\mathrm{fin}} Δ fin in the full map and must be calculated before a physical coefficient is compared with data.
We next state the local inverse-problem criterion used to decide whether a matching calculation actually determines the microscopic coordinates. Let N N N be the number of selected observable components and write M ∗ ( q ) : = M μ , R ( Γ ∗ , q ) \mathcal{M}_*(\mathbf{q}):=\mathcal{M}_{\mu,\mathcal{R}}(\Gamma_*,\mathbf{q}) M ∗ ( q ) := M μ , R ( Γ ∗ , q ) . In local coordinates, the matching Jacobian is
J A α ( q ∗ ) : = ∂ M ∗ , A ∂ q α ∣ q = q ∗ , 1 ≤ A ≤ N , 1 ≤ α ≤ d . (453) J_{A\alpha}(\mathbf{q}_*):=\left.\frac{\partial\mathcal{M}_{*,A}}{\partial q^\alpha}\right|_{\mathbf{q}=\mathbf{q}_*},\qquad1\le A\le N,\qquad1\le\alpha\le d. \tag*{(453)} J A α ( q ∗ ) := ∂ q α ∂ M ∗ , A q = q ∗ , 1 ≤ A ≤ N , 1 ≤ α ≤ d . (453)
Theorem 18 (Conditional matching and local identifiability). Assume the following.
(M1) The selection functional has a unique minimizer Γ ∗ \Gamma_\ast Γ ∗ modulo package equivalence, and the selected vacuum determines a point q ∗ ∈ Q Γ ∗ \mathbf{q}_\ast\in Q_{\Gamma_\ast} q ∗ ∈ Q Γ ∗ independently of the measured values of the observables under consideration.
(M2) The regulator, residual operator basis, and renormalization prescription R \mathcal{R} R are fixed, and the perturbative and heavy-mode expansions in (451) have a specified error bound.
(M3) The point q ∗ \mathbf{q}_\ast q ∗ lies in the interior of a d d d -dimensional smooth coordinate chart of the physical quotient. In this chart, the map M ∗ : Q Γ ∗ → R N \mathcal{M}_\ast: Q_{\Gamma_\ast} \to\mathbb{R}^{N} M ∗ : Q Γ ∗ → R N is C 1 C^1 C 1 , after dimensionful coordinates have been expressed in the fixed reference units.
Then the following statements hold.
(i) The tree-level, loop, and threshold expressions define one matching map, with the tree component given by (449) and the corrections separated as in (451) .
(ii) If N = d N=d N = d and det J ( q ∗ ) ≠ 0 \det J(\mathbf{q}_\ast) \ne0 det J ( q ∗ ) = 0 , there are neighborhoods U U U of q ∗ \mathbf{q}_\ast q ∗ and V V V of M ∗ ( q ∗ ) \mathcal{M}_\ast(\mathbf{q}_\ast) M ∗ ( q ∗ ) for which M ∗ : U → V \mathcal{M}_\ast: U \to V M ∗ : U → V has a unique C 1 C^1 C 1 inverse. Thus the selected matching data determine the microscopic coordinates locally, modulo the package equivalence already removed in defining Q Γ ∗ Q_{\Gamma_\ast} Q Γ ∗ .
(iii) If N ≥ d N \ge d N ≥ d and rank J ( q ∗ ) = d \operatorname{rank} J(\mathbf{q}_\ast)=d rank J ( q ∗ ) = d , a set of d d d observable components has a nonsingular Jacobian minor. Those components locally identify q ∗ \mathbf{q}_\ast q ∗ ; the remaining components are consistency relations and can be tested without refitting the identified coordinates.
(iv) If rank J ( q ∗ ) < d \operatorname{rank} J(\mathbf{q}_\ast)<d rank J ( q ∗ ) < d , there is a nonzero tangent direction h ∈ ker J ( q ∗ ) h \in\ker J(\mathbf{q}_\ast) h ∈ ker J ( q ∗ ) with
M ∗ ( q ∗ + ε h ) = M ∗ ( q ∗ ) + o ( ε ) . (454) \mathcal{M}_\ast(\mathbf{q}_\ast+\varepsilon h)=\mathcal{M}_\ast(\mathbf{q}_\ast)+o(\varepsilon). \tag*{(454)} M ∗ ( q ∗ + ε h ) = M ∗ ( q ∗ ) + o ( ε ) . (454)
The linearized matching problem is then non-identifiable. Higher-order terms can remove this degeneracy only after an additional nonlinear calculation; rank deficiency alone cannot be reported as a unique parameter determination.
Proof. The first assertion is a bookkeeping identity: the tree term is obtained by evaluating the selected invariant pairing on the low-energy operators, the renormalization-group term is the integral of the beta function between the two displayed scales, and the threshold term is the finite contribution of the modes removed at μ m a t c h \mu_{\mathrm{match}} μ match . The separation is meaningful once the scheme and operator basis in (M2) have been fixed. The one-loop formula follows directly from μ d ( g a 2 ) / d μ = ε l o o p b a / ( 8 π 2 ) + O ( ε l o o p 2 g a 2 ) \mu\,\mathrm{d}(g_a^2)/\mathrm{d}\mu=\varepsilon_{\mathrm{loop}}b_a/(8\pi^2)+O(\varepsilon_{\mathrm{loop}}^2g_a^2) μ d ( g a 2 ) / d μ = ε loop b a / ( 8 π 2 ) + O ( ε loop 2 g a 2 ) , which is the derivative of the beta-function convention stated above. The heavy-scalar part is the Gaussian elimination identity already proved in Theorem 41; its contribution is therefore exactly the first term in (452) . For (ii), the hypotheses make M ∗ \mathcal{M}_* M ∗ a C 1 C^1 C 1 map between two finite-dimensional spaces of equal dimension. The inverse-function theorem applied at q ∗ \mathbf{q}_* q ∗ gives neighborhoods U , V U,V U , V and a unique C 1 C^1 C 1 inverse because det J ( q ∗ ) ≠ 0 \det J(\mathbf{q}_*) \ne0 det J ( q ∗ ) = 0 . Package-equivalent directions have already been quotiented out, so this is the claimed physical local identifiability.
For (iii), full column rank provides a set of d d d rows whose d × d d \times d d × d minor is invertible. Apply the inverse-function theorem to the submap made of those rows. Its local inverse determines all coordinates in q ∗ \mathbf{q}_* q ∗ ; composing the remaining rows with that inverse gives well-defined local consistency relations. These relations are not used to choose a second set of fitted coordinates.
For (iv), choose a nonzero h h h in the kernel of J ( q ∗ ) J(\mathbf{q}_*) J ( q ∗ ) . The first-order Taylor formula for the C 1 C^1 C 1 map gives M ∗ ( q ∗ + ε h ) = M ∗ ( q ∗ ) + ε J ( q ∗ ) h + o ( ε ) = M ∗ ( q ∗ ) + o ( ε ) \mathcal{M}_*(\mathbf{q}_*+\varepsilon h)=\mathcal{M}_*(\mathbf{q}_*)+\varepsilon J(\mathbf{q}_*)h+o(\varepsilon)=\mathcal{M}_*(\mathbf{q}_*)+o(\varepsilon) M ∗ ( q ∗ + ε h ) = M ∗ ( q ∗ ) + ε J ( q ∗ ) h + o ( ε ) = M ∗ ( q ∗ ) + o ( ε ) , proving the stated linearized non-identifiability. No assertion about an exact nonlinear family is made without further hypotheses. □
The theorem also fixes the status of a numerical statement. If (M1) fixes q ∗ \mathbf{q}_* q ∗ without using the measured values of O \mathbf{O} O , every component of M ∗ ( q ∗ ) \mathcal{M}_*(\mathbf{q}_*) M ∗ ( q ∗ ) is a prediction in principle, subject to the stated truncation error. If instead r r r coordinates are calibrated from r r r observables, the calibrated coordinates are empirical inputs; only independent relations among the remaining observables are predictions. More invariantly, a relation R 0 \mathcal{R}_0 R 0 is parameter-free on the selected class when
R 0 ( M μ , R ( Γ ∗ , q ) ) = 0 for every admissible q retained by selection. (455) \mathcal{R}_0(\mathcal{M}_{\mu,\mathcal{R}}(\Gamma_*,\mathbf{q}))=0 \quad\text{for every admissible }\mathbf{q}\text{ retained by selection.} \tag*{(455)} R 0 ( M μ , R ( Γ ∗ , q )) = 0 for every admissible q retained by selection. (455)
Thus a successful fit, by itself, is a reconstruction. A prediction requires either a uniquely selected q ∗ \mathbf{q}_* q ∗ or a relation such as (455) that survives all still-allowed microscopic coordinates and the calculated loop and threshold corrections.
Gauge invariance and stochastic covariance
Let g g g denote a smooth deterministic gauge transformation, possibly depending on time. For the Cartan bundle specified above, g g g is at least an H H H -valued transformation; a transformation by the full G ~ \widetilde{G} G requires an additional extension of the bundle and is not assumed automatically. The master connection transforms as
A ↦ g A g − 1 + g d g − 1 , (456) A \mapsto gAg^{-1}+g\,dg^{-1}, \tag*{(456)} A ↦ g A g − 1 + g d g − 1 , (456)
and the curvature transforms covariantly,
F ↦ g F g − 1 . (457) F \mapsto gFg^{-1}. \tag*{(457)} F ↦ g F g − 1 . (457)
If the invariant pairing satisfies (46) and the extension data satisfy (49) , then
⟨ F , F ⟩ g (458) \langle F,F\rangle_g \tag*{(458)} ⟨ F , F ⟩ g (458)
is gauge invariant. The matter and information terms are gauge invariant provided their representations and covariant derivatives are chosen consistently.
The stochastic mobility must transform as a tensor on configuration space. If R g R_g R g denotes the tangent action of a gauge transformation, assume that the chosen field-space metric is R g R_g R g -isometric and let R g ∗ R_g^* R g ∗ denote the corresponding dual action on cotangent vectors. Covariance then requires
M ( g ⋅ X ) = R g M ( X ) R g ∗ . (459) \mathrm{M}(g\cdot\mathbf{X})=R_g\mathrm{M}(\mathbf{X})R_g^*. \tag*{(459)} M ( g ⋅ X ) = R g M ( X ) R g ∗ . (459)
Here the right-hand side uses the cotangent action on the dual slot; in the metric-identified tangent notation this is equivalently
M ^ ( g ⋅ X ) = R g M ^ ( X ) R g − 1 , σ ( g ⋅ X ) σ ( g ⋅ X ) ∗ = R g ( σ σ ∗ ) ( X ) R g − 1 . (460) \widehat{\mathrm{M}}(g\cdot\mathbf{X})=R_g\widehat{\mathrm{M}}(\mathbf{X})R_g^{-1},\qquad\sigma(g\cdot\mathbf{X})\sigma(g\cdot\mathbf{X})^*=R_g(\sigma\sigma^*)(\mathbf{X})R_g^{-1}. \tag*{(460)} M ( g ⋅ X ) = R g M ( X ) R g − 1 , σ ( g ⋅ X ) σ ( g ⋅ X ) ∗ = R g ( σ σ ∗ ) ( X ) R g − 1 . (460)
This is a covariance condition on the second-order symbol. For a smooth deterministic coordinate transformation y = Υ ( t , x ) y=\Upsilon(t,x) y = Υ ( t , x ) , Itô’s formula also gives the transformed drift
b ~ a = ∂ t Υ a + ∂ i Υ a b i + 1 2 a i j ∂ i ∂ j Υ a . (461) \widetilde{b}^{a}=\partial_t\Upsilon^{a}+\partial_i\Upsilon^{a}b^i+\frac{1}{2}a^{ij}\partial_i\partial_j\Upsilon^{a}. \tag*{(461)} b a = ∂ t Υ a + ∂ i Υ a b i + 2 1 a ij ∂ i ∂ j Υ a . (461)
The time-derivative term is present even for an affine time-dependent transformation; the quadratic-variation term is present for a nonlinear transformation even when it is time independent. A stochastic gauge transformation requires additional Itô–Wentzell and cross-variation terms and is not covered by this deterministic change of coordinates. Stratonovich notation absorbs the latter into the ordinary chain rule, but retains the former. A mathematically complete construction must either work on the quotient configuration space or introduce a gauge-fixing and BRST/BV treatment. The covariance condition alone does not complete the global quotient construction.
7.5. Well-posedness requirements. An infinite-dimensional stochastic equation requires an analytic configuration space in addition to its geometric notation. In particular, a realization of (62) must specify the operator domains, the boundary conditions, the treatment of gauge orbits, the covariance of the noise, and the sense in which the stochastic integral is defined. A quotient by gauge orbits need not be a Hilbert space. The Hilbert-space statements below apply to a specified Hilbert realization of a gauge slice, or to quotient coordinates for which that structure has actually been established.
On a finite-dimensional slice, write M N M_N M N for the tangent mobility and S N S_N S N for the projected action. In Euclidean coordinates with Lebesgue reference volume, the reversible Itô approximation is
d X t N = [ − M N ∇ S N + θ div M N ] ( X t N ) d t + 2 θ M N ( X t N ) d W t N . (462) \mathrm{d}X_t^N=\left[-M_N\nabla S_N+\theta\operatorname{div}M_N\right](X_t^N)\,\mathrm{d}t+\sqrt{2\theta M_N(X_t^N)}\,\mathrm{d}W_t^N. \tag*{(462)} d X t N = [ − M N ∇ S N + θ div M N ] ( X t N ) d t + 2 θ M N ( X t N ) d W t N . (462)
Here ( div M N ) i = ∑ j ∂ j M N i j (\operatorname{div}M_N)^i=\sum_j\partial_jM_N^{ij} ( div M N ) i = ∑ j ∂ j M N ij . For constant mobility the divergence vanishes. With a positive reference density r N r_N r N , it is replaced by r N − 1 div ( r N M N ) r_N^{-1}\operatorname{div}(r_NM_N) r N − 1 div ( r N M N ) . Thus a projection of the formal gradient and a projection of the noise do not, by themselves, specify a reversible approximation for a variable mobility or a nonconstant reference density.
A useful first estimate is
sup N E [ sup t ≤ T V ( X t N ) + ∫ 0 T D ( X t N ) d t ] < ∞ . (463) \sup_N \mathbb{E}\left[\sup_{t\leq T}\mathcal{V}(X_t^N)+\int_0^T \mathcal{D}(X_t^N)\,\mathrm{d}t\right]<\infty. \tag*{(463)} N sup E [ t ≤ T sup V ( X t N ) + ∫ 0 T D ( X t N ) d t ] < ∞. (463)
Its consequences depend on what the functionals control. A bound on the Hilbert norm alone does not give compactness in an infinite-dimensional Hilbert space. Spatial compactness, control of time increments, and identification of the limiting martingales are distinct requirements. The general stochastic-evolution and invariant-measure framework is developed in [53] and [54]; the martingale-problem and measure-theoretic background is given in [148] and [19].
A complete finite-regulator equilibrium construction
The scalar confinement estimate can be extended to all variables of a specified bosonic regulator. This supplies an existence and uniqueness result in which the coefficients and the reference measure are fixed. It also separates the finite-regulator problem from the continuum geometry problem. In this construction the lattice geometry is fixed; the coframe is not an additional integration variable.
Let V \mathcal{V} V be a finite set of n n n vertices, let E \mathcal{E} E be a finite set of oriented links, and choose finitely many oriented closed plaquette paths P \mathcal{P} P . Set
K = S O ( 5 ) × S U ( 3 ) × S U ( 2 ) × U ( 1 ) , K = SO(5) \times SU(3) \times SU(2) \times U(1), K = S O ( 5 ) × S U ( 3 ) × S U ( 2 ) × U ( 1 ) ,
X l a t = K E × R ν , ν l a t = d U d s . (464) \mathcal{X}_{\mathrm{lat}} = K^{\mathcal{E}} \times\mathbb{R}^{\nu}, \qquad\nu_{\mathrm{lat}} = \mathrm{d}U\,\mathrm{d}s. \tag*{(464)} X lat = K E × R ν , ν lat = d U d s . (464)
Here d U \mathrm{d}U d U is product Haar probability and d s \mathrm{d}s d s is ordinary Lebesgue measure in the declared dimensionless scalar coordinates. For each plaquette p p p , let U p U_p U p be the ordered link product, using an inverse when its orientation opposes that of a link. Choose a base vertex v ( p ) v(p) v ( p ) , fixed positive coefficients β a \beta_a β a , and unitary representations R a R_a R a of dimensions d a d_a d a for the four factors. Define
k p ( U ) = ∑ a β a [ d a − Re tr R a ( U p , a ) ] , k_p(U) = \sum_a \beta_a\left[d_a-\operatorname{Re}\operatorname{tr}R_a(U_{p,a})\right], k p ( U ) = a ∑ β a [ d a − Re tr R a ( U p , a ) ] ,
0 ≤ k p ( U ) ≤ k p , max : = 2 ∑ a β a d a . (465) 0\leq k_p(U)\leq k_{p,\max}:=2\sum_a\beta_a d_a. \tag*{(465)} 0 ≤ k p ( U ) ≤ k p , m a x := 2 a ∑ β a d a . (465)
The lower bound follows by summing 1 − cos ϑ 1-\cos\vartheta 1 − cos ϑ over the eigenphases of a unitary matrix. These are smooth gauge-invariant functions. Their plaquette form is a declared regulator of curvature energy; no convergence to the dynamical-coframe Cartan theory is inferred from this definition.
Fix f , m , q , θ > 0 f,m,q,\theta>0 f , m , q , θ > 0 , α σ , s 0 ∈ R \alpha_\sigma,s_0\in\mathbb{R} α σ , s 0 ∈ R , and nonnegative edge weights a i j a_{ij} a ij . The edge sum below counts each unordered scalar edge once:
S l a t ( U , s ) = ∑ p ∈ P e − α σ s v ( p ) k p ( U ) + f 2 2 ∑ { i , j } a i j ( s i − s j ) 2 + m 2 f 2 2 ∑ i ∈ V ( s i − s 0 ) 2 . (466) S_{\mathrm{lat}}(U,s)=\sum_{p\in\mathcal{P}}e^{-\alpha_\sigma s_{v(p)}}k_p(U)+\frac{f^2}{2}\sum_{\{i,j\}}a_{ij}(s_i-s_j)^2
+\frac{m^2f^2}{2}\sum_{i\in\mathcal{V}}(s_i-s_0)^2. \tag*{(466)} S lat ( U , s ) = p ∈ P ∑ e − α σ s v ( p ) k p ( U ) + 2 f 2 { i , j } ∑ a ij ( s i − s j ) 2 + 2 m 2 f 2 i ∈ V ∑ ( s i − s 0 ) 2 . (466)
The scalar weight and quadratic stabilizer are precisely the mechanisms in Section 5.4.1. Higgs and fermion coordinates are omitted from this bosonic realization; their measures cannot be deduced from the following construction.
Equip K E K^{\mathcal{E}} K E with a fixed product bi-invariant Riemannian metric. Write ∇ U , Δ U \nabla_U,\Delta_U ∇ U , Δ U for its gradient and Laplacian, with Δ U \Delta_U Δ U nonpositive in the L 2 L^2 L 2 pairing. The generator is
L l a t F = θ Δ U F − ⟨ ∇ U S l a t , ∇ U F ⟩ + q ( θ Δ s F − ∇ s S l a t ⋅ ∇ s F ) . (467) \mathcal{L}_{\mathrm{lat}}F=\theta\Delta_UF-\langle\nabla_US_{\mathrm{lat}},\nabla_UF\rangle+q(\theta\Delta_sF-\nabla_sS_{\mathrm{lat}}\cdot\nabla_sF). \tag*{(467)} L lat F = θ Δ U F − ⟨ ∇ U S lat , ∇ U F ⟩ + q ( θ Δ s F − ∇ s S lat ⋅ ∇ s F ) . (467)
This is an intrinsic generator. In local group coordinates its Itô drift includes the metric-volume term; it is not obtained by dropping the correction in (80).
Theorem 19 (Finite-regulator equilibrium). The diffusion with generator (468) has a unique nonexplosive realization from every initial point. Its unique invariant probability, among all probability measures on X l a t \mathcal{X}_{\mathrm{lat}} X lat , is
π l a t ( d U d s ) = Z l a t − 1 e − S l a t ( U , s ) / θ d U d s , 0 < Z l a t < ∞ . (468) \pi_{\mathrm{lat}}(dU\,ds)=Z_{\mathrm{lat}}^{-1}e^{-S_{\mathrm{lat}}(U,s)/\theta}\,dU\,ds,\qquad0<Z_{\mathrm{lat}}<\infty. \tag*{(468)} π lat ( d U d s ) = Z lat − 1 e − S lat ( U , s ) / θ d U d s , 0 < Z lat < ∞. (468) It is reversible. Every polynomial scalar moment and every fixed linear-exponential scalar moment is finite under this probability.
Proof. Put y i = s i − s 0 y_i=s_i-s_0 y i = s i − s 0 . Positivity of every summand gives
0 < Z l a t ≤ ∫ R n exp ( − m 2 f 2 2 θ ∣ y ∣ 2 ) d y = ( 2 π θ m 2 f 2 ) n / 2 . (469) 0<Z_{\mathrm{lat}}\leq\int_{\mathbb{R}^n}\exp\left(-\frac{m^2f^2}{2\theta}|y|^2\right)\,dy=\left(\frac{2\pi\theta}{m^2f^2}\right)^{n/2}. \tag*{(469)} 0 < Z lat ≤ ∫ R n exp ( − 2 θ m 2 f 2 ∣ y ∣ 2 ) d y = ( m 2 f 2 2 π θ ) n /2 . (469)
For strict positivity, restrict to a scalar cube of positive volume: the continuous action is bounded there because the group factor is compact, so the integrand has a strictly positive lower bound. Multiplication by ∣ y ∣ r |y|^r ∣ y ∣ r or e b ⋅ y e^{b\cdot y} e b ⋅ y preserves integrability of the bounding Gaussian. Division by Z l a t > 0 Z_{\mathrm{lat}}>0 Z lat > 0 proves the moment assertions.
Smooth coefficients give a pathwise unique local diffusion on this finite-dimensional manifold. To prove that its lifetime is infinite, use V ( U , s ) = 1 + ∣ y ∣ 2 V(U,s)=1+|y|^2 V ( U , s ) = 1 + ∣ y ∣ 2 . A direct differentiation yields
L l a t V = 2 q θ n − 2 q m 2 f 2 ∣ y ∣ 2 − 2 q f 2 ∑ { i , j } a i j ( y i − y j ) 2 + 2 q ∑ p α σ y v ( p ) e − α σ ( s 0 + y v ( p ) ) k p ( U ) ≤ A 0 − 2 q m 2 f 2 ∣ y ∣ 2 , (470) \mathcal{L}_{\mathrm{lat}}V=2q\theta n-2qm^2f^2|y|^2-2qf^2\sum_{\{i,j\}}a_{ij}(y_i-y_j)^2
+2q\sum_p\alpha_\sigma y_{v(p)}e^{-\alpha_\sigma(s_0+y_{v(p)})}k_p(U)\leq A_0-2qm^2f^2|y|^2, \tag*{(470)} L lat V = 2 q θ n − 2 q m 2 f 2 ∣ y ∣ 2 − 2 q f 2 { i , j } ∑ a ij ( y i − y j ) 2 + 2 q p ∑ α σ y v ( p ) e − α σ ( s 0 + y v ( p ) ) k p ( U ) ≤ A 0 − 2 q m 2 f 2 ∣ y ∣ 2 , (470)
where
A 0 = 2 q θ n + 2 q e − α σ s 0 e ∑ p k p , max . (471) A_0 = 2q\theta n + \frac{2q e^{-\alpha_\sigma s_0}}{e}\sum_p k_{p,\max}. \tag*{(471)} A 0 = 2 q θ n + e 2 q e − α σ s 0 p ∑ k p , m a x . (471)
Indeed, z e − z ≤ e − 1 ze^{-z} \le e^{-1} z e − z ≤ e − 1 for every real z z z : it is negative for z < 0 z < 0 z < 0 , and differentiation gives its maximum at z = 1 z = 1 z = 1 on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) . This also covers α σ = 0 \alpha_\sigma= 0 α σ = 0 , when the corresponding force is zero.
Let τ R = inf { t : ∣ y t ∣ ≥ R } \tau_R = \inf\{t : |y_t| \ge R\} τ R = inf { t : ∣ y t ∣ ≥ R } with R > ∣ y 0 ∣ R > |y_0| R > ∣ y 0 ∣ . Before τ R \tau_R τ R , coefficients and derivatives are bounded on a compact set, so the stopped stochastic integral has mean zero. Dynkin’s formula and (470) imply
E V ( U t ∧ τ R , s t ∧ τ R ) ≤ V ( U 0 , s 0 i n i t ) + A 0 t . (472) \mathbb{E}V(U_{t\wedge\tau_R},s_{t\wedge\tau_R}) \le V(U_0,s_0^{\mathrm{init}}) + A_0t. \tag*{(472)} E V ( U t ∧ τ R , s t ∧ τ R ) ≤ V ( U 0 , s 0 init ) + A 0 t . (472)
Here s 0 i n i t s_0^{\mathrm{init}} s 0 init denotes the initial vector, not the scalar center s 0 s_0 s 0 . Continuity gives V ≥ 1 + R 2 V \ge1 + R^2 V ≥ 1 + R 2 on { τ R ≤ t } \{\tau_R \le t\} { τ R ≤ t } , and therefore
P ( τ R ≤ t ) ≤ V ( U 0 , s 0 i n i t ) + A 0 t 1 + R 2 ⟶ 0. (473) \mathbb{P}(\tau_R \le t) \le\frac{V(U_0,s_0^{\mathrm{init}}) + A_0t}{1 + R^2} \longrightarrow0. \tag*{(473)} P ( τ R ≤ t ) ≤ 1 + R 2 V ( U 0 , s 0 init ) + A 0 t ⟶ 0. (473)
The compact group coordinates cannot escape. Thus the local lifetime is infinite for every initial point, which completes the existence and pathwise-uniqueness argument.
We next verify invariance without relying solely on a formal stationarity equation. For compactly supported smooth functions F , G F,G F , G , integration by parts on the group and Euclidean factors gives
∫ F L l a t G d π l a t = − θ ∫ [ ⟨ ∇ U F , ∇ U G ⟩ + q ∇ s F ⋅ ∇ s G ] d π l a t . (474) \int F\mathcal{L}_{\mathrm{lat}}G\,\mathrm{d}\pi_{\mathrm{lat}} = -\theta\int\left[\langle\nabla_U F,\nabla_U G\rangle+ q\nabla_s F\cdot\nabla_s G\right]\mathrm{d}\pi_{\mathrm{lat}}. \tag*{(474)} ∫ F L lat G d π lat = − θ ∫ [ ⟨ ∇ U F , ∇ U G ⟩ + q ∇ s F ⋅ ∇ s G ] d π lat . (474)
On the bounded smooth domain K E × { ∣ y ∣ < R } K^{\mathcal{E}}\times\{|y|<R\} K E × { ∣ y ∣ < R } , the killed diffusion has a symmetric Dirichlet transition semigroup P t ( R ) P_t^{(R)} P t ( R ) for this divergence-form operator. This follows from the Dirichlet integration-by-parts identity and uniqueness of its uniformly elliptic heat equation. For nonnegative bounded functions, the killed expectations increase to the full diffusion expectations as R → ∞ R\to\infty R → ∞ . Monotone convergence passes symmetry to P t P_t P t . Nonexplosion gives P t 1 = 1 P_t1=1 P t 1 = 1 . Taking one argument equal to 1 1 1 in the resulting semigroup identity proves
∫ P t F d π l a t = ∫ F P t 1 d π l a t = ∫ F d π l a t . (475) \int P_tF\,\mathrm{d}\pi_{\mathrm{lat}} = \int FP_t1\,\mathrm{d}\pi_{\mathrm{lat}} = \int F\,\mathrm{d}\pi_{\mathrm{lat}}. \tag*{(475)} ∫ P t F d π lat = ∫ F P t 1 d π lat = ∫ F d π lat . (475)
The extension from bounded nonnegative functions to bounded measurable functions follows by positive and negative parts. Thus the probability in (468) is invariant and reversible.
Finally, the manifold is connected and the diffusion matrix is positive definite. On every bounded domain its smooth transition density is strictly positive between interior points for positive time, by the strong maximum principle for the uniformly elliptic heat equation. Any two points lie in a common connected bounded domain. The full transition kernel consequently has a strictly positive density relative to π l a t \pi_{\mathrm{lat}} π lat ; local heat-kernel existence and this exhaustion are within the finite-dimensional diffusion theory of [148].
If η \eta η is another invariant probability, invariance and absolute continuity of each positive-time transition give η = h π l a t \eta=h\pi_{\mathrm{lat}} η = h π lat for some h ≥ 0 h\ge0 h ≥ 0 with ∫ h d π l a t = 1 \int h\,\mathrm{d}\pi_{\mathrm{lat}}=1 ∫ h d π lat = 1 . Semigroup symmetry extends to a nonnegative L 1 L^1 L 1 argument by applying it to truncations and then using monotone convergence. Reversibility implies P t h = h P_t h=h P t h = h almost everywhere: test this identity against a bounded function, use invariance of η \eta η , and interchange the semigroup by symmetry. For the strictly concave bounded function φ ( r ) = r / ( 1 + r ) \varphi(r)=r/(1+r) φ ( r ) = r / ( 1 + r ) on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) , Jensen’s inequality gives P t φ ( h ) ≤ φ ( P t h ) = φ ( h ) P_t\varphi(h)\le\varphi(P_t h)=\varphi(h) P t φ ( h ) ≤ φ ( P t h ) = φ ( h ) almost everywhere. Both sides have equal π l a t \pi_{\mathrm{lat}} π lat integral by invariance. Equality therefore holds almost everywhere. Equality in the strict Jensen inequality forces h h h to be constant almost surely under a positive-time transition from any point where equality holds. Such a transition is equivalent to π l a t \pi_{\mathrm{lat}} π lat , so h h h is constant π l a t \pi_{\mathrm{lat}} π lat -almost everywhere. Its integral is one, hence h = 1 h=1 h = 1 and η = π l a t \eta=\pi_{\mathrm{lat}} η = π lat . □
The lattice gauge group has finite Haar volume, so gauge invariance does not create the infinite group-volume obstruction of an unreduced continuum functional integral. Gauge-invariant observables may also be pushed to the finite quotient. The theorem supplies no estimates uniform in the number of lattice sites, no integration over a degenerating coframe, and no interacting fermion measure. Those are separate questions when this regulator is compared with the original continuum geometry.
A concrete four-dimensional interacting regulator and its thermodynamic reduction
The preceding construction is abstract until one declares a finite branch on which every integration variable, coefficient, and boundary condition is visible. This section supplies such a branch. Its purpose is deliberately limited: it gives a fully specified four-dimensional Euclidean regulator with interacting gauge, coframe, and order-parameter variables, and proves the finite-regulator thermodynamic theorem on that branch. It does not identify this regulator with the complete Lorentzian gravitational theory, and it does not claim a regulator-independent quantum limit.
Finite four-dimensional state space and mother-data registry
Let Λ h \Lambda_h Λ h be a finite periodic hypercubic lattice approximating the four-dimensional torus T 4 \mathbb{T}^4 T 4 . Write V h V_h V h , E h E_h E h , and P h P_h P h for its vertices, oriented links, and oriented plaquettes. The compact gauge group used at the finite regulator is
K = S O ( 5 ) × S U ( 3 ) × S U ( 2 ) × U ( 1 ) , X h = K E h × R 16 ∣ V h ∣ × R ∣ V h ∣ . (476) K=\mathrm{SO}(5)\times\mathrm{SU}(3)\times\mathrm{SU}(2)\times\mathrm{U}(1),\qquad\mathcal{X}_h=K^{E_h}\times\mathbb{R}^{16|V_h|}\times\mathbb{R}^{|V_h|}. \tag*{(476)} K = SO ( 5 ) × SU ( 3 ) × SU ( 2 ) × U ( 1 ) , X h = K E h × R 16∣ V h ∣ × R ∣ V h ∣ . (476)
The first factor contains link variables U ℓ U_\ell U ℓ . The second contains the four-by-four coframe coefficients e v e_v e v at each vertex, flattened into R 16 \mathbb{R}^{16} R 16 , and the last contains the scalar/order-parameter values σ v \sigma_v σ v . The coframe variables in this section are Euclidean regulator coordinates. The oriented nondegenerate sector and its Lorentzian continuation are treated as additional matching conditions below. In this particular finite model, e v e_v e v and σ v \sigma_v σ v are invariant under link gauge transformations. The differences e v − e w e_v-e_w e v − e w below are ordinary differences, not covariant differences of local Lorentz vectors. Thus the sixteen variables at a vertex are coordinates for a proposed geometric sector; identifying them with a gravitational coframe requires a further covariant construction. The finite thermal argument does not assume that identification.
All numerical coefficients are recorded before any evolution or comparison with data. The registry is
R h = ( β a , α , η , κ e , m e , λ e , f , m σ , θ , e ⋆ , σ ⋆ , Γ h ) , (477) \mathcal{R}_h=(\beta_a,\alpha,\eta,\kappa_e,m_e,\lambda_e,f,m_\sigma,\theta,e_\star,\sigma_\star,\Gamma_h), \tag*{(477)} R h = ( β a , α , η , κ e , m e , λ e , f , m σ , θ , e ⋆ , σ ⋆ , Γ h ) , (477)
where β a , κ e , m e , λ e , f , m σ , θ > 0 \beta_a,\kappa_e,m_e,\lambda_e,f,m_\sigma,\theta>0 β a , κ e , m e , λ e , f , m σ , θ > 0 , η , α ≥ 0 \eta,\alpha\geq0 η , α ≥ 0 , and Γ h \Gamma_h Γ h is a fixed positive mobility matrix. In the unified-field interpretation each entry of R h \mathcal{R}_h R h must be the value of a specified pairing, regulator, vacuum datum, or bath coupling in the mother data. A value selected after inspecting an observable is a fit and is not counted as a mother prediction. The graph, the positive symmetric edge weights a v w a_{vw} a v w , and the unitary representations R a R_a R a of dimensions d a d_a d a are also fixed data. Each geometric plaquette is assigned one orientation and one anchor vertex v ( p ) v(p) v ( p ) .
For a plaquette p p p let U p , a U_{p,a} U p , a be the ordered product in the a a a th group factor and define
k p ( U ) = ∑ a β a [ d a − Re tr R a ( U p , a ) ] ≥ 0. (478) k_p(U)=\sum_a\beta_a\left[d_a-\operatorname{Re}\operatorname{tr}R_a(U_{p,a})\right]\geq0. \tag*{(478)} k p ( U ) = a ∑ β a [ d a − Re tr R a ( U p , a ) ] ≥ 0. (478)
Let
w v ( e , σ ) = exp ( − α σ v ) [ 1 + η ∥ e v − e ⋆ ∥ 2 1 + ∥ e v − e ⋆ ∥ 2 ] . (479) w_v(e,\sigma)=\exp(-\alpha\sigma_v)\left[1+\eta\frac{\lVert e_v-e_\star\rVert^2}{1+\lVert e_v-e_\star\rVert^2}\right]. \tag*{(479)} w v ( e , σ ) = exp ( − α σ v ) [ 1 + η 1 + ∥ e v − e ⋆ ∥ 2 ∥ e v − e ⋆ ∥ 2 ] . (479)
The coframe factor lies in [ 1 , 1 + η ] [1,1+\eta] [ 1 , 1 + η ] . The full weight is positive but is unbounded as σ v → − ∞ \sigma_v\to-\infty σ v → − ∞ when α > 0 \alpha>0 α > 0 . Its multiplication by k p ≥ 0 k_p\geq0 k p ≥ 0 preserves the lower bound on the action; boundedness of the full weight is not needed for the partition integral. With each unoriented edge counted once, set
S h ( U , e , σ ) = ∑ p ∈ P h w v ( p ) ( e , σ ) k p ( U ) + κ e 2 ∑ { v , w } ∥ e v − e w ∥ 2 + m e 2 2 ∑ v ∥ e v − e ⋆ ∥ 2 + λ e 4 ∑ v ∥ e v − e ⋆ ∥ 4 + f 2 2 ∑ { v , w } a v w ( σ v − σ w ) 2 + m σ 2 f 2 2 ∑ v ( σ v − σ ⋆ ) 2 . (480) \begin{aligned}
S_h(U,e,\sigma)={}&\sum_{p\in P_h}w_{v(p)}(e,\sigma)k_p(U)+\frac{\kappa_e}{2}\sum_{\{v,w\}}\lVert e_v-e_w\rVert^2\\
&+\frac{m_e^2}{2}\sum_v\lVert e_v-e_\star\rVert^2+\frac{\lambda_e}{4}\sum_v\lVert e_v-e_\star\rVert^4\\
&+\frac{f^2}{2}\sum_{\{v,w\}}a_{vw}(\sigma_v-\sigma_w)^2+\frac{m_\sigma^2f^2}{2}\sum_v(\sigma_v-\sigma_\star)^2. \tag*{(480)}
\end{aligned} S h ( U , e , σ ) = p ∈ P h ∑ w v ( p ) ( e , σ ) k p ( U ) + 2 κ e { v , w } ∑ ∥ e v − e w ∥ 2 + 2 m e 2 v ∑ ∥ e v − e ⋆ ∥ 2 + 4 λ e v ∑ ∥ e v − e ⋆ ∥ 4 + 2 f 2 { v , w } ∑ a v w ( σ v − σ w ) 2 + 2 m σ 2 f 2 v ∑ ( σ v − σ ⋆ ) 2 . (480)
This is a specified finite candidate for matching to the positive norm-square branch of the extended curvature. The plaquette term is the gauge-curvature block, the coframe terms are the regulated geometric block, and the scalar terms are the order-parameter block. The common weight in (480) is the retained mixed interaction. Equality with the primitive mother norm is a matching condition on the selected pairing and regulator; it is not obtained merely by reusing the word “action”. If that matching condition is established, the displayed terms must all be accounted for by the same primitive pairing. The finite theorem below holds for the displayed action without presupposing that the matching has been proved.
Exact finite generator and Gibbs law
Equip K E h K^{E_h} K E h with the product bi-invariant metric and use the Euclidean metric on the coframe and scalar factors. Let ∇ h \nabla_h ∇ h and div h \operatorname{div}_h div h denote the corresponding gradient and divergence. For the fixed temperature parameter θ > 0 \theta> 0 θ > 0 define
L h F = θ div h ( Γ h ∇ h F ) − ⟨ Γ h ∇ h S h , ∇ h F ⟩ h , Γ h = Γ h T ≻ 0. (481) \mathcal{L}_h F = \theta\operatorname{div}_h(\Gamma_h \nabla_h F) - \langle\Gamma_h \nabla_h S_h, \nabla_h F \rangle_h,\qquad\Gamma_h = \Gamma_h^{T} \succ0. \tag*{(481)} L h F = θ div h ( Γ h ∇ h F ) − ⟨ Γ h ∇ h S h , ∇ h F ⟩ h , Γ h = Γ h T ≻ 0. (481)
The intrinsic form includes the metric-volume drift; writing an Itô equation in local coordinates without that drift would change the invariant measure. Here a fixed matrix on the group factors means its components in a fixed product frame of left-invariant and Euclidean vector fields. More generally, Γ h \Gamma_h Γ h may be a smooth symmetric bundle map satisfying γ − I ≤ Γ h ≤ γ + I \gamma_- I \le\Gamma_h \le\gamma_+ I γ − I ≤ Γ h ≤ γ + I for fixed positive constants. These are assumptions at a fixed regulator, not bounds uniform in h h h .
Gauge covariance imposes a further restriction on the mobility. Let Φ g \Phi_g Φ g be the site-gauge action on the links, acting trivially on the coframe and scalar coordinates of this particular model. In addition to the gauge-invariant action and reference volume, a covariant generator requires
Γ h ( Φ g x ) = D Φ g ( x ) Γ h ( x ) D Φ g ( x ) ∗ . (482) \Gamma_h(\Phi_g x) = D\Phi_g(x)\,\Gamma_h(x)\,D\Phi_g(x)^*. \tag*{(482)} Γ h ( Φ g x ) = D Φ g ( x ) Γ h ( x ) D Φ g ( x ) ∗ . (482)
where the adjoint uses the specified product metric. This condition makes L h ( F ∘ Φ g ) = ( L h F ) ∘ Φ g \mathcal{L}_h(F \circ\Phi_g) = (\mathcal{L}_h F) \circ\Phi_g L h ( F ∘ Φ g ) = ( L h F ) ∘ Φ g by invariance of the gradient and divergence. It is not implied by positive definiteness. For example, in an SU(2) link block, a left-frame matrix diag ( 1 , 2 , 3 ) \operatorname{diag}(1,2,3) diag ( 1 , 2 , 3 ) is positive but a right gauge rotation can exchange its first two principal axes. The resulting covariance is then diag ( 2 , 1 , 3 ) \operatorname{diag}(2,1,3) diag ( 2 , 1 , 3 ) , so the generator is not gauge covariant. A sufficient choice is a block-diagonal mobility with a fixed positive scalar multiple of the identity on each link-factor tangent block and an arbitrary fixed positive symmetric block on the Euclidean variables. In this sufficient choice, all mixed blocks involving a link factor vanish. The equilibrium theorem below holds for the full stated class of positive mobilities on the unreduced finite space; its gauge-covariant interpretation uses the subclass satisfying (482) . Invariance of a Gibbs law alone does not justify passage of the dynamics to gauge equivalence classes.
Theorem 20 (Finite four-dimensional interacting equilibrium). For every fixed finite lattice, every registry satisfying (478) and the fixed geometric data above, the diffusion generated by (482) is nonexplosive and has the unique invariant probability
π h ( d U d e d σ ) = Z h − 1 exp [ − S h ( U , e , σ ) / θ ] d U d e d σ , 0 < Z h < ∞ . (483) \pi_h(dU\,de\,d\sigma)=Z_h^{-1}\exp[-S_h(U,e,\sigma)/\theta]\,dU\,de\,d\sigma,\qquad0<Z_h<\infty. \tag*{(483)} π h ( d U d e d σ ) = Z h − 1 exp [ − S h ( U , e , σ ) / θ ] d U d e d σ , 0 < Z h < ∞. (483) It is reversible with respect to π h \pi_h π h , and all polynomial moments of the Euclidean variables are finite.
Proof. The compact gauge factor has finite Haar volume. Since k p ≥ 0 k_p\ge0 k p ≥ 0 and the quadratic and quartic coframe terms are nonnegative, the integrand in (483) is bounded above by a product of Gaussian–quartic integrable factors in ( e , σ ) (e,\sigma) ( e , σ ) . Restricting these variables to a compact box gives a strictly positive lower bound, so 0 < Z h < ∞ 0<Z_h<\infty 0 < Z h < ∞ . Smoothness and uniform ellipticity on each bounded coordinate domain give local existence and pathwise uniqueness. For nonexplosion, we use an energy argument that does not require an unproved comparison between exponential derivatives and polynomial confinement. Write z = ( e − e ⋆ , σ − σ ⋆ ) z=(e-e_\star,\sigma-\sigma_\star) z = ( e − e ⋆ , σ − σ ⋆ ) and choose smooth cutoffs χ R ( z ) = χ ( z / R ) \chi_R(z)=\chi(z/R) χ R ( z ) = χ ( z / R ) with 0 ≤ χ R ≤ 1 0\le\chi_R\le1 0 ≤ χ R ≤ 1 , χ R → 1 \chi_R\to1 χ R → 1 , and ∣ ∇ χ R ∣ ≤ C / R |\nabla\chi_R|\le C/R ∣∇ χ R ∣ ≤ C / R . The gradient form on L 2 ( π h ) L^2(\pi_h) L 2 ( π h ) is closable because its density is smooth and positive and its coefficients are locally uniformly elliptic. For its closure E h \mathcal{E}_h E h , finite total mass gives
∥ 1 − χ R ∥ L 2 ( π h ) → 0 , E h ( χ R , χ R ) ≤ θ γ + C 2 / R 2 → 0. (484) \|1-\chi_R\|_{L^2(\pi_h)}\to0,\qquad\mathcal{E}_h(\chi_R,\chi_R)\le\theta\gamma+C^2/R^2\to0. \tag*{(484)} ∥1 − χ R ∥ L 2 ( π h ) → 0 , E h ( χ R , χ R ) ≤ θ γ + C 2 / R 2 → 0. (484)
Consequently 1 belongs to the closed form domain with zero energy. The associated symmetric semigroup preserves 1, so its diffusion has no killing or explosion for π h \pi_h π h -almost every initial state. Local smooth ellipticity upgrades this to every starting point. Indeed, the minimal diffusion has an absolutely continuous transition law before its lifetime; the Markov property at time ε > 0 \varepsilon>0 ε > 0 then rules out an explosion after ε \varepsilon ε , and local existence rules out explosion at time zero. The local strong solution therefore extends globally. This is the conservative-form criterion of [75]. Integration by parts gives
∫ F L h G d π h = − θ ∫ ⟨ ∇ h F , Γ h ∇ h G ⟩ h d π h , (485) \int F L_hG\,d\pi_h=-\theta\int\langle\nabla_hF,\Gamma_h\nabla_hG\rangle_h\,d\pi_h, \tag*{(485)} ∫ F L h G d π h = − θ ∫ ⟨ ∇ h F , Γ h ∇ h G ⟩ h d π h , (485)
which proves reversibility and invariance. Strict positivity of the finite dimensional heat kernel and connectedness imply uniqueness by the standard irreducibility argument. □
Second-law identity on the declared branch
For a law ρ t ≪ π h \rho_t\ll\pi_h ρ t ≪ π h , let r t = d ρ t / d μ h r_t=d\rho_t/d\mu_h r t = d ρ t / d μ h and f t = d ρ t / d π h f_t=d\rho_t/d\pi_h f t = d ρ t / d π h , where μ h = d U d e d σ \mu_h=dU\,de\,d\sigma μ h = d U d e d σ . Define F h [ ρ ] = ∫ S h d ρ + θ ∫ log r d ρ F_h[\rho]=\int S_h\,d\rho+\theta\int\log r\,d\rho F h [ ρ ] = ∫ S h d ρ + θ ∫ log r d ρ . Then
Table 3. Synthetic finite-regulator quadrature check. The numbers are not experimental data and are not a continuum test.
Theorem 21 (Finite-regulator entropy production). For every smooth positive solution of the Fokker–Planck equation generated by (482), with finite Fisher information,
F h [ ρ t ] − F h [ π h ] = θ D ( ρ t ∥ π h ) , (486) F_h[\rho_t]-F_h[\pi_h]=\theta D(\rho_t\Vert\pi_h), \tag*{(486)} F h [ ρ t ] − F h [ π h ] = θ D ( ρ t ∥ π h ) , (486) d d t D ( ρ t ∥ π h ) = − θ ∫ r t ( ∇ h log ρ t π h ) T Γ h ( ∇ h log ρ t π h ) d μ h ≤ 0. (487) \frac{\mathrm{d}}{\mathrm{d}t}D(\rho_t\Vert\pi_h)=-\theta\int r_t\left(\nabla_h\log\frac{\rho_t}{\pi_h}\right)^{\mathsf T}\Gamma_h\left(\nabla_h\log\frac{\rho_t}{\pi_h}\right)\,\mathrm{d}\mu_h\le0. \tag*{(487)} d t d D ( ρ t ∥ π h ) = − θ ∫ r t ( ∇ h log π h ρ t ) T Γ h ( ∇ h log π h ρ t ) d μ h ≤ 0. (487) For a physical heat interpretation, additionally assume θ = k B T \theta=k_{\mathrm B}T θ = k B T and that S h S_h S h is the effective energy in fixed units. The Euclidean regulator and its sampling parameter do not by themselves specify a physical bath or physical time. If the bath coupling obeys local detailed balance and the heat convention is Q ˙ b a t h = − d E [ S h ] / d t \dot{Q}_{\mathrm{bath}}=-\mathrm{d}E[S_h]/\mathrm{d}t Q ˙ bath = − d E [ S h ] / d t , then
S ˙ t o t = − T − 1 d F h [ ρ t ] d t ≥ 0. (488) \dot{S}_{\mathrm{tot}}=-T^{-1}\frac{\mathrm{d}F_h[\rho_t]}{\mathrm{d}t}\ge0. \tag*{(488)} S ˙ tot = − T − 1 d t d F h [ ρ t ] ≥ 0. (488)
Proof. (485) gives the symmetric Dirichlet form. Applying it to f t = ρ t / π h f_t=\rho_t/\pi_h f t = ρ t / π h yields the second line of (488); positive definiteness of Γ h \Gamma_h Γ h gives the sign. The first line is the exact Gibbs identity and follows by expanding log ( d ρ t / d π h ) \log(\mathrm{d}\rho_t/\mathrm{d}\pi_h) log ( d ρ t / d π h ) . The final statement uses the first-law balance and local detailed balance; it is not inferred from Gibbs invariance alone.
Reproducible finite diagnostic
As a numerical check of the analytic identity, we evaluated the one-plaquette U(1) subbranch with one coframe and one scalar coordinate. The potential was the restriction of (481) with ( θ , β , α , η , m 2 , λ e , m 2 0 ) = ( 0.8 , 1 , 0.3 , 0.2 , 1 , 0.1 , 1 ) (\theta,\beta,\alpha,\eta,m_2,\lambda_e,m_2^0)=(0.8,1,0.3,0.2,1,0.1,1) ( θ , β , α , η , m 2 , λ e , m 2 0 ) = ( 0.8 , 1 , 0.3 , 0.2 , 1 , 0.1 , 1 ) and Γ = diag ( 0.7 , 0.5 , 0.4 ) \Gamma=\operatorname{diag}(0.7,0.5,0.4) Γ = diag ( 0.7 , 0.5 , 0.4 ) . A tensor-product quadrature on 64 × 161 × 161 64\times161\times161 64 × 161 × 161 points was applied to a smooth test function g ( ϑ , e , σ ) = sin ϑ + 0.2 e 2 + 0.1 σ e g(\vartheta,e,\sigma)=\sin\vartheta+0.2e^2+0.1\sigma e g ( ϑ , e , σ ) = sin ϑ + 0.2 e 2 + 0.1 σ e . The two sides of the Dirichlet identity were computed independently: The angular reference measure in this table is d ϑ \mathrm{d}\vartheta d ϑ with total mass 2 π 2\pi 2 π ; normalizing Haar measure changes Z h Z_h Z h but not either expectation. The discrepancy is a floating-point residual, not a certified bound on quadrature or truncation error. The calculation checks the finite integration-by-parts identity only; it supplies no evidence for a regulator-independent or quantum theory.
Relation to Lorentzian and continuum dynamics
The use of polynomially confined matrix coordinates leaves a geometric integrability issue even at fixed resolution. A matrix can be invertible almost surely while its inverse has no finite first moment.
Proposition 22 (Inverse-coframe moments in the finite reference law). For the Gibbs law (484), at every vertex v v v and for every r ≥ 1 r \ge1 r ≥ 1 ,
E π h ∣ det e v ∣ − r = ∞ , E π h ∥ e v − 1 ∥ F r = ∞ . \mathbb{E}_{\pi_h}\lvert\det e_v\rvert^{-r}=\infty,\qquad\mathbb{E}_{\pi_h}\lVert e_v^{-1}\rVert_{\mathrm{F}}^r=\infty. E π h ∣ det e v ∣ − r = ∞ , E π h ∥ e v − 1 ∥ F r = ∞. These statements persist on conditioning on det e v > 0 \det e_v>0 det e v > 0 .
Proof. Choose a rank-three matrix E 0 E_0 E 0 . At least one cofactor is nonzero, so the determinant has nonzero differential there. The inverse function theorem gives local coordinates ( u , y ) ∈ R × R 15 (u,y) \in\mathbb{R} \times\mathbb{R}^{15} ( u , y ) ∈ R × R 15 with u = det e v u=\det e_v u = det e v , with a smooth nonzero change-of-volume factor. Restrict y y y and all remaining noncompact variables to small compact boxes and the group variables to a set of positive Haar measure. The Gibbs density is continuous and strictly positive on this compact set, including u = 0 u=0 u = 0 . Its contribution to the first expectation is therefore bounded below by c ∫ − ε ε ∣ u ∣ − r d u = ∞ c\int_{-\varepsilon}^{\varepsilon}|u|^{-r}\,du=\infty c ∫ − ε ε ∣ u ∣ − r d u = ∞ . The selected cofactor remains bounded away from zero on a smaller box. The inverse-matrix formula consequently bounds one inverse entry below in magnitude by c ′ / ∣ u ∣ c'/|u| c ′ /∣ u ∣ , proving the second assertion. On the positive orientation sector the same argument uses ∫ 0 ε u − r d u \int_0^\varepsilon u^{-r}\,du ∫ 0 ε u − r d u ; conditioning divides by a finite positive probability and cannot remove the divergence. □
The singular matrices themselves have zero probability. The divergence comes from their neighborhoods and is compatible with finiteness of all polynomial moments proved above. Interpreting this reference law as a measure on fluctuating nondegenerate geometry therefore requires additional control near singular coframes. A change of measure or action for that purpose must be matched to the mother data and its quantum positivity checked separately.
On this fixed four-dimensional lattice, the same declared action supplies the gauge-curvature energy, the coframe/order-parameter coupling, the Gibbs law, and the entropy-production identity. It is not a proof of the full unified theory. In particular, the following implications remain to be established before a continuum or quantum claim is made:
the oriented nondegenerate coframe sector and Lorentzian continuation must be selected without changing the registry;
the nonlinear Hamiltonian and gauge constraints must close and be preserved for all lattice modes;
estimates uniform in h h h must give a regulator-independent limit;
a physical interacting Hilbert space, positive inner product, and anomaly-free quantum constraint algebra must be constructed.
The finite theorem connects the declared energy, probability law, and dissipation on the same state space. The next section examines which additional structures can be constructed explicitly and which require further matching to the gravitational theory.
A discrete finite four-dimensional certification branch
The continuous compact-group branch introduced above and the discrete branch below serve different purposes. The former carries the coframe and compact gauge variables used for the proposed geometric interface. The present Z 2 \mathbb{Z}_2 Z 2 branch is a smaller, completely enumerable certification model. It supplies an exact finite state space, an explicit stochastic kernel, and an independently reproducible no-refit check. No identification of the two regulators, and no inference about a continuum limit, is made.
The preceding structural discussion becomes mathematically testable only after one model, one state space, and one parameter registry have been fixed. We therefore give a complete finite example. The example is not introduced as a continuum limit in disguise: every sum below is finite, and every conclusion is stated at the regulator level.
Lattice, fields, and gauge constraint
Fix the periodic lattice
Λ 3 = ( Z / 3 Z ) 4 , ∣ Λ 3 ∣ = 81 , \Lambda_3 = (\mathbb{Z}/3\mathbb{Z})^4,\qquad|\Lambda_3| = 81, Λ 3 = ( Z /3 Z ) 4 , ∣ Λ 3 ∣ = 81 ,
with four positive coordinate directions labelled by μ ∈ { 0 , 1 , 2 , 3 } \mu\in\{0,1,2,3\} μ ∈ { 0 , 1 , 2 , 3 } . There are therefore 324 oriented links. A raw configuration is
ω = ( U , σ , τ ) , U x , μ , σ x , τ x ∈ { + 1 , − 1 } . \omega= (U,\sigma,\tau),\qquad U_{x,\mu},\sigma_x,\tau_x \in\{+1,-1\}. ω = ( U , σ , τ ) , U x , μ , σ x , τ x ∈ { + 1 , − 1 } .
The finite gauge group is G = { η : Λ 3 → { + 1 , − 1 } } \mathcal{G} = \{\eta:\Lambda_3 \to\{+1,-1\}\} G = { η : Λ 3 → { + 1 , − 1 }} , acting by
U x , μ ↦ η x U x , μ η x + μ ^ , σ x ↦ η x σ x , τ x ↦ η x τ x . U_{x,\mu} \mapsto\eta_x U_{x,\mu}\eta_{x+\hat{\mu}},\qquad\sigma_x \mapsto\eta_x\sigma_x,\qquad\tau_x \mapsto\eta_x\tau_x. U x , μ ↦ η x U x , μ η x + μ ^ , σ x ↦ η x σ x , τ x ↦ η x τ x .
The physical finite state space is the orbit quotient
Ω p h y s = Ω / G . \Omega_{\mathrm{phys}} = \Omega/\mathcal{G}. Ω phys = Ω/ G .
The quotient is finite and nonempty because the raw state space is finite and nonempty. All observables used below are orbit invariant and hence are well-defined on this quotient.
Gauge-invariant action
Define the three sufficient statistics
P ( U ) = ∑ x ∈ Λ 3 ∑ 0 ≤ μ < ν ≤ 3 U x , μ U x + μ ^ , ν U x + ν ^ , μ U x , ν , (489) P(U) = \sum_{x\in\Lambda_3}\sum_{0\leq\mu<\nu\leq3} U_{x,\mu}U_{x+\hat{\mu},\nu}U_{x+\hat{\nu},\mu}U_{x,\nu}, \tag*{(489)} P ( U ) = x ∈ Λ 3 ∑ 0 ≤ μ < ν ≤ 3 ∑ U x , μ U x + μ ^ , ν U x + ν ^ , μ U x , ν , (489)
H ( U , σ ) = ∑ x ∈ Λ 3 ∑ μ = 0 3 σ x U x , μ σ x + μ ^ , (490) H(U,\sigma) = \sum_{x\in\Lambda_3}\sum_{\mu=0}^{3}\sigma_x U_{x,\mu}\sigma_{x+\hat{\mu}}, \tag*{(490)} H ( U , σ ) = x ∈ Λ 3 ∑ μ = 0 ∑ 3 σ x U x , μ σ x + μ ^ , (490)
R ( σ , τ ) = ∑ x ∈ Λ 3 σ x τ x . (491) R(\sigma,\tau)=\sum_{x\in\Lambda_3}\sigma_x\tau_x. \tag*{(491)} R ( σ , τ ) = x ∈ Λ 3 ∑ σ x τ x . (491)
Each factor of η \eta η appears twice in every summand of (490) and (491), and the two factors cancel because η x 2 = 1 \eta_x^2=1 η x 2 = 1 . The same cancellation is immediate in (492). Thus P P P , H H H , R R R and the action
S β , κ , ρ ( ω ) = − β P ( U ) − κ H ( U , σ ) − ρ R ( σ , τ ) S_{\beta,\kappa,\rho}(\omega)=-\beta P(U)-\kappa H(U,\sigma)-\rho R(\sigma,\tau) S β , κ , ρ ( ω ) = − β P ( U ) − κ H ( U , σ ) − ρR ( σ , τ )
are constant on gauge orbits. The Lean quotient construction in the finite gauge–Higgs verification module formalizes precisely the descent step: an orbit-invariant function is lifted to a function on Ω p h y s \Omega_{\mathrm{phys}} Ω phys .
Finite Gibbs law
For θ = ( β , κ , ρ ) \theta=(\beta,\kappa,\rho) θ = ( β , κ , ρ ) , define
Z θ = ∑ ω ∈ Ω p h y s exp [ − S θ ( ω ) ] , (492) Z_\theta=\sum_{\omega\in\Omega_{\mathrm{phys}}}\exp[-S_\theta(\omega)], \tag*{(492)} Z θ = ω ∈ Ω phys ∑ exp [ − S θ ( ω )] , (492)
π θ ( ω ) = Z θ − 1 exp [ − S θ ( ω ) ] . (493) \pi_\theta(\omega)=Z_\theta^{-1}\exp[-S_\theta(\omega)]. \tag*{(493)} π θ ( ω ) = Z θ − 1 exp [ − S θ ( ω )] . (493)
The quotient is finite and nonempty, and every summand in Z θ Z_\theta Z θ is strictly positive. Consequently,
Z θ > 0 , π θ ( ω ) > 0 , ∑ ω ∈ Ω p h y s π θ ( ω ) = 1. Z_\theta>0,\qquad\pi_\theta(\omega)>0,\qquad\sum_{\omega\in\Omega_{\mathrm{phys}}}\pi_\theta(\omega)=1. Z θ > 0 , π θ ( ω ) > 0 , ω ∈ Ω phys ∑ π θ ( ω ) = 1.
These statements are proved by the finite Gibbs lemmas in the Lean module; they do not use a thermodynamic limit or an integrability estimate.
Reversible finite dynamics and entropy balance
Let q ( i , j ) ≥ 0 q(i,j)\geq0 q ( i , j ) ≥ 0 be a symmetric proposal weight on Ω p h y s \Omega_{\mathrm{phys}} Ω phys . Set
w θ ( i , j ) = q ( i , j ) exp ( S θ ( i ) − S θ ( j ) 2 ) . w_\theta(i,j)=q(i,j)\exp\left(\frac{S_\theta(i)-S_\theta(j)}{2}\right). w θ ( i , j ) = q ( i , j ) exp ( 2 S θ ( i ) − S θ ( j ) ) .
The rate is nonnegative and satisfies detailed balance by direct algebra:
exp [ − S θ ( i ) ] w θ ( i , j ) = q ( i , j ) exp [ − S θ ( i ) + S θ ( j ) 2 ] (494) \exp[-S_\theta(i)]w_\theta(i,j)=q(i,j)\exp\left[-\frac{S_\theta(i)+S_\theta(j)}{2}\right] \tag*{(494)} exp [ − S θ ( i )] w θ ( i , j ) = q ( i , j ) exp [ − 2 S θ ( i ) + S θ ( j ) ] (494)
= exp [ − S θ ( j ) ] w θ ( j , i ) . (495) =\exp[-S_\theta(j)]w_\theta(j,i). \tag*{(495)} = exp [ − S θ ( j )] w θ ( j , i ) . (495)
The common factor Z θ − 1 Z_\theta^{-1} Z θ − 1 therefore gives the same identity for π θ \pi_\theta π θ . A probability vector p t p_t p t evolves according to
p ˙ i ( t ) = ∑ j [ p j ( t ) w θ ( j , i ) − p i ( t ) w θ ( i , j ) ] . \dot{p}_i(t)=\sum_j\left[p_j(t)w_\theta(j,i)-p_i(t)w_\theta(i,j)\right]. p ˙ i ( t ) = j ∑ [ p j ( t ) w θ ( j , i ) − p i ( t ) w θ ( i , j ) ] .
For the relative entropy
H ( p t ∣ π θ ) = ∑ i p i ( t ) log p i ( t ) π θ ( i ) , \mathcal{H}(p_t\mid\pi_\theta)=\sum_i p_i(t)\log\frac{p_i(t)}{\pi_\theta(i)}, H ( p t ∣ π θ ) = i ∑ p i ( t ) log π θ ( i ) p i ( t ) ,
the finite-state calculation gives
d d t H ( p t ∣ π θ ) = − 1 2 ∑ i , j ( p i w i j − p j w j i ) log p i w i j p j w j i (496) \frac{\mathrm{d}}{\mathrm{d}t}\mathcal{H}(p_t \mid\pi_\theta) = -\frac{1}{2}\sum_{i,j}(p_iw_{ij}-p_jw_{ji})\log\frac{p_iw_{ij}}{p_jw_{ji}} \tag*{(496)} d t d H ( p t ∣ π θ ) = − 2 1 i , j ∑ ( p i w ij − p j w j i ) log p j w j i p i w ij (496)
= − D θ ( p t ) ≤ 0 , (497) =-D_\theta(p_t)\le0, \tag*{(497)} = − D θ ( p t ) ≤ 0 , (497)
where D θ ≥ 0 D_\theta\ge0 D θ ≥ 0 follows from ( a − b ) ( log a − log b ) ≥ 0 (a-b)(\log a-\log b)\ge0 ( a − b ) ( log a − log b ) ≥ 0 for positive a , b a,b a , b . The Lean theorem for the finite entropy balance makes the dependence on the master-equation derivative, positivity, normalization, and symmetric-support hypotheses explicit. Thus the entropy sign is derived from the finite dynamics; it is not inserted as an independent axiom.
Parameter registration
Let c 0 c_0 c 0 be the all-plus configuration. Let c 1 c_1 c 1 flip one link, c 2 c_2 c 2 flip one σ \sigma σ -spin, and c 3 c_3 c 3 flip one τ \tau τ -spin. An independent lattice enumeration gives
Δ ( P , H , R ) c 1 = ( − 12 , − 2 , 0 ) , (498) \Delta(P,H,R)_{c_1}=(-12,-2,0), \tag*{(498)} Δ ( P , H , R ) c 1 = ( − 12 , − 2 , 0 ) , (498)
Δ ( P , H , R ) c 2 = ( 0 , − 16 , − 2 ) , (499) \Delta(P,H,R)_{c_2}=(0,-16,-2), \tag*{(499)} Δ ( P , H , R ) c 2 = ( 0 , − 16 , − 2 ) , (499)
Δ ( P , H , R ) c 3 = ( 0 , 0 , − 2 ) . (500) \Delta(P,H,R)_{c_3}=(0,0,-2). \tag*{(500)} Δ ( P , H , R ) c 3 = ( 0 , 0 , − 2 ) . (500)
The first number follows because one link belongs to six plaquettes in four dimensions and each affected plaquette changes by − 2 -2 − 2 . The second follows because one site belongs to eight oriented links. The third is the single on-site cross-field factor. The Python checker records these counts and rejects the registry if any one differs.
Because the partition function cancels in a probability ratio,
y i : = log π θ ( c i ) π θ ( c 0 ) y_i:=\log\frac{\pi_\theta(c_i)}{\pi_\theta(c_0)} y i := log π θ ( c 0 ) π θ ( c i )
satisfies
( y 1 y 2 y 3 ) = ( − 12 − 2 0 0 − 16 − 2 0 0 − 2 ) ( β κ ρ ) . \begin{pmatrix} y_1 \\ y_2 \\ y_3 \end{pmatrix}
=
\begin{pmatrix}
-12 & -2 & 0 \\
0 & -16 & -2 \\
0 & 0 & -2
\end{pmatrix}
\begin{pmatrix} \beta\\ \kappa\\ \rho\end{pmatrix}. y 1 y 2 y 3 = − 12 0 0 − 2 − 16 0 0 − 2 − 2 β κ ρ .
The determinant is − 384 -384 − 384 , so the registry is injective. Explicitly,
ρ = − y 3 2 , κ = − y 2 + 2 ρ 16 , β = − y 1 + 2 κ 12 . \rho=-\frac{y_3}{2},\qquad\kappa=-\frac{y_2+2\rho}{16},\qquad\beta=-\frac{y_1+2\kappa}{12}. ρ = − 2 y 3 , κ = − 16 y 2 + 2 ρ , β = − 12 y 1 + 2 κ .
The Lean declarations registry_injective and registry_inverse prove these identities over R \mathbb{R} R .
Frozen no-refit prediction
In an empirical application, a calibration split would be used once to estimate the three ratios in (9.5); after solving for θ ^ \hat{\theta} θ ^ , the parameters would be frozen. In the finite check reported below, the ratios are supplied by the exact registry construction, so no statistical calibration sample is used. For a held-out observable O O O , the prediction is
E ^ [ O ] = ∑ ω ∈ Ω p h y s O ( ω ) exp [ − S θ ^ ( ω ) ] Z θ ^ . \widehat{\mathbb{E}}[O] = \frac{\sum_{\omega\in\Omega_{\mathrm{phys}}} O(\omega)\exp[-S_{\widehat{\theta}}(\omega)]}{Z_{\widehat{\theta}}}. E [ O ] = Z θ ∑ ω ∈ Ω phys O ( ω ) exp [ − S θ ( ω )] .
The primary declared observable is the plaquette mean P / 486 P/486 P /486 ; the matter-link and cross-field means are secondary observables. The Lean registry theorems prove that equal registered ratios imply equal parameters and hence equality of the resulting finite expectations. The numerical report must still provide Monte–Carlo uncertainty and held-out error; the exact finite theorem alone is not an experimental confirmation. In the numerical check below, “held-out” denotes an independent simulation chain, not an independent physical data set.
9.7. Uniformization, nonnegativity, and normalization. The entropy identity above was initially stated for a strictly positive, normalized solution of the master equation. The finite construction below first establishes a nonnegative, normalized law from the jump rates. Let
r i = ∑ j q ( i , j ) , K i j = δ i j ( 1 − r i λ ) + ( 1 − δ i j ) q ( i , j ) λ . r_i = \sum_j q(i,j), \qquad K_{ij} = \delta_{ij}\left(1-\frac{r_i}{\lambda}\right) + (1-\delta_{ij})\frac{q(i,j)}{\lambda}. r i = j ∑ q ( i , j ) , K ij = δ ij ( 1 − λ r i ) + ( 1 − δ ij ) λ q ( i , j ) .
where λ > 0 \lambda> 0 λ > 0 , q ( i , j ) ≥ 0 q(i,j) \ge0 q ( i , j ) ≥ 0 , q ( i , i ) = 0 q(i,i) = 0 q ( i , i ) = 0 , and r i ≤ λ r_i \le\lambda r i ≤ λ . The finite generator is
L i j = { − r i , i = j , q ( i , j ) , i ≠ j , L = λ ( K − I ) . L_{ij} =
\begin{cases}
-r_i, & i=j,\\
q(i,j), & i\ne j,
\end{cases}
\qquad L = \lambda(K-I). L ij = { − r i , q ( i , j ) , i = j , i = j , L = λ ( K − I ) .
The verified generator–uniformization bridge proves
K i j ≥ 0 , ∑ j K i j = 1 , λ ( K i j − δ i j ) = L i j . K_{ij} \ge0,\qquad\sum_j K_{ij}=1,\qquad\lambda(K_{ij}-\delta_{ij})=L_{ij}. K ij ≥ 0 , j ∑ K ij = 1 , λ ( K ij − δ ij ) = L ij .
For an initial probability vector p 0 p_0 p 0 and t ≥ 0 t \ge0 t ≥ 0 , define the Poisson-uniformized law, with a = λ t ≥ 0 a=\lambda t \ge0 a = λ t ≥ 0 , by
p t ( j ) = ∑ n = 0 ∞ e − a a n n ! ( p 0 K n ) j . p_t(j) = \sum_{n=0}^{\infty} e^{-a}\frac{a^n}{n!}(p_0K^n)_j. p t ( j ) = n = 0 ∑ ∞ e − a n ! a n ( p 0 K n ) j .
The Lean module ‘UniformizedFiniteMarkov20261118‘ proves termwise nonnegativity, summability, and
p t ( j ) ≥ 0 , ∑ j p t ( j ) = 1. p_t(j) \ge0,\qquad\sum_j p_t(j)=1. p t ( j ) ≥ 0 , j ∑ p t ( j ) = 1.
The remaining analytic bridge is termwise differentiation of (9.7), which yields p ˙ t = p t L \dot{p}_t=p_tL p ˙ t = p t L (or the transpose convention for column vectors). It is a finite-series estimate and is kept separate from the completed positivity and mass proof. The present formal result therefore establishes nonnegativity and unit mass, but does not by itself establish strict positivity. Strict positivity for positive times requires an irreducibility condition together with a full-support initial law; alternatively, the entropy identity can be stated with the standard convention 0 log 0 = 0 0 \log0 = 0 0 log 0 = 0 .
Independent frozen-parameter check
The registry equations were solved once before either Monte–Carlo chain was inspected. We used
( β , κ , ρ ) = ( 0.08 , 0.06 , 0.04 ) , ( y 1 , y 2 , y 3 ) = ( − 1.08 , − 1.04 , − 0.08 ) , (\beta,\kappa,\rho)=(0.08,0.06,0.04), \qquad(y_1,y_2,y_3)=(-1.08,-1.04,-0.08), ( β , κ , ρ ) = ( 0.08 , 0.06 , 0.04 ) , ( y 1 , y 2 , y 3 ) = ( − 1.08 , − 1.04 , − 0.08 ) ,
and froze this parameter vector. Two independent chains, with seeds 2026111801 and 2026111802, each retained 12000 samples after 3000 burn-in steps and thinning by 10. The differences between the chains were
Table 7.
All three intervals contain zero. This verifies reproducibility of the frozen finite regulator at the reported Monte–Carlo precision. It is not an experimental confirmation: no public physical observable has yet been mapped to the three discrete lattice statistics.
Limits of the result
This section closes one finite interacting regulator. It does not prove that the family has a regulator-independent continuum limit, that the infinite volume limit is unique, or that the construction yields a positive physical Hilbert space for interacting Lorentzian gravity. Those are separate claims requiring uniform estimates and a reconstruction theorem. The present model is valuable precisely because every finite statement is explicit, auditable, and falsifiable before those harder limits are attempted.
Identifiable finite models, constrained motion, and positive quantum evolution
A common variational structure becomes predictive only after its parameters, admissible motions, and observation rules have been fixed. Recovering coefficients from responses does not select those coefficients from first principles. Preserving a constraint does not identify it with a gravitational constraint. Likewise, constructing positive quantum evolution at fixed spatial resolution leaves a separate problem of removing that resolution. We develop explicit constructions that provide a starting point for these three distinct extensions.
What the finite responses determine
Fix the graph, its edge weights and representations, the reference fields e ⋆ , σ ⋆ e_{\star},\sigma_{\star} e ⋆ , σ ⋆ , and units of field amplitude, energy, and time. Consider (477) with constant mobility in the specified product frame. We ask whether different coefficients can give the same complete responses. This inverse problem within a selected model differs from whether primitive mother data select one model and one registry.
An actual configuration variation avoids treating plaquette energies as independent coordinates. Choose a link ℓ \ell ℓ contained in n ℓ > 0 n_{\ell} > 0 n ℓ > 0 distinct plaquettes and X a X_a X a in one Lie-algebra factor, with c a = tr [ d R a ( X a ) † d R a ( X a ) ] > 0 c_a = \operatorname{tr}[dR_a(X_a)^{\dagger}dR_a(X_a)] > 0 c a = tr [ d R a ( X a ) † d R a ( X a )] > 0 . Vary U ℓ , a ( t ) = exp ( t X a ) U_{\ell,a}(t) = \exp(tX_a) U ℓ , a ( t ) = exp ( t X a ) , set all other links to the identity, and take constant fields σ = s \sigma= s σ = s and ∥ e − e ⋆ ∥ 2 / ( 1 + ∥ e − e ⋆ ∥ 2 ) = r \lVert e-e_{\star}\rVert^2/(1+\lVert e-e_{\star}\rVert^2) = r ∥ e − e ⋆ ∥ 2 / ( 1 + ∥ e − e ⋆ ∥ 2 ) = r . Then
Q a ( r , s ) : = 1 n ℓ c a d 2 d t 2 S h ( U ( t ) , e , s ) ∣ t = 0 = β a e − α s ( 1 + η r ) . (501) Q_a(r,s) := \left.\frac{1}{n_{\ell}c_a}\frac{\mathrm{d}^2}{\mathrm{d}t^2}S_h(U(t),e,s)\right|_{t=0} = \beta_a e^{-\alpha s}(1+\eta r). \tag*{(501)} Q a ( r , s ) := n ℓ c a 1 d t 2 d 2 S h ( U ( t ) , e , s ) t = 0 = β a e − α s ( 1 + η r ) . (501)
Indeed, either orientation of an incident plaquette gives d a − Re tr R a ( e ± t X a ) = c a t 2 / 2 + O ( t 4 ) d_a-\operatorname{Re}\operatorname{tr}R_a(e^{\pm tX_a}) = c_a t^2/2+O(t^4) d a − Re tr R a ( e ± t X a ) = c a t 2 /2 + O ( t 4 ) ; the remaining energy terms are independent of t t t . Consequently
β a = Q a ( 0 , 0 ) , α = − ∂ s log Q a ( 0 , s ) ∣ s = 0 , η = ∂ r log Q a ( r , 0 ) ∣ r = 0 + . (502) \beta_a = Q_a(0,0), \qquad\alpha= -\left.\partial_s\log Q_a(0,s)\right|_{s=0}, \qquad\eta= \left.\partial_r\log Q_a(r,0)\right|_{r=0+}. \tag*{(502)} β a = Q a ( 0 , 0 ) , α = − ∂ s log Q a ( 0 , s ) ∣ s = 0 , η = ∂ r log Q a ( r , 0 ) ∣ r = 0 + . (502)
The last derivative can equivalently be obtained from a unit-vector coframe variation: ∂ t 2 Q a ( r ( e ⋆ + t v ) , 0 ) ∣ t = 0 = 2 β a η \left.\partial_t^2 Q_a(r(e_{\star}+tv),0)\right|_{t=0}=2\beta_a\eta ∂ t 2 Q a ( r ( e ⋆ + t v ) , 0 ) t = 0 = 2 β a η . No extension to negative r r r is required.
At identity links and reference fields, let L 0 , L a L_0,L_a L 0 , L a be the unweighted and weighted graph Laplacians, with each edge counted once. The noncompact Hessians and an onsite fourth derivative are
H e = ( κ e L 0 + m e 2 I ) ⊗ I 16 , H σ = f 2 L a + m σ 2 I , ∂ e v , i 4 S h = 6 λ e . (503) H_e=(\kappa_eL_0+m_e^2I)\otimes I_{16}, \qquad H_{\sigma}=f^2L_a+m_{\sigma}^2I, \qquad\partial_{e_v,i}^4S_h=6\lambda_e. \tag*{(503)} H e = ( κ e L 0 + m e 2 I ) ⊗ I 16 , H σ = f 2 L a + m σ 2 I , ∂ e v , i 4 S h = 6 λ e . (503)
The mixed gauge energy vanishes at identity links. An off-diagonal edge entry of H e H_e H e identifies κ e \kappa_e κ e , and a row sum identifies m e 2 m_e^2 m e 2 . A scalar edge entry with known positive weight identifies f 2 f^2 f 2 ; a row sum then identifies m σ 2 m_{\sigma}^2 m σ 2 . Positivity fixes the square roots. This argument does not require a nondegenerate full gauge Hessian.
For temperature and mobility, let p h = d π h / d μ h p_h=\mathrm{d}\pi_h/\mathrm{d}\mu_h p h = d π h / d μ h be the Gibbs density relative to the fixed volume and choose S h ( x ) ≠ S h ( y ) S_h(x)\ne S_h(y) S h ( x ) = S h ( y ) . Then
θ = − S h ( x ) − S h ( y ) log p h ( x ) − log p h ( y ) , 1 2 lim Δ t ↓ 0 Cov ( Δ Y ∣ Y t = y ) Δ t = θ Γ h ( y ) . (504) \theta=-\frac{S_h(x)-S_h(y)}{\log p_h(x)-\log p_h(y)}, \qquad\frac{1}{2}\lim_{\Delta t\downarrow0}\frac{\operatorname{Cov}(\Delta Y\mid Y_t=y)}{\Delta t}=\theta\Gamma_h(y). \tag*{(504)} θ = − log p h ( x ) − log p h ( y ) S h ( x ) − S h ( y ) , 2 1 Δ t ↓ 0 lim Δ t Cov ( Δ Y ∣ Y t = y ) = θ Γ h ( y ) . (504)
The covariance uses the fixed local frame. Equivalently, the carré du champ, defined as [ L h ( F G ) − F L h G − G L h F ] / 2 [L_h(FG)-FL_hG-GL_hF]/2 [ L h ( F G ) − F L h G − G L h F ] /2 , equals θ ⟨ ∇ F , Γ h ∇ G ⟩ \theta\langle\nabla F,\Gamma_h\nabla G\rangle θ ⟨ ∇ F , Γ h ∇ G ⟩ . These are ideal response and density probes; estimation from noisy finite observations remains a separate inverse problem.
Proposition 23 (Finite response identifiability). If the fixed graph has the stated link and edge probes and all responses are given in fixed units, the coefficients ( β a , α , η , κ e , m e , λ e , f , m σ , θ , Γ h ) (\beta_a,\alpha,\eta,\kappa_e,m_e,\lambda_e,f,m_{\sigma},\theta,\Gamma_h) ( β a , α , η , κ e , m e , λ e , f , m σ , θ , Γ h ) are uniquely determined by the preceding probes. The graph, reference fields, representations, and functional form are held fixed.
Proof. Equation (502) recovers the gauge coefficients and shared couplings. Edge entries, row sums, and fourth derivatives in (503) recover the confinement and edge coefficients. Since f > 0 f>0 f > 0 , division recovers m σ 2 m_{\sigma}^{2} m σ 2 . Equation (504) then gives θ \theta θ and Γ h \Gamma_{h} Γ h . Two registries with the same probes therefore coincide. This does not select the graph, vacuum, or group from primitive data. □
Absolute energy calibration is essential. For every c > 0 c>0 c > 0 ,
( S h , θ , Γ h ) ⟼ ( c S h , c θ , Γ h / c ) (505) (S_h,\theta,\Gamma_h)\longmapsto(cS_h,c\theta,\Gamma_h/c) \tag*{(505)} ( S h , θ , Γ h ) ⟼ ( c S h , c θ , Γ h / c ) (505)
preserves both π h \pi_h π h and L h \mathcal{L}_h L h . It is realized within the registry by multiplying β a \beta_a β a , κ e \kappa_e κ e , m σ 2 m_{\sigma}^{2} m σ 2 , λ e \lambda_e λ e , f 2 f^{2} f 2 by c c c , leaving m σ 2 m_{\sigma}^{2} m σ 2 , α \alpha α , η \eta η fixed, and changing θ \theta θ , Γ h \Gamma_h Γ h as displayed. Probability laws alone cannot distinguish this family. One independent energy normalization removes this ambiguity; it does not prove unique selection of every constant by the mother theory.
For quadratic potentials one may instead use the Ornstein–Uhlenbeck relations A ∗ = − Γ h H ∗ A_*=-\Gamma_hH_* A ∗ = − Γ h H ∗ and C ∗ = θ H − 1 C_*=\theta H^{-1} C ∗ = θ H − 1 . The covariance relation is not exact for a nonlinear Gibbs equilibrium. For example, V ( x ) = m 2 x 2 / 2 + λ x 4 / 4 V(x)=m^{2}x^{2}/2+\lambda x^{4}/4 V ( x ) = m 2 x 2 /2 + λ x 4 /4 gives, by integration by parts, m 2 E x 2 + λ E x 4 = θ m^{2}\mathbb{E}x^{2}+\lambda\mathbb{E}x^{4}=\theta m 2 E x 2 + λ E x 4 = θ . For λ > 0 \lambda>0 λ > 0 , setting E x 2 = θ / m 2 \mathbb{E}x^{2}=\theta/m^{2} E x 2 = θ / m 2 is incorrect.
10.2. Finite response measurements and stable prediction. The ideal inverse formulas identify coefficients from exact derivatives. An experiment, however, returns a finite collection of responses with uncertainty. We therefore retain the response law already obtained from the finite action and ask what three measurements can determine. Fix one gauge factor, suppress its index, choose h > 0 h>0 h > 0 and 0 < r < 1 0<r<1 0 < r < 1 , and write
Q ( r ′ , s ′ ) = β exp ( − α s ′ ) ( 1 + η r ′ ) , β > 0 , α , η ≥ 0. (506) Q(r',s')=\beta\exp(-\alpha s')(1+\eta r'),\qquad\beta>0,\quad\alpha,\eta\geq0. \tag*{(506)} Q ( r ′ , s ′ ) = β exp ( − α s ′ ) ( 1 + η r ′ ) , β > 0 , α , η ≥ 0. (506)
All response logarithms below use numerical values in one fixed positive response unit; the same unit is used for β \beta β . Ratios and the final four-response contrast are independent of this choice of unit. The calibration settings are ( 0 , 0 ) (0,0) ( 0 , 0 ) , ( 0 , h ) (0,h) ( 0 , h ) , and ( r , 0 ) (r,0) ( r , 0 ) . Here Q Q Q remains the normalized link-response coefficient defined in (501) ; a measurement protocol must control the error in extracting that coefficient as well as the error in recording it. Finite settings in ( r , s ) (r,s) ( r , s ) do not by themselves remove the link-derivative measurement requirement.
Let their measured positive responses be Y 0 Y_0 Y 0 , Y h Y_h Y h , Y r Y_r Y r . Assume a deterministic simultaneous log-error bound with δ ≥ 0 \delta\geq0 δ ≥ 0
Y 0 = β e e 0 , Y h = β e − α h + e h , Y r = β ( 1 + η r ) e e r , ∣ e 0 ∣ , ∣ e h ∣ , ∣ e r ∣ ≤ δ . (507) Y_0=\beta e^{e_0},\qquad Y_h=\beta e^{-\alpha h+e_h},\qquad Y_r=\beta(1+\eta r)e^{e_r},\qquad|e_0|,|e_h|,|e_r|\leq\delta. \tag*{(507)} Y 0 = β e e 0 , Y h = β e − α h + e h , Y r = β ( 1 + η r ) e e r , ∣ e 0 ∣ , ∣ e h ∣ , ∣ e r ∣ ≤ δ . (507)
This hypothesis may be supplied by a metrological error budget or by an appropriately simultaneous confidence event. Three separate confidence statements are not automatically one event of the same confidence level.
Put
z 0 = log Y 0 , z h = log Y h , z r = log Y r , b = log β , v = log ( 1 + η r ) . (508) z_0=\log Y_0,\quad z_h=\log Y_h,\quad z_r=\log Y_r,\quad b=\log\beta,\quad v=\log(1+\eta r). \tag*{(508)} z 0 = log Y 0 , z h = log Y h , z r = log Y r , b = log β , v = log ( 1 + η r ) . (508)
The observation map becomes affine in ( b , α , v ) (b,\alpha,v) ( b , α , v ) :
( z 0 , z h , z r ) = ( b , b − α h , b + v ) + ( e 0 , e h , e r ) , α , v ≥ 0. (509) (z_0,z_h,z_r)=(b,b-\alpha h,b+v)+(e_0,e_h,e_r),\qquad\alpha,v\geq0. \tag*{(509)} ( z 0 , z h , z r ) = ( b , b − α h , b + v ) + ( e 0 , e h , e r ) , α , v ≥ 0. (509)
This coordinate choice makes both positivity and common calibration uncertainty visible.
Proposition 24 (Exact feasible coefficients from three finite settings). Define
B − : = max { z 0 − δ , z h − δ } , B + : = min { z 0 + δ , z r + δ } . (510) B_-:=\max\{z_0-\delta,z_h-\delta\},\qquad B_+:=\min\{z_0+\delta,z_r+\delta\}. \tag*{(510)} B − := max { z 0 − δ , z h − δ } , B + := min { z 0 + δ , z r + δ } . (510) A parameter triple satisfying (507) and (508) exists if and only if B − ≤ B + B_-\leq B_+ B − ≤ B + . When it exists, the exact joint feasible set consists of
B − ≤ b ≤ B + , A − ( b ) ≤ α ≤ A + ( b ) , V − ( b ) ≤ v ≤ V + ( b ) , B_-\leq b\leq B_+,\qquad A_-(b)\leq\alpha\leq A_+(b),\qquad V_-(b)\leq v\leq V_+(b), B − ≤ b ≤ B + , A − ( b ) ≤ α ≤ A + ( b ) , V − ( b ) ≤ v ≤ V + ( b ) , A − ( b ) : = max { 0 , ( b − z h − δ ) / h } , A + ( b ) : = ( b − z h + δ ) / h , A_-(b):=\max\{0,(b-z_h-\delta)/h\},\qquad A_+(b):=(b-z_h+\delta)/h, A − ( b ) := max { 0 , ( b − z h − δ ) / h } , A + ( b ) := ( b − z h + δ ) / h , V − ( b ) : = max { 0 , z r − δ − b } , V + ( b ) : = z r + δ − b , (511) V_-(b):=\max\{0,z_r-\delta-b\},\qquad V_+(b):=z_r+\delta-b, \tag*{(511)} V − ( b ) := max { 0 , z r − δ − b } , V + ( b ) := z r + δ − b , (511) with β = e b \beta=e^b β = e b and η = ( e v − 1 ) / r \eta=(e^v-1)/r η = ( e v − 1 ) / r . In particular the exact marginal intervals are
β ∈ [ e B − , e B + ] , \beta\in[e^{B_-},e^{B_+}], β ∈ [ e B − , e B + ] , α ∈ [ max { 0 , ( B − − z h − δ ) / h } , ( B + − z h + δ ) / h ] , \alpha\in[\max\{0,(B_--z_h-\delta)/h\},(B_+-z_h+\delta)/h], α ∈ [ max { 0 , ( B − − z h − δ ) / h } , ( B + − z h + δ ) / h ] , η ∈ [ exp ( max { 0 , z r − δ − B + } ) − 1 r , exp ( z r + δ − B − ) − 1 r ] . (512) \eta\in\left[
\frac{\exp(\max\{0,z_r-\delta-B_+\})-1}{r},
\frac{\exp(z_r+\delta-B_-)-1}{r}
\right]. \tag*{(512)} η ∈ [ r exp ( max { 0 , z r − δ − B + }) − 1 , r exp ( z r + δ − B − ) − 1 ] . (512) These marginal intervals are individually sharp. Their Cartesian product generally contains triples outside the joint feasible set.
Proof. The first observation gives b ∈ [ z 0 − δ , z 0 + δ ] b\in[z_0-\delta,z_0+\delta] b ∈ [ z 0 − δ , z 0 + δ ] . The second gives b − z h − δ ≤ α h ≤ b − z h + δ b-z_h-\delta\leq\alpha h\leq b-z_h+\delta b − z h − δ ≤ α h ≤ b − z h + δ . Its intersection with α ≥ 0 \alpha\geq0 α ≥ 0 is nonempty exactly when b ≥ z h − δ b\geq z_h-\delta b ≥ z h − δ . The third gives z r − δ − b ≤ v ≤ z r + δ − b z_r-\delta-b\leq v\leq z_r+\delta-b z r − δ − b ≤ v ≤ z r + δ − b . Its intersection with v ≥ 0 v\geq0 v ≥ 0 is nonempty exactly when b ≤ z r + δ b\leq z_r+\delta b ≤ z r + δ . Their combined requirement is precisely (511). For every b b b in this interval, the two remaining intervals are nonempty and independent conditional on b b b . Conversely, every point in these intervals supplies errors e 0 = z 0 − b e_0=z_0-b e 0 = z 0 − b , e h = z h − b + α h e_h=z_h-b+\alpha h e h = z h − b + α h , and e r = z r − b − v e_r=z_r-b-v e r = z r − b − v in [ − δ , δ ] [-\delta,\delta] [ − δ , δ ] . Thus the description is exact. Both endpoints for α \alpha α increase with b b b , whereas both endpoints for v v v decrease with b b b . Taking their extrema on [ B − , B + ] [B_-,B_+] [ B − , B + ] and using the monotonicity of the exponential proves (513). □ \square □
An empty set has a direct meaning: the stated response family, sign conditions and simultaneous error budget cannot all explain the three measurements. It does not identify which of these inputs failed. A nonempty set also does not establish that the physical system realizes the response family; an unused response setting tests that additional claim.
Proposition 25 (A positive estimator and a uniform target-response bound). Define
b ^ = z 0 , α ^ = max { 0 , ( z 0 − z h ) / h } , \widehat{b}=z_0,\qquad\widehat{\alpha}=\max\{0,(z_0-z_h)/h\}, b = z 0 , α = max { 0 , ( z 0 − z h ) / h } , v ^ = max { 0 , z r − z 0 } , η ^ = ( e v ^ − 1 ) / r . (513) \widehat{v}=\max\{0,z_r-z_0\},\qquad\widehat{\eta}=(e^{\widehat{v}}-1)/r. \tag*{(513)} v = max { 0 , z r − z 0 } , η = ( e v − 1 ) / r . (513) For an unused setting 0 ≤ r ∗ < 1 0\le r_\ast<1 0 ≤ r ∗ < 1 , s ∗ ∈ R s_\ast\in\mathbb{R} s ∗ ∈ R , let t = r ∗ / r t=r_\ast/r t = r ∗ / r and set
Q ^ ∗ = e b ^ − α ^ s ∗ ( 1 + η ^ r ∗ ) , L t = { 0 , t = 0 , max { 1 , t } , t > 0. (514) \widehat{Q}_\ast=e^{\widehat{b}-\widehat{\alpha}s_\ast}(1+\widehat{\eta}r_\ast),\qquad
L_t=\begin{cases}
0, & t=0,\\
\max\{1,t\}, & t>0.
\end{cases} \tag*{(514)} Q ∗ = e b − α s ∗ ( 1 + η r ∗ ) , L t = { 0 , max { 1 , t } , t = 0 , t > 0. (514) Under (507) ,
∣ b ^ − b ∣ ≤ δ , ∣ α ^ − α ∣ ≤ 2 δ / h , ∣ v ^ − v ∣ ≤ 2 δ , |\widehat{b}-b|\le\delta,\qquad|\widehat{\alpha}-\alpha|\le2\delta/h,\qquad|\widehat{v}-v|\le2\delta, ∣ b − b ∣ ≤ δ , ∣ α − α ∣ ≤ 2 δ / h , ∣ v − v ∣ ≤ 2 δ , ∣ η ^ − η ∣ ≤ 1 + η r r ( e 2 δ − 1 ) , |\widehat{\eta}-\eta|\le\frac{1+\eta r}{r}(e^{2\delta}-1), ∣ η − η ∣ ≤ r 1 + η r ( e 2 δ − 1 ) , ∣ log Q ^ ∗ Q ( r ∗ , s ∗ ) ∣ ≤ E ∗ : = δ + 2 δ ∣ s ∗ ∣ h + 2 δ L t . (515) \left|\log\frac{\widehat{Q}_\ast}{Q(r_\ast,s_\ast)}\right|
\le E_\ast:=\delta+\frac{2\delta|s_\ast|}{h}+2\delta L_t. \tag*{(515)} log Q ( r ∗ , s ∗ ) Q ∗ ≤ E ∗ := δ + h 2 δ ∣ s ∗ ∣ + 2 δ L t . (515) Consequently e − E ∗ ≤ Q ^ ∗ / Q ( r ∗ , s ∗ ) ≤ e E ∗ e^{-E_\ast}\le\widehat{Q}_\ast/Q(r_\ast,s_\ast)\le e^{E_\ast} e − E ∗ ≤ Q ∗ / Q ( r ∗ , s ∗ ) ≤ e E ∗ . No division by η \eta η or its estimate is required, so the result also holds at η = 0 \eta=0 η = 0 .
Proof. The raw slope errors are ( e 0 − e h ) / h (e_0-e_h)/h ( e 0 − e h ) / h and e r − e 0 e_r-e_0 e r − e 0 . Projection onto [ 0 , ∞ ) [0,\infty) [ 0 , ∞ ) does not increase distance from a nonnegative true value, which proves the first three bounds. The estimate ∣ v ^ − v ∣ ≤ 2 δ |\widehat{v}-v|\le2\delta ∣ v − v ∣ ≤ 2 δ yields ∣ e v ^ − e v ∣ ≤ e v ( e 2 δ − 1 ) |e^{\widehat{v}}-e^v|\le e^v(e^{2\delta}-1) ∣ e v − e v ∣ ≤ e v ( e 2 δ − 1 ) and hence the coefficient bound. For the target define g t ( v ) = log ( 1 − t + t e v ) g_t(v)=\log(1-t+te^v) g t ( v ) = log ( 1 − t + t e v ) on v ≥ 0 v\ge0 v ≥ 0 . Its argument is positive because 1 − t + t e v = 1 + t ( e v − 1 ) ≥ 1 1-t+te^v=1+t(e^v-1)\ge1 1 − t + t e v = 1 + t ( e v − 1 ) ≥ 1 . If t = 0 t=0 t = 0 then g t = 0 g_t=0 g t = 0 . Otherwise
g t ′ ( v ) = t e v 1 − t + t e v , 0 ≤ g t ′ ( v ) ≤ max { 1 , t } . (516) g_t'(v)=\frac{te^v}{1-t+te^v},\qquad0\le g_t'(v)\le\max\{1,t\}. \tag*{(516)} g t ′ ( v ) = 1 − t + t e v t e v , 0 ≤ g t ′ ( v ) ≤ max { 1 , t } . (516)
For 0 < t ≤ 1 0<t\le1 0 < t ≤ 1 the upper bound follows by adding the nonnegative 1 − t 1-t 1 − t to the denominator; for t ≥ 1 t\ge1 t ≥ 1 , the inequality g t ′ ≤ t g_t'\le t g t ′ ≤ t is equivalent to ( t − 1 ) ( e v − 1 ) ≥ 0 (t-1)(e^v-1)\ge0 ( t − 1 ) ( e v − 1 ) ≥ 0 . The target logarithm is b − α s ∗ + g t ( v ) b-\alpha s_\ast+g_t(v) b − α s ∗ + g t ( v ) . Applying the triangle inequality and the preceding Lipschitz bound proves the assertion. □
The clipped estimate is always physically signed, although it need not belong to the joint feasible set for a particular data realization. When feasibility itself is part of the inference, use (511) . Assuming B − ≤ B + B_-\le B_+ B − ≤ B + , for s ∗ ≥ 0 s_\ast\ge0 s ∗ ≥ 0 the exact minimum and maximum of the target logarithm over that set reduce to
ℓ ∗ − = min b ∈ [ B − , B + ] { b − s ∗ A + ( b ) + g t ( V − ( b ) ) } , \ell_\ast^-=\min_{b\in[B_-,B_+]}\{b-s_\ast A_+(b)+g_t(V_-(b))\}, ℓ ∗ − = b ∈ [ B − , B + ] min { b − s ∗ A + ( b ) + g t ( V − ( b ))} ,
ℓ ∗ + = max b ∈ [ B − , B + ] { b − s ∗ A − ( b ) + g t ( V + ( b ) ) } . (517) \ell^{+}_{*}=\max_{b\in[B_{-},B_{+}]} \{b-s_{*}A_{-}(b)+g_t(V_{+}(b))\}. \tag*{(517)} ℓ ∗ + = b ∈ [ B − , B + ] max { b − s ∗ A − ( b ) + g t ( V + ( b ))} . (517)
For s ∗ < 0 s_{*}<0 s ∗ < 0 , exchange A + A_{+} A + and A − A_{-} A − in these two expressions. The functions are continuous on a compact interval, so the extrema are attained. Monotonicity in α \alpha α and v v v proves that [ e ℓ ∗ − , e ℓ ∗ + ] [e^{\ell^{-}_{*}},e^{\ell^{+}_{*}}] [ e ℓ ∗ − , e ℓ ∗ + ] is the exact prediction interval. This one-dimensional optimization retains the shared-baseline uncertainty that is lost by independently combining marginal endpoints.
An especially simple unused setting is the fourth corner ( r , h ) (r,h) ( r , h ) . The response law itself gives the exact relation
Q ( r , h ) = Q ( 0 , h ) Q ( r , 0 ) Q ( 0 , 0 ) . (518) Q(r,h)=\frac{Q(0,h)Q(r,0)}{Q(0,0)}. \tag*{(518)} Q ( r , h ) = Q ( 0 , 0 ) Q ( 0 , h ) Q ( r , 0 ) . (518)
Thus Y h Y r / Y 0 Y_hY_r/Y_0 Y h Y r / Y 0 is a prediction formed entirely from the three calibration measurements. Its log error relative to the true fourth response is e h + e r − e 0 e_h+e_r-e_0 e h + e r − e 0 , of magnitude at most 3 δ 3\delta 3 δ . If an independent measurement Y ∗ Y_{*} Y ∗ at ( r , h ) (r,h) ( r , h ) has log error ∣ e ∗ ∣ ≤ δ |e_{*}|\leq\delta ∣ e ∗ ∣ ≤ δ , the observable contrast obeys
∣ log Y ∗ − log Y h − log Y r + log Y 0 ∣ ≤ 4 δ . (519) |\log Y_{*}-\log Y_h-\log Y_r+\log Y_0|\leq4\delta. \tag*{(519)} ∣ log Y ∗ − log Y h − log Y r + log Y 0 ∣ ≤ 4 δ . (519)
No parameter is refitted to Y ∗ Y_{*} Y ∗ . A violation rejects the conjunction of the separable response law and its stated error budget. Agreement tests this structural relation, which is also shared by other separable response families, and cannot uniquely identify the mother functional. With independent log errors of known variance σ e 2 \sigma_e^2 σ e 2 , the contrast variance is 4 σ e 2 4\sigma_e^2 4 σ e 2 ; a Gaussian test additionally requires centered Gaussian errors (or independently known biases removed) and a test fixed before observing the target. If data are correlated, their full covariance replaces this variance.
What the measurement spacings control. Increasing h h h reduces the deterministic slope uncertainty 2 δ / h 2\delta/h 2 δ / h provided the same log-error budget remains justified. This proviso matters: an instrument with fixed additive error has increasing relative error when Q ( 0 , h ) Q(0,h) Q ( 0 , h ) becomes small. The factor 1 / r 1/r 1/ r likewise shows why a very small coframe perturbation is poor for estimating η \eta η in absolute units. Taking r r r close to one is not free: the defining coframe displacement has squared norm r / ( 1 − r ) r/(1-r) r / ( 1 − r ) and therefore becomes large. Admissible perturbation size, signal-to-noise ratio and the range of the response model must all enter the design.
The common baseline also creates a statistical correlation. If the three log errors are independent, centered, and have variance σ e 2 \sigma_e^2 σ e 2 , then the covariance of the raw, unclipped coordinates ( z 0 , ( z 0 − z h ) / h , z r − z 0 ) (z_0,(z_0-z_h)/h,z_r-z_0) ( z 0 , ( z 0 − z h ) / h , z r − z 0 ) is exactly
σ e 2 ( 1 1 / h − 1 1 / h 2 / h 2 − 1 / h − 1 − 1 / h 2 ) . (520) \sigma_e^2
\begin{pmatrix}
1 & 1/h & -1 \\
1/h & 2/h^2 & -1/h \\
-1 & -1/h & 2
\end{pmatrix}. \tag*{(520)} σ e 2 1 1/ h − 1 1/ h 2/ h 2 − 1/ h − 1 − 1/ h 2 . (520)
Clipping changes this covariance near the sign boundaries. Equation (520) must therefore not be used as the exact covariance of the clipped estimator. A deterministic error budget and a probabilistic covariance model are separate statements; either requires support from the specified measurement procedure.
These results extend the three-coefficient response law to finite uncertainty. They do not select the graph or representation, establish the physical availability of the probes, identify the entire remaining registry, or prove unique selection of fundamental constants.
Constraint tangency and the physical quotient.
A constraint selects admissible states; gauge symmetry identifies admissible states with the same physical meaning. A dynamics must respect both operations. Let C = ( C 1 , … , C m ) \mathcal{C}=(C_1,\ldots,C_m) C = ( C 1 , … , C m ) be C 3 C^3 C 3 on an open subset of R n \mathbb{R}^n R n , with G = D C G=D\mathcal{C} G = D C of full row rank near S = { C = 0 } S=\{\mathcal{C}=0\} S = { C = 0 } . Set
P T = I − G T ( G G T ) − 1 G , v = P T b , B k = P T σ k , (521) P_T=I-G^{\mathsf{T}}(GG^{\mathsf{T}})^{-1}G,\qquad v=P_Tb,\qquad B_k=P_T\sigma_k, \tag*{(521)} P T = I − G T ( G G T ) − 1 G , v = P T b , B k = P T σ k , (521)
and assume sufficient smoothness for the local Stratonovich equation
d Y t = v ( Y t ) d t + ∑ k B k ( Y t ) ∘ d W t k . (522) \mathrm{d}Y_t=v(Y_t)\mathrm{d}t+\sum_k B_k(Y_t)\circ\mathrm{d}W_t^k. \tag*{(522)} d Y t = v ( Y t ) d t + k ∑ B k ( Y t ) ∘ d W t k . (522)
Proposition 26 (Constraint preservation). If Y 0 ∈ S Y_0\in S Y 0 ∈ S , the process remains in S S S until its lifetime or exit from the regular neighborhood. Its Itô drift is v + 1 2 ∑ k D B k B k v+\frac{1}{2}\sum_k DB_kB_k v + 2 1 ∑ k D B k B k .
Proof. Since G P T = 0 GP_T=0 G P T = 0 , the Stratonovich chain rule gives d C ( Y t ) = 0 \mathrm{d}\mathcal{C}(Y_t)=0 d C ( Y t ) = 0 . For a direct Itô check, differentiate D C i B k = 0 D\mathcal{C}_iB_k=0 D C i B k = 0 along B k B_k B k :
D 2 C i [ B k , B k ] + D C i ( D B k B k ) = 0. (523) D^2\mathcal{C}_i[B_k,B_k]+D\mathcal{C}_i(DB_kB_k)=0. \tag*{(523)} D 2 C i [ B k , B k ] + D C i ( D B k B k ) = 0. (523) The curvature drift cancels the quadratic-variation term. Projecting only the noise in an Itô equation would generally omit this drift. □
Hamiltonian first-class closure is a separate requirement:
{ C α , C β } = f α β γ C γ , { C α , H } = u α γ C γ . (524) \{C_\alpha,C_\beta\}=f_{\alpha\beta}^{\ \ \gamma}C_\gamma,\qquad\{C_\alpha,H\}=u_\alpha^{\ \gamma}C_\gamma. \tag*{(524)} { C α , C β } = f α β γ C γ , { C α , H } = u α γ C γ . (524)
It makes J ∇ H J\nabla H J ∇ H tangent on S S S . Enforcing tangency by a projection does not prove either bracket identity. For a smooth quotient q : S → S / G q:S\to S/\mathcal{G} q : S → S / G , physical evolution further requires
L ( f ∘ q ) = ( L p h y s f ) ∘ q for every smooth quotient function f . (525) \mathcal{L}(f\circ q)=(\mathcal{L}_{\mathrm{phys}}f)\circ q\quad\text{for every smooth quotient function }f. \tag*{(525)} L ( f ∘ q ) = ( L phys f ) ∘ q for every smooth quotient function f . (525)
Gauge invariance of H H H and tangent ellipticity alone do not guarantee this: the diffusion and drift must respect gauge orbits. Nor does pure tangent noise preserve a Gibbs law without a matching dissipative drift and the appropriate surface and quotient volumes.
For a nonlinear example, take canonical coordinates ( q 1 , q 2 , p 1 , p 2 ) (q_1,q_2,p_1,p_2) ( q 1 , q 2 , p 1 , p 2 ) and
C = q 1 p 2 − q 2 p 1 , H = 1 2 ∣ p ∣ 2 + ω 2 2 ∣ q ∣ 2 + λ 4 ∣ q ∣ 4 , λ ≥ 0. (526) C=q_1p_2-q_2p_1,\qquad H=\frac{1}{2}|p|^2+\frac{\omega^2}{2}|q|^2+\frac{\lambda}{4}|q|^4,\qquad\lambda\ge0. \tag*{(526)} C = q 1 p 2 − q 2 p 1 , H = 2 1 ∣ p ∣ 2 + 2 ω 2 ∣ q ∣ 2 + 4 λ ∣ q ∣ 4 , λ ≥ 0. (526)
Direct differentiation gives { C , H } = { C , C } = 0 \{C,H\}=\{C,C\}=0 { C , H } = { C , C } = 0 and ∣ D C ∣ 2 = ∣ q ∣ 2 + ∣ p ∣ 2 |DC|^2=|q|^2+|p|^2 ∣ D C ∣ 2 = ∣ q ∣ 2 + ∣ p ∣ 2 . Away from the origin the constraint is regular. Simultaneous rotations of q , p q,p q , p are orthogonal, and P T P_T P T is equivariant. For b = J ∇ H b=J\nabla H b = J ∇ H and isotropic ambient noise, the full projected generator commutes with rotations and descends to the regular quotient. On q ≠ 0 q\ne0 q = 0 , the coordinates r = ∣ q ∣ r=|q| r = ∣ q ∣ , p r = q ⋅ p / r p_r=q\cdot p/r p r = q ⋅ p / r give H = p r 2 / 2 + ω 2 r 2 / 2 + λ r 4 / 4 H=p_r^2/2+\omega^2r^2/2+\lambda r^4/4 H = p r 2 /2 + ω 2 r 2 /2 + λ r 4 /4 on C = 0 C=0 C = 0 . This verifies nonlinear tangency and projectability in a concrete model; its rotational constraint is not the ADM constraint system.
The limit required for entropy dissipation
Convergence of linear generators does not by itself control dissipation along nonlinear densities. Embed probabilities π N \pi_N π N in a common Polish space, with π N ⇒ π \pi_N\Rightarrow\pi π N ⇒ π . Let E N \mathcal{E}_N E N be symmetric, strongly local diffusion forms, including the temperature coefficient. For ρ s N = f s N π N \rho_s^N=f_s^N\pi_N ρ s N = f s N π N , their chain rule gives
I N ( ρ s N ) = 4 E N ( f s N , f s N ) . (527) \mathcal{I}_N(\rho_s^N)=4\mathcal{E}_N\left(\sqrt{f_s^N},\sqrt{f_s^N}\right). \tag*{(527)} I N ( ρ s N ) = 4 E N ( f s N , f s N ) . (527)
This expression is not asserted for arbitrary jump forms.
Theorem 27 (Conditional transfer of entropy dissipation). Fix T > 0 T>0 T > 0 . Assume the finite laws obey integrated entropy inequalities, their initial entropies converge to D ( ρ 0 ∥ π ) < ∞ D(\rho_0\Vert\pi)<\infty D ( ρ 0 ∥ π ) < ∞ , and ρ T N ⇒ ρ T \rho_T^N\Rightarrow\rho_T ρ T N ⇒ ρ T . Assume that for almost every s ∈ ( 0 , T ) s\in(0,T) s ∈ ( 0 , T ) the limiting density ρ s = f s π \rho_s=f_s\pi ρ s = f s π is identified and f s N \sqrt{f_s^N} f s N converges to f s \sqrt{f_s} f s in the weak comparison topology for the Mosco lower-bound condition. Then
D ( ρ T ∥ π ) + 4 ∫ 0 T E ( f s , f s ) d s ≤ D ( ρ 0 ∥ π ) . (528) D(\rho_T\Vert\pi)+4\int_0^T\mathcal{E}\left(\sqrt{f_s},\sqrt{f_s}\right)\,ds\le D(\rho_0\Vert\pi). \tag*{(528)} D ( ρ T ∥ π ) + 4 ∫ 0 T E ( f s , f s ) d s ≤ D ( ρ 0 ∥ π ) . (528)
Proof. The variational formula D ( ρ ∥ π ) = sup g ∈ C b { ∫ g d ρ − log ∫ e g d π } D(\rho\Vert\pi)=\sup_{g\in C_b}\{\int g\,d\rho-\log\int e^g\,d\pi\} D ( ρ ∥ π ) = sup g ∈ C b { ∫ g d ρ − log ∫ e g d π } gives joint lower semicontinuity at T T T . Square-root convergence and the Mosco lower bound give E ( f s , f s ) ≤ lim inf N E N ( f s N , f s N ) \mathcal{E}(\sqrt{f_s},\sqrt{f_s})\le\liminf_N\mathcal{E}_N(\sqrt{f_s^N},\sqrt{f_s^N}) E ( f s , f s ) ≤ lim inf N E N ( f s N , f s N ) almost everywhere in s s s . Fatou’s lemma passes this through the integral. Add the lower bounds and use initial entropy convergence. □
For varying Hilbert spaces the comparison maps are part of the hypothesis; see [107]. Weak convergence of laws does not alone identify their square roots in this topology. The required compactness, or a direct integrated Fisher lower bound, must be established for the chosen sequence. The theorem specifies a passage to the limit; it does not supply uniform estimates for the full dynamical-coframe theory. It concerns thermal probability evolution, not quantum reconstruction.
Positive quantum evolution at fixed spatial resolution
Quantum positivity can be achieved constructively before taking a spatial limit. The following is an alternative temporal regularization: its time-crossing kernel replaces the temporal plaquette weight of (481). Equality of the two measures is not assumed. In particular, anchoring a weight to one vertex of a time-crossing plaquette need not preserve time reflection.
Fix a finite three-dimensional spatial lattice and set Q h = K E h s p × R 17 ∣ V h s p ∣ Q_h = K^{E_h^{\mathrm{sp}}} \times\mathbb{R}^{17|V_h^{\mathrm{sp}}|} Q h = K E h sp × R 17∣ V h sp ∣ , with product volume μ h s p \mu_h^{\mathrm{sp}} μ h sp . Let V h ≥ 0 V_h \ge0 V h ≥ 0 be the spatial part of the finite action, including its shared interaction and confinement. Then ∫ e − a V h d μ h s p < ∞ \int e^{-aV_h}\,d\mu_h^{\mathrm{sp}} < \infty ∫ e − a V h d μ h sp < ∞ for every a > 0 a > 0 a > 0 . Fix all field units and use ℏ = 1 \hbar= 1 ℏ = 1 , so a a a has inverse-energy units. For a kinetic coefficient ν > 0 \nu> 0 ν > 0 with corresponding units, define on H h = L 2 ( Q h , μ h s p ) \mathcal{H}_h = L^2(Q_h,\mu_h^{\mathrm{sp}}) H h = L 2 ( Q h , μ h sp )
H 0 , h = − ν Δ Q h , M a = e − a V h / 2 , T a , h = M a e − a H 0 , h M a . (529) H_{0,h} = -\nu\Delta_{Q_h}, \qquad M_a = e^{-aV_h/2}, \qquad T_{a,h} = M_a e^{-aH_{0,h}} M_a. \tag*{(529)} H 0 , h = − ν Δ Q h , M a = e − a V h /2 , T a , h = M a e − a H 0 , h M a . (529)
The heat kernel propagates configurations between time slices; M a M_a M a assigns half the potential weight to each endpoint. This symmetric assignment yields positivity without assuming it for an arbitrary Gibbs weight.
Proposition 28 (An interacting quantum model with a spatial cutoff). At fixed spatial lattice, T a , h T_{a,h} T a , h is an injective positive self-adjoint trace-class contraction. It defines H a , h = − a − 1 log T a , h ≥ 0 H_{a,h} = -a^{-1}\log T_{a,h} \ge0 H a , h = − a − 1 log T a , h ≥ 0 on a dense domain. Bounded positive-time cylinder observables on an even periodic time lattice satisfy reflection positivity, and the link-gauge-invariant subspace is preserved. Also, for every t > 0 t > 0 t > 0 ,
( T t / n , h ) n ⟶ e − t H h strongly , H h = H 0 , h ∔ V h , (530) (T_{t/n,h})^n \longrightarrow e^{-tH_h}\ \text{strongly}, \qquad H_h = H_{0,h} \dotplus V_h, \tag*{(530)} ( T t / n , h ) n ⟶ e − t H h strongly , H h = H 0 , h ∔ V h , (530) where ∔ \dotplus ∔ denotes the closed nonnegative form sum. The spatial regulator is fixed throughout this time-continuum limit.
Proof. Set B = e − a H 0 , h / 2 M a B = e^{-aH_{0,h}/2}M_a B = e − a H 0 , h /2 M a . Then T a , h = B ∗ B T_{a,h} = B^*B T a , h = B ∗ B and ⟨ ψ , T a , h ψ ⟩ = ∥ B ψ ∥ 2 ≥ 0 \langle\psi,T_{a,h}\psi\rangle= \lVert B\psi\rVert^2 \ge0 ⟨ ψ , T a , h ψ ⟩ = ∥ B ψ ∥ 2 ≥ 0 . Both factors are injective contractions. The heat kernel p a p_a p a has bounded diagonal on the compact group factor and constant diagonal ( 4 π ν a ) − d / 2 (4\pi\nu a)^{-d/2} ( 4 π ν a ) − d /2 on the Euclidean factor. Hence
∥ B ∥ H S 2 = ∫ e − a V h ( x ) p a ( x , x ) d μ h s p ( x ) < ∞ . (531) \lVert B\rVert_{\mathrm{HS}}^2 = \int e^{-aV_h(x)}p_a(x,x)\,d\mu_h^{\mathrm{sp}}(x) < \infty. \tag*{(531)} ∥ B ∥ HS 2 = ∫ e − a V h ( x ) p a ( x , x ) d μ h sp ( x ) < ∞. (531)
This makes T a , h T_{a,h} T a , h trace class. Spectral calculus gives its densely defined self-adjoint nonnegative logarithmic Hamiltonian.
The transfer kernel is real, symmetric, and nonnegative. Reflect an even periodic time lattice through two opposite bonds. Multiplication by bounded observables and transfer along one half assemble into an operator A A A ; integrating the reflected half gives the quadratic form
Tr ( A ∗ T a , h A T a , h ) = ∥ T a , h 1 / 2 A T a , h 1 / 2 ∥ H S 2 ≥ 0. (532) \operatorname{Tr}(A^*T_{a,h}AT_{a,h}) = \lVert T_{a,h}^{1/2}AT_{a,h}^{1/2}\rVert_{\mathrm{HS}}^2 \ge0. \tag*{(532)} Tr ( A ∗ T a , h A T a , h ) = ∥ T a , h 1/2 A T a , h 1/2 ∥ HS 2 ≥ 0. (532)
Linear combinations of cylinder observables give the corresponding linear combination of A A A . The normalization Tr ( T a , h 2 n ) \operatorname{Tr}(T_{a,h}^{2n}) Tr ( T a , h 2 n ) is finite and positive.
The finite site gauge group acts isometrically by left and right link multiplication. It preserves V h V_h V h : spatial plaquette traces are invariant, and e , σ e,\sigma e , σ are singlets in this model. Thus the group commutes with both transfer factors. Haar averaging projects orthogonally onto the invariant subspace, preserved by the transfer and Hamiltonian operators.
Finally, ν ∫ ∣ ∇ ψ ∣ 2 d μ h s p + ∫ V h ∣ ψ ∣ 2 d μ h s p \nu\int|\nabla\psi|^2\,\mathrm{d}\mu_h^{\mathrm{sp}}+\int V_h|\psi|^2\,\mathrm{d}\mu_h^{\mathrm{sp}} ν ∫ ∣∇ ψ ∣ 2 d μ h sp + ∫ V h ∣ ψ ∣ 2 d μ h sp is densely defined, closed, and nonnegative. On this complete product manifold the smooth nonnegative potential gives essential self-adjointness on smooth compactly supported functions. The contraction product formula [151] then yields (530) , with symmetric half-potentials at the endpoints. No bound uniform in spatial lattice size has been used. □
This is interacting quantum mechanics of finitely many field coordinates, with a positive inner product and link gauge symmetry. It does not impose the gravitational Hamiltonian and momentum constraints. The selection of ν \nu ν , V h V_h V h by mother data, and their matching to a physical Lorentzian action, remain separate questions. Removing the spatial regulator requires uniform estimates and renormalized limiting observables. Time reflection here acts on Euclidean time slices, not on the relaxation parameter of the earlier stochastic sampling process.
Positivity of a classical density is insufficient even if it is reflection symmetric. The density proportional to exp [ − ( x 2 + y 2 + 2 c x y ) / 2 ] \exp[-(x^2+y^2+2cxy)/2] exp [ − ( x 2 + y 2 + 2 c x y ) /2 ] , 0 < c < 1 0<c<1 0 < c < 1 , is positive and invariant under x ↔ y x\leftrightarrow y x ↔ y . For F ( x ) = x F(x)=x F ( x ) = x , however,
E [ ( Θ F ) F ] = E [ x y ] = − c 1 − c 2 < 0. (533) \mathbb{E}[(\Theta F)F]=\mathbb{E}[xy]=-\frac{c}{1-c^2}<0. \tag*{(533)} E [( Θ F ) F ] = E [ x y ] = − 1 − c 2 c < 0. (533)
Truncation of F F F supplies bounded counterexamples as well. This explains why the transfer construction is a genuine extra requirement.
Finite calculations and the next matching problem
In a single-gauge-factor example, group the coordinates as u = ( β , α , η , κ e , m 2 ) u=(\beta,\alpha,\eta,\kappa_e,m_2) u = ( β , α , η , κ e , m 2 ) and v = ( λ e , f 2 , m σ 2 , θ , γ 1 , γ 2 , γ 3 ) v=(\lambda_e,f^2,m_\sigma^2,\theta,\gamma_1,\gamma_2,\gamma_3) v = ( λ e , f 2 , m σ 2 , θ , γ 1 , γ 2 , γ 3 ) . The first five responses equal u u u ; the remaining seven are ( 6 λ e , f 2 , f 2 m σ 2 , θ − 1 , θ γ 1 , θ γ 2 , θ γ 3 ) (6\lambda_e,f^2,f^2m_\sigma^2,\theta^{-1},\theta\gamma_1,\theta\gamma_2,\theta\gamma_3) ( 6 λ e , f 2 , f 2 m σ 2 , θ − 1 , θ γ 1 , θ γ 2 , θ γ 3 ) . Their Jacobian determinant is − 6 f 2 θ ≠ 0 -6f^2\theta\ne0 − 6 f 2 θ = 0 . This checks the selected inverse map, not selection of primitive constants.
For a transfer calculation choose H = − 1 2 ∂ x 2 + x 2 / 2 + 0.1 x 4 H=-\frac{1}{2}\partial_x^2+x^2/2+0.1x^4 H = − 2 1 ∂ x 2 + x 2 /2 + 0.1 x 4 on [ − 6 , 6 ] [-6,6] [ − 6 , 6 ] with Dirichlet boundaries. A centered finite-difference kinetic matrix and its exact matrix exponential define T a T_a T a . We compare − a − 1 log λ max ( T a ) -a^{-1}\log\lambda_{\max}(T_a) − a − 1 log λ m a x ( T a ) with the smallest eigenvalue of the unsplit Hamiltonian on the same grid, separating time-splitting from spatial discretization error. Units are dimensionless and data synthetic.
Table 4. Synthetic transfer calculation with 200 interior points on [ − 6 , 6 ] [-6, 6] [ − 6 , 6 ] . The reference is the unsplit matrix on the same grid; no continuum error bound is inferred.
Figure 5. Synthetic quartic-oscillator calculation. Left: transfer energy error relative to the unsplit Hamiltonian on each identical grid. Right: spatial-grid variation of the unsplit energy. These finite diagnostics are not observations of gravity or a quantum-field spatial continuum limit.
Further work now has a definite order. First derive or independently calibrate the registry and freeze it before predicting a held-out observable. Next derive the actual Lorentzian constraints and prove propagation and projectability. Then establish uniform estimates and the nonlinear density convergence required by the entropy theorem. Finally match a positive quantum construction to that same Lorentzian branch and remove its spatial regulator. The finite results above provide components of this programme, with the remaining implications kept explicit.
Matching a shared temporal weight to canonical quantum dynamics.
The positive transfer operator above leaves a specific matching question: which kinetic operator represents the same field-dependent weight as the regulated gauge action? A constant configuration-space diffusion does not in general answer that question. We can resolve it in a retained compact gauge mode coupled to the scalar. This calculation keeps a genuine shared interaction and makes the temporal normalization explicit.
Freeze the coframe coordinates at their reference values. Let q ∈ S 1 q \in S^1 q ∈ S 1 denote a retained U ( 1 ) U(1) U ( 1 ) holonomy angle and shift the scalar reference to σ ⋆ = 0 \sigma_\star= 0 σ ⋆ = 0 , absorbing the constant weight into both β \beta β and the temporal gauge inertia I I I . In fixed dimensionless reference units, with ℏ = 1 \hbar= 1 ℏ = 1 , define
w ( σ ) = e − α σ , V ( q , σ ) = β w ( σ ) ( 1 − cos q ) + 1 2 K σ σ 2 , β , K σ > 0. (534) w(\sigma) = e^{-\alpha\sigma}, \qquad V(q,\sigma) = \beta w(\sigma)(1-\cos q) + \frac{1}{2}K_\sigma\sigma^2, \qquad\beta, K_\sigma> 0. \tag*{(534)} w ( σ ) = e − α σ , V ( q , σ ) = β w ( σ ) ( 1 − cos q ) + 2 1 K σ σ 2 , β , K σ > 0. (534)
Here K σ K_\sigma K σ is the scalar onsite coefficient m σ 2 f 2 m_\sigma^2 f^2 m σ 2 f 2 in the fixed normalization. Temporal coefficients require their own regulator scaling: write I = lim a ↓ 0 a β t ( a ) > 0 I = \lim_{a\downarrow0} a\beta_t(a) > 0 I = lim a ↓ 0 a β t ( a ) > 0 for the gauge coefficient and J > 0 J > 0 J > 0 for the analogously normalized scalar inertia. These limits are matching conditions on temporal edge coefficients. The unscaled four-dimensional plaquette coefficient alone does not determine them without the spatial cell volume and field normalization.
The local Lorentzian mechanical action associated with these temporal coefficients is
L L = 1 2 I w ( σ ) q ˙ 2 + 1 2 J σ ˙ 2 − V ( q , σ ) , H L = p q 2 2 I w ( σ ) + p σ 2 2 J + V . (535) L_L = \frac{1}{2}Iw(\sigma)\dot{q}^{\,2} + \frac{1}{2}J\dot{\sigma}^{\,2} - V(q,\sigma), \qquad H_L = \frac{p_q^2}{2Iw(\sigma)} + \frac{p_\sigma^2}{2J} + V. \tag*{(535)} L L = 2 1 I w ( σ ) q ˙ 2 + 2 1 J σ ˙ 2 − V ( q , σ ) , H L = 2 I w ( σ ) p q 2 + 2 J p σ 2 + V . (535)
Both inertias are positive. The momenta are p q = I w q ˙ p_q = Iw\dot{q} p q = I w q ˙ and p σ = J σ ˙ p_\sigma= J\dot{\sigma} p σ = J σ ˙ , so the Legendre map is invertible at every finite scalar value. Its Euler–Lagrange equations read
I w q ¨ + I w ′ σ ˙ q ˙ + V q = 0 , J σ ¨ − 1 2 I w ′ q ˙ 2 + V σ = 0. (536) Iw\ddot{q} + Iw'\dot{\sigma}\dot{q} + V_q = 0, \qquad J\ddot{\sigma} - \frac{1}{2}Iw'\dot{q}^{\,2} + V_\sigma= 0. \tag*{(536)} I w q ¨ + I w ′ σ ˙ q ˙ + V q = 0 , J σ ¨ − 2 1 I w ′ q ˙ 2 + V σ = 0. (536)
The two terms containing w ′ w' w ′ exchange energy between the retained gauge and scalar modes. Direct differentiation of H L H_L H L along the canonical equations gives H ˙ L = 0 \dot{H}_L = 0 H ˙ L = 0 . The Wick substitution t = − i τ t = -i\tau t = − i τ gives the positive kinetic Euclidean action S E = ∫ [ I w ( q ′ ) 2 / 2 + J ( σ ′ ) 2 / 2 + V ] d τ S_E = \int[Iw(q')^2/2 + J(\sigma')^2/2 + V]d\tau S E = ∫ [ I w ( q ′ ) 2 /2 + J ( σ ′ ) 2 /2 + V ] d τ . This is an explicit finite-mode continuation; it is not an analytic continuation theorem for the full fluctuating-coframe measure.
The kinetic mismatch and its correction. Choose flat canonical volume d q d σ dq\,d\sigma d q d σ and the symmetric divergence ordering. The corresponding differential expression is
H = − 1 2 I w ( σ ) ∂ q 2 − 1 2 J ∂ σ 2 + V . (537) H = -\frac{1}{2Iw(\sigma)}\partial_q^2 - \frac{1}{2J}\partial_\sigma^2 + V. \tag*{(537)} H = − 2 I w ( σ ) 1 ∂ q 2 − 2 J 1 ∂ σ 2 + V . (537)
Since w w w is independent of q q q , the gauge kinetic term is exactly − ∂ q [ ( 2 I w ) − 1 ∂ q ] -\partial_q[(2Iw)^{-1}\partial_q] − ∂ q [( 2 I w ) − 1 ∂ q ] . There is no missing scalar drift in this chosen flat-volume ordering. A constant gauge coefficient ν = ( 2 I w ( 0 ) ) − 1 \nu= (2Iw(0))^{-1} ν = ( 2 I w ( 0 ) ) − 1 instead changes the kinetic energy of a Fourier mode e i n q e^{inq} e in q at a fixed scalar value by
δ E n ( σ ) = n 2 2 I ( w ( σ ) − 1 − w ( 0 ) − 1 ) . (538) \delta E_n(\sigma) = \frac{n^2}{2I}\left(w(\sigma)^{-1} - w(0)^{-1}\right). \tag*{(538)} δ E n ( σ ) = 2 I n 2 ( w ( σ ) − 1 − w ( 0 ) − 1 ) . (538)
Thus agreement at the reference vacuum does not imply agreement on the interacting branch. On ∣ σ ∣ ≤ S |\sigma| \le S ∣ σ ∣ ≤ S , ∣ n ∣ ≤ N |n| \le N ∣ n ∣ ≤ N , the exact operator norm of this kinetic difference is
∥ δ H q ∥ = N 2 2 I ( e α S − 1 ) ( α ≥ 0 ) . (539) \lVert\delta H_q\rVert= \frac{N^2}{2I}(e^{\alpha S}-1) \quad(\alpha\ge0). \tag*{(539)} ∥ δ H q ∥ = 2 I N 2 ( e α S − 1 ) ( α ≥ 0 ) . (539)
This is a kinetic norm bound on a declared finite Fourier band, not a bound for full gravitational field evolution.
A positive transfer operator with the corrected coefficient. First restrict the scalar to ( − S , S ) (-S,S) ( − S , S ) with Dirichlet boundary conditions, keeping the angle periodic. This additional field-amplitude regulator makes the coefficient smooth and uniformly elliptic. Let K q = − ( 2 I w ) − 1 ∂ q 2 K_q=-(2Iw)^{-1}\partial_q^2 K q = − ( 2 I w ) − 1 ∂ q 2 and K σ = − ( 2 J ) − 1 ∂ σ 2 K_\sigma=-(2J)^{-1}\partial_\sigma^2 K σ = − ( 2 J ) − 1 ∂ σ 2 . Define
T a = e − a V / 2 e − a K σ / 2 e − a K q e − a K σ / 2 e − a V / 2 . (540) T_a=e^{-aV/2}e^{-aK_\sigma/2}e^{-aK_q}e^{-aK_\sigma/2}e^{-aV/2}. \tag*{(540)} T a = e − aV /2 e − a K σ /2 e − a K q e − a K σ /2 e − aV /2 . (540)
The gauge heat operator acts independently at each scalar value and has the explicit fiber kernel
p a ( q ′ − q ; σ ) = 1 2 π ∑ n ∈ Z e − a n 2 / ( 2 I w ( σ ) ) e i n ( q ′ − q ) . (541) p_a(q'-q;\sigma)=\frac{1}{2\pi}\sum_{n\in\mathbb{Z}}e^{-an^2/(2Iw(\sigma))}e^{in(q'-q)}. \tag*{(541)} p a ( q ′ − q ; σ ) = 2 π 1 n ∈ Z ∑ e − a n 2 / ( 2 I w ( σ )) e in ( q ′ − q ) . (541)
Its Gaussian image representation has variance a / ( I w ) a/(Iw) a / ( I w ) and includes the normalizing factor ( I w / ( 2 π a ) ) 1 / 2 (Iw/(2\pi a))^{1/2} ( I w / ( 2 π a ) ) 1/2 . Both the exponent and this field-dependent factor matter in comparison with a configuration integral.
Proposition 29 (Matched positive evolution on the retained branch). For fixed S < ∞ S<\infty S < ∞ , the operator (541) is an injective positive self-adjoint trace-class contraction. Its even periodic-time cylinder measure is reflection positive. For every t > 0 t>0 t > 0 , T t / n n → e − t H S T_{t/n}^n\to e^{-tH_S} T t / n n → e − t H S strongly, where H S H_S H S is the Dirichlet realization of (538). After restoring ℏ \hbar ℏ in the kinetic terms, its leading semiclassical symbol is the Hamiltonian in (536).
Proof. Put B = e − a K q / 2 e − a K σ / 2 e − a V / 2 B=e^{-aK_q/2}e^{-aK_\sigma/2}e^{-aV/2} B = e − a K q /2 e − a K σ /2 e − aV /2 . Every factor is an injective contraction, and T a = B ∗ B T_a=B^*B T a = B ∗ B . For the trace, expand in angle Fourier modes and let s a / 2 s_{a/2} s a /2 be the Dirichlet scalar heat kernel. Since V ≥ 0 V\ge0 V ≥ 0 ,
∥ B ∥ H S 2 ≤ ∫ − S S ∫ − S S ∣ s a / 2 ( σ ′ , σ ) ∣ 2 ∑ n ∈ Z e − a n 2 / ( 2 I w ( σ ′ ) ) d σ d σ ′ < ∞ . (542) \lVert B\rVert_{\mathrm{HS}}^2\le\int_{-S}^{S}\int_{-S}^{S}\lvert s_{a/2}(\sigma',\sigma)\rvert^2\sum_{n\in\mathbb{Z}}e^{-an^2/(2Iw(\sigma'))}\,d\sigma\,d\sigma'<\infty. \tag*{(542)} ∥ B ∥ HS 2 ≤ ∫ − S S ∫ − S S ∣ s a /2 ( σ ′ , σ ) ∣ 2 n ∈ Z ∑ e − a n 2 / ( 2 I w ( σ ′ )) d σ d σ ′ < ∞. (542)
Boundedness above and below of w w w on this interval controls the Gaussian sum, and the scalar heat operator is Hilbert–Schmidt. The same folding argument as (533) proves reflection positivity because the kernel is nonnegative and T a ≥ 0 T_a\ge0 T a ≥ 0 . The uniformly elliptic Dirichlet operator on the compact cylinder is self-adjoint, with the smooth Dirichlet functions as a core. The symmetric contraction product formula gives the stated time limit. Taking the principal symbol, or performing the Legendre calculation in (536), identifies the kinetic coefficient I w Iw I w .
The amplitude restriction can also be removed for this two-coordinate Hamiltonian. On S 1 × R S^{1} \times\mathbb{R} S 1 × R , close the nonnegative form
q [ ψ ] = ∫ [ ∣ ∂ q ψ ∣ 2 2 I w + ∣ ∂ σ ψ ∣ 2 2 J + V ∣ ψ ∣ 2 ] d q d σ (543) q[\psi] = \int\left[\frac{|\partial_q\psi|^2}{2I w}+\frac{|\partial_\sigma\psi|^2}{2J}+V|\psi|^2\right]\,dq\,d\sigma \tag*{(543)} q [ ψ ] = ∫ [ 2 I w ∣ ∂ q ψ ∣ 2 + 2 J ∣ ∂ σ ψ ∣ 2 + V ∣ ψ ∣ 2 ] d q d σ (543)
initially on smooth functions with compact scalar support. Such a closure exists by local ellipticity and positivity. The increasing union of the finite-interval Dirichlet form domains is dense in this closed form by its definition. Monotone convergence of closed forms therefore gives strong resolvent, and then strong positive-time semigroup, convergence as S → ∞ S \to\infty S → ∞ , with the natural zero extensions. This constructs a positive Hamiltonian after removing an amplitude cutoff at fixed spatial and modal content. It does not remove a spatial field regulator or prove the remaining relativistic reconstruction requirements.
What normalization of a Wilson temporal link changes. The same matching can be checked against the actual compact-link interaction rather than only a quadratic approximation. At fixed σ \sigma σ the temporal Wilson factor and its row integral are
r a ( δ ; σ ) = e − z ( 1 − cos δ ) , z = I w ( σ ) a , c a ( σ ) = 2 π e − z I 0 ( z ) . (544) r_a(\delta;\sigma)=e^{-z(1-\cos\delta)},\qquad z=\frac{I w(\sigma)}{a},\qquad c_a(\sigma)=2\pi e^{-z}I_0(z). \tag*{(544)} r a ( δ ; σ ) = e − z ( 1 − c o s δ ) , z = a I w ( σ ) , c a ( σ ) = 2 π e − z I 0 ( z ) . (544)
Here I n ( z ) = ( 2 π ) − 1 ∫ − π π e z cos δ e − i n δ d δ I_n(z)=(2\pi)^{-1}\int_{-\pi}^{\pi}e^{z\cos\delta}e^{-in\delta}\,d\delta I n ( z ) = ( 2 π ) − 1 ∫ − π π e z c o s δ e − in δ d δ is the modified Bessel coefficient. The normalized convolution r a / c a r_a/c_a r a / c a has positive Fourier multipliers I n ( z ) / I 0 ( z ) ≤ 1 I_n(z)/I_0(z)\leq1 I n ( z ) / I 0 ( z ) ≤ 1 . For fixed Fourier band and a compact scalar interval, their energy satisfies, uniformly as a ↓ 0 a\downarrow0 a ↓ 0 ,
− a − 1 log I n ( z ) I 0 ( z ) = n 2 2 I w + a n 2 4 I 2 w 2 + O ( a 2 ) . (545) -a^{-1}\log\frac{I_n(z)}{I_0(z)}=\frac{n^2}{2I w}+\frac{a n^2}{4I^2w^2}+O(a^2). \tag*{(545)} − a − 1 log I 0 ( z ) I n ( z ) = 2 I w n 2 + 4 I 2 w 2 a n 2 + O ( a 2 ) . (545)
The fixed-order Bessel expansion is given in [123], formula 10.40.1 of its online companion. For completeness, the first two logarithmic terms follow by expanding I n ( z ) = e z ( 2 π z ) − 1 / 2 [ 1 − ( 4 n 2 − 1 ) / ( 8 z ) + ( 4 n 2 − 1 ) ( 4 n 2 − 9 ) / ( 128 z 2 ) + O ( z − 3 ) ] I_n(z)=e^z(2\pi z)^{-1/2}[1-(4n^2-1)/(8z)+(4n^2-1)(4n^2-9)/(128z^2)+O(z^{-3})] I n ( z ) = e z ( 2 π z ) − 1/2 [ 1 − ( 4 n 2 − 1 ) / ( 8 z ) + ( 4 n 2 − 1 ) ( 4 n 2 − 9 ) / ( 128 z 2 ) + O ( z − 3 )] . Subtracting the logarithmic series for n = 0 n=0 n = 0 gives log ( I n / I 0 ) = − n 2 / ( 2 z ) − n 2 / ( 4 z 2 ) + O ( z − 3 ) \log(I_n/I_0)=-n^2/(2z)-n^2/(4z^2)+O(z^{-3}) log ( I n / I 0 ) = − n 2 / ( 2 z ) − n 2 / ( 4 z 2 ) + O ( z − 3 ) .
The unnormalized factor is not equivalent to the normalized one when σ \sigma σ fluctuates. Relative to the reference row integral, the additional multiplication factor can be written e − a U a ( σ ) e^{-aU_a(\sigma)} e − a U a ( σ ) , with
U a ( σ ) = − a − 1 log c a ( σ ) c a ( 0 ) = 1 2 a log w ( σ ) w ( 0 ) − 1 8 I ( 1 w ( σ ) − 1 w ( 0 ) ) + O ( a ) . (546) \begin{aligned}
U_a(\sigma)=-a^{-1}\log\frac{c_a(\sigma)}{c_a(0)} \\
&=\frac{1}{2a}\log\frac{w(\sigma)}{w(0)}-\frac{1}{8I}\left(\frac{1}{w(\sigma)}-\frac{1}{w(0)}\right)+O(a). \tag*{(546)}
\end{aligned} U a ( σ ) = − a − 1 log c a ( 0 ) c a ( σ ) = 2 a 1 log w ( 0 ) w ( σ ) − 8 I 1 ( w ( σ ) 1 − w ( 0 ) 1 ) + O ( a ) . (546)
For the exponential weight the leading term is − α σ / ( 2 a ) -\alpha\sigma/(2a) − α σ / ( 2 a ) . This term is field dependent and generally divergent at fixed nonzero σ \sigma σ as a ↓ 0 a\downarrow0 a ↓ 0 . It cannot be discarded as a constant partition normalization. One must specify the temporal measure or a compensating local factor. The compact-interval asymptotic statement does not assert a limiting vacuum for the unnormalized full field model. Section 10.13 specifies a symmetric endpoint arrangement and calculates its exact compensating factor, including the finite term left by leading-power normalization alone.
Quantum ordering is a separate declared choice. The metric kinetic interpretation uses d s Q 2 = I w ( σ ) d q 2 + J d σ 2 ds_Q^2=Iw(\sigma)dq^2+Jd\sigma^2 d s Q 2 = I w ( σ ) d q 2 + J d σ 2 . Its natural volume is proportional to w d q d σ \sqrt{w}\,dq\,d\sigma w d q d σ . Quantization by − Δ Q / 2 + V -\Delta_Q/2+V − Δ Q /2 + V , followed by the unitary map ψ ↦ w 1 / 4 ψ \psi\mapsto w^{1/4}\psi ψ ↦ w 1/4 ψ to flat volume, gives
U ( − 1 2 Δ Q + V ) U − 1 = H + V o r d , V o r d = ( log w ) ′ ′ 8 J + [ ( log w ) ′ ] 2 32 J . (547) U\left(-\tfrac{1}{2}\Delta_Q+V\right)U^{-1}=H+V_{\mathrm{ord}},\qquad V_{\mathrm{ord}}=\frac{(\log w)^{\prime\prime}}{8J}+\frac{[(\log w)^{\prime}]^2}{32J}. \tag*{(547)} U ( − 2 1 Δ Q + V ) U − 1 = H + V ord , V ord = 8 J ( log w ) ′′ + 32 J [( log w ) ′ ] 2 . (547)
To check the formula, the scalar part of Δ Q \Delta_Q Δ Q is J − 1 [ ∂ σ 2 + 1 2 ( log w ) ′ ∂ σ ] J^{-1}[\partial_\sigma^2+\tfrac{1}{2}(\log w)'\partial_\sigma] J − 1 [ ∂ σ 2 + 2 1 ( log w ) ′ ∂ σ ] ; conjugation removes the first derivative and gives the displayed potential. For w = e − α σ w=e^{-\alpha\sigma} w = e − α σ , V o r d = α 2 / ( 32 J ) V_{\mathrm{ord}}=\alpha^2/(32J) V ord = α 2 / ( 32 J ) is constant, so the two choices have identical energy gaps in this reduced example. For more general retained weights this need not be a constant. The classical action alone therefore does not settle the quantum ordering.
The nonabelian lattice extension and its exact Gauss constraints. The kinetic correction extends to the full compact link group at a fixed spatial graph. Keep e , σ e,\sigma e , σ as the gauge singlets specified in the finite branch, and choose positive temporal weights w ℓ ( e , σ ) w_\ell(e,\sigma) w ℓ ( e , σ ) from the same retained weight. For source and target vertices s ( ℓ ) , t ( ℓ ) s(\ell),t(\ell) s ( ℓ ) , t ( ℓ ) , introduce a temporal gauge multiplier A 0 , v A_{0,v} A 0 , v and the covariant velocity
D t U ℓ = U ˙ ℓ + A 0 , s ( ℓ ) U ℓ − U ℓ A 0 , t ( ℓ ) , L g = ∑ ℓ I ℓ w ℓ 2 ∥ U ℓ − 1 D t U ℓ ∥ 2 − V h . (548) D_tU_\ell=\dot{U}_\ell+A_{0,s(\ell)}U_\ell-U_\ell A_{0,t(\ell)},\qquad L_g=\sum_\ell\frac{I_\ell w_\ell}{2}\lVert U_\ell^{-1}D_tU_\ell\rVert^2-V_h. \tag*{(548)} D t U ℓ = U ˙ ℓ + A 0 , s ( ℓ ) U ℓ − U ℓ A 0 , t ( ℓ ) , L g = ℓ ∑ 2 I ℓ w ℓ ∥ U ℓ − 1 D t U ℓ ∥ 2 − V h . (548)
Add the declared positive kinetic terms for the singlet variables. The group norm is bi-invariant. Under a time-dependent site transformation, U ℓ ↦ g s U ℓ g t − 1 U_\ell\mapsto g_sU_\ell g_t^{-1} U ℓ ↦ g s U ℓ g t − 1 and A 0 , v ↦ g v A 0 , v g v − 1 − g ˙ v g v − 1 A_{0,v}\mapsto g_vA_{0,v}g_v^{-1}-\dot{g}_vg_v^{-1} A 0 , v ↦ g v A 0 , v g v − 1 − g ˙ v g v − 1 , the covariant velocity transforms as D t U ℓ ↦ g s ( D t U ℓ ) g t − 1 D_tU_\ell\mapsto g_s(D_tU_\ell)g_t^{-1} D t U ℓ ↦ g s ( D t U ℓ ) g t − 1 . Thus the spatial Wilson potential and the full kinetic expression are gauge invariant even when the weights fluctuate.
Let G = K V h G=K^{V_h} G = K V h act by this link transformation. On the cotangent phase space, choose the equivariant moment-map convention { G [ ξ ] , G [ ζ ] } = G [ [ ξ , ζ ] ] \{G[\xi],G[\zeta]\}=G[[\xi,\zeta]] { G [ ξ ] , G [ ζ ]} = G [[ ξ , ζ ]] . The components of G G G are the sums of outgoing left and incoming minus right link momenta, with this fixed momentum convention. Variation of A 0 A_0 A 0 imposes G v α = 0 G_v^\alpha=0 G v α = 0 .
Proposition 30 (Finite lattice Gauss closure and positive quantization). For the continuous-time lattice action (549), the Gauss constraints are first class and preserved by its Hamiltonian:
{ G v a , G u b } = δ v u f a b c G v c , { G v a , H h } = 0. (549) \{G_v^a,G_u^b\}=\delta_{vu}f^{ab}{}_cG_v^c,\qquad\{G_v^a,H_h\}=0. \tag*{(549)} { G v a , G u b } = δ v u f ab c G v c , { G v a , H h } = 0. (549)
At a fixed finite scalar/coframe amplitude box with Dirichlet boundary, replace the constant link kinetic coefficient by
K g = − ∑ ℓ 1 2 I ℓ w ℓ ( e , σ ) Δ K , ℓ . (550) K_g=-\sum_{\ell}\frac{1}{2I_{\ell}w_{\ell}(e,\sigma)}\Delta_{K,\ell}. \tag*{(550)} K g = − ℓ ∑ 2 I ℓ w ℓ ( e , σ ) 1 Δ K , ℓ . (550)
The symmetric transfer construction using K g K_g K g , the singlet kinetic operator, and V h V_h V h is positive and preserves the gauge-invariant Hilbert subspace. On smooth gauge vectors, the quantum Gauss generators close without a central term.
Proof. The cotangent lift of a site action has the canonical moment map obtained by pairing each link momentum with the corresponding infinitesimal link variation. Left and right actions on the same link commute, actions on different links commute, and the action at a vertex has Lie bracket [ ξ v , ζ v ] [\xi_v,\zeta_v] [ ξ v , ζ v ] . The canonical moment-map brackets consequently give the first equation in (549) . Invariance of the Hamiltonian gives the second. The covariant time action therefore preserves Gauss constraints; it is not an unconstrained evolution with an added numerical projection.
For quantization, let U ( g ) \mathcal{U}(g) U ( g ) be the unitary pullback representation on product Haar and singlet volume. Define G ^ v a = i d U ( T v a ) \widehat{G}^{a}_{v}=i\,d\mathcal{U}(T^{a}_{v}) G v a = i d U ( T v a ) , so that U ( e t T v a ) = e − i t G ^ v a \mathcal{U}(e^{tT^{a}_{v}})=e^{-it\widehat{G}^{a}_{v}} U ( e t T v a ) = e − i t G v a . Differentiating the representation identity gives [ G ^ v a , G ^ u b ] = i δ v u f a b c G ^ v c [\widehat{G}^{a}_{v},\widehat{G}^{b}_{u}]=i\delta_{vu}f^{ab}{}_{c}\widehat{G}^{c}_{v} [ G v a , G u b ] = i δ v u f ab c G v c on smooth vectors. This calculation contains no central term. The bi-invariant link Laplacian, the singlet weights, and V h V_h V h commute with the site action. Each transfer factor therefore commutes with the Haar projection Π G = ∫ G U ( g ) d g \Pi_G=\int_G\mathcal{U}(g)\,dg Π G = ∫ G U ( g ) d g . Its range is precisely the invariant subspace, equivalently the joint kernel of the infinitesimal generators for the connected group used here. Positivity follows from the same B ∗ B B^*B B ∗ B factorization as (540) . On the finite amplitude box the link diffusion coefficients have a common positive lower bound; the compact-group heat trace is uniformly finite. Together with the singlet Dirichlet heat trace this gives the trace-class estimate as in (542) .
This proves the full finite-graph internal Gauss algebra for this selected continuous-time action and its invariant quantum subspace. The lapse, shift, and dynamical spacetime coframe have not been introduced in this construction. Accordingly the hypersurface-deformation algebra of gravity, spatial regulator removal, and their quantum realization remain additional requirements.
Calculations with the same parameter registry. The following synthetic calculations fix I = J = β = K σ = 1 I=J=\beta=K_{\sigma}=1 I = J = β = K σ = 1 , α = 0.3 \alpha=0.3 α = 0.3 and σ ⋆ = 0 \sigma_{\star}=0 σ ⋆ = 0 before any comparison. Fourier modes − N , … , N -N,\ldots,N − N , … , N represent the angle; centered scalar differences use Dirichlet data at σ = ± 6 \sigma=\pm6 σ = ± 6 . The matched kinetic operator and the reference-frozen constant coefficient use identical potentials and identical numerical spaces. No parameter is refitted between them. The scalar mean in the table is the quantum ground-state expectation, not a classical thermal mean.
Table 5. Synthetic coupled spectrum on 200 scalar interior points and angle modes − 6 , … , 6 -6,\ldots,6 − 6 , … , 6 . Parameters, potential, and discretization are identical in both rows. Grid and Fourier refinements are recorded separately.
Figure 6. Synthetic retained-mode matching. Left: normalized Wilson-link energy error relative to the matched heat energy, for n = 1 n=1 n = 1 and fixed scalar values. Right: the field-dependent normalization term at σ = 1 \sigma=1 σ = 1 compared with its calculated leading divergence. These are tests of a finite shared-weight branch, not gravitational observations.
The calculation resolves a specific part of the earlier matching problem: on this retained interacting branch the common temporal weight fixes the classical inertia and the principal quantum kinetic coefficient, and a positive transfer operator implements both. It also identifies the measure normalization and ordering data still required. Gravitational coframe variation, the full nonlinear constraint algebra, and a spatial quantum continuum are not supplied by freezing those degrees of freedom.
10.8. Dynamical geometry, material clocks, and the physical inner product. The retained gauge–scalar interaction provides a controlled setting in which to compare temporal measures and kinetic operators. A dynamical spacetime requires an additional distinction. A coframe determines lengths and causal cones, whereas a material reference system assigns readings to events. These roles can interact without being identified. We now give the local geometric construction, an exactly soluble constrained quantum example, and a sufficient estimate for taking limits of constraint products.
Geometry and reference readings. Use a , b a, b a , b for Lorentz indices, μ , ν \mu, \nu μ , ν for spacetime indices, and I , J I, J I , J for labels of four scalar reference fields. With signature ( − , + , + , + ) (-,+,+,+) ( − , + , + , + ) , an invertible coframe and its matrix inverse satisfy
g μ ν = η a b e μ a e ν b , g μ ν = η a b e a μ e b ν , e μ a e b μ = δ b a . (551) g_{\mu\nu}=\eta_{ab}e^a_{\mu}e^b_{\nu},\qquad g^{\mu\nu}=\eta^{ab}e^\mu_a e^\nu_b,\qquad e^a_{\mu}e^\mu_b=\delta^a_b. \tag*{(551)} g μν = η ab e μ a e ν b , g μν = η ab e a μ e b ν , e μ a e b μ = δ b a . (551)
The inverse is a matrix inverse; it is not obtained by raising an index with an independently prescribed background metric. Four scalar fields X I X^I X I do not replace this coframe. In particular, if e a = d X a e^a=dX^a e a = d X a and the Jacobian has full rank, the X a X^a X a are local coordinates and g = η a b d X a d X b g=\eta_{ab}\,dX^a dX^b g = η ab d X a d X b is locally flat. Thus this identification excludes general curved metrics. The statement concerns the equality e a = d X a e^a=dX^a e a = d X a : a common nonconstant factor e a = Ω d X a e^a=\Omega\,dX^a e a = Ω d X a can give a curved, conformally flat metric. Also, d e a = 0 de^a=0 d e a = 0 alone is not a torsion condition when an independent connection is present, since T a = d e a + ω b a ∧ e b T^a=de^a+\omega^a_{\ b}\wedge e^b T a = d e a + ω b a ∧ e b .
To use T = X 0 T=X^0 T = X 0 as a local clock, its level sets must be spacelike. On an injective reference chart the relevant conditions and normal are
g μ ν ∂ μ T ∂ ν T < 0 , det ( ∂ μ X I ) ≠ 0 , g^{\mu\nu}\partial_\mu T\partial_\nu T<0,\qquad\det(\partial_\mu X^I)\ne0, g μν ∂ μ T ∂ ν T < 0 , det ( ∂ μ X I ) = 0 ,
n μ = − ∂ μ T − g α β ∂ α T ∂ β T , h μ ν = g μ ν + n μ n ν . (552) n_\mu=-\frac{\partial_\mu T}{\sqrt{-g^{\alpha\beta}\partial_\alpha T\partial_\beta T}},\qquad h_{\mu\nu}=g_{\mu\nu}+n_\mu n_\nu. \tag*{(552)} n μ = − − g α β ∂ α T ∂ β T ∂ μ T , h μν = g μν + n μ n ν . (552)
The sign of n n n is chosen to give the desired future orientation. The weaker condition d T ≠ 0 dT\ne0 d T = 0 allows both spacelike and null gradients: in Minkowski space T = x 1 T=x^1 T = x 1 and T = t + x 1 T=t+x^1 T = t + x 1 are elementary examples. If a clock ceases to define a chart, one may need another clock or another patch. Such failure does not by itself exclude the underlying physical configuration.
The inverse map x = x ( X ) x=x(X) x = x ( X ) defines a relational scalar and metric by
O r e l ( X ) = O ( x ( X ) ) , g I J r e l ( X ) = g μ ν ( x ( X ) ) ∂ x μ ∂ X I ∂ x ν ∂ X J . (553) O_{\mathrm{rel}}(X)=O(x(X)),\qquad g^{\mathrm{rel}}_{IJ}(X)=g_{\mu\nu}(x(X))\frac{\partial x^\mu}{\partial X^I}\frac{\partial x^\nu}{\partial X^J}. \tag*{(553)} O rel ( X ) = O ( x ( X )) , g I J rel ( X ) = g μν ( x ( X )) ∂ X I ∂ x μ ∂ X J ∂ x ν . (553)
They are invariant under simultaneous spacetime diffeomorphisms of the fields, within the chosen chart. A coframe component evaluated at the same relational event still transforms under local Lorentz rotations. Its use as a fully invariant observable requires a frame choice or an additional Lorentz-invariant construction. Relational quantum descriptions can be related by reduction maps under specified clock and spectral assumptions [97]; this does not assert an equivalence for every gravitational constraint system.
Reference matter and its gravitational response. A physical clock carries energy. One explicit extension of the two-derivative gravitational branch is a scalar reference action
S X = − ∫ − g [ 1 2 Q I J g μ ν ∂ μ X I ∂ ν X J + U X ( X ) ] d 4 x , Q I J = Q J I > 0. (554) S_X=-\int\sqrt{-g}\left[\frac{1}{2}Q_{IJ}g^{\mu\nu}\partial_\mu X^I\partial_\nu X^J+U_X(X)\right]d^4x,\qquad Q_{IJ}=Q_{JI}>0. \tag*{(554)} S X = − ∫ − g [ 2 1 Q I J g μν ∂ μ X I ∂ ν X J + U X ( X ) ] d 4 x , Q I J = Q J I > 0. (554)
Here Q Q Q is constant. In four spacetime dimensions, with c = ℏ = 1 c = \hbar= 1 c = ℏ = 1 , one may normalize X I X^I X I to have mass dimension one, Q Q Q to be dimensionless, and U X U_X U X to have dimension four. The readings X I X^I X I need not have the units of coordinate time or length. Positive Q Q Q gives ordinary healthy scalar kinetic terms; the Lorentz metric on coframes is not a target metric for these four scalars. This action is a declared matter extension. Identifying its coefficients with primitive mother data is a separate matching condition.
Variation gives Q I J □ g X J − ∂ I U X = 0 Q_{IJ}\Box_g X^J - \partial_I U_X = 0 Q I J □ g X J − ∂ I U X = 0 and
T μ ν X = Q I J ∂ μ X I ∂ ν X J − g μ ν ( 1 2 Q I J g α β ∂ α X I ∂ β X J + U X ) . (555) T^X_{\mu\nu}=Q_{IJ}\partial_\mu X^I\partial_\nu X^J-g_{\mu\nu}\left(\frac{1}{2}Q_{IJ}g^{\alpha\beta}\partial_\alpha X^I\partial_\beta X^J+U_X\right). \tag*{(555)} T μν X = Q I J ∂ μ X I ∂ ν X J − g μν ( 2 1 Q I J g α β ∂ α X I ∂ β X J + U X ) . (555)
The ADM Legendre transform gives the two constraint contributions
H X = Q I J P I P J 2 h + h 2 Q I J h i j ∂ i X I ∂ j X J + h U X , (556) \mathcal{H}_X=\frac{Q^{IJ}P_I P_J}{2\sqrt{h}}+\frac{\sqrt{h}}{2}Q_{IJ}h^{ij}\partial_iX^I\partial_jX^J+\sqrt{h}U_X, \tag*{(556)} H X = 2 h Q I J P I P J + 2 h Q I J h ij ∂ i X I ∂ j X J + h U X , (556)
H i , X = P I ∂ i X I . \mathcal{H}_{i,X}=P_I\partial_iX^I. H i , X = P I ∂ i X I .
Indeed P I = h Q I J ( X ˙ J − N i ∂ i X J ) / N P_I=\sqrt{h}Q_{IJ}(\dot{X}^J-N^i\partial_iX^J)/N P I = h Q I J ( X ˙ J − N i ∂ i X J ) / N , so substitution into P I X ˙ I − L X P_I\dot{X}^I-L_X P I X ˙ I − L X gives N H X + N i H i , X N\mathcal{H}_X+N^i\mathcal{H}_{i,X} N H X + N i H i , X . For the Einstein–Hilbert branch this stress enters M P 2 ( G μ ν + Λ g μ ν ) = T μ ν X + T μ ν o t h e r M_{\mathrm{P}}^2(G_{\mu\nu}+\Lambda g_{\mu\nu})=T^X_{\mu\nu}+T^{\mathrm{other}}_{\mu\nu} M P 2 ( G μν + Λ g μν ) = T μν X + T μν other . Reference matter therefore changes the constraints rather than merely labelling a pre-existing solution. Neither a free scalar action nor the ADM decomposition guarantees a globally valid clock, and the lapse, shift, and Lorentz gauge directions do not acquire independent physical momenta merely because a coframe is used.
Gauge averaging also affects how semiclassical geometry should be tested. Consider the four proper spatial rotations
R 0 = diag ( 1 , 1 , 1 ) , R 1 = diag ( 1 , − 1 , − 1 ) , R_0=\operatorname{diag}(1,1,1),\qquad R_1=\operatorname{diag}(1,-1,-1), R 0 = diag ( 1 , 1 , 1 ) , R 1 = diag ( 1 , − 1 , − 1 ) ,
R 2 = diag ( − 1 , 1 , − 1 ) , R 3 = diag ( − 1 , − 1 , 1 ) . (557) R_2=\operatorname{diag}(-1,1,-1),\qquad R_3=\operatorname{diag}(-1,-1,1). \tag*{(557)} R 2 = diag ( − 1 , 1 , − 1 ) , R 3 = diag ( − 1 , − 1 , 1 ) . (557)
Each coframe e ( r ) = diag ( 1 , R r ) e_{(r)}=\operatorname{diag}(1,R_r) e ( r ) = diag ( 1 , R r ) is invertible and induces the same Minkowski metric. Nevertheless, 1 4 ∑ r e ( r ) = diag ( 1 , 0 , 0 , 0 ) \frac{1}{4}\sum_r e_{(r)}=\operatorname{diag}(1,0,0,0) 4 1 ∑ r e ( r ) = diag ( 1 , 0 , 0 , 0 ) is degenerate. A degenerate mean of a gauge-dependent coframe therefore need not signal degenerate physical geometry. Metric, volume, or relational observables give appropriate geometric tests. A semiclassical Einstein equation also requires a specified state, renormalized stress, and control of geometric fluctuations: in general G ( ⟨ e ⟩ ) ≠ ⟨ G ( e ) ⟩ G(\langle e\rangle)\ne\langle G(e)\rangle G (⟨ e ⟩) = ⟨ G ( e )⟩ .
Ordering and a nontrivial test of constraint covariance. Coordinate covariance and a symmetric inner product restrict quantization but need not select it uniquely. On a compact smooth configuration manifold with metric γ \gamma γ and without boundary, the operators
H ξ = − ℏ 2 2 m ( Δ γ − ξ R γ ) + V , ξ ∈ R , (558) H_\xi=-\frac{\hbar^2}{2m}\left(\Delta_\gamma-\xi\mathcal{R}_\gamma\right)+V,\qquad\xi\in\mathbb{R}, \tag*{(558)} H ξ = − 2 m ℏ 2 ( Δ γ − ξ R γ ) + V , ξ ∈ R , (558)
have the same principal symbol and are coordinate covariant. For smooth real V V V , each has a self-adjoint realization with domain H 2 H^2 H 2 in L 2 ( d v o l γ ) L^2(\mathrm{dvol}_{\gamma}) L 2 ( dvol γ ) . The continuous curvature parameter is not fixed by these requirements. Earlier gravitational ordering analyses likewise distinguish ordering conditions from algebraic anomalies [34]. In a separate companion paper, Christodoulakis and Zanelli obtained a formally consistent constraint algebra using a minimal modification of canonical commutation relations [35]. That construction changes the algebraic input; it does not by itself provide a regulator-independent representation, a physical inner product, or a quantization of the present interacting mother action. Formal symmetry, self-adjoint realization, and a common invariant domain for products are separate requirements.
In the metric Einstein–Yang–Mills benchmark, let C i C_i C i denote the gauge-covariant momentum constraint, G A G_A G A the Gauss constraint with internal gauge label A A A , and V i = C i + A i A G A V_i=C_i+A_i^A G_A V i = C i + A i A G A the ordinary spatial diffeomorphism constraint. With G A = − D i E A i G_A=-D_iE_A^i G A = − D i E A i and the usual canonical Poisson bracket, the Hamiltonian bracket is
{ H [ N ] , H [ M ] } = C [ Z ] = V [ Z ] − G [ Z i A i ] , Z i = h i j ( N ∂ j M − M ∂ j N ) . (559) \{H[N],H[M]\}=C[Z]=V[Z]-G[Z^i A_i],\qquad Z^i=h^{ij}(N\partial_jM-M\partial_jN). \tag*{(559)} { H [ N ] , H [ M ]} = C [ Z ] = V [ Z ] − G [ Z i A i ] , Z i = h ij ( N ∂ j M − M ∂ j N ) . (559)
The right-hand side is a momentum constraint, with a field-dependent structure function. It is not another scalar Hamiltonian constraint. This distinction persists when a quantum representative and an ordering for the right-hand side are chosen.
Let Φ \Phi Φ be a common invariant dense test domain on which the required products exist. If C ^ α ψ = 0 \widehat{C}_{\alpha}\psi=0 C α ψ = 0 for every constraint, then [ C ^ α , C ^ β ] ψ = 0 [\widehat{C}_{\alpha},\widehat{C}_{\beta}]\psi=0 [ C α , C β ] ψ = 0 already follows algebraically. Thus a calculation only on this common kernel cannot establish the required spacetime covariance. One must also establish a nonzero physical space and examine the commutator on a suitable domain, or its matrix elements in controlled semiclassical states, with the correct structure functions and regulator estimates. A term that adds independent restrictions may change the physical sector; its effect cannot be decided by simply discarding it on the kernel.
Changing canonical commutators is one possible additional hypothesis, not a consequence of the ordering problem. It must define a consistent adjoint algebra and a representation. For example, the proposal [ q i , p j ] = i ℏ f ( q ) δ i j [q_i,p_j]=i\hbar f(q)\delta_{ij} [ q i , p j ] = i ℏ f ( q ) δ ij together with [ q i , q j ] = [ p i , p j ] = 0 [q_i,q_j]=[p_i,p_j]=0 [ q i , q j ] = [ p i , p j ] = 0 gives the Jacobi expression
[ p i , [ p j , q k ] ] + [ p j , [ q k , p i ] ] + [ q k , [ p i , p j ] ] = ℏ 2 f ( δ i k ∂ j f − δ j k ∂ i f ) . (560) [p_i,[p_j,q_k]]+[p_j,[q_k,p_i]]+[q_k,[p_i,p_j]]=\hbar^2 f(\delta_{ik}\partial_j f-\delta_{jk}\partial_i f). \tag*{(560)} [ p i , [ p j , q k ]] + [ p j , [ q k , p i ]] + [ q k , [ p i , p j ]] = ℏ 2 f ( δ ik ∂ j f − δ j k ∂ i f ) . (560)
For nonconstant positive f f f in dimension at least two this generally fails. Alternative regulators, counterterms, reduced quantization, or other constraint constructions must be assessed on their own assumptions.
From a positive form to physical states. For a unitary representation of a compact internal gauge group, normalized Haar averaging gives an orthogonal projection P i n v = ∫ U ( g ) d g P_{\mathrm{inv}}=\int U(g)\,\mathrm{d}g P inv = ∫ U ( g ) d g . Hence
q ( u , v ) = ⟨ u , P i n v v ⟩ , q ( u , u ) = ∥ P i n v u ∥ 2 ≥ 0 , q(u,v)=\langle u,P_{\mathrm{inv}}v\rangle,\qquad q(u,u)=\lVert P_{\mathrm{inv}}u\rVert^2\ge0, q ( u , v ) = ⟨ u , P inv v ⟩ , q ( u , u ) = ∥ P inv u ∥ 2 ≥ 0 ,
H p h y s ≃ H a u x / ker P i n v ≃ Ran P i n v . (561) \mathcal{H}_{\mathrm{phys}}\simeq\mathcal{H}_{\mathrm{aux}}/\ker P_{\mathrm{inv}}\simeq\operatorname{Ran}P_{\mathrm{inv}}. \tag*{(561)} H phys ≃ H aux / ker P inv ≃ Ran P inv . (561)
Nonzero auxiliary vectors can have zero physical norm. The quotient removes these vectors and gives a positive definite inner product. For noncompact constraints, a rigging form instead requires a test space, a justified integral or distributional prescription, positivity, compatibility with physical observables, and completion of the null quotient. Uniqueness is an additional question. The Giulini–Marolf theorem assumes a finite-dimensional Lie group, an appropriate L 1 L^1 L 1 test space invariant under their group algebra, and a nonzero averaging form; under those conditions any compatible rigging map, if it exists, is proportional to group averaging [80]. Convergence alone is not that theorem. The field-dependent structure functions in (560) do not provide an ordinary constraint group with Haar measure for all gravitational constraints.
An explicit noncompact example shows both the construction and its scope. Set ℏ = 1 \hbar=1 ℏ = 1 , choose a Hermitian d × d d\times d d × d matrix H s H_s H s , and take
H a u x = L 2 ( R T ) ⊗ C d , Φ = S ( R ) ⊗ C d , C = − i ∂ T + H s . (562) \mathcal{H}_{\mathrm{aux}}=L^2(\mathbb{R}_T)\otimes\mathbb{C}^d,\qquad\Phi=\mathcal{S}(\mathbb{R})\otimes\mathbb{C}^d,\qquad C=-i\partial_T+H_s. \tag*{(562)} H aux = L 2 ( R T ) ⊗ C d , Φ = S ( R ) ⊗ C d , C = − i ∂ T + H s . (562)
In an orthonormal eigenbasis H s ∣ ℓ ⟩ = E ℓ ∣ ℓ ⟩ H_s\lvert\ell\rangle=E_\ell\lvert\ell\rangle H s ∣ ℓ ⟩ = E ℓ ∣ ℓ ⟩ , use the unitary Fourier transform in T T T . Group averaging is then exactly
η ( u ) [ v ] = ∫ R d s 2 π ⟨ u , e i s C v ⟩ = ∑ ℓ = 1 d u ~ ℓ ( − E ℓ ) v ~ ℓ ( − E ℓ ) ‾ . (563) \eta(u)[v]=\int_{\mathbb{R}}\frac{\mathrm{d}s}{2\pi}\langle u,e^{isC}v\rangle=\sum_{\ell=1}^{d}\widetilde{u}_\ell(-E_\ell)\overline{\widetilde{v}_\ell(-E_\ell)}. \tag*{(563)} η ( u ) [ v ] = ∫ R 2 π d s ⟨ u , e i s C v ⟩ = ℓ = 1 ∑ d u ℓ ( − E ℓ ) v ℓ ( − E ℓ ) . (563)
For these test vectors the matrix coefficient is Schwartz in s s s , so the integral is absolutely convergent. Fourier inversion proves the displayed evaluation formula. It is positive semidefinite. Its null space consists precisely of the vectors for which all evaluations u ~ ℓ ( − E ℓ ) \widetilde{u}_\ell(-E_\ell) u ℓ ( − E ℓ ) vanish. Evaluation maps onto C d \mathbb{C}^d C d , since one may prescribe these finitely many values with Schwartz functions. The completed physical space is therefore C d \mathbb{C}^d C d . A corresponding history is Ψ ( T ) = e − i H s T χ \Psi(T)=e^{-iH_sT}\chi Ψ ( T ) = e − i H s T χ , and C Ψ = 0 C\Psi=0 C Ψ = 0 follows by direct differentiation. For χ ≠ 0 \chi\ne0 χ = 0 this history is not square integrable in the clock variable, even though its physical norm ∥ χ ∥ \lVert\chi\rVert ∥ χ ∥ is finite. This example realizes a deparametrized clock constraint; it does not realize the full gravitational hypersurface-deformation algebra.
The sign of an inner product must also be tied to its construction. For the relativistic mass-shell constraint C = p 0 2 − ∣ p ∣ 2 − m 2 C=p_0^2-\lvert\mathbf{p}\rvert^2-m^2 C = p 0 2 − ∣ p ∣ 2 − m 2 , m > 0 m>0 m > 0 , put ω p = ∣ p ∣ 2 + m 2 \omega_{\mathbf{p}}=\sqrt{\lvert\mathbf{p}\rvert^2+m^2} ω p = ∣ p ∣ 2 + m 2 . Averaging with δ ( C ) \delta(C) δ ( C ) gives the positive shell form
q ( u , v ) = ∫ d 3 p 2 ω p [ u ( ω p , p ) ‾ v ( ω p , p ) + u ( − ω p , p ) ‾ v ( − ω p , p ) ] . (564) q(u,v)=\int\frac{\mathrm{d}^3p}{2\omega_{\mathbf{p}}}\left[\overline{u(\omega_{\mathbf{p}},\mathbf{p})}v(\omega_{\mathbf{p}},\mathbf{p})+\overline{u(-\omega_{\mathbf{p}},\mathbf{p})}v(-\omega_{\mathbf{p}},\mathbf{p})\right]. \tag*{(564)} q ( u , v ) = ∫ 2 ω p d 3 p [ u ( ω p , p ) v ( ω p , p ) + u ( − ω p , p ) v ( − ω p , p ) ] . (564)
The conserved Klein–Gordon charge form has opposite signs on the two frequency sectors. These are different forms; positivity of one does not turn the other into a positive form. Enlarging a test space while retaining a negative-norm vector cannot remove that negative norm. Similarly, positivity of a temporal transfer operator B † B B^\dagger B B † B does not identify its Hilbert space with a physical constraint quotient without a reconstruction or intertwining map.
Spatial refinement and products of constraints. Clock resolution, temporal step size, spatial field truncation, and amplitude cutoffs are different parameters. Finer clock readings do not add missing spatial field modes. A relational resolution can describe spatial refinement only after an explicit map to physical geometry and observables has been given. Projective consistency, a renormalization group fixed point, and convergence of a chosen sequence are also distinct properties. A fixed point is one possible route to a continuum theory; failure to find one in a truncation is not a nonexistence proof.
Even convergent matrix elements are insufficient for constraint products. Let { e 0 , e 1 , … } \{e_0,e_1,\ldots\} { e 0 , e 1 , … } be the standard basis of ℓ 2 \ell^2 ℓ 2 and set
A n = ∣ e 0 ⟩ ⟨ e n ∣ + ∣ e n ⟩ ⟨ e 0 ∣ , B n = i ( ∣ e n ⟩ ⟨ e 0 ∣ − ∣ e 0 ⟩ ⟨ e n ∣ ) . (565) A_n = \lvert e_0\rangle\langle e_n\rvert+ \lvert e_n\rangle\langle e_0\rvert,\qquad B_n = i\left(\lvert e_n\rangle\langle e_0\rvert-\lvert e_0\rangle\langle e_n\rvert\right). \tag*{(565)} A n = ∣ e 0 ⟩ ⟨ e n ∣ + ∣ e n ⟩ ⟨ e 0 ∣ , B n = i ( ∣ e n ⟩ ⟨ e 0 ∣ − ∣ e 0 ⟩ ⟨ e n ∣ ) . (565)
Both self-adjoint sequences have norm one and converge weakly to zero. However, [ A n , B n ] = 2 i ( P 0 − P n ) [A_n,B_n]=2i(P_0-P_n) [ A n , B n ] = 2 i ( P 0 − P n ) converges weakly to 2 i P 0 2iP_0 2 i P 0 , where P j = ∣ e j ⟩ ⟨ e j ∣ P_j=\lvert e_j\rangle\langle e_j\rvert P j = ∣ e j ⟩ ⟨ e j ∣ . Thus the commutator of the limits differs from the limit of the commutators.
The following stronger estimate gives a sufficient replacement.
Proposition 31 (Control of commutators on a scale of spaces). Let V 2 ↪ V 1 ↪ V 0 V_2\hookrightarrow V_1\hookrightarrow V_0 V 2 ↪ V 1 ↪ V 0 be Banach spaces. Suppose A n , B n , A , B A_n,B_n,A,B A n , B n , A , B act consistently on both successive levels V 2 → V 1 V_2\to V_1 V 2 → V 1 and V 1 → V 0 V_1\to V_0 V 1 → V 0 , with every such operator norm at most M M M . If, on both levels, ∥ A n − A ∥ ≤ ϵ A ( n ) \lVert A_n-A\rVert\le\epsilon_A(n) ∥ A n − A ∥ ≤ ϵ A ( n ) and ∥ B n − B ∥ ≤ ϵ B ( n ) \lVert B_n-B\rVert\le\epsilon_B(n) ∥ B n − B ∥ ≤ ϵ B ( n ) , then
∥ [ A n , B n ] − [ A , B ] ∥ V 2 → V 0 ≤ 2 M ( ϵ A ( n ) + ϵ B ( n ) ) . (566) \lVert[A_n,B_n]-[A,B]\rVert_{V_2\to V_0}\le2M\left(\epsilon_A(n)+\epsilon_B(n)\right). \tag*{(566)} ∥[ A n , B n ] − [ A , B ] ∥ V 2 → V 0 ≤ 2 M ( ϵ A ( n ) + ϵ B ( n ) ) . (566)
Proof. Write A n B n − A B = ( A n − A ) B n + A ( B n − B ) A_nB_n-AB=(A_n-A)B_n+A(B_n-B) A n B n − A B = ( A n − A ) B n + A ( B n − B ) . Composition between the successive spaces bounds this difference by M ( ϵ A + ϵ B ) M(\epsilon_A+\epsilon_B) M ( ϵ A + ϵ B ) . Apply the same argument to B n A n − B A B_nA_n-BA B n A n − B A and use the triangle inequality. □ \square □
For changing Hilbert spaces the operators must first be transported by specified comparison maps. For unbounded gravitational constraints, the required domains and bounds remain to be established. The proposition is a conditional analytic criterion, rather than an estimate already proved for those constraints.
Quantization data and comparison of branches. The shared-weight example shows concretely why a path-integral measure and a kinetic expression must be matched. In general the quantization data also include the regulator, counterterms, boundary conditions, operator domains, and state preparation.
A measure alone need not select an ordering. Terms of order ℏ 2 \hbar^2 ℏ 2 need not be negligible at low energy, and equations of motion need not uniquely select a boundary state or a reference-field preparation.
Agreement of one relational expectation value is not physical equivalence. For example, H 1 = diag ( 0 , 1 ) H_1 = \operatorname{diag}(0,1) H 1 = diag ( 0 , 1 ) and H 2 = diag ( 0 , 2 ) H_2 = \operatorname{diag}(0,2) H 2 = diag ( 0 , 2 ) have the same ground-state expectation of σ z \sigma_z σ z but different excitation gaps. Exact equivalence requires a unitary map W W W that preserves the physical observable representation and evolution:
W π 1 ( O ) W − 1 = π 2 ( O ) , W U 1 ( t ) = U 2 ( t ) W . (567) W\pi_1(O)W^{-1} = \pi_2(O), \qquad WU_1(t) = U_2(t)W. \tag*{(567)} W π 1 ( O ) W − 1 = π 2 ( O ) , W U 1 ( t ) = U 2 ( t ) W . (567)
An effective low-energy comparison can instead specify a family of observables and states and bound their differences after comparison. Distinct vacua, superselection sectors, and inequivalent quantizations must be distinguished in that analysis. Their presence leaves the conditional branch constructions meaningful, while preventing a claim that the primitive mother data have already selected a unique physical quantum theory.
These results separate three tasks that a material-clock formulation must connect: dynamical geometric constraints, a positive physical state space, and spatial cutoff removal. The local reference construction and the clock model complete explicit parts of this connection. The full nonlinear gravitational operator algebra and its interacting continuum representation require additional constructions beyond the finite gauge sector.
Logarithmic coframes and integrability near degenerate geometry
An invertible coframe can approach a singular matrix while every matrix entry remains bounded. Inverse metric observables can therefore fail to be integrable even when the partition function is finite. Proposition 22 establishes this obstruction for the smooth positive matrix-entry density of the finite mother branch. Logarithmic coordinates locate the relevant tail precisely and provide an explicit alternative measure with finite inverse moments. The distinction between changing coordinates and changing the probability law is essential.
The polar change of variables. Work at one vertex in fixed dimensionless Euclidean coframe coordinates. If a dimensional coframe is used, it must first be divided by a specified reference scale. Every e ∈ G L + ( 4 , R ) e \in\mathrm{GL}^{+}(4,\mathbb{R}) e ∈ GL + ( 4 , R ) has the unique polar representation e = R exp S e = R\exp S e = R exp S , where R ∈ S O ( 4 ) R \in\mathrm{SO}(4) R ∈ SO ( 4 ) and S = S T S = S^{T} S = S T . Let d S F dS_F d S F be Lebesgue measure in the ten orthonormal Frobenius coordinates, namely S i i S_{ii} S ii and 2 S i j \sqrt{2}S_{ij} 2 S ij for i < j i<j i < j , and normalize Haar measure d R dR d R to have mass one. For eigenvalues λ 1 , … , λ 4 \lambda_1,\ldots,\lambda_4 λ 1 , … , λ 4 of S S S , the matrix-entry volume is
d e = J ( S ) d R d S F , J ( S ) = c 4 e 4 tr S ∏ i < j sinh ( λ i − λ j ) λ i − λ j . (568) de = J(S)\,dR\,dS_F,\qquad J(S) = c_4 e^{4\operatorname{tr} S}\prod_{i<j}\frac{\sinh(\lambda_i-\lambda_j)}{\lambda_i-\lambda_j}. \tag*{(568)} d e = J ( S ) d R d S F , J ( S ) = c 4 e 4 tr S i < j ∏ λ i − λ j sinh ( λ i − λ j ) . (568)
Here c 4 > 0 c_4>0 c 4 > 0 depends only on the volume normalizations, and each ratio at a repeated eigenvalue has value one. To derive the formula, first write e = R P e=RP e = R P with P > 0 P>0 P > 0 . At diagonal P P P with entries s i s_i s i , the differential d e = Ω P + d P de=\Omega P+dP d e = Ω P + d P , Ω T = − Ω \Omega^{\mathsf{T}}=-\Omega Ω T = − Ω , has Jacobian proportional to ∏ i < j ( s i + s j ) \prod_{i<j}(s_i+s_j) ∏ i < j ( s i + s j ) . Indeed the two off-diagonal entries for each pair ( i , j ) (i,j) ( i , j ) have determinant s i + s j s_i+s_j s i + s j . Next, the differential of P = exp S P=\exp S P = exp S multiplies the diagonal direction i i i by e λ i e^{\lambda_i} e λ i and the symmetric direction ( i , j ) (i,j) ( i , j ) by ( e λ i − e λ j ) / ( λ i − λ j ) (e^{\lambda_i}-e^{\lambda_j})/(\lambda_i-\lambda_j) ( e λ i − e λ j ) / ( λ i − λ j ) . Multiplying these factors and using e 2 λ i − e 2 λ j = 2 e λ i + λ j sinh ( λ i − λ j ) e^{2\lambda_i}-e^{2\lambda_j}=2e^{\lambda_i+\lambda_j}\sinh(\lambda_i-\lambda_j) e 2 λ i − e 2 λ j = 2 e λ i + λ j sinh ( λ i − λ j ) gives (569).
In particular, S = 0 S=0 S = 0 means e = R e=R e = R and det e = 1 \det e=1 det e = 1 . No inverse-coframe singularity occurs there. The singular values of e e e are e λ i e^{\lambda_i} e λ i , so the dangerous direction is λ min → − ∞ \lambda_{\min}\to-\infty λ m i n → − ∞ . For the illustrative matrix Gaussian d μ ∝ e − a tr ( e T e ) d e d\mu\propto e^{-a\operatorname{tr}(e^{\mathsf{T}}e)}de d μ ∝ e − a tr ( e T e ) d e , a > 0 a>0 a > 0 , conditioned on positive orientation, integration over the angular coordinates gives the joint logarithmic singular-value density
p a ( λ ) ∝ e − a ∑ i e 2 λ i e ∑ i λ i ∏ i < j ∣ e 2 λ i − e 2 λ j ∣ . (569) p_a(\lambda)\propto e^{-a\sum_i e^{2\lambda_i}}e^{\sum_i\lambda_i}\prod_{i<j}\left|e^{2\lambda_i}-e^{2\lambda_j}\right|. \tag*{(569)} p a ( λ ) ∝ e − a ∑ i e 2 λ i e ∑ i λ i i < j ∏ e 2 λ i − e 2 λ j . (569)
The ordinary real-symmetric spectral volume contributes ∣ Δ ( λ ) ∣ = ∏ i < j ∣ λ i − λ j ∣ |\Delta(\lambda)|=\prod_{i<j}|\lambda_i-\lambda_j| ∣Δ ( λ ) ∣ = ∏ i < j ∣ λ i − λ j ∣ ; these factors cancel the denominators in (569). The eigenvalues are not independent.
Keep λ 2 , λ 3 , λ 4 \lambda_2,\lambda_3,\lambda_4 λ 2 , λ 3 , λ 4 in compact disjoint intervals. As λ 1 → − ∞ \lambda_1\to-\infty λ 1 → − ∞ , the density in (570) is bounded above and below by positive multiples of e λ 1 e^{\lambda_1} e λ 1 , uniformly on smaller such intervals. Both ( det e ) − r (\det e)^{-r} ( det e ) − r and ∥ e − 1 ∥ r \lVert e^{-1}\rVert^r ∥ e − 1 ∥ r then have a lower bound proportional to e − r λ 1 e^{-r\lambda_1} e − r λ 1 . Consequently their integrals contain
∫ − ∞ − L e ( 1 − r ) λ 1 d λ 1 = ∞ , r ≥ 1. (570) \int_{-\infty}^{-L}e^{(1-r)\lambda_1}\,d\lambda_1=\infty,\qquad r\ge1. \tag*{(570)} ∫ − ∞ − L e ( 1 − r ) λ 1 d λ 1 = ∞ , r ≥ 1. (570)
Thus the logarithmic change of variables preserves, and makes visible, the original divergence. In the interacting mother measure the exact pullback is instead J ( S ) e − S h ( U , R e S , σ ) / θ J(S)e^{-S_h(U,Re^S,\sigma)/\theta} J ( S ) e − S h ( U , R e S , σ ) / θ , with the remaining fields and angular variables retained. It need not have the factorized spectral form of (570).
For comparison, a bounded neighborhood of S = 0 S=0 S = 0 gives an elementary finite integral. Ordering the four eigenvalues and integrating the Vandermonde polynomial yields
∫ [ 0 , ε ] 4 ∣ Δ ( λ ) ∣ d 4 λ = ε 10 1050 . (571) \int_{[0,\varepsilon]^4}|\Delta(\lambda)|\,d^4\lambda=\frac{\varepsilon^{10}}{1050}. \tag*{(571)} ∫ [ 0 , ε ] 4 ∣Δ ( λ ) ∣ d 4 λ = 1050 ε 10 . (571)
The power ten is the sum of four integration dimensions and the six Vandermonde factors. The coefficient follows from 4 ! ∫ 0 < x 1 < x 2 < x 3 < x 4 < 1 ∏ i < j ( x j − x i ) d 4 x = 1 / 1050 4!\int_{0<x_1<x_2<x_3<x_4<1}\prod_{i<j}(x_j-x_i)\,d^4x=1/1050 4 ! ∫ 0 < x 1 < x 2 < x 3 < x 4 < 1 ∏ i < j ( x j − x i ) d 4 x = 1/1050 . For r ≥ 0 r\ge0 r ≥ 0 , inserting e − r ∑ i λ i e^{-r\sum_i\lambda_i} e − r ∑ i λ i bounds the integral between e − 4 r ε ε 10 / 1050 e^{-4r\varepsilon}\varepsilon^{10}/1050 e − 4 r ε ε 10 /1050 and ε 10 / 1050 \varepsilon^{10}/1050 ε 10 /1050 . Even without the Vandermonde factor the bounded integrand is integrable on this cube. A weight vanishing at S = 0 S=0 S = 0 is therefore unnecessary for inverse-moment control and cannot repair the negative tail by its behavior at the origin alone.
A different measure with explicit inverse moments. Consider the declared logarithmic Gaussian law
d ν a ( R , S ) = ( a π ) 5 e − a tr S 2 d S F d R , a > 0. (572) d\nu_a(R,S)=\left(\frac{a}{\pi}\right)^5 e^{-a\operatorname{tr}S^2}\,dS_F\,dR,\qquad a>0. \tag*{(572)} d ν a ( R , S ) = ( π a ) 5 e − a tr S 2 d S F d R , a > 0. (572)
It is not the pullback of the matrix Gaussian above. Relative to matrix-entry volume its density contains J ( S ) − 1 J(S)^{-1} J ( S ) − 1 . This change suppresses nearly singular configurations rather than merely renaming them. The Euclidean construction is invariant under fixed left and right S O ( 4 ) \mathrm{SO}(4) SO ( 4 ) rotations. It is not, on that account, a Lorentz-invariant probability law or a solution of the gravitational constraints.
Proposition 32 (Inverse moments of a logarithmic Gaussian coframe). For (573) and every r ≥ 0 r\geq0 r ≥ 0 ,
E ν a ( det e ) − r = e r 2 / a , E ν a ∥ e − 1 ∥ F r ≤ 2 r + 5 e r 2 / ( 2 a ) < ∞ . (573) \mathbb{E}_{\nu_a}(\det e)^{-r}=e^{r^2/a},\qquad\mathbb{E}_{\nu_a}\lVert e^{-1}\rVert_F^r\leq2^{r+5}e^{r^2/(2a)}<\infty. \tag*{(573)} E ν a ( det e ) − r = e r 2 / a , E ν a ∥ e − 1 ∥ F r ≤ 2 r + 5 e r 2 / ( 2 a ) < ∞. (573) If d ν a , W = Z W − 1 e − W d ν a d\nu_{a,W}=Z_W^{-1}e^{-W}d\nu_a d ν a , W = Z W − 1 e − W d ν a , where W ≥ − C W\geq-C W ≥ − C and 0 < Z W < ∞ 0<Z_W<\infty 0 < Z W < ∞ , each corresponding moment is at most e C / Z W e^C/Z_W e C / Z W times its value or bound under ν a \nu_a ν a .
Proof. The ten orthonormal matrix coordinates give the partition integral ( π / a ) 5 (\pi/a)^5 ( π / a ) 5 . Since det e = e tr S \det e=e^{\operatorname{tr}S} det e = e tr S , complete the square:
− a tr S 2 − r tr S = − a tr ( S + r 2 a I 4 ) 2 + r 2 a . (574) -a\operatorname{tr}S^2-r\operatorname{tr}S=-a\operatorname{tr}\left(S+\frac{r}{2a}I_4\right)^2+\frac{r^2}{a}. \tag*{(574)} − a tr S 2 − r tr S = − a tr ( S + 2 a r I 4 ) 2 + a r 2 . (574)
Translation invariance of d S F dS_F d S F proves the determinant formula. Moreover ∥ e − 1 ∥ F ≤ 2 e ∥ S ∥ F \lVert e^{-1}\rVert_F\leq2e^{\lVert S\rVert_F} ∥ e − 1 ∥ F ≤ 2 e ∥ S ∥ F and r ρ ≤ a ρ 2 / 2 + r 2 / ( 2 a ) r\rho\leq a\rho^2/2+r^2/(2a) r ρ ≤ a ρ 2 /2 + r 2 / ( 2 a ) . Integrating the resulting Gaussian with precision a / 2 a/2 a /2 rather than a a a contributes 2 5 2^5 2 5 , giving the second bound. Finally e − W ≤ e C e^{-W}\leq e^C e − W ≤ e C proves the assertion for the tilt. □
The parameter a a a remains free: integrability holds for every a > 0 a>0 a > 0 . For example, E tr S 2 = 5 / a \mathbb{E}\operatorname{tr}S^2=5/a E tr S 2 = 5/ a and Var ( log det e ) = 2 / a \operatorname{Var}(\log\det e)=2/a Var ( log det e ) = 2/ a offer possible calibration relations, but do not determine their observed values. A dynamical realization of the un-tilted law is equally explicit. If B s F B_s^F B s F is Brownian motion in the ten orthonormal symmetric coordinates, then
d S s = − 2 θ a S s d s + 2 θ d B s F , θ > 0 , (575) dS_s=-2\theta aS_s\,ds+\sqrt{2\theta}\,dB_s^F,\qquad\theta>0, \tag*{(575)} d S s = − 2 θ a S s d s + 2 θ d B s F , θ > 0 , (575)
has invariant density proportional to e − a tr S 2 e^{-a\operatorname{tr}S^2} e − a tr S 2 . An independent rotation diffusion with invariant Haar measure gives (573). Every finite-time sample has an invertible coframe. Here s s s is an auxiliary diffusion parameter; its identification with a physical clock requires further dynamics.
These estimates also explain the limit of a finite-lattice construction. For N v N_v N v vertices with marginal law (573), no independence assumption is needed for the union bound
P { min v det e v < δ } ≤ min { 1 , N v exp [ − a 4 ( log 1 δ ) 2 ] } , 0 < δ < 1. (576) \mathbb{P}\left\{\min_v\det e_v<\delta\right\}\leq\min\left\{1,N_v\exp\left[-\frac{a}{4}\left(\log\frac{1}{\delta}\right)^2\right]\right\},\qquad0<\delta<1. \tag*{(576)} P { v min det e v < δ } ≤ min { 1 , N v exp [ − 4 a ( log δ 1 ) 2 ] } , 0 < δ < 1. (576)
Indeed Markov’s inequality gives N v δ e r 2 / a N_v\delta e^{r^2/a} N v δ e r 2 / a , minimized at r = ( a / 2 ) log ( 1 / δ ) r=(a/2)\log(1/\delta) r = ( a /2 ) log ( 1/ δ ) . The bound deteriorates as N v N_v N v grows. This loss of an upper bound is not itself evidence of degeneration. § 14.12 compares different joint laws with the same marginals and proves a bound for a fixed list of averaged observations. § 14.10 embeds the logarithmic reference in a finite locally rotation-covariant interaction, retaining the interaction-dependent normalization in its moment estimates. Individual inverse moments therefore do not establish a uniform nondegenerate continuum geometry; spatial regularity, correlations, and regulator-uniform estimates remain necessary.
Finite-dimensional numerical illustration. Table 6 gives synthetic values for a = 1 a=1 a = 1 . The determinant moments are exact. For the second Frobenius moment, one can reduce the spectral integral to one-dimensional Gaussian quadratures. Define the antisymmetric matrix, with indices 0 ≤ i , j ≤ 3 0\le i,j\le3 0 ≤ i , j ≤ 3 ,
Table 6. Synthetic inverse moments for a single logarithmic Gaussian coframe with a = 1 a = 1 a = 1 . The matrix-entry Gaussian has divergent values in all four rows. Numerical values refer to the different law ν 1 \nu_1 ν 1 .
A i j ( t ) = ∫ x < y ( x i y j − y i x j ) e − x 2 − y 2 ( 1 + t e − 2 x ) ( 1 + t e − 2 y ) d x d y . (577) A_{ij}(t)=\int_{x<y}(x^iy^j-y^ix^j)e^{-x^2-y^2}(1+te^{-2x})(1+te^{-2y})\,dx\,dy. \tag*{(577)} A ij ( t ) = ∫ x < y ( x i y j − y i x j ) e − x 2 − y 2 ( 1 + t e − 2 x ) ( 1 + t e − 2 y ) d x d y . (577)
Expansion of the Vandermonde determinant on the ordered chamber gives Z ( t ) = 4 ! Pf A ( t ) \mathcal{Z}(t)=4!\operatorname{Pf}A(t) Z ( t ) = 4 ! Pf A ( t ) , where Pf A = A 01 A 23 − A 02 A 13 + A 03 A 12 \operatorname{Pf}A=A_{01}A_{23}-A_{02}A_{13}+A_{03}A_{12} Pf A = A 01 A 23 − A 02 A 13 + A 03 A 12 . Thus Z ( 0 ) = 3 π / 2 \mathcal{Z}(0)=3\pi/2 Z ( 0 ) = 3 π /2 and
E ν 1 ∥ e − 1 ∥ F 2 = d d t log Z ( t ) ∣ t = 0 ≃ 31.1192891. (578) \mathbb{E}_{\nu_1}\lVert e^{-1}\rVert_F^2=\left.\frac{d}{dt}\log\mathcal{Z}(t)\right|_{t=0}\simeq31.1192891. \tag*{(578)} E ν 1 ∥ e − 1 ∥ F 2 = d t d log Z ( t ) t = 0 ≃ 31.1192891. (578)
The substitution x = u − v x=u-v x = u − v , y = u + v y=u+v y = u + v , v > 0 v>0 v > 0 , factors each polynomial term into a full-line and a half-line Gaussian moment. The derivative in (578) is justified by Gaussian integrability; Z ( t ) \mathcal{Z}(t) Z ( t ) is a generating polynomial near zero, not a claim of a positive density for negative t t t .
An independent calculation in matrix coordinates uses twelve scrambled Sobol replicates of 2 17 2^{17} 2 17 points each. Writing S = ( q / 2 ) I 4 + S 0 S=(q/2)I_4+S_0 S = ( q /2 ) I 4 + S 0 , with tr S 0 = 0 \operatorname{tr}S_0=0 tr S 0 = 0 , q ∼ N ( 0 , 1 / 2 ) q\sim N(0,1/2) q ∼ N ( 0 , 1/2 ) independent of S 0 S_0 S 0 , integrates the trace exactly: the factor for the r r r th inverse norm is e r 2 / 16 e^{r^2/16} e r 2 /16 . The remaining nine-dimensional calculation gives 4.9246643 for the first moment and 31.1112707 for the second. Replicate-based standard error estimates are 0.0000451 and 0.002582, respectively. The second-moment difference from (578) is −0.0080184, approximately 0.026% or 3.1 estimated standard errors. The replicate estimates do not provide rigorous integration bounds. Figure 7 illustrates the different negative tails responsible for the integrability change.
Figure 7. Synthetic conditional spectral slices, weighted by e − 2 λ 1 e^{-2\lambda_1} e − 2 λ 1 , with the other three logarithmic singular values fixed at log ( 1 , 2 , 3 ) \log(1,2,3) log ( 1 , 2 , 3 ) . Each unweighted slice is normalized to one at λ 1 = − 1 \lambda_1=-1 λ 1 = − 1 . The matrix-entry Gaussian grows after inverse-square weighting toward λ 1 = − ∞ \lambda_1=-\infty λ 1 = − ∞ , whereas the logarithmic Gaussian decays. These slices are neither marginal densities nor observational data.
Constraint reduction and quantum consistency beyond moment bounds
Finite inverse moments make geometric observables integrable. They do not determine the number of propagating degrees of freedom or the sign of a physical inner product. Those questions concern the action, its constraints, and its quantization. A finite mechanical example separates the constraint issue from the measure issue.
Second-class constraints can remove a higher-derivative mode. On a regular irreducible constraint stratum with N N N canonical pairs, F F F independent first-class constraints and S c S_c S c independent second-class constraints, the reduced number of canonical pairs is
N p h y s = N − F − 1 2 S c . (579) N_{\mathrm{phys}} = N - F - \frac{1}{2}S_c. \tag*{(579)} N phys = N − F − 2 1 S c . (579)
Each first-class constraint removes one constraint direction and one gauge direction. Each second-class constraint removes one phase-space dimension. A degenerate velocity Hessian produces primary constraints, but preservation can generate further constraints or fix multipliers. The entire Dirac–Bergmann procedure, on a specified rank stratum, is needed before applying (580). The higher-derivative scalar–tensor analysis of Langlois and Noui provides a field-theoretic example of this distinction [108].
Consider the mechanical Lagrangian, with ω x , ω y > 0 \omega_x,\omega_y > 0 ω x , ω y > 0 ,
L = 1 2 ( q ˙ + x ¨ ) 2 + 1 2 x ˙ 2 − 1 2 ω x 2 x 2 − 1 2 ω y 2 ( q + x ˙ ) 2 . (580) L = \frac{1}{2}(\dot{q}+\ddot{x})^2 + \frac{1}{2}\dot{x}^2 - \frac{1}{2}\omega_x^2x^2 - \frac{1}{2}\omega_y^2(q+\dot{x})^2. \tag*{(580)} L = 2 1 ( q ˙ + x ¨ ) 2 + 2 1 x ˙ 2 − 2 1 ω x 2 x 2 − 2 1 ω y 2 ( q + x ˙ ) 2 . (580)
Introduce Q = x ˙ Q=\dot{x} Q = x ˙ and a multiplier enforcing this identity. After eliminating the multiplier’s auxiliary canonical pair, the coordinates are ( x , Q , q ) (x,Q,q) ( x , Q , q ) and
H = 1 2 p q 2 − 1 2 Q 2 + p x Q + 1 2 ω x 2 x 2 + 1 2 ω y 2 ( q + Q ) 2 , ϕ = p Q − p q = 0. (581) H = \frac{1}{2}p_q^2 - \frac{1}{2}Q^2 + p_xQ + \frac{1}{2}\omega_x^2x^2 + \frac{1}{2}\omega_y^2(q+Q)^2,\qquad\phi= p_Q-p_q=0. \tag*{(581)} H = 2 1 p q 2 − 2 1 Q 2 + p x Q + 2 1 ω x 2 x 2 + 2 1 ω y 2 ( q + Q ) 2 , ϕ = p Q − p q = 0. (581)
The two momenta p Q , p q p_Q,p_q p Q , p q coincide because the Hessian in ( Q ˙ , q ˙ ) (\dot{Q},\dot{q}) ( Q ˙ , q ˙ ) is the rank-one matrix with every entry equal to one. Canonical brackets give
χ = { ϕ , H } = Q − p x , { ϕ , χ } = − 1 , N p h y s = 3 − 0 − 2 / 2 = 2. (582) \chi= \{\phi,H\}=Q-p_x,\qquad\{\phi,\chi\}=-1,\qquad N_{\mathrm{phys}}=3-0-2/2=2. \tag*{(582)} χ = { ϕ , H } = Q − p x , { ϕ , χ } = − 1 , N phys = 3 − 0 − 2/2 = 2. (582)
Preserving χ \chi χ fixes the multiplier of ϕ \phi ϕ ; it generates no gauge freedom. On the constraint surface put y = q + Q y=q+Q y = q + Q . The canonical one-form reduces to p x d x + p q d y p_x\,dx+p_q\,dy p x d x + p q d y , and the Hamiltonian becomes
H r e d = 1 2 ( p x 2 + p q 2 + ω x 2 x 2 + ω y 2 y 2 ) . (583) H_{\mathrm{red}}=\frac{1}{2}(p_x^2+p_q^2+\omega_x^2x^2+\omega_y^2y^2). \tag*{(583)} H red = 2 1 ( p x 2 + p q 2 + ω x 2 x 2 + ω y 2 y 2 ) . (583)
Two second-class constraints remove the extra pair and leave two stable oscillators, although there are no first-class constraints. For a coframe–connection action the analogous calculation must retain the independent connection until its constraints are classified. Elimination is justified by a map preserving the reduced symplectic structure and dynamics, not solely by equality of expressions after setting constraints to zero. Quantization adds domain and ordering requirements to this classical equivalence.
Real characteristic speeds are not a complete evolution estimate. Strong hyperbolicity of a first-order reduction requires a uniformly controlled diagonalization of its principal symbol, or an appropriate uniformly positive symmetrizer, in addition to real characteristic speeds. The constant system ∂ t u = A ∂ x u \partial_tu=A\partial_xu ∂ t u = A ∂ x u with
A = ( 1 1 0 1 ) , e i k A t = e i k t [ I + i k t ( 0 1 0 0 ) ] (584) A=\begin{pmatrix}1&1\\0&1\end{pmatrix},\qquad e^{ikAt}=e^{ikt}\left[I+ikt\begin{pmatrix}0&1\\0&0\end{pmatrix}\right] \tag*{(584)} A = ( 1 0 1 1 ) , e ik A t = e ik t [ I + ik t ( 0 0 1 0 ) ] (584)
has real eigenvalues but growth proportional to ∣ k ∣ t |k|t ∣ k ∣ t , precluding a frequency-uniform estimate at the same Sobolev order. Gauge-dependent characteristic speeds can occur in legitimate gauge evolution equations. Their presence alone neither removes a mode nor proves a physical instability. Gauge fixing, the constraint subsystem, principal-symbol estimates, and boundary conditions must be considered together.
A vanishing physical expectation can miss an anomaly. Suppose provisionally that a physical subspace is represented by an orthogonal projection P P P in an auxiliary Hilbert space. Even vanishing of all compressed matrix elements only states P A P = 0 PAP=0 P A P = 0 . For example,
P = ( 1 0 0 0 ) , A = ( 0 1 1 0 ) , P A P = 0 , A ( 1 0 ) = ( 0 1 ) . (585) P=\begin{pmatrix}1&0\\0&0\end{pmatrix},\qquad A=\begin{pmatrix}0&1\\1&0\end{pmatrix},\qquad PAP=0,\qquad A\begin{pmatrix}1\\0\end{pmatrix}=\begin{pmatrix}0\\1\end{pmatrix}. \tag*{(585)} P = ( 1 0 0 0 ) , A = ( 0 1 1 0 ) , P A P = 0 , A ( 1 0 ) = ( 0 1 ) . (585)
The operator takes a physical vector outside the physical subspace. Moreover, commutators vanish algebraically on a common constraint kernel whenever the products exist, and an empty physical space satisfies every such test vacuously. In refined algebraic quantization physical states can instead be distributions, as in (562) ; an auxiliary orthogonal projection onto their kernel is then not available.
A substantive sufficient estimate uses a common domain and the correct classical structure functions. For the benchmark (559) , write, at fixed ℏ > 0 \hbar>0 ℏ > 0 ,
A h [ N , M ] = ( i ℏ ) − 1 [ H ^ h [ N ] , H ^ h [ M ] ] − C [ Z ] ^ h , ∥ A h [ N , M ] ψ ∥ V 0 ≤ ϵ h ( N , M ) ∥ ψ ∥ V 2 , ϵ h ( N , M ) ⟶ 0. (586) A_h[N,M]=(i\hbar)^{-1}\left[\widehat{H}_h[N],\widehat{H}_h[M]\right]-\widehat{C[Z]}_h,
\qquad
\left\|A_h[N,M]\psi\right\|_{V_0}\leq\epsilon_h(N,M)\left\|\psi\right\|_{V_2},\qquad\epsilon_h(N,M)\longrightarrow0. \tag*{(586)} A h [ N , M ] = ( i ℏ ) − 1 [ H h [ N ] , H h [ M ] ] − C [ Z ] h , ∥ A h [ N , M ] ψ ∥ V 0 ≤ ϵ h ( N , M ) ∥ ψ ∥ V 2 , ϵ h ( N , M ) ⟶ 0. (586)
The ordering of C [ Z ] ^ h \widehat{C[Z]}_h C [ Z ] h must include the metric dependence of Z Z Z . Comparison maps are needed when the regulated spaces differ. Proposition 31 supplies one sufficient product estimate, but not these gravitational bounds. This off-shell test must be accompanied by a nontrivial positive physical construction with the intended classical observables. An O ( ℏ ) O(\hbar) O ( ℏ ) remainder can describe a semiclassical approximation; at fixed ℏ \hbar ℏ it is not a vanishing cutoff error or a proof of exact anomaly cancellation.
Group averaging has similarly specific hypotheses. Under the Giulini–Marolf assumptions, including their L 1 L^1 L 1 test-space condition, invariance under the relevant group algebra, and a nonzero averaging form, a compatible rigging map is unique up to normalization if it exists [80]. Their integrability requirements include the specified modularly weighted measures for nonunimodular groups. Mere convergence of one Haar integral, or an arbitrary convergent regularization, does not supply these hypotheses. Louko and Molgado exhibit a convergent indefinite averaging form for the ( 1 , 1 ) (1,1) ( 1 , 1 ) oscillator representation; under the uniqueness hypotheses it rules out a rigging map for their chosen test space and observable structure [115]. This is not a universal nonexistence statement for every quantization of that classical system. Likewise, the relational trinity of [97] uses particular clock–system constraints and spectral assumptions; it is not a theorem for all ordinary Lie-group constraints, much less the full field-dependent gravitational algebra.
Higher-derivative poles, physical positivity, and effective theory. The BFV formulation organizes gauge constraints and their ghost sector. It does not by itself remove every negative-residue propagating mode. In the quadratic-gravity calculation of Bellorin, Bórquez, and Droguett, the stated sign choice gives a massive transverse-traceless tensor and one scalar with negative norm, while the massive transverse vector has positive norm [14]. These are statements about that calculation; classifying the present model requires its own constraint reduction and physical observables.
A fakeon prescription changes the quantization of selected poles rather than deriving their removal from the classical constraint algebra. Fakeons can contribute internally to tree and loop amplitudes, but are excluded from physical asymptotic states and the corresponding unitarity cuts. The construction uses a specified nonanalytic continuation, with average continuation across fake thresholds [3]. Its physical cutting equations sum over physical states. It therefore does not dispense with positive physical probabilities, nor does it supply a positive physical space for an arbitrary mother action. It also does not preserve ordinary microscopic causality: the one-loop gravity analysis explicitly finds a causality violation associated with the fakeon scale [4]. A perturbative fakeon construction and a reflection-positive Euclidean reconstruction impose different conditions; neither may be inferred merely by choosing a propagator pole.
The low-energy alternative admits a simple quantitative statement. For a Euclidean scalar polarization with G ℓ ( p ) = [ p 2 ( 1 + ℓ 2 p 2 ) ] − 1 G_{\ell}(p) = [p^{2}(1+\ell^{2}p^{2})]^{-1} G ℓ ( p ) = [ p 2 ( 1 + ℓ 2 p 2 ) ] − 1 , let G ℓ ( n ) ( p ) = p − 2 ∑ j = 0 n ( − ℓ 2 p 2 ) j G_{\ell}^{(n)}(p) = p^{-2}\sum_{j=0}^{n}(-\ell^{2}p^{2})^{j} G ℓ ( n ) ( p ) = p − 2 ∑ j = 0 n ( − ℓ 2 p 2 ) j . At p 2 > 0 p^{2}>0 p 2 > 0 , the finite geometric-series identity gives
G ℓ − G ℓ ( n ) = ( − ℓ 2 p 2 ) n + 1 p 2 ( 1 + ℓ 2 p 2 ) , ∣ G ℓ − G ℓ ( n ) ∣ ∣ G ℓ ∣ ≤ ( ℓ 2 Λ 2 ) n + 1 , 0 < p 2 ≤ Λ 2 . (587) G_{\ell}-G_{\ell}^{(n)}=\frac{(-\ell^{2}p^{2})^{n+1}}{p^{2}(1+\ell^{2}p^{2})},\qquad\frac{\lvert G_{\ell}-G_{\ell}^{(n)}\rvert}{\lvert G_{\ell}\rvert}\leq(\ell^{2}\Lambda^{2})^{n+1},\qquad0<p^{2}\leq\Lambda^{2}. \tag*{(587)} G ℓ − G ℓ ( n ) = p 2 ( 1 + ℓ 2 p 2 ) ( − ℓ 2 p 2 ) n + 1 , ∣ G ℓ ∣ ∣ G ℓ − G ℓ ( n ) ∣ ≤ ( ℓ 2 Λ 2 ) n + 1 , 0 < p 2 ≤ Λ 2 . (587)
The bound is useful when ℓ Λ < 1 \ell\Lambda<1 ℓ Λ < 1 . It bounds this propagator expansion, not an interacting amplitude without further estimates, and not the negative part of a physical inner product. Treating higher-derivative operators perturbatively in an effective theory avoids using its truncated action as an exact high-energy propagator; it does not establish a positive ultraviolet completion of that propagator.
Power counting must use the actual high-momentum behavior. For a connected four-dimensional graph, with propagator decay p − r e p^{-r_e} p − r e on internal line e e e and d v d_v d v derivatives at vertex v v v ,
D = 4 L + ∑ v d v − ∑ e r e , L = I − V + 1. (588) D=4L+\sum_{v}d_v-\sum_{e}r_e,\qquad L=I-V+1. \tag*{(588)} D = 4 L + v ∑ d v − e ∑ r e , L = I − V + 1. (588)
Two-derivative Einstein power counting gives D = 2 L + 2 D=2L+2 D = 2 L + 2 ; if all relevant propagators decay as p − 4 p^{-4} p − 4 and all vertices carry at most four derivatives, it gives D ≤ 4 D\leq4 D ≤ 4 . Mixed matter and gauge sectors require their own line and vertex counts. Neither estimate proves positive physical spectral weights. A negative component of a gauge-fixed auxiliary propagator also cannot alone decide physical unitarity: positivity tests concern physical observables under the stated spectral assumptions. Finally, a Euclidean renormalization-group fixed point does not supply Lorentzian constraint closure, causal reconstruction, or the physical inner product without the corresponding additional results.
Constraint-preserving diffusions and pathwise clock readings
The geometric constraint problem also appears before quantization: noise tangent to a curved surface produces a normal Itô drift. This drift is fixed by the curvature of the constraint surface. Keeping it explicit separates geometric invariance from the choice of an optimal approximation to an unconstrained stochastic model.
For a smooth regular constraint surface M = { C A = 0 } \mathcal{M}=\{C_A=0\} M = { C A = 0 } and an Itô diffusion d Z = b d s + ∑ r σ r d W r dZ=b\,ds+\sum_r \sigma_r\,dW^r d Z = b d s + ∑ r σ r d W r , Itô’s formula gives the local invariance conditions on M \mathcal{M} M :
D C A σ r = 0 , D C A b + 1 2 ∑ r D 2 C A [ σ r , σ r ] = 0. (589) D C_A \sigma_r=0,\qquad D C_A b+\frac{1}{2}\sum_r D^2 C_A[\sigma_r,\sigma_r]=0. \tag*{(589)} D C A σ r = 0 , D C A b + 2 1 r ∑ D 2 C A [ σ r , σ r ] = 0. (589)
Under smooth local existence and uniqueness assumptions, these conditions ensure invariance up to the lifetime of the solution. They are independent of a particular projection convention. Stratonovich, Itô-vector and Itô-jet projections are all manifold-valued constructions when defined intrinsically. Armstrong, Brigo, and Ferrucci identify different local approximation criteria: in their setting the Itô-vector projection minimizes the stated leading weak error after fixing the diffusion approximation, whereas the Itô-jet projection minimizes the stated mean-square error relative to the metric-projected process [6]. These criteria are distinct from constraint preservation. A numerical constraint residual can compare discretizations but cannot by itself choose a unique continuous projection.
On the unit sphere S d − 1 \mathbb{S}^{d-1} S d − 1 , d ≥ 2 d\ge2 d ≥ 2 , put P x = I − x x T P_x=I-xx^{\mathsf{T}} P x = I − x x T and let U U U be smooth. The intrinsic Stratonovich diffusion and its equivalent Itô equation are
d X = − P X ∇ U d s + 2 θ P X ∘ d W , d X = [ − P X ∇ U − θ ( d − 1 ) X ] d s + 2 θ P X d W . (590) dX=-P_X\nabla U\,ds+\sqrt{2\theta}\,P_X\circ dW,
\\
dX=[-P_X\nabla U-\theta(d-1)X]\,ds+\sqrt{2\theta}\,P_X\,dW. \tag*{(590)} d X = − P X ∇ U d s + 2 θ P X ∘ d W , d X = [ − P X ∇ U − θ ( d − 1 ) X ] d s + 2 θ P X d W . (590)
Here an arbitrary smooth ambient extension of U U U can be used; its projected gradient on the sphere is intrinsic. For the diffusion columns V r ( x ) = P x e r V_r(x)=P_x e_r V r ( x ) = P x e r , differentiation gives ∑ r D V r ( x ) V r ( x ) = − ( d − 1 ) x \sum_r D V_r(x)V_r(x)=-(d-1)x ∑ r D V r ( x ) V r ( x ) = − ( d − 1 ) x on the sphere. Therefore the Stratonovich-to-Itô correction is − θ ( d − 1 ) x -\theta(d-1)x − θ ( d − 1 ) x . Since x T P x = 0 x^{\mathsf{T}}P_x=0 x T P x = 0 and tr P x = d − 1 \operatorname{tr}P_x=d-1 tr P x = d − 1 ,
d ∣ X ∣ 2 = [ − 2 θ ( d − 1 ) + 2 θ tr P X ] d s = 0. (591) d|X|^2=[-2\theta(d-1)+2\theta\operatorname{tr}P_X]\,ds=0. \tag*{(591)} d ∣ X ∣ 2 = [ − 2 θ ( d − 1 ) + 2 θ tr P X ] d s = 0. (591)
The generator is L = θ Δ S − ∇ S U ⋅ ∇ S L=\theta\Delta_{\mathbb{S}}-\nabla_{\mathbb{S}}U\cdot\nabla_{\mathbb{S}} L = θ Δ S − ∇ S U ⋅ ∇ S . Integration by parts on the compact sphere gives the invariant law Z − 1 e − U / θ d vol S Z^{-1}e^{-U/\theta}\,d\operatorname{vol}_{\mathbb{S}} Z − 1 e − U / θ d vol S , and, for smooth f , g f,g f , g ,
∫ f L g d ν = − θ ∫ ∇ S f ⋅ ∇ S g d ν . (592) \int fLg\,d\nu=-\theta\int\nabla_{\mathbb{S}}f\cdot\nabla_{\mathbb{S}}g\,d\nu. \tag*{(592)} ∫ f Lg d ν = − θ ∫ ∇ S f ⋅ ∇ S g d ν . (592)
This example verifies both the constraint and the Gibbs reference. A timestepping rule must still be checked separately: a generic Euler step does not preserve the sphere exactly, and geometric projection of a step does not automatically preserve the invariant law.
Finally, a clock function T ( Z ) T(Z) T ( Z ) has Itô differential and quadratic variation
d T = ( L T ) d s + ∑ r D T σ r d W r , [ T ] s = ∫ 0 s ∑ r ∣ D T σ r ∣ 2 d u . (593) dT = (\mathcal{L}T)\,ds + \sum_r DT\sigma_r\,dW^r,\qquad[T]_s = \int_0^s \sum_r \lvert DT\sigma_r\rvert^2\,du. \tag*{(593)} d T = ( L T ) d s + r ∑ D T σ r d W r , [ T ] s = ∫ 0 s r ∑ ∣ D T σ r ∣ 2 d u . (593)
A positive drift does not ensure increasing readings along individual paths: T s = s + ε W s T_s=s+\varepsilon W_s T s = s + ε W s fails to be monotone for every ε > 0 \varepsilon>0 ε > 0 . A continuous monotone path has finite variation and zero quadratic variation. Thus monotone diffusion clocks necessarily satisfy D T σ r = 0 DT\sigma_r=0 D T σ r = 0 almost everywhere along their paths. Together with L T ≥ 0 \mathcal{L}T\ge0 L T ≥ 0 , this condition is sufficient for nondecreasing readings; a uniform positive lower bound gives strictly increasing readings while the chart remains valid. The spacetime condition that d T dT d T is timelike, the probabilistic condition of pathwise monotonicity, and the identification of a physical clock with an auxiliary diffusion parameter are three separate requirements.
The finite calculations above give explicit measure, reduction, and stochastic-invariance results. Connecting them to an interacting quantum geometry requires a single action and measure to satisfy the complete gravitational constraints, regulator-uniform evolution and moment estimates, and a nontrivial positive physical quantum construction. Those connections are additional physical and analytic requirements; the logarithmic Gaussian example alone does not select them.
A finite covariant extension with controlled inverse coframes
The preceding constructions isolate three physical requirements: a coframe must transform with the local reference frame, inverse geometric observables must be integrable, and the time evolution must use the declared interaction. These requirements can be satisfied together in a finite Euclidean model. The construction below changes both the coframe representation and its probability law. It is an extension to be matched to the original mother branch, rather than a change of coordinates in that branch.
Take a fixed finite graph with oriented edges, a finite list of plaquettes, and four-dimensional internal Euclidean frames. At vertex v v v , set e v = R v exp S v e_v=R_v\exp S_v e v = R v exp S v , with R v ∈ S O ( 4 ) R_v\in\mathrm{SO}(4) R v ∈ SO ( 4 ) and S v ∈ S y m ( 4 ) S_v\in\mathrm{Sym}(4) S v ∈ Sym ( 4 ) ; let σ v ∈ R \sigma_v\in\mathbb{R} σ v ∈ R . On edge ( v , w ) (v,w) ( v , w ) take U v w ∈ S O ( 4 ) U_{vw}\in\mathrm{SO}(4) U v w ∈ SO ( 4 ) . For local rotations L v L_v L v , prescribe
R v ↦ L v R v , U v w ↦ L v U v w L w − 1 , S v ↦ S v , σ v ↦ σ v . (594) R_v\mapsto L_vR_v,\qquad U_{vw}\mapsto L_vU_{vw}L_w^{-1},\qquad S_v\mapsto S_v,\qquad\sigma_v\mapsto\sigma_v. \tag*{(594)} R v ↦ L v R v , U v w ↦ L v U v w L w − 1 , S v ↦ S v , σ v ↦ σ v . (594)
Thus e v − U v w e w e_v-U_{vw}e_w e v − U v w e w transforms as L v ( e v − U v w e w ) L_v(e_v-U_{vw}e_w) L v ( e v − U v w e w ) . The graph and coordinate column indices remain fixed: local internal rotations are being restored, not arbitrary changes of spacetime coordinates. Normalize each Haar measure and put independent logarithmic Gaussians ν a \nu_a ν a of (573) on the coframes, and normalized densities proportional to e − b σ v 2 e^{-b\sigma_v^2} e − b σ v 2 on the scalars, with a , b > 0 a,b>0 a , b > 0 . Call the product reference measure ν a , b , h \nu_{a,b,h} ν a , b , h . For nonnegative κ \kappa κ , f 2 f^2 f 2 , η \eta η and α ≥ 0 \alpha\geq0 α ≥ 0 , define
W h = κ 2 ∑ ( v , w ) ∥ e v − U v w e w ∥ F 2 + ∑ p ω v ( p ) ( p ) k p ( U ) + f 2 2 ∑ ( v , w ) ( σ v − σ w ) 2 , (595) W_h=\frac{\kappa}{2}\sum_{(v,w)}\lVert e_v-U_{vw}e_w\rVert_F^2+\sum_p\omega_{v(p)}(p)k_p(U)+\frac{f^2}{2}\sum_{(v,w)}(\sigma_v-\sigma_w)^2, \tag*{(595)} W h = 2 κ ( v , w ) ∑ ∥ e v − U v w e w ∥ F 2 + p ∑ ω v ( p ) ( p ) k p ( U ) + 2 f 2 ( v , w ) ∑ ( σ v − σ w ) 2 , (595)
ω v = e − α σ v ( 1 + η tr S v 2 1 + tr S v 2 ) , k p ( U ) = 4 − tr U p . \omega_v=e^{-\alpha\sigma_v}\left(1+\eta\frac{\operatorname{tr}S_v^2}{1+\operatorname{tr}S_v^2}\right),\qquad k_p(U)=4-\operatorname{tr}U_p. ω v = e − α σ v ( 1 + η 1 + tr S v 2 tr S v 2 ) , k p ( U ) = 4 − tr U p .
Here v ( p ) v(p) v ( p ) is a declared base vertex of each plaquette and U p U_p U p is its oriented holonomy. Since U p ∈ S O ( 4 ) U_p\in\mathrm{SO}(4) U p ∈ SO ( 4 ) , k p ≥ 0 k_p\geq0 k p ≥ 0 . All quantities in this construction are dimensionless after fixing a coframe scale, a scalar scale, and an energy unit.
Proposition 33 (A finite covariant Gibbs law). The law d μ h = Z h − 1 e − W h d ν a , b , h d\mu_h=Z_h^{-1}e^{-W_h}d\nu_{a,b,h} d μ h = Z h − 1 e − W h d ν a , b , h is well defined, locally S O ( 4 ) \mathrm{SO}(4) SO ( 4 ) invariant, and has all positive inverse-coframe moments at each vertex. More precisely, for r > 0 r>0 r > 0 ,
0 < Z h ≤ 1 , E μ h ( det e v ) − r ≤ Z h − 1 e r 2 / a , E μ h ∥ e v − 1 ∥ F ≤ Z h − 1 2 r + 5 e r 2 / ( 2 a ) . (596) 0<Z_h\leq1,\qquad\mathbb{E}_{\mu_h}(\det e_v)^{-r}\leq Z_h^{-1}e^{r^2/a},\qquad\mathbb{E}_{\mu_h}\lVert e_v^{-1}\rVert_F\leq Z_h^{-1}2^{r+5}e^{r^2/(2a)}. \tag*{(596)} 0 < Z h ≤ 1 , E μ h ( det e v ) − r ≤ Z h − 1 e r 2 / a , E μ h ∥ e v − 1 ∥ F ≤ Z h − 1 2 r + 5 e r 2 / ( 2 a ) . (596) On the product Riemannian configuration space, its gradient Dirichlet form has a conservative reversible realization and only constants have zero Dirichlet energy. It therefore has a unique invariant probability density relative to μ h \mu_h μ h .
Proof. Every summand of W h W_h W h is nonnegative and smooth. The integrand e − W h e^{-W_h} e − W h is strictly positive at every finite configuration, proving 0 < Z h ≤ 1 0<Z_h\leq1 0 < Z h ≤ 1 . Equation (594) , orthogonal invariance of the Frobenius norm, and conjugation invariance of the plaquette trace give invariance of W h W_h W h . Left invariance of Haar measure gives invariance of the reference law. For a nonnegative observable F F F , E μ h F ≤ Z h − 1 E ν a , b , h F \mathbb{E}_{\mu_h}F\leq Z_h^{-1}\mathbb{E}_{\nu_{a,b,h}}F E μ h F ≤ Z h − 1 E ν a , b , h F . Apply the previously calculated logarithmic Gaussian moments to obtain (596) .
Write Φ h = a ∑ v tr S v 2 + b ∑ v σ v 2 + W h \Phi_h=a\sum_v\operatorname{tr}S_v^2+b\sum_v\sigma_v^2+W_h Φ h = a ∑ v tr S v 2 + b ∑ v σ v 2 + W h . The smooth positive density proportional to e − Φ h e^{-\Phi_h} e − Φ h defines the closed gradient form E h ( f , f ) = θ ∫ ∣ ∇ f ∣ 2 d μ h \mathcal{E}_h(f,f)=\theta\int|\nabla f|^2d\mu_h E h ( f , f ) = θ ∫ ∣∇ f ∣ 2 d μ h . Choose smooth radial cutoffs in the noncompact coordinates, equal to one on a ball of radius R R R , vanishing outside radius 2 R 2R 2 R , and with gradient bounded by C / R C/R C / R . They converge to one in L 2 ( μ h ) L^2(\mu_h) L 2 ( μ h ) and their energies are at most θ C 2 / R 2 \theta C^2/R^2 θ C 2 / R 2 . Hence the constant function belongs to the form domain with zero energy, and the associated symmetric Markov semigroup is conservative. A function of zero energy has zero weak gradient; connectedness and the strictly positive density imply that it is constant. This gives irreducibility and uniqueness of the invariant probability density. The generator on compactly supported smooth functions is
L h = θ Δ − θ ∇ Φ h ⋅ ∇ , d d t [ θ ∫ f t log f t d μ h ] = − θ 2 ∫ ∣ ∇ f t ∣ 2 f t d μ h . (597) \mathcal{L}_h = \theta\Delta- \theta\nabla\Phi_h \cdot\nabla,\qquad\frac{d}{dt}\left[\theta\int f_t \log f_t\,d\mu_h\right] = -\theta^2 \int\frac{|\nabla f_t|^2}{f_t}\,d\mu_h. \tag*{(597)} L h = θ Δ − θ ∇ Φ h ⋅ ∇ , d t d [ θ ∫ f t log f t d μ h ] = − θ 2 ∫ f t ∣∇ f t ∣ 2 d μ h . (597)
The identity follows by integration by parts for positive smooth densities with finite dissipation; approximation extends the integrated inequality to the usual finite-entropy domain. □
The thermodynamic content of this model is specific. The reference energy is E h = θ Φ h E_h=\theta\Phi_h E h = θ Φ h , the chosen mobility is the identity in the specified product metric, and its covariance is 2 θ 2\theta 2 θ times that mobility. (597) is therefore a free-energy balance for this declared isothermal dynamics. Identification with measured bath entropy still uses the heat convention and local detailed balance developed earlier. Neither the equilibrium density alone nor local rotation invariance fixes the mobility, the clock, or a , b a,b a , b .
There is also an exact internal constraint statement. On the cotangent bundle, the infinitesimal site action now moves R v R_v R v as well as the links. With the moment-map convention of Proposition 30, its Gauss generator pairs the momenta with these variations:
G [ ξ ] = ∑ ( v , w ) p U v w ( ξ v U v w − U v w ξ w ) + ∑ v p R v ( ξ v R v ) . (598) G[\xi] = \sum_{(v,w)} p_{U_{vw}}\left(\xi_v U_{vw}-U_{vw}\xi_w\right) + \sum_v p_{R_v}\left(\xi_v R_v\right). \tag*{(598)} G [ ξ ] = ( v , w ) ∑ p U v w ( ξ v U v w − U v w ξ w ) + v ∑ p R v ( ξ v R v ) . (598)
Cotangent-lift equivariance gives the same internal Lie algebra and any invariant Hamiltonian preserves G = 0 G=0 G = 0 . Omitting the second sum would omit the coframe charge. These are internal rotation constraints on a fixed graph; lapse and shift constraints have not been constructed here. Moreover, the positive Euclidean action above has not been derived by reducing the full mother group. Its matching, Lorentzian continuation, and regulator dependence remain distinct questions. In particular, Z h − 1 Z_h^{-1} Z h − 1 in (597) need not be uniformly bounded as the graph grows.
Exact temporal normalization and the retained inertia
A second connection concerns the time coordinate. A field-dependent normalization repeated at each time slice can change the limiting Hamiltonian, even when it seems harmless in a single integral. The retained U ( 1 ) U(1) U ( 1 ) mode allows this effect to be computed without an expansion of the full field theory. Put w = e − α σ w=e^{-\alpha\sigma} w = e − α σ , w ( 0 ) = 1 w(0)=1 w ( 0 ) = 1 , and let ϵ > 0 \epsilon>0 ϵ > 0 denote the temporal spacing. The Wilson row integral is
c ϵ ( σ ) = ∫ − π π e − I w ( σ ) ( 1 − cos q ) / ϵ d q = 2 π e − z I 0 ( z ) , z = I w ( σ ) ϵ , R ϵ = c ϵ ( σ ) c ϵ ( 0 ) . (599) c_\epsilon(\sigma)=\int_{-\pi}^{\pi}e^{-I w(\sigma)(1-\cos q)/\epsilon}\,dq=2\pi e^{-z}I_0(z),\qquad z=\frac{I w(\sigma)}{\epsilon},\qquad R_\epsilon=\frac{c_\epsilon(\sigma)}{c_\epsilon(0)}. \tag*{(599)} c ϵ ( σ ) = ∫ − π π e − I w ( σ ) ( 1 − c o s q ) / ϵ d q = 2 π e − z I 0 ( z ) , z = ϵ I w ( σ ) , R ϵ = c ϵ ( 0 ) c ϵ ( σ ) . (599)
I 0 I_0 I 0 is the modified Bessel function. To specify how this row factor enters a transfer operator, consider the symmetric arrangement T r a w = R ϵ 1 / 2 T n o r m R ϵ 1 / 2 T_{\mathrm{raw}}=R_\epsilon^{1/2}T_{\mathrm{norm}}R_\epsilon^{1/2} T raw = R ϵ 1/2 T norm R ϵ 1/2 , where each R ϵ R_{\epsilon} R ϵ acts by multiplication. This is a declared kernel arrangement; an arbitrary mother kernel need not factor this way.
Proposition 34 (Endpoint compensation for one compact mode). In this arrangement the exact compensating kernel multiplier is
N ϵ ( σ , σ ′ ) = R ϵ ( σ ) − 1 / 2 R ϵ ( σ ′ ) − 1 / 2 . (600) N_{\epsilon}(\sigma,\sigma')=R_{\epsilon}(\sigma)^{-1/2}R_{\epsilon}(\sigma')^{-1/2}. \tag*{(600)} N ϵ ( σ , σ ′ ) = R ϵ ( σ ) − 1/2 R ϵ ( σ ′ ) − 1/2 . (600) Uniformly on each fixed compact scalar interval,
log R ϵ = − 1 2 log w + ϵ 8 I ( w − 1 − 1 ) + O ( ϵ 2 ) . (601) \log R_{\epsilon}=-\frac{1}{2}\log w+\frac{\epsilon}{8I}(w^{-1}-1)+O(\epsilon^2). \tag*{(601)} log R ϵ = − 2 1 log w + 8 I ϵ ( w − 1 − 1 ) + O ( ϵ 2 ) . (601) Thus its leading endpoint multiplier is w ( σ ) 1 / 4 w ( σ ′ ) 1 / 4 w(\sigma)^{1/4}w(\sigma')^{1/4} w ( σ ) 1/4 w ( σ ′ ) 1/4 .
Proof. Multiplication by (601) cancels the two endpoint factors algebraically. The large-z z z formula I 0 ( z ) = e z ( 2 π z ) − 1 / 2 ( 1 + 1 / ( 8 z ) + O ( z − 2 ) ) I_0(z)=e^z(2\pi z)^{-1/2}(1+1/(8z)+O(z^{-2})) I 0 ( z ) = e z ( 2 π z ) − 1/2 ( 1 + 1/ ( 8 z ) + O ( z − 2 )) gives (602), since w w w is bounded above and below on the specified interval. Taking minus one half of this logarithm gives the quarter power at each endpoint. □
For a constant scalar value, the uncompensated effective contribution U ϵ = − ϵ − 1 log R ϵ U_{\epsilon}=-\epsilon^{-1}\log R_{\epsilon} U ϵ = − ϵ − 1 log R ϵ is
U ϵ = − α σ 2 ϵ − e α σ − 1 8 I + O ( ϵ ) . (602) U_{\epsilon}=-\frac{\alpha\sigma}{2\epsilon}-\frac{e^{\alpha\sigma}-1}{8I}+O(\epsilon). \tag*{(602)} U ϵ = − 2 ϵ α σ − 8 I e α σ − 1 + O ( ϵ ) . (602)
A half power of w w w at each endpoint overcompensates and leaves the opposite divergence + α σ / ( 2 ϵ ) +\alpha\sigma/(2\epsilon) + α σ / ( 2 ϵ ) . A quarter power at each endpoint cancels the divergence but leaves the finite term − ( e α σ − 1 ) / ( 8 I ) -(e^{\alpha\sigma}-1)/(8I) − ( e α σ − 1 ) / ( 8 I ) . Exact compensation cancels the entire declared row factor. For d d d independent Gaussian kinetic coordinates with the same scalar weight, the leading endpoint exponent is d / 4 d/4 d /4 ; with a general kinetic matrix it is the corresponding determinant factor. The exponent must therefore be computed for the retained variables.
For example, the synthetic values I = 1 I=1 I = 1 , α = 0.3 \alpha=0.3 α = 0.3 , σ = 1 \sigma=1 σ = 1 give the following direct Bessel evaluations. The last column concerns only the row contribution, not the complete Hamiltonian.
Table 8.
Normalization does not remove the matched inertia I w ( σ ) Iw(\sigma) I w ( σ ) in (535) . Rescaling the angle by a scalar-dependent factor would change its periodicity and introduce mixed derivatives when the scalar moves. Nor does the factor J ( Δ σ ) 2 / ( 2 ϵ ) J(\Delta\sigma)^2/(2\epsilon) J ( Δ σ ) 2 / ( 2 ϵ ) in a short-time action imply a Hamiltonian inertia proportional to ϵ − 1 \epsilon^{-1} ϵ − 1 : the factor ϵ − 1 \epsilon^{-1} ϵ − 1 arises from integrating a squared velocity over a time step. Temporal normalization, kinetic matching, and spatial regulator removal are separate operations.
Figure 8. Synthetic normalization calculation for one retained compact mode. Left: the uncompensated and half-power choices have opposite divergent potentials. Right: quarter-power compensation leaves a finite potential, whereas the exact Bessel ratio removes the declared row factor.
Nondegeneracy at a stated observation scale
An observation averages fields over a finite region. This gives a useful nondegeneracy question that is weaker than controlling the smallest coframe determinant at every regulator vertex. Fix the coordinate reference volume used to define D v = det e v > 0 D_v = \det e_v > 0 D v = det e v > 0 , and deterministic weights b j v ≥ 0 b_{jv} \ge0 b j v ≥ 0 with ∑ v b j v = 1 \sum_v b_{jv} = 1 ∑ v b j v = 1 for j = 1 , … , M j = 1,\ldots,M j = 1 , … , M . Define D ˉ j = ∑ v b j v D v \bar{D}_j = \sum_v b_{jv}D_v D ˉ j = ∑ v b j v D v ; these are averages of a density relative to that fixed volume, not coordinate-independent scalars without an observation prescription.
Proposition 35 (Finite-resolution determinant bound). Suppose, for every r > 0 r > 0 r > 0 and vertex v v v , E D v − r ≤ C h e r 2 / a \mathbb{E}D_v^{-r} \le C_h e^{r^2/a} E D v − r ≤ C h e r 2 / a , with C h ≥ 1 C_h \ge1 C h ≥ 1 . Then, for 0 < δ < 1 0 < \delta< 1 0 < δ < 1 ,
E D ˉ j − r ≤ C h e r 2 / a , P { min j D ˉ j < δ } ≤ min { 1 , M C h e − a log 2 ( 1 / δ ) / 4 } . (603) \mathbb{E}\bar{D}_j^{-r} \le C_h e^{r^2/a}, \qquad\mathbb{P}\left\{\min_j \bar{D}_j < \delta\right\} \le\min\left\{1,MC_h e^{-a\log^2(1/\delta)/4}\right\}. \tag*{(603)} E D ˉ j − r ≤ C h e r 2 / a , P { j min D ˉ j < δ } ≤ min { 1 , M C h e − a l o g 2 ( 1/ δ ) /4 } . (603) No independence assumption is required.
Proof. Convexity of x ↦ x − r x \mapsto x^{-r} x ↦ x − r gives D ˉ j − r ≤ ∑ v b j v D v − r \bar{D}_j^{-r} \le\sum_v b_{jv}D_v^{-r} D ˉ j − r ≤ ∑ v b j v D v − r . Markov’s inequality and a union bound then give M C h δ r e r 2 / a MC_h\delta^r e^{r^2/a} M C h δ r e r 2 / a . Its exponent r log δ + r 2 / a r\log\delta+ r^2/a r log δ + r 2 / a is minimized at r = ( a / 2 ) log ( 1 / δ ) r = (a/2)\log(1/\delta) r = ( a /2 ) log ( 1/ δ ) , proving the assertion. □
For the product logarithmic law, C h = 1 C_h = 1 C h = 1 . For the interacting model of Proposition 33, C h = Z h − 1 C_h = Z_h^{-1} C h = Z h − 1 is available and must be retained. If D ˉ ( x ) \bar{D}(x) D ˉ ( x ) is also defined for continuously located observations, assume its Lipschitz constant is at most K L K_L K L outside an event of probability ζ \zeta ζ . An η \eta η -net with M ( η ) M(\eta) M ( η ) centers then gives, for every r > 0 r>0 r > 0 ,
P { inf x D ‾ ( x ) < δ } ≤ min { 1 , ζ + M ( η ) C h ( δ + K L η ) e r 2 / a } . (604) \mathbb{P}\left\{\inf_x \overline{D}(x)<\delta\right\}\leq\min\left\{1,\ \zeta+M(\eta)C_h\left(\delta+K_L\eta\right)e^{r^2/a}\right\}. \tag*{(604)} P { x inf D ( x ) < δ } ≤ min { 1 , ζ + M ( η ) C h ( δ + K L η ) e r 2 / a } . (604)
Indeed an observation below δ \delta δ has a net center below δ + K L η \delta+K_L\eta δ + K L η on the Lipschitz event. Spatial regularity and uniform control of C h C_h C h would still have to be proved for a continuum claim.
The need for joint control can be seen without asymptotics. Under the one-vertex logarithmic law, log D \log D log D is Gaussian with variance 2 / a 2/a 2/ a . For a = 1 a=1 a = 1 and δ = 0.1 \delta=0.1 δ = 0.1 , the one-vertex probability is approximately 0.05174405. For N = 10 N=10 N = 10 independent vertices the minimum falls below δ \delta δ with probability 0.41216455; for N = 100 N=100 N = 100 the probability is 0.99507328. Perfectly correlated copies keep the one-vertex probability for every N N N . The same marginals therefore do not determine the large-graph event. A union upper bound that grows beyond one proves neither degeneration nor nondegeneration. Finally, e = diag ( z , z − 1 , 1 , 1 ) e=\operatorname{diag}(z,z^{-1},1,1) e = diag ( z , z − 1 , 1 , 1 ) has determinant one while ∥ e − 1 ∥ F 2 = z 2 + z − 2 + 2 \lVert e^{-1}\rVert_F^2=z^2+z^{-2}+2 ∥ e − 1 ∥ F 2 = z 2 + z − 2 + 2 diverges as z → ∞ z\to\infty z → ∞ . Determinant control must be supplemented by shape control when inverse metrics are the observable of interest.
Equilibrium preservation does not select a physical clock
The equilibrium reference constrains admissible dynamics, but a reference density does not determine the rate of physical evolution. This is particularly relevant when a thermodynamic model is matched to a field-theoretic clock. Let π ∝ e − U / θ \pi\propto e^{-U/\theta} π ∝ e − U / θ , and write the reversible diffusion generator in fixed Euclidean coordinates as
L f = θ π − 1 ∇ ⋅ ( π Γ ∇ f ) = θ ∇ ⋅ ( Γ ∇ f ) − Γ ∇ U ⋅ ∇ f . (605) Lf=\theta\pi^{-1}\nabla\cdot(\pi\Gamma\nabla f)=\theta\nabla\cdot(\Gamma\nabla f)-\Gamma\nabla U\cdot\nabla f. \tag*{(605)} L f = θ π − 1 ∇ ⋅ ( π Γ∇ f ) = θ ∇ ⋅ ( Γ∇ f ) − Γ∇ U ⋅ ∇ f . (605)
Here Γ \Gamma Γ is symmetric positive definite. Under the time change d t = r ( X s ) d s dt=r(X_s)ds d t = r ( X s ) d s , with smooth 0 < c ≤ r ≤ C 0<c\leq r\leq C 0 < c ≤ r ≤ C , the new generator and stationary law are
L r = r − 1 L , d π r = Z r − 1 r d π , Z r = ∫ r d π . (606) L_r=r^{-1}L,\qquad d\pi_r=Z_r^{-1}r\,d\pi,\qquad Z_r=\int r\,d\pi. \tag*{(606)} L r = r − 1 L , d π r = Z r − 1 r d π , Z r = ∫ r d π . (606)
To verify stationarity, integrate L r f L_rf L r f against π r \pi_r π r and cancel r r r . More explicitly, choose Γ r = Γ / r \Gamma_r=\Gamma/r Γ r = Γ/ r and U r = U − θ log r U_r=U-\theta\log r U r = U − θ log r . The derivative of r − 1 r^{-1} r − 1 in the divergence term cancels the derivative of − θ log r -\theta\log r − θ log r in the drift, leaving L / r L/r L / r . The Dirichlet form becomes E r = E / Z r \mathcal{E}_r=\mathcal{E}/Z_r E r = E / Z r . Although U r U_r U r represents the new occupation density, this algebra does not show that a mechanical energy has changed solely because a clock has been relabelled. Physical heat and work must use the calibrated generator and the declared bath model.
For a concrete obstruction, take L a = θ ∂ y 2 − 2 θ a y ∂ y L_a=\theta\partial_y^2-2\theta ay\partial_y L a = θ ∂ y 2 − 2 θ a y ∂ y and r ( y ) = e − ( a ′ − a ) y 2 r(y)=e^{-(a'-a)y^2} r ( y ) = e − ( a ′ − a ) y 2 with a , a ′ > 0 a,a'>0 a , a ′ > 0 . The stationary density after time change is still Gaussian, proportional to e − a ′ y 2 e^{-a'y^2} e − a ′ y 2 , but its diffusion coefficient is θ / r ( y ) \theta/r(y) θ / r ( y ) and is nonconstant if a ′ ≠ a a'\ne a a ′ = a . Thus the Gaussian Gibbs family is preserved while the constant-diffusion Ornstein–Uhlenbeck family is not. Although this r r r need not be bounded, its stationary mean is finite and positive; ergodicity of the original process gives ∫ 0 s r ( X u ) d u → ∞ \int_{0}^{s} r(X_u)\,\mathrm{d}u \to\infty ∫ 0 s r ( X u ) d u → ∞ , so the time change remains conservative. Even a constant r r r leaves a global clock-scale freedom. Selecting a physical time therefore requires a dynamical matching or a clock observable beyond membership in a Gibbs family.
Backreaction from the same shared weight.
The retained mechanical model now gives a quantitative test of how a common weight affects an observable. A changing scalar changes both the gauge kinetic energy and its restoring force. The leading frequency at a fixed scalar value is consequently independent of that weight, while energy exchanged with a moving scalar gives an amplitude-dependent correction. This distinction determines which experiment the model would require.
Write K = K σ > 0 K=K_{\sigma}>0 K = K σ > 0 , x = σ − σ 0 x=\sigma-\sigma_{0} x = σ − σ 0 , and w 0 = e − α σ 0 w_{0}=e^{-\alpha\sigma_{0}} w 0 = e − α σ 0 . The previous branch has σ 0 = 0 \sigma_{0}=0 σ 0 = 0 ; a displaced reference requires an explicitly included preparation or stabilizing potential K ( σ − σ 0 ) 2 / 2 K(\sigma-\sigma_{0})^{2}/2 K ( σ − σ 0 ) 2 /2 . With ω 0 2 = β / I \omega_{0}^{2}=\beta/I ω 0 2 = β / I and Ω 2 = K / J \Omega^{2}=K/J Ω 2 = K / J , the exact equations are
q ¨ − α x ˙ q ˙ + ω 0 2 sin q = 0 , (607) \ddot{q}-\alpha\dot{x}\dot{q}+\omega_{0}^{2}\sin q=0, \tag*{(607)} q ¨ − α x ˙ q ˙ + ω 0 2 sin q = 0 , (607)
J x ¨ + K x = α w 0 e − α x [ β ( 1 − cos q ) − I q ˙ 2 2 ] . (608) J\ddot{x}+Kx=\alpha w_{0}e^{-\alpha x}\left[\beta(1-\cos q)-\frac{I\dot{q}^{2}}{2}\right]. \tag*{(608)} J x ¨ + K x = α w 0 e − α x [ β ( 1 − cos q ) − 2 I q ˙ 2 ] . (608)
Their conserved energy is
E = 1 2 I w 0 e − α x q ˙ 2 + 1 2 J x ˙ 2 + β w 0 e − α x ( 1 − cos q ) + 1 2 K x 2 . (609) E=\frac{1}{2}Iw_{0}e^{-\alpha x}\dot{q}^{2}+\frac{1}{2}J\dot{x}^{2}+\beta w_{0}e^{-\alpha x}(1-\cos q)+\frac{1}{2}Kx^{2}. \tag*{(609)} E = 2 1 I w 0 e − α x q ˙ 2 + 2 1 J x ˙ 2 + β w 0 e − α x ( 1 − cos q ) + 2 1 K x 2 . (609)
The scalar source contains a potential contribution and a kinetic contribution with opposite signs. Substitution of q = A cos ω 0 t q=A\cos\omega_{0}t q = A cos ω 0 t gives F cos 2 ω 0 t F\cos2\omega_{0}t F cos 2 ω 0 t to order A 2 A^{2} A 2 , where
F = 1 2 α β w 0 A 2 , Δ = K − 4 J ω 0 2 , B = F / Δ . (610) F=\frac{1}{2}\alpha\beta w_{0}A^{2},\qquad\Delta=K-4J\omega_{0}^{2},\qquad B=F/\Delta. \tag*{(610)} F = 2 1 α β w 0 A 2 , Δ = K − 4 J ω 0 2 , B = F /Δ. (610)
Its mean vanishes at this order. For Δ ≠ 0 \Delta\ne0 Δ = 0 , the periodic particular solution is B cos 2 ω 0 t B\cos2\omega_{0}t B cos 2 ω 0 t , whereas zero scalar position and velocity initially give
x a ( t ) = B ( cos 2 ω 0 t − cos Ω t ) . (611) x_{a}(t)=B(\cos2\omega_{0}t-\cos\Omega t). \tag*{(611)} x a ( t ) = B ( cos 2 ω 0 t − cos Ω t ) . (611)
At Ω = 2 ω 0 \Omega=2\omega_{0} Ω = 2 ω 0 , direct resonant forcing instead gives
x a ( t ) = α β w 0 A 2 8 J ω 0 t sin 2 ω 0 t . x_{a}(t)=\frac{\alpha\beta w_{0}A^{2}}{8J\omega_{0}}t\sin2\omega_{0}t. x a ( t ) = 8 J ω 0 α β w 0 A 2 t sin 2 ω 0 t .
The static denominator K K K cannot replace Δ \Delta Δ in an oscillatory response, and proximity to resonance requires a different approximation.
For a smooth small-amplitude periodic branch admitting a regular expansion, the order-A 3 A^{3} A 3 equation for q q q contains the fundamental forcing ω 0 2 A 3 / 8 + α B A ω 0 2 \omega_{0}^{2}A^{3}/8+\alpha BA\omega_{0}^{2} ω 0 2 A 3 /8 + α B A ω 0 2 . The first term follows from cos 3 t = ( 3 cos t + cos 3 t ) / 4 \cos^{3}t=(3\cos t+\cos3t)/4 cos 3 t = ( 3 cos t + cos 3 t ) /4 ; the second follows from 2 sin 2 t sin t = cos t − cos 3 t 2\sin2t\sin t=\cos t-\cos3t 2 sin 2 t sin t = cos t − cos 3 t in α x ˙ q ˙ \alpha\dot{x}\dot{q} α x ˙ q ˙ . Removing the secular fundamental term therefore gives the coefficient
ω − ω 0 ω 0 = − A 2 16 − α 2 β ω 0 A 2 4 Δ + O ( A 4 ) . (612) \frac{\omega-\omega_0}{\omega_0}=-\frac{A^2}{16}-\frac{\alpha^2\beta\omega_0A^2}{4\Delta}+O(A^4). \tag*{(612)} ω 0 ω − ω 0 = − 16 A 2 − 4Δ α 2 β ω 0 A 2 + O ( A 4 ) . (612)
This coefficient calculation presumes such a regular branch and excludes the displayed resonance. It does not establish its global existence or a uniform remainder for arbitrary preparations. A finite-time estimate for an explicitly prepared trajectory can, however, be proved directly.
Proposition 36 (Finite-time response estimate). Assume I , J , β , K , ω 0 > 0 I,J,\beta,K,\omega_0>0 I , J , β , K , ω 0 > 0 , α ≥ 0 \alpha\geq0 α ≥ 0 , 0 < A < π 0<A<\pi 0 < A < π and Δ ≠ 0 \Delta\ne0 Δ = 0 . Let the exact solution start at ( q , q ˙ , x , x ˙ ) = ( A , 0 , 0 , 0 ) (q,\dot q,x,\dot x)=(A,0,0,0) ( q , q ˙ , x , x ˙ ) = ( A , 0 , 0 , 0 ) , and set y a = ( A cos ω 0 t , − A ω 0 sin ω 0 t , x a , x ˙ a ) y_a=(A\cos\omega_0t,-A\omega_0\sin\omega_0t,x_a,\dot x_a) y a = ( A cos ω 0 t , − A ω 0 sin ω 0 t , x a , x ˙ a ) . Define E 0 = β ω 0 ( 1 − cos A ) E_0=\beta\omega_0(1-\cos A) E 0 = β ω 0 ( 1 − cos A ) and
S e = 2 E 0 K , ω − = ω 0 e − α S e , Q e = arccos ( 1 − E 0 β ω − ) , S_e=\sqrt{\frac{2E_0}{K}},\qquad
\omega_-=\omega_0e^{-\alpha S_e},\qquad
Q_e=\arccos\left(1-\frac{E_0}{\beta\omega_-}\right), S e = K 2 E 0 , ω − = ω 0 e − α S e , Q e = arccos ( 1 − β ω − E 0 ) , V e = 2 E 0 I ω − , Z e = 2 E 0 J , S a = 2 ∣ B ∣ , Z a = ∣ B ∣ ( 2 ω 0 + Ω ) . (613) V_e=\sqrt{\frac{2E_0}{I\omega_-}},\qquad
Z_e=\sqrt{\frac{2E_0}{J}},\qquad
S_a=2|B|,\qquad
Z_a=|B|(2\omega_0+\Omega). \tag*{(613)} V e = I ω − 2 E 0 , Z e = J 2 E 0 , S a = 2∣ B ∣ , Z a = ∣ B ∣ ( 2 ω 0 + Ω ) . (613) Require E 0 < 2 β ω − E_0<2\beta\omega_- E 0 < 2 β ω − and put
Q = max ( Q e , A ) , V = max ( V e , A ω 0 ) , S = max ( S e , S a ) , Q=\max(Q_e,A),\qquad V=\max(V_e,A\omega_0),\qquad S=\max(S_e,S_a), Q = max ( Q e , A ) , V = max ( V e , A ω 0 ) , S = max ( S e , S a ) , Z = max ( Z e , Z a ) , ω + = ω 0 e α S . (614) Z=\max(Z_e,Z_a),\qquad\omega_+=\omega_0e^{\alpha S}. \tag*{(614)} Z = max ( Z e , Z a ) , ω + = ω 0 e α S . (614) In the norm ∥ y ∥ E 2 = ω 0 2 q 2 + v 2 + Ω 2 x 2 + z 2 \|y\|_E^2=\omega_0^2q^2+v^2+\Omega^2x^2+z^2 ∥ y ∥ E 2 = ω 0 2 q 2 + v 2 + Ω 2 x 2 + z 2 , the exact error obeys, for every finite t ≥ 0 t\geq0 t ≥ 0 ,
∥ y ( t ) − y a ( t ) ∥ E ≤ η e L E t − 1 L E , (615) \|y(t)-y_a(t)\|_E\leq\eta\frac{e^{L_Et}-1}{L_E}, \tag*{(615)} ∥ y ( t ) − y a ( t ) ∥ E ≤ η L E e L E t − 1 , (615) with the continuous interpretation η t \eta t η t if L E = 0 L_E=0 L E = 0 , where
η v = ω 0 2 A 3 6 + α Z a A ω 0 , \eta_v=\frac{\omega_0^2A^3}{6}+\alpha Z_aA\omega_0, η v = 6 ω 0 2 A 3 + α Z a A ω 0 , η z = α β ω 0 J [ A 2 2 ( e α S a − 1 ) + e α S a A 4 24 ] , η = ( η v 2 + η z 2 ) 1 / 2 , (616) \eta_z=\frac{\alpha\beta\omega_0}{J}\left[\frac{A^2}{2}(e^{\alpha S_a}-1)+e^{\alpha S_a}\frac{A^4}{24}\right],
\qquad
\eta=(\eta_v^2+\eta_z^2)^{1/2}, \tag*{(616)} η z = J α β ω 0 [ 2 A 2 ( e α S a − 1 ) + e α S a 24 A 4 ] , η = ( η v 2 + η z 2 ) 1/2 , (616) L E 2 = ( ω 0 Q 2 2 ) 2 + ( α Z ) 2 + ( α V ) 2 + ( α β ω + Q J ω 0 ) 2 L_E^2=\left(\frac{\omega_0Q^2}{2}\right)^2+(\alpha Z)^2+(\alpha V)^2+\left(\frac{\alpha\beta\omega_+Q}{J\omega_0}\right)^2 L E 2 = ( 2 ω 0 Q 2 ) 2 + ( α Z ) 2 + ( α V ) 2 + ( J ω 0 α β ω + Q ) 2 + ( α I ω + V J ) 2 + [ α 2 ω + ( β Q 2 + I V 2 ) 2 J Ω ] 2 . (617) \qquad+\left(\frac{\alpha I\omega_+V}{J}\right)^2+\left[\frac{\alpha^2\omega_+(\beta Q^2+IV^2)}{2J\Omega}\right]^2. \tag*{(617)} + ( J α I ω + V ) 2 + [ 2 J Ω α 2 ω + ( β Q 2 + I V 2 ) ] 2 . (617)
Proof. Energy conservation gives ∣ x ∣ ≤ S e |x|\leq S_e ∣ x ∣ ≤ S e , ∣ x ˙ ∣ ≤ Z e |\dot x|\leq Z_e ∣ x ˙ ∣ ≤ Z e and ∣ q ˙ ∣ ≤ V e |\dot q|\leq V_e ∣ q ˙ ∣ ≤ V e . A continuous trajectory from A ∈ ( 0 , π ) A\in(0,\pi) A ∈ ( 0 , π ) cannot reach either q = π q=\pi q = π or q = − π q=-\pi q = − π , because its potential energy there is at least 2 β ω − > E 0 2\beta\omega_->E_0 2 β ω − > E 0 . On this connected potential well it therefore satisfies ∣ q ∣ ≤ Q e |q|\leq Q_e ∣ q ∣ ≤ Q e . The approximate trajectory lies in the enlarged box (615), so every line segment between the two trajectories lies in that box. These bounds also prevent finite-time escape of the exact solution. Separate the first-order vector field as y ˙ = A y + N ( y ) \dot{y}=Ay+N(y) y ˙ = A y + N ( y ) , where A A A is the direct sum of the two harmonic oscillators and
N ( y ) = ( 0 , α z v + ω 0 2 ( q − sin q ) , 0 , α ω 0 e − α x J [ β ( 1 − cos q ) − I v 2 / 2 ] ) . (618) N(y)=\left(0,\alpha zv+\omega_{0}^{2}(q-\sin q),0,\frac{\alpha\omega_{0}e^{-\alpha x}}{J}\left[\beta(1-\cos q)-Iv^{2}/2\right]\right). \tag*{(618)} N ( y ) = ( 0 , α z v + ω 0 2 ( q − sin q ) , 0 , J α ω 0 e − α x [ β ( 1 − cos q ) − I v 2 /2 ] ) . (618)
The linear group is an isometry in ∥ ⋅ ∥ E \|\cdot\|_{E} ∥ ⋅ ∥ E . The six possibly nonzero entries of the weighted Jacobian of N N N have the bounds whose squares appear in (617) . Here one uses ∣ sin q ∣ ≤ ∣ q ∣ |\sin q|\leq|q| ∣ sin q ∣ ≤ ∣ q ∣ and 1 − cos q ≤ q 2 / 2 1-\cos q\leq q^{2}/2 1 − cos q ≤ q 2 /2 . Their Frobenius norm bounds the operator norm, giving a Lipschitz constant L E L_{E} L E on the box.
Only the velocity equations have a residual at y a y_{a} y a . The bound ∣ q − sin q ∣ ≤ ∣ q ∣ 3 / 6 |q-\sin q|\leq|q|^{3}/6 ∣ q − sin q ∣ ≤ ∣ q ∣ 3 /6 gives η v \eta_{v} η v . For the scalar equation,
β ( 1 − cos q a ) − I v a 2 / 2 = 1 2 β A 2 cos 2 ω 0 t + ρ ( t ) , ∣ ρ ( t ) ∣ ≤ β A 4 / 24. (619) \beta(1-\cos q_{a})-Iv_{a}^{2}/2=\frac{1}{2}\beta A^{2}\cos2\omega_{0}t+\rho(t),\qquad|\rho(t)|\leq\beta A^{4}/24. \tag*{(619)} β ( 1 − cos q a ) − I v a 2 /2 = 2 1 β A 2 cos 2 ω 0 t + ρ ( t ) , ∣ ρ ( t ) ∣ ≤ β A 4 /24. (619)
Multiply by e − α x a e^{-\alpha x_{a}} e − α x a and use ∣ e − α x a − 1 ∣ ≤ e α S a − 1 |e^{-\alpha x_{a}}-1|\leq e^{\alpha S_{a}}-1 ∣ e − α x a − 1∣ ≤ e α S a − 1 to obtain η z \eta_{z} η z . Variation of constants then bounds the error by ∫ 0 t ( L E ∥ y ( s ) − y a ( s ) ∥ E + η ) d s \int_{0}^{t}(L_{E}\|y(s)-y_{a}(s)\|_{E}+\eta)\,ds ∫ 0 t ( L E ∥ y ( s ) − y a ( s ) ∥ E + η ) d s . Gronwall’s inequality gives (615) because the initial error is zero. □
The estimate controls a complete trajectory in fixed time units, not only the scalar amplitude. It can be conservative over long observation windows because it grows exponentially. A small frequency correction therefore still needs a phase-error estimate, a sufficiently long calibrated measurement, and detector noise below the signal. In particular, (615) supplies no automatic experimental resolution.
Synthetic comparisons without changing the parameters. Fix I = J = β = 1 I=J=\beta=1 I = J = β = 1 , K = 9 K=9 K = 9 , α = 0.3 \alpha=0.3 α = 0.3 , and σ 0 = 0 \sigma_{0}=0 σ 0 = 0 . For zero scalar initial data, integrate the complete equations (608)–(609) for T = 2 π T=2\pi T = 2 π . An eighth-order DOP853 calculation uses relative tolerance 2 ⋅ 10 − 12 2\cdot10^{-12} 2 ⋅ 1 0 − 12 , absolute tolerance 2 ⋅ 10 − 14 2\cdot10^{-14} 2 ⋅ 1 0 − 14 , and maximum step 0.025. The table compares the largest error on 2001 equally spaced samples with the analytic bound at T T T , which also bounds all earlier times.
Table 9.
The maximum sampled relative energy drift is below 7.5 ⋅ 10 − 14 7.5\cdot10^{-14} 7.5 ⋅ 1 0 − 14 . This is a numerical conservation check; floating-point integration is not an interval enclosure of the exact trajectory.
A separate long-time calculation uses the same coefficients and ( q , q ˙ , x , x ˙ ) = ( A , 0 , B , 0 ) (q,\dot{q},x,\dot{x})=(A,0,B,0) ( q , q ˙ , x , x ˙ ) = ( A , 0 , B , 0 ) , which matches the periodic preparation through order A 2 A^{2} A 2 . DOP853 tolerances are 2 ⋅ 10 − 11 2\cdot10^{-11} 2 ⋅ 1 0 − 11 and 2 ⋅ 10 − 13 2\cdot10^{-13} 2 ⋅ 1 0 − 13 , the maximum step is 0.15, and the duration is 240 π 240\pi 240 π ; frequency is measured from successive peaks after 40 π 40\pi 40 π . For A = 0.05 , 0.10 , 0.20 A=0.05,0.10,0.20 A = 0.05 , 0.10 , 0.20 , respectively, the measured frequencies are 0.999832510370, 0.999330165599, and 0.997322636213. Equation (612) gives 0.999832500000, 0.999330000000 0.999330000000 0.999330000000 , and 0.997320000000 0.997320000000 0.997320000000 . The coefficients approach ( 1 − ω ) / A 2 = 0.067 (1 - \omega)/A^2 = 0.067 ( 1 − ω ) / A 2 = 0.067 , whereas the frozen-scalar pendulum has leading coefficient 1 / 16 = 0.0625 1/16 = 0.0625 1/16 = 0.0625 . Its exact comparator is π / [ 2 K ( sin 2 ( A / 2 ) ) ] \pi/[2K(\sin^2(A/2))] π / [ 2 K ( sin 2 ( A /2 ))] , where K ( m ) = ∫ 0 π / 2 ( 1 − m sin 2 u ) − 1 / 2 d u K(m) = \int_{0}^{\pi/2}(1 - m\sin^2 u)^{-1/2}\mathrm{d}u K ( m ) = ∫ 0 π /2 ( 1 − m sin 2 u ) − 1/2 d u . No coefficient is fitted between these comparisons. The two initial preparations must be distinguished: the finite-time theorem above addresses the zero-scalar preparation, not the long-time peak estimator.
Figure 9. Synthetic shared-weight response. Left: frequencies from the coupled equations, compared with the frozen-scalar pendulum and the perturbative coefficient, using the same registered parameters. Right: the zero-scalar preparation at A = 0.1 A = 0.1 A = 0.1 , compared with the analytic finite-time error bound. These simulations are finite-model calculations, not observations of a new interaction.
What the finite construction requires of the field theory
The finite results give concrete tests for an extension to fluctuating geometry. Such an extension must identify its independent fields and reduce their complete constrained spectrum. Up to ten broken symmetry directions can be represented in a vierbein in suitable spontaneous Lorentz- and diffeomorphism-breaking models [18]; this is not a count of ten propagating massless particles. Gauge directions, first- and second-class constraints, and kinetic degeneracies change the physical count. Massive negative-norm or tachyonic modes must be tested as well as massless polarizations. The teleparallel equivalent of general relativity is equivalent to GR under its defining assumptions [117]; extra modes in modified teleparallel theories cannot be assigned to TEGR itself. An independent Palatini connection can be auxiliary. Its spectrum depends on its action and cannot be inferred from a universal “two plus five” count.
Constraint closure must likewise refer to the specified generators. Loll’s lattice calculation concerns the spatial diffeomorphism commutator in a particular construction and its small-spacing behavior [114]. It does not preclude exact finite internal Gauss closure, and it does not supply the missing Hamiltonian and mixed brackets of the present mother theory. If a deformed gravitational algebra is used, its generators, structure functions, Jacobi identities, error topology, and common continuum limit must all be specified. These requirements cannot be replaced by the fact that a finite symmetry exists.
Finally, a local one-loop expansion cannot decide a far negative log-coframe tail without a uniform remainder and a coercive estimate there. Nor does a nonzero projection of a constraint onto a Hessian nullspace establish ghost removal: the full reduced Hamiltonian and its inner product are needed. A fakeon prescription is a distinct physical choice involving analytic continuation and microscopic causality [3] and [4]. Discrete spatial momentum alone is not a no-go theorem for that prescription; lattice dispersion functions can have analytic extensions and time can remain continuous. Conversely, the present finite construction does not implement such a prescription. The earlier negative-residue reflection test for the unchanged higher-derivative tensor kernel remains applicable. A positive transfer operator for the retained compact model cannot remove that obstruction in a different physical sector. The new finite estimates identify what can be matched and calculated now, while leaving a precise spectral and dynamical task for the complete theory.
A conditional Galerkin continuum theorem
We give explicit sufficient hypotheses for a compactness and identification argument. They include uniform control of a stronger spatial norm and are stronger than the assumptions used in many variational SPDE existence theorems. This strength makes the convergence topology and every passage to a limit unambiguous.
Let V ↪ H ≡ H ∗ ↪ V ∗ V \hookrightarrow H \equiv H^{*} \hookrightarrow V^{*} V ↪ H ≡ H ∗ ↪ V ∗ be a Gelfand triple of separable Hilbert spaces with dense continuous embeddings and compact embedding V ↪ H V \hookrightarrow H V ↪ H . Let U U U be another separable Hilbert space. Choose nested finite-dimensional spaces H N ⊂ V H_N \subset V H N ⊂ V with dense union in V V V , and let P N : H → H N P_N : H \to H_N P N : H → H N be their orthogonal projections. Assume P N v → v P_N v \to v P N v → v in V V V for every v ∈ V v \in V v ∈ V . The uniform boundedness principle then gives sup N ∥ P N ∥ V → V < ∞ \sup_N \lVert P_N\rVert_{V\to V} < \infty sup N ∥ P N ∥ V → V < ∞ . Extend P N P_N P N to V ∗ V^{*} V ∗ by ⟨ P N ℓ , v ⟩ V ∗ , V = ⟨ ℓ , P N v ⟩ V ∗ , V \langle P_N\ell, v\rangle_{V^{*},V} = \langle\ell, P_Nv\rangle_{V^{*},V} ⟨ P N ℓ , v ⟩ V ∗ , V = ⟨ ℓ , P N v ⟩ V ∗ , V .
Let
B : V → V ∗ , Σ : V → L 2 ( U , H ) (620) B : V \to V^{*}, \qquad\Sigma: V \to\mathcal{L}_2(U,H) \tag*{(620)} B : V → V ∗ , Σ : V → L 2 ( U , H ) (620)
be Borel maps, where L 2 \mathcal{L}_2 L 2 denotes the Hilbert–Schmidt class. The drift B B B includes any correction required by the reference measure and the mobility. For the constant-mobility, flat-reference version of (62), it is − M ^ grad G S M -\widehat{M}\operatorname{grad}_{G} S_M − M grad G S M on its domain. Assume that the finite-dimensional equations
d X t N = B N ( X t N ) d t + Σ N ( X t N ) d W t , X 0 N ∈ H N , (621) \mathrm{d}X_t^N = B_N(X_t^N)\,\mathrm{d}t + \Sigma_N(X_t^N)\,\mathrm{d}W_t, \qquad X_0^N \in H_N, \tag*{(621)} d X t N = B N ( X t N ) d t + Σ N ( X t N ) d W t , X 0 N ∈ H N , (621)
have non-explosive continuous solutions. The driving process is cylindrical on U U U , and Σ N ( x ) ∈ L 2 ( U , H N ) \Sigma_N(x) \in\mathcal{L}_2(U,H_N) Σ N ( x ) ∈ L 2 ( U , H N ) . For R < ∞ R < \infty R < ∞ , set
r N , R B = sup x ∈ H N ∥ x ∥ V ≤ R ∥ B N ( x ) − P N B ( x ) ∥ V ∗ , r N , R Σ = sup x ∈ H N ∥ x ∥ V ≤ R ∥ Σ N ( x ) − P N Σ ( x ) ∥ L 2 ( U , H ) . (622) r^B_{N,R} = \sup_{\substack{x\in H_N\\ \lVert x\rVert_V\le R}} \lVert B_N(x)-P_NB(x)\rVert_{V^*},
\qquad
r^\Sigma_{N,R} = \sup_{\substack{x\in H_N\\ \lVert x\rVert_V\le R}} \lVert\Sigma_N(x)-P_N\Sigma(x)\rVert_{\mathcal{L}_2(U,H)}. \tag*{(622)} r N , R B = x ∈ H N ∥ x ∥ V ≤ R sup ∥ B N ( x ) − P N B ( x ) ∥ V ∗ , r N , R Σ = x ∈ H N ∥ x ∥ V ≤ R sup ∥ Σ N ( x ) − P N Σ ( x ) ∥ L 2 ( U , H ) . (622)
The initial laws converge weakly to a probability ν 0 \nu_0 ν 0 on H H H . One sufficient, stronger initial-data condition on a common space is
E ∥ X 0 N − P N X 0 ∥ H 2 ⟶ 0 , Law ( X 0 ) = ν 0 , E ∥ X 0 ∥ H 2 < ∞ . (623) \mathbb{E}\lVert X^N_0-P_NX_0\rVert_H^2\longrightarrow0,
\qquad
\operatorname{Law}(X_0)=\nu_0,
\qquad
\mathbb{E}\lVert X_0\rVert_H^2<\infty. \tag*{(623)} E ∥ X 0 N − P N X 0 ∥ H 2 ⟶ 0 , Law ( X 0 ) = ν 0 , E ∥ X 0 ∥ H 2 < ∞. (623)
Theorem 37 (Conditional Galerkin-to-continuum convergence). Suppose that the preceding setup and the following assumptions hold for every T < ∞ T<\infty T < ∞ .
(G1) For every R < ∞ R<\infty R < ∞ , r N , R B + r N , R Σ → 0 r^B_{N,R}+r^\Sigma_{N,R}\to0 r N , R B + r N , R Σ → 0 . On the ball K R = { v ∈ V : ∥ v ∥ V ≤ R } K_R=\{v\in V:\lVert v\rVert_V\le R\} K R = { v ∈ V : ∥ v ∥ V ≤ R } , the norms ∥ B ( v ) ∥ V ∗ \lVert B(v)\rVert_{V^*} ∥ B ( v ) ∥ V ∗ and ∥ Σ ( v ) ∥ L 2 ( U , H ) \lVert\Sigma(v)\rVert_{\mathcal{L}_2(U,H)} ∥ Σ ( v ) ∥ L 2 ( U , H ) are bounded. Whenever v n , v ∈ K R v_n,v\in K_R v n , v ∈ K R and v n → v v_n\to v v n → v in H H H , one has
⟨ B ( v n ) , ϕ ⟩ V ∗ , V → ⟨ B ( v ) , ϕ ⟩ V ∗ , V ( ϕ ∈ V ) , ∥ Σ ( v n ) − Σ ( v ) ∥ L 2 ( U , H ) → 0. (624) \langle B(v_n),\phi\rangle_{V^*,V}\to\langle B(v),\phi\rangle_{V^*,V}\quad(\phi\in V),
\qquad
\lVert\Sigma(v_n)-\Sigma(v)\rVert_{\mathcal{L}_2(U,H)}\to0. \tag*{(624)} ⟨ B ( v n ) , ϕ ⟩ V ∗ , V → ⟨ B ( v ) , ϕ ⟩ V ∗ , V ( ϕ ∈ V ) , ∥ Σ ( v n ) − Σ ( v ) ∥ L 2 ( U , H ) → 0. (624) (G2) For some δ > 0 \delta>0 δ > 0 and C T < ∞ C_T<\infty C T < ∞ , independent of N N N ,
sup N E [ sup t ≤ T ∥ X t N ∥ V 2 + δ + ( ∫ 0 T ∥ B N ( X s N ) ∥ V ∗ d s ) 2 + δ + ( ∫ 0 T ∥ Σ N ( X s N ) ∥ L 2 ( U , H ) 2 d s ) 1 + δ / 2 ] ≤ C T . (625) \sup_N\mathbb{E}\left[
\sup_{t\le T}\lVert X^N_t\rVert_V^{2+\delta}
+\left(\int_0^T\lVert B_N(X^N_s)\rVert_{V^*}\,ds\right)^{2+\delta}
+\left(\int_0^T\lVert\Sigma_N(X^N_s)\rVert_{\mathcal{L}_2(U,H)}^2\,ds\right)^{1+\delta/2}
\right]\le C_T. \tag*{(625)} N sup E t ≤ T sup ∥ X t N ∥ V 2 + δ + ( ∫ 0 T ∥ B N ( X s N ) ∥ V ∗ d s ) 2 + δ + ( ∫ 0 T ∥ Σ N ( X s N ) ∥ L 2 ( U , H ) 2 d s ) 1 + δ /2 ≤ C T . (625) In particular, this supplies the spatial localization
sup N P ( sup t ≤ T ∥ X t N ∥ V > R ) ≤ C T R − ( 2 + δ ) . (626) \sup_N\mathbb{P}\left(\sup_{t\le T}\lVert X^N_t\rVert_V>R\right)\le C_TR^{-(2+\delta)}. \tag*{(626)} N sup P ( t ≤ T sup ∥ X t N ∥ V > R ) ≤ C T R − ( 2 + δ ) . (626) (G3) For some p > 2 p>2 p > 2 , α > 0 \alpha>0 α > 0 , and C T , p < ∞ C_{T,p}<\infty C T , p < ∞ ,
sup N E ∥ X t N − X s N ∥ V ∗ p ≤ C T , p ∣ t − s ∣ 1 + α , 0 ≤ s ≤ t ≤ T . (627) \sup_N\mathbb{E}\lVert X^N_t-X^N_s\rVert_{V^*}^p\le C_{T,p}\lvert t-s\rvert^{1+\alpha},
\qquad0\le s\le t\le T. \tag*{(627)} N sup E ∥ X t N − X s N ∥ V ∗ p ≤ C T , p ∣ t − s ∣ 1 + α , 0 ≤ s ≤ t ≤ T . (627) Then the laws of X N X^N X N are tight in C ( [ 0 , T ] ; H ) C([0,T];H) C ([ 0 , T ] ; H ) . Every subsequential limit solves the martingale problem for
d X t = B ( X t ) d t + Σ ( X t ) d W t , Law ( X 0 ) = ν 0 , (628) dX_t=B(X_t)\,dt+\Sigma(X_t)\,dW_t,
\qquad
\operatorname{Law}(X_0)=\nu_0, \tag*{(628)} d X t = B ( X t ) d t + Σ ( X t ) d W t , Law ( X 0 ) = ν 0 , (628) in the following precise sense. For every ϕ ∈ V \phi\in V ϕ ∈ V ,
M t ϕ = ( X t − X 0 , ϕ ) H − ∫ 0 t ⟨ B ( X s ) , ϕ ⟩ V ∗ , V d s (629) M^\phi_t=(X_t-X_0,\phi)_H-\int_0^t\langle B(X_s),\phi\rangle_{V^*,V}\,ds \tag*{(629)} M t ϕ = ( X t − X 0 , ϕ ) H − ∫ 0 t ⟨ B ( X s ) , ϕ ⟩ V ∗ , V d s (629)
is a continuous square-integrable martingale, and for φ , ψ ∈ V \varphi,\psi\in V φ , ψ ∈ V ,
[ M φ , M ψ ] t = ∫ 0 t ( Σ ( X s ) ∗ φ , Σ ( X s ) ∗ ψ ) U d s . (630) [M^\varphi,M^\psi]_t=\int_0^t\bigl(\Sigma(X_s)^*\varphi,\Sigma(X_s)^*\psi\bigr)_U\,\mathrm{d}s. \tag*{(630)} [ M φ , M ψ ] t = ∫ 0 t ( Σ ( X s ) ∗ φ , Σ ( X s ) ∗ ψ ) U d s . (630)
These identities define the weak stochastic equation through its drift and covariance. If a representation by a cylindrical Wiener process is desired, the usual Hilbert-space martingale representation may be performed on an enlarged probability space. If this martingale problem is unique in law in the stated solution class, the entire sequence converges to that law.
Proof. We first prove compactness in a Polish path space. The moment estimate in (627) gives tight bounds for the V ∗ V^* V ∗ -Hölder seminorm with any exponent 0 < β < α / p 0<\beta<\alpha/p 0 < β < α / p , by the moment form of the Kolmogorov criterion. Together with (626) , this places the paths, with arbitrarily high probability, in sets of the form
sup t ≤ T ∥ u ( t ) ∥ V ≤ R , ∥ u ( t ) − u ( s ) ∥ V ∗ ≤ L ∣ t − s ∣ β . (631) \sup_{t\leq T}\|u(t)\|_V\leq R,\qquad\|u(t)-u(s)\|_{V^*}\leq L|t-s|^\beta. \tag*{(631)} t ≤ T sup ∥ u ( t ) ∥ V ≤ R , ∥ u ( t ) − u ( s ) ∥ V ∗ ≤ L ∣ t − s ∣ β . (631)
For z ∈ V z\in V z ∈ V , the pivot-space identity and duality give
∥ z ∥ H 2 = ⟨ z , z ⟩ V ∗ , V ≤ ∥ z ∥ V ∗ ∥ z ∥ V . (632) \|z\|_H^2=\langle z,z\rangle_{V^*,V}\leq\|z\|_{V^*}\|z\|_V. \tag*{(632)} ∥ z ∥ H 2 = ⟨ z , z ⟩ V ∗ , V ≤ ∥ z ∥ V ∗ ∥ z ∥ V . (632)
Consequently these paths have the common H H H -modulus ∥ u ( t ) − u ( s ) ∥ H 2 ≤ 2 R L ∣ t − s ∣ β \|u(t)-u(s)\|_H^2\leq2RL|t-s|^\beta ∥ u ( t ) − u ( s ) ∥ H 2 ≤ 2 R L ∣ t − s ∣ β . The closed ball K R K_R K R is compact in H H H : compactness gives relative compactness, and weak lower semicontinuity of the V V V -norm makes it closed in H H H . The metric-space Arzelà–Ascoli theorem now gives relative compactness of the path sets in C ( [ 0 , T ] ; H ) C([0,T];H) C ([ 0 , T ] ; H ) . Their closures are compact, so the laws are tight.
Choose a weakly convergent subsequence. Since C ( [ 0 , T ] ; H ) C([0,T];H) C ([ 0 , T ] ; H ) is Polish, its Skorokhod representation gives copies, still denoted X N X^N X N , X X X , such that X N → X X^N\to X X N → X uniformly in H H H , almost surely. No common Wiener process is asserted for these copies. The map
u ⟼ sup t ≤ T ∥ u ( t ) ∥ V (633) u\longmapsto\sup_{t\leq T}\|u(t)\|_V \tag*{(633)} u ⟼ t ≤ T sup ∥ u ( t ) ∥ V (633)
is lower semicontinuous on this path space, where the V V V -norm is extended by + ∞ +\infty + ∞ off V V V . This follows by taking the supremum of the lower semicontinuous functions u ↦ ∥ u ( t ) ∥ V u\mapsto\|u(t)\|_V u ↦ ∥ u ( t ) ∥ V . The limit therefore inherits the V V V -moment bound in (G2). In particular, its paths are bounded in V V V almost surely.
Fix φ ∈ V \varphi\in V φ ∈ V and R < ∞ R<\infty R < ∞ . On the event that both paths remain in K R K_R K R , decompose the drift discrepancy as
⟨ B N ( X s N ) , φ ⟩ − ⟨ B ( X s ) , φ ⟩ = ⟨ B N ( X s N ) − P N B ( X s N ) , φ ⟩ + ⟨ B ( X s N ) , P N φ − φ ⟩ + ⟨ B ( X s N ) − B ( X s ) , φ ⟩ . (634) \begin{aligned}
\langle B_N(X_s^N),\varphi\rangle-\langle B(X_s),\varphi\rangle
={}&\langle B_N(X_s^N)-P_NB(X_s^N),\varphi\rangle\\
&+\langle B(X_s^N),P_N\varphi-\varphi\rangle
+\langle B(X_s^N)-B(X_s),\varphi\rangle. \tag*{(634)}
\end{aligned} ⟨ B N ( X s N ) , φ ⟩ − ⟨ B ( X s ) , φ ⟩ = ⟨ B N ( X s N ) − P N B ( X s N ) , φ ⟩ + ⟨ B ( X s N ) , P N φ − φ ⟩ + ⟨ B ( X s N ) − B ( X s ) , φ ⟩ . (634)
The first term is bounded by r N , R B ∥ φ ∥ V r^B_{N,R}\|\varphi\|_V r N , R B ∥ φ ∥ V , and the second by C R ∥ P N φ − φ ∥ V C_R\|P_N\varphi-\varphi\|_V C R ∥ P N φ − φ ∥ V . The scalar map v ↦ ⟨ B ( v ) , φ ⟩ v\mapsto\langle B(v),\varphi\rangle v ↦ ⟨ B ( v ) , φ ⟩ is continuous on the compact H H H -space K R K_R K R , hence uniformly continuous there. Uniform convergence of the paths makes the last term tend to zero uniformly in time. The probability of the complementary event is at most 2 C T R − ( 2 + δ ) 2C_T R^{-(2+\delta)} 2 C T R − ( 2 + δ ) , using the inherited bound for the limit. First taking N → ∞ N \to\infty N → ∞ , then R → ∞ R \to\infty R → ∞ , proves convergence in probability of the drift integrals, uniformly in their upper limit.
The same argument applies to the covariance. On K R K_R K R the image Σ ( K R ) \Sigma(K_R) Σ ( K R ) is compact in L 2 ( U , H ) L_2(U,H) L 2 ( U , H ) . The operators A ↦ P N A A \mapsto P_N A A ↦ P N A are contractions on that space and converge strongly to the identity; they consequently converge uniformly on this compact image. Using the noise residual, we obtain
sup s ≤ T ∥ Σ N ( X s N ) − Σ ( X s ) ∥ L 2 ( U , H ) ⟶ 0 (635) \sup_{s \le T}\left\|\Sigma_N(X_s^N)-\Sigma(X_s)\right\|_{L_2(U,H)} \longrightarrow0 \tag*{(635)} s ≤ T sup Σ N ( X s N ) − Σ ( X s ) L 2 ( U , H ) ⟶ 0 (635)
in probability after localization. Products of the adjoint coefficients against φ , ψ \varphi,\psi φ , ψ therefore converge in their time integrals. This passage concerns covariance integrals; it does not identify stochastic integrals across the new probability space.
We next pass the martingale identities themselves to the limit. Let M N φ M_N^\varphi M N φ be the coordinate martingale defined as in (630), with B N , X N B_N,X^N B N , X N in place of B , X B,X B , X . The finite-dimensional stochastic equation and the Burkholder–Davis–Gundy inequality imply
sup N E sup t ≤ T ∣ M N φ ( t ) ∣ 2 + δ ≤ C δ , T ∥ φ ∥ H 2 + δ sup N E ( ∫ 0 T ∥ Σ N ( X s N ) ∥ 2 2 d s ) 1 + δ / 2 < ∞ . (636) \sup_N \mathbb{E}\sup_{t \le T}\left|M_N^\varphi(t)\right|^{2+\delta}
\le C_{\delta,T}\|\varphi\|_H^{2+\delta}\sup_N \mathbb{E}\left(\int_0^T\left\|\Sigma_N(X_s^N)\right\|_2^2\,ds\right)^{1+\delta/2}<\infty. \tag*{(636)} N sup E t ≤ T sup ∣ M N φ ( t ) ∣ 2 + δ ≤ C δ , T ∥ φ ∥ H 2 + δ N sup E ( ∫ 0 T Σ N ( X s N ) 2 2 d s ) 1 + δ /2 < ∞. (636)
Thus these martingales are uniformly integrable, as are their pairwise products. Their covariance integrals are uniformly integrable by the last term of (G2). The drift convergence and uniform path convergence identify the limit M φ M^\varphi M φ .
For $0 \le s < t \le T$, let F F F be a bounded continuous function of the path restricted to [ 0 , s ] [0,s] [ 0 , s ] . Each approximate law satisfies
E [ ( M N φ ( t ) − M N φ ( s ) ) F ( X N ∣ [ 0 , s ] ) ] = 0. (637) \mathbb{E}\left[(M_N^\varphi(t)-M_N^\varphi(s))F(X^N|_{[0,s]})\right]=0. \tag*{(637)} E [ ( M N φ ( t ) − M N φ ( s )) F ( X N ∣ [ 0 , s ] ) ] = 0. (637)
Uniform integrability permits passage to the limit in this identity. Apply the same argument to M N φ M N ψ − ∫ 0 ⋅ ( Σ N ∗ φ , Σ N ∗ ψ ) U d r M_N^\varphi M_N^\psi-\int_0^\cdot(\Sigma_N^*\varphi,\Sigma_N^*\psi)_U\,dr M N φ M N ψ − ∫ 0 ⋅ ( Σ N ∗ φ , Σ N ∗ ψ ) U d r . A monotone-class argument extends these identities from bounded continuous past-path functions to all bounded functions measurable with respect to the past. Hence M φ M^\varphi M φ is a martingale and (631) holds. A countable dense set in V V V , followed by approximation using the integrability bounds, gives the assertions for every test vector. The initial law is ν 0 \nu_0 ν 0 , because evaluation at time zero is continuous. This constructs a limiting martingale solution; its existence was not an assumption. Finally, uniqueness in law makes all subsequential limits equal, which proves convergence of the entire sequence. □
For completeness, a convergence-in-probability upgrade requires more than pathwise uniqueness stated without reference to the approximations. On a common stochastic basis, one must prove joint tightness of ( X N , X M , W ) (X^N, X^M, W) ( X N , X M , W ) in a Polish realization of the noise and show that every joint subsequential limit consists of two solutions driven by the same limiting noise with the same initial condition. Pathwise uniqueness then puts every joint limit on the diagonal. The Gyöngy–Krylov criterion gives convergence in probability in C ( [ 0 , T ] ; H ) C([0,T];H) C ([ 0 , T ] ; H ) . These joint identification hypotheses are additional to convergence of the marginal martingale problems.
For the master model, the preceding theorem is a conditional criterion. Its coefficient continuity, moment estimates, noise factorization, and uniqueness hypothesis must be verified for a specified realization. In particular, a formal Galerkin projection or a finite-dimensional calculation does not establish any of those analytic properties.
Infinite-dimensional invariant measures
A finite-time continuum limit and a stationary continuum law answer different questions. We now fix a time-homogeneous Markov realization of
d X t = B ( X t ) d t + Σ ( X t ) d W t (638) \mathrm{d}X_t = B(X_t)\,\mathrm{d}t + \Sigma(X_t)\,\mathrm{d}W_t \tag*{(638)} d X t = B ( X t ) d t + Σ ( X t ) d W t (638)
on H H H , with transition semigroup P t f ( x ) = E [ f ( X t x ) ] P_t f(x) = \mathbb{E}[f(X_t^x)] P t f ( x ) = E [ f ( X t x )] . The next result states exactly the Lyapunov and coupling assumptions used in the existence and uniqueness arguments.
Theorem 38 (Conditional existence and uniqueness of an infinite-dimensional invariant measure). Assume (I1)–(I4).
(I1) The chosen realization is a conservative Markov family with continuous H H H -paths, and P t P_t P t maps C b ( H ) C_b(H) C b ( H ) into C b ( H ) C_b(H) C b ( H ) .
(I2) There is a lower semicontinuous function W : H → [ 0 , ∞ ] \mathcal{W}: H \to[0,\infty] W : H → [ 0 , ∞ ] , finite at some x ∗ x_\ast x ∗ , with compact sublevel sets and constants c H > 0 c_H > 0 c H > 0 , C H ≥ 0 C_H \ge0 C H ≥ 0 such that
W ( x ) ≥ c H ∥ x ∥ H 2 − C H . (639) \mathcal{W}(x) \ge c_H\lVert x\rVert_H^2 - C_H. \tag*{(639)} W ( x ) ≥ c H ∥ x ∥ H 2 − C H . (639) (I3) For each initial point with finite W \mathcal{W} W , the process W ( X t ) \mathcal{W}(X_t) W ( X t ) has the local decomposition
W ( X t ) = W ( X 0 ) + ∫ 0 t a s d s + M t , (640) \mathcal{W}(X_t) = \mathcal{W}(X_0) + \int_0^t a_s\,\mathrm{d}s + M_t, \tag*{(640)} W ( X t ) = W ( X 0 ) + ∫ 0 t a s d s + M t , (640) with a local martingale M M M , and
a s ≤ c 0 − c 1 W ( X s ) for almost every s , c 0 ≥ 0 , c 1 > 0. (641) a_s \le c_0 - c_1\mathcal{W}(X_s) \quad\text{for almost every }s,\qquad c_0 \ge0,\qquad c_1 > 0. \tag*{(641)} a s ≤ c 0 − c 1 W ( X s ) for almost every s , c 0 ≥ 0 , c 1 > 0. (641) A justified generalized Itô formula may supply this decomposition; lower semicontinuity of W \mathcal{W} W alone does not.
(I4) There are increasing stopping times τ n → ∞ \tau_n \to\infty τ n → ∞ almost surely such that M t ∧ τ n M_{t\wedge\tau_n} M t ∧ τ n is a true martingale with mean zero and all terms in the stopped decomposition are integrable for finite t t t . The exhaustion τ n → ∞ \tau_n \to\infty τ n → ∞ is part of this assumption and must be proved when constructing a realization.
Then there exists an invariant probability π \pi π , with ∫ W d π ≤ c 0 / c 1 \int\mathcal{W}\,\mathrm{d}\pi\le c_0/c_1 ∫ W d π ≤ c 0 / c 1 and finite second moment, satisfying
∫ H P t f d π = ∫ H f d π , t ≥ 0 , f ∈ C b ( H ) . (642) \int_H P_t f\,\mathrm{d}\pi= \int_H f\,\mathrm{d}\pi,\qquad t \ge0,\quad f \in C_b(H). \tag*{(642)} ∫ H P t f d π = ∫ H f d π , t ≥ 0 , f ∈ C b ( H ) . (642)
For the uniqueness assertion assume, in addition:
(I5) Solutions from any coupling of two initial laws with finite second moment admit a common-noise variational realization. Their paths belong to V V V for almost every time, the variational Itô formula for their squared H H H -distance holds, and
2 ⟨ B ( x ) − B ( y ) , x − y ⟩ V ∗ , V + ∥ Σ ( x ) − Σ ( y ) ∥ L 2 ( U , H ) 2 ≤ − 2 κ ∥ x − y ∥ H 2 , x , y ∈ V , (643) 2\langle B(x)-B(y),x-y\rangle_{V^*,V}
+\|\Sigma(x)-\Sigma(y)\|_{L^2(U,H)}^2
\le-2\kappa\|x-y\|_H^2,\qquad x,y\in V, \tag*{(643)} 2 ⟨ B ( x ) − B ( y ) , x − y ⟩ V ∗ , V + ∥Σ ( x ) − Σ ( y ) ∥ L 2 ( U , H ) 2 ≤ − 2 κ ∥ x − y ∥ H 2 , x , y ∈ V , (643)
for some κ > 0 \kappa>0 κ > 0 . The semigroup preserves finite second moments.
Then
W 2 ( P t ∗ μ , P t ∗ ν ) ≤ e − κ t W 2 ( μ , ν ) (644) W_2(P_t^*\mu,P_t^*\nu)\le e^{-\kappa t}W_2(\mu,\nu) \tag*{(644)} W 2 ( P t ∗ μ , P t ∗ ν ) ≤ e − κ t W 2 ( μ , ν ) (644)
for all probability measures μ , ν \mu,\nu μ , ν with finite second moment. Consequently π \pi π is the unique invariant probability measure on H H H . The Wasserstein estimate concerns finite-second-moment laws, but the uniqueness conclusion does not impose a moment condition on a competing invariant probability.
Proof. Start at x ∗ x_* x ∗ . Stop the decomposition in (I3), apply (I4), and use nonnegativity of W \mathcal{W} W . This gives
E W ( X t ∧ τ n ) + c 1 E ∫ 0 t ∧ τ n W ( X s ) d s ≤ W ( x ∗ ) + c 0 t . (645) \mathbb{E}\mathcal{W}(X_{t\wedge\tau_n})+c_1\mathbb{E}\int_0^{t\wedge\tau_n}\mathcal{W}(X_s)\,\mathrm{d}s\le\mathcal{W}(x_*)+c_0t. \tag*{(645)} E W ( X t ∧ τ n ) + c 1 E ∫ 0 t ∧ τ n W ( X s ) d s ≤ W ( x ∗ ) + c 0 t . (645)
For each fixed t t t , exhaustion implies X t ∧ τ n = X t X_{t\wedge\tau_n}=X_t X t ∧ τ n = X t for all sufficiently large n n n , almost surely. Fatou’s lemma for the first term and monotone convergence for the integral therefore yield
E W ( X t ) + c 1 E ∫ 0 t W ( X s ) d s ≤ W ( x ∗ ) + c 0 t . (646) \mathbb{E}\mathcal{W}(X_t)+c_1\mathbb{E}\int_0^t\mathcal{W}(X_s)\,\mathrm{d}s\le\mathcal{W}(x_*)+c_0t. \tag*{(646)} E W ( X t ) + c 1 E ∫ 0 t W ( X s ) d s ≤ W ( x ∗ ) + c 0 t . (646)
This argument needs no unproved uniform integrability of the terminal Lyapunov values. It does need the exhausting stopping times: a continuous H H H -path need not remain in bounded sublevels of an arbitrary extended lower semicontinuous function.
Define the time-averaged laws
μ T = 1 T ∫ 0 T Law ( X s x ∗ ) d s . (647) \mu_T = \frac{1}{T}\int_0^T \operatorname{Law}(X_s^{x_*})\,\mathrm{d}s. \tag*{(647)} μ T = T 1 ∫ 0 T Law ( X s x ∗ ) d s . (647)
Tonelli’s theorem and the preceding estimate give
∫ H W d μ T ≤ W ( x ∗ ) c 1 T + c 0 c 1 . (648) \int_H \mathcal{W}\,\mathrm{d}\mu_T \le\frac{\mathcal{W}(x_*)}{c_1T}+\frac{c_0}{c_1}. \tag*{(648)} ∫ H W d μ T ≤ c 1 T W ( x ∗ ) + c 1 c 0 . (648)
For K R = { W ≤ R } K_R=\{\mathcal{W}\le R\} K R = { W ≤ R } , Markov’s inequality consequently gives
μ T ( H ∖ K R ) ≤ 1 R ( W ( x ∗ ) c 1 T + c 0 c 1 ) . (649) \mu_T(H\setminus K_R)\le\frac{1}{R}\left(\frac{\mathcal{W}(x_*)}{c_1T}+\frac{c_0}{c_1}\right). \tag*{(649)} μ T ( H ∖ K R ) ≤ R 1 ( c 1 T W ( x ∗ ) + c 1 c 0 ) . (649)
The sets K R K_R K R are compact, so { μ T : T ≥ 1 } \{\mu_T:T\ge1\} { μ T : T ≥ 1 } is tight. Prokhorov’s theorem supplies T n → ∞ T_n\to\infty T n → ∞ and μ T n ⇒ π \mu_{T_n}\Rightarrow\pi μ T n ⇒ π . Lower semicontinuity and the Portmanteau theorem show
∫ H W d π ≤ lim inf n → ∞ ∫ H W d μ T n ≤ c 0 / c 1 . (650) \int_H \mathcal{W}\,\mathrm{d}\pi\le\liminf_{n\to\infty}\int_H \mathcal{W}\,\mathrm{d}\mu_{T_n}\le c_0/c_1. \tag*{(650)} ∫ H W d π ≤ n → ∞ lim inf ∫ H W d μ T n ≤ c 0 / c 1 . (650)
Together with (640), this proves the claimed finite second moment.
For f ∈ C b ( H ) f\in C_b(H) f ∈ C b ( H ) and t ≥ 0 t\ge0 t ≥ 0 , the semigroup property gives the exact boundary-term identity
∫ H P t f d μ T − ∫ H f d μ T = 1 T [ ∫ T T + t P s f ( x ∗ ) d s − ∫ 0 t P s f ( x ∗ ) d s ] . (651) \int_H P_tf\,\mathrm{d}\mu_T-\int_H f\,\mathrm{d}\mu_T
=\frac{1}{T}\left[\int_T^{T+t}P_sf(x_*)\,\mathrm{d}s-\int_0^tP_sf(x_*)\,\mathrm{d}s\right]. \tag*{(651)} ∫ H P t f d μ T − ∫ H f d μ T = T 1 [ ∫ T T + t P s f ( x ∗ ) d s − ∫ 0 t P s f ( x ∗ ) d s ] . (651)
Its absolute value is at most 2 t ∥ f ∥ ∞ / T 2t\|f\|_\infty/T 2 t ∥ f ∥ ∞ / T . The Feller property makes both integrands on the left bounded and continuous as functions of the initial state. Weak convergence therefore proves (643). Bounded continuous functions determine Borel probability measures on H H H ; hence this is invariance of the probability measure.
Now impose (I5), put D t = X t − Y t D_t=X_t-Y_t D t = X t − Y t , and use the same driving Wiener process. The variational Itô identity reads
d ∥ D t ∥ H 2 = [ 2 ⟨ B ( X t ) − B ( Y t ) , D t ⟩ + ∥ Σ ( X t ) − Σ ( Y t ) ∥ L 2 2 ] d t + d N t , \mathrm{d}\|D_t\|_H^2
=\left[2\langle B(X_t)-B(Y_t),D_t\rangle+\|\Sigma(X_t)-\Sigma(Y_t)\|_{\mathcal{L}_2}^2\right]\mathrm{d}t+\mathrm{d}N_t, d ∥ D t ∥ H 2 = [ 2 ⟨ B ( X t ) − B ( Y t ) , D t ⟩ + ∥Σ ( X t ) − Σ ( Y t ) ∥ L 2 2 ] d t + d N t ,
N t = 2 ∫ 0 t ( Σ ( X s ) ∗ D s − Σ ( Y s ) ∗ D s , d W s ) U . (652) N_t=2\int_0^t\left(\Sigma(X_s)^*D_s-\Sigma(Y_s)^*D_s,\mathrm{d}W_s\right)_U. \tag*{(652)} N t = 2 ∫ 0 t ( Σ ( X s ) ∗ D s − Σ ( Y s ) ∗ D s , d W s ) U . (652)
All pairings are evaluated at times for which the paths lie in V V V . No continuous extension of the unbounded drift pairing to H H H is required. Multiplication by e 2 κ t e^{2\kappa t} e 2 κ t and (644) show that e 2 κ t ∥ D t ∥ H 2 e^{2\kappa t}\|D_t\|_H^2 e 2 κ t ∥ D t ∥ H 2 has nonpositive drift and a local martingale part. Stop that martingale at a localizing sequence, take expectations, and apply Fatou’s lemma to the nonnegative left-hand side as the stopping times increase to infinity. We obtain
E ∥ X t − Y t ∥ H 2 ≤ e − 2 κ t E ∥ X 0 − Y 0 ∥ H 2 . (653) \mathbb{E}\|X_t-Y_t\|_H^2\le e^{-2\kappa t}\mathbb{E}\|X_0-Y_0\|_H^2. \tag*{(653)} E ∥ X t − Y t ∥ H 2 ≤ e − 2 κ t E ∥ X 0 − Y 0 ∥ H 2 . (653)
For any initial coupling, the joint law of X t , Y t X_t,Y_t X t , Y t is a coupling of the two evolved laws. Taking the infimum over initial couplings and then square roots proves (644) . If π 1 , π 2 \pi_1,\pi_2 π 1 , π 2 are invariant and have finite second moment,
W 2 ( π 1 , π 2 ) ≤ e − κ t W 2 ( π 1 , π 2 ) . (654) W_2(\pi_1,\pi_2) \le e^{-\kappa t}W_2(\pi_1,\pi_2). \tag*{(654)} W 2 ( π 1 , π 2 ) ≤ e − κ t W 2 ( π 1 , π 2 ) . (654)
For t > 0 t>0 t > 0 the multiplier is strictly less than one, so the finite distance must vanish. Thus π 1 = π 2 \pi_1=\pi_2 π 1 = π 2 .
To remove the moment restriction on a competing invariant law, fix x ∈ H x\in H x ∈ H . Since the constructed π \pi π has finite second moment,
W 2 ( δ x , π ) 2 = ∫ H ∥ x − y ∥ H 2 π ( d y ) < ∞ . (655) W_2(\delta_x,\pi)^2=\int_H\|x-y\|_H^2\,\pi(\mathrm{d}y)<\infty. \tag*{(655)} W 2 ( δ x , π ) 2 = ∫ H ∥ x − y ∥ H 2 π ( d y ) < ∞. (655)
For any bounded Lipschitz function f : H → R f:H\to\mathbb{R} f : H → R , integrate ∣ f ( u ) − f ( v ) ∣ ≤ Lip ( f ) ∥ u − v ∥ H |f(u)-f(v)|\le\operatorname{Lip}(f)\|u-v\|_H ∣ f ( u ) − f ( v ) ∣ ≤ Lip ( f ) ∥ u − v ∥ H against a coupling and apply Cauchy–Schwarz. Taking the infimum over couplings and using invariance of π \pi π gives
∣ P t f ( x ) − π ( f ) ∣ ≤ Lip ( f ) W 2 ( P t ∗ δ x , π ) ≤ Lip ( f ) e − κ t W 2 ( δ x , π ) ⟶ 0. (656) |P_t f(x)-\pi(f)|\le\operatorname{Lip}(f)W_2(P_t^*\delta_x,\pi)\le\operatorname{Lip}(f)e^{-\kappa t}W_2(\delta_x,\pi)\longrightarrow0. \tag*{(656)} ∣ P t f ( x ) − π ( f ) ∣ ≤ Lip ( f ) W 2 ( P t ∗ δ x , π ) ≤ Lip ( f ) e − κ t W 2 ( δ x , π ) ⟶ 0. (656)
Let η \eta η be any invariant probability on H H H , with no moment assumption. Invariance gives η ( f ) = ∫ H P t f ( x ) η ( d x ) \eta(f)=\int_H P_t f(x)\,\eta(\mathrm{d}x) η ( f ) = ∫ H P t f ( x ) η ( d x ) . The integrand converges pointwise to π ( f ) \pi(f) π ( f ) , and its absolute value is bounded by ∥ f ∥ ∞ \|f\|_\infty ∥ f ∥ ∞ . Dominated convergence therefore yields η ( f ) = π ( f ) \eta(f)=\pi(f) η ( f ) = π ( f ) . Bounded Lipschitz functions determine probability measures on the Polish space H H H , so η = π \eta=\pi η = π . □
The existence proof uses compactness and time averaging; the uniqueness proof uses the additional contraction. A spectral gap or a logarithmic Sobolev inequality for one prescribed equilibrium does not, without further irreducibility or accessibility assumptions, rule out a different invariant probability supported on another invariant component. Alternative uniqueness arguments must state their own state-space, regularity, and irreducibility hypotheses.
Reference measures and reversible continuum dynamics
There is no infinite-dimensional analogue of translation-invariant Lebesgue probability. A continuum Gibbs formula must specify a reference probability and the generator compatible with it. Let { e j } j ≥ 1 \{e_j\}_{j\ge1} { e j } j ≥ 1 be an orthonormal basis of H H H , and let γ \gamma γ be the centered Gaussian probability with covariance C e j = c j e j Ce_j=c_je_j C e j = c j e j , where c j > 0 c_j>0 c j > 0 and ∑ j c j < ∞ \sum_j c_j<\infty ∑ j c j < ∞ . For a measurable interaction U U U , suppose
π G ( d x ) = Z − 1 e − U ( x ) / θ γ ( d x ) , 0 < Z = ∫ H e − U ( x ) / θ γ ( d x ) < ∞ . (657) \pi_G(\mathrm{d}x)=Z^{-1}e^{-U(x)/\theta}\gamma(\mathrm{d}x),\qquad0<Z=\int_H e^{-U(x)/\theta}\gamma(\mathrm{d}x)<\infty. \tag*{(657)} π G ( d x ) = Z − 1 e − U ( x ) / θ γ ( d x ) , 0 < Z = ∫ H e − U ( x ) / θ γ ( d x ) < ∞. (657)
The interaction U U U is the part of the energy not already encoded in γ \gamma γ . If the full formal action is placed in this density, the reference Gaussian contributes an additional quadratic energy; the resulting model must be interpreted accordingly.
Here is a precise reversible construction. Choose diagonal, constant mobilities m j > 0 m_j>0 m j > 0 . For smooth cylindrical functions define
E ( f , g ) = θ ∫ H ∑ j ≥ 1 m j ∂ j f ∂ j g d π G . (658) \mathcal{E}(f,g)=\theta\int_H\sum_{j\geq1}m_j\partial_j f\,\partial_j g\,\mathrm{d}\pi_G. \tag*{(658)} E ( f , g ) = θ ∫ H j ≥ 1 ∑ m j ∂ j f ∂ j g d π G . (658)
Each integrand has only finitely many nonzero derivatives. Assume the cylindrical domain, including constants, is dense in L 2 ( π G ) L^2(\pi_G) L 2 ( π G ) , the displayed integrals are finite, and the form is closable. Its closure is a symmetric Dirichlet form: the contraction property follows from the ordinary chain rule and passage to the closure. Let Q t Q_t Q t be its associated self-adjoint Markov semigroup. Since 1 1 1 belongs to the form domain and E ( 1 , g ) = 0 \mathcal{E}(1,g)=0 E ( 1 , g ) = 0 for every form-domain g g g , the definition of the associated generator gives 1 1 1 in its operator domain and L G 1 = 0 \mathcal{L}_G1=0 L G 1 = 0 . Consequently Q t 1 = 1 Q_t1=1 Q t 1 = 1 , and self-adjointness proves
∫ H Q t f d π G = ( Q t f , 1 ) L 2 ( π G ) = ( f , Q t 1 ) L 2 ( π G ) = ∫ H f d π G . (659) \int_H Q_t f\,\mathrm{d}\pi_G=(Q_t f,1)_{L^2(\pi_G)}=(f,Q_t1)_{L^2(\pi_G)}=\int_H f\,\mathrm{d}\pi_G. \tag*{(659)} ∫ H Q t f d π G = ( Q t f , 1 ) L 2 ( π G ) = ( f , Q t 1 ) L 2 ( π G ) = ∫ H f d π G . (659)
This establishes invariance at the semigroup level.
If the coordinate weak derivatives of U \mathcal{U} U and the relevant integration-by-parts identities exist with the required integrability, the generator on its cylindrical operator domain is
L G f = ∑ j m j [ ∂ j j f − ( ∂ j U + θ x j c j ) ∂ j f ] . (660) \mathcal{L}_G f=\sum_j m_j\left[\partial_{jj}f-\left(\partial_j\mathcal{U}+\frac{\theta x_j}{c_j}\right)\partial_j f\right]. \tag*{(660)} L G f = j ∑ m j [ ∂ j j f − ( ∂ j U + c j θ x j ) ∂ j f ] . (660)
To check the coefficient, the one-coordinate Gaussian identity gives
∫ ∂ j h d π G = ∫ h ( x j c j + ∂ j U θ ) d π G . (661) \int\partial_j h\,\mathrm{d}\pi_G=\int h\left(\frac{x_j}{c_j}+\frac{\partial_j\mathcal{U}}{\theta}\right)\mathrm{d}\pi_G. \tag*{(661)} ∫ ∂ j h d π G = ∫ h ( c j x j + θ ∂ j U ) d π G . (661)
Apply this with h = f ∂ j g h=f\partial_jg h = f ∂ j g , multiply by θ m j \theta m_j θ m j , and sum over the finitely many active coordinates. The result is − ∫ f L G g d π G = E ( f , g ) -\int f\mathcal{L}_Gg\,\mathrm{d}\pi_G=\mathcal{E}(f,g) − ∫ f L G g d π G = E ( f , g ) . Thus the reference drift − θ m j x j / c j -\theta m_jx_j/c_j − θ m j x j / c j is essential. The expression C − 1 x C^{-1}x C − 1 x need not be H H H -valued at a typical Gaussian sample, so (661) is initially a cylindrical or variational identity.
To identify Q t Q_t Q t with the transition semigroup of a specified SPDE, one must also establish that the constructed process solves that SPDE and that the relevant martingale problem is well posed. The identity ∫ L f d π G = 0 \int\mathcal{L}f\,\mathrm{d}\pi_G=0 ∫ L f d π G = 0 on a merely formal test class, without this generator or semigroup justification, is insufficient. For the four-dimensional master field, normalizability, the gauge construction, closability, and identification remain conditions on the chosen continuum realization.
10.21. An explicit infinite-dimensional equilibrium example. The analytic requirements are nonempty. The following realization gives an exact continuum process and stationary measure, while keeping its relation to the nonlinear