betaFeedback

WorksPhysics and Astronomy

A Finite-Regulator Variational–Stochastic Programme: Geometry, Gauge Fields, Matter, and the Arrow of Time

Abstract

We develop a finite-regulator variational–stochastic programme organized by four principles: structured admissible change; compatible transport, dissipation and fluctuations; selection relative to a reference path law; and consistent observation across descriptions. The central question is which relationships between motion, statistics and response follow from shared dynamical data. An exactly solved finite model shows how one transition operator gives thermal relaxation, growth-weighted population dynamics and Bayesian updating. Its unweighted occupancy, selected population endpoints and driven histories have distinct limiting distributions, all determined by the same specified rates. Equilibrium and kinetic inputs then predict a selection response. A martingale constraint gives a financial probability selection, while an underdamped oscillator connects Hamiltonian motion, thermal relaxation and fluctuation–response through one matched reservoir. These examples expose the common operations and the additional meanings supplied by each application. For the regulated field law we derive Gibbs invariance, relative free-energy balance and a quadratic path-irreversibility rate under specified boundary, covariance and reversal conditions. Matched coframe and connection variations recover Einstein–higher-curvature and Yang–Mills equations; quantified remainders delimit their usual low-energy forms. Canonical Yang–Mills dynamics require constant kinetic coefficients. Stable-pole, stress Ward-identity and universal-coupling assumptions connect matter masses, WKB trajectories and tensor-sector inertial–gravitational mass equality. A boundary-dependent Schur criterion characterizes conditional Machian closure. Additive vacuum shifts show why equilibrium alone does not select the cosmological constant, and the declared 331 candidate is a realization rather than a unique group-selection theorem. Coefficient-identification bounds and observation theorems determine which reduced quantities shared calibrated inputs fix. Exact Haar–Gaussian Gram bounds identify six lattice-action coefficients, including a volume-uniform normalized estimate on a periodic family; hidden states, sampling and operator choice delimit this identification. Finite fluid, quantum-noise and scalar-source models supply further balance and response relations. The matched finite-Galerkin fluid projection has injection, transfer and dissipation balances without establishing a continuum cascade. Bell-local, measurement-independent classical models obey CHSH, and shared noise parameters constrain coherence and repeated-readout correlations. For a specified metric common-weight scalar branch, a nonlinear conformal-slope bound, massive-fluid balance and coercive static-response estimate connect its coupling to a source-conditioned circular-radiation threshold. The trace-response illustration lies below three published binary-summary errors; it neither supplies nuclear sensitivities nor excludes arbitrary scalar theories. The construction distinguishes a shared variational organization from a uniquely selected microscopic theory. Physical entropy production requires explicit reservoirs and local detailed balance; quantum completion requires positive physical states and measurement rules, with reflection positivity additionally needed for Euclidean reconstruction. Synthetic and retrospective calculations illustrate conditional predictions. Regulator-independent quantization, unique group–matter selection, independently fixed fundamental parameters, the Yang–Mills mass gap and continuum turbulence remain separate research problems.

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 GG, the elementary variational problem is

inf⁡Q≪P{D(Q∥P)−EQG}.(1)\inf_{Q \ll P}\{\mathcal{D}(Q\Vert P)-\mathbb{E}_{Q}G\}. \tag*{(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 LL be the row generator of an irreducible continuous-time Markov chain: Lij≥0L_{ij} \ge0 for i≠ji \ne j and ∑jLij=0\sum_j L_{ij} = 0. Column distributions obey p˙=LTp\dot{p} = L^{\mathsf{T}}p. Assume detailed balance with a positive probability vector π\pi:

aij:=πiLij=πjLji(i≠j).a_{ij} := \pi_i L_{ij} = \pi_j L_{ji} \qquad(i \ne j).

Write ui=pi/πiu_i = p_i/\pi_i. Then p˙i=∑j≠iaij(uj−ui)\dot{p}_i = \sum_{j \ne i} a_{ij}(u_j-u_i). Consequently,

ddtD(p∥π)=−∑i<jaij(ui−uj)(log⁡ui−log⁡uj)≤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)}

Indeed, differentiation gives ∑ip˙ilog⁡ui\sum_i \dot{p}_i \log u_i, because ∑ip˙i=0\sum_i \dot{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), with Λ(x,x)=x\Lambda(x,x)=x, and let eie_i be the coordinate vectors. The positive semidefinite operator

K(p)=∑i<jaijΛ(ui,uj)(ei−ej)(ei−ej)TK(p) = \sum_{i<j}a_{ij}\Lambda(u_i,u_j)(e_i-e_j)(e_i-e_j)^{\mathsf{T}}

satisfies p˙=−K(p)∇pD(p∥π)\dot{p}=-K(p)\nabla_pD(p\mathbin{\|}\pi). It acts in the simplex tangent space because K(p)1=0K(p)\mathbf{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 EiE_i and θ=kBT>0\theta=k_{\mathrm{B}}T>0, set πi=Z−1e−Ei/θ\pi_i=Z^{-1}e^{-E_i/\theta}. Then θD(p∥π)=∑ipiEi+θ∑ipilog⁡pi+θlog⁡Z\theta D(p\mathbin{\|}\pi)=\sum_i p_iE_i+\theta\sum_i p_i\log p_i+\theta\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, whereas the conductances also fix the time scale.

One variational identity for a reweighted law.

Let P\mathbb{P} be the reference path law on a finite horizon. For a bounded dimensionless path score GG, put ZG=EPeGZ_G = \mathbb{E}_{\mathbb{P}}e^G and define dPG=eGdP/ZGd\mathbb{P}_G = e^Gd\mathbb{P}/Z_G. For any Q≪P\mathbb{Q} \ll\mathbb{P} with finite relative entropy,

D(Q∥P)−EQG=D(Q∥PG)−log⁡ZG.(3)D(\mathbb{Q}\Vert\mathbb{P})-\mathbb{E}_{\mathbb{Q}}G = D(\mathbb{Q}\Vert\mathbb{P}_G)-\log Z_G. \tag*{(3)}

This follows by substituting log⁡(dPG/dP)=G−log⁡ZG\log(d\mathbb{P}_G/d\mathbb{P}) = G-\log Z_G in the definition of relative entropy. Nonnegativity of relative entropy proves that PG\mathbb{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 fif_i and a positive terminal likelihood ℓi\ell_i, choose

G=∫0TfJt dt+log⁡ℓJT,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)}

Rates LijL_{ij} and fif_i have units of inverse time; ℓi\ell_i is a dimensionless likelihood. The forward and backward equations are

n(t)=eBTtp0,h(t)=eB(T−t)ℓ,ZG=n(t)Th(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).

The partition function is independent of the intermediate time because n˙=BTn\dot n=B^{\mathsf{T}}n and h˙=−Bh\dot h=-Bh. The marginal of the fixed-horizon reweighted law is

piG,T(t)=ni(t)hi(t)ZG,LijG(t)=Lijhj(t)hi(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)}

To obtain the jump rate, condition on a transition i→ji\to j during a short interval: its reference weight is Lij dtL_{ij}\,\mathrm{d}t, and its remaining future weight changes from hih_i to hjh_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 p0ihi(0)/ZGp_{0i}h_i(0)/Z_G. If the initial distribution must remain p0p_0, the minimization is instead performed separately conditional on J0=iJ_0=i. Its density is eG/hJ0(0)e^G/h_{J_0}(0), its value is −∑ip0ilog⁡hi(0)-\sum_i p_{0i}\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 LL and type ii has a constant net reproductive rate fif_i. In a continuous-time multitype branching realization, the expected counts satisfy n˙=LTn+diag⁡(f)n\dot n=L^{\mathsf{T}}n+\operatorname{diag}(f)n. Normalizing these expected counts gives

pi(t)=ni(t)∑jnj(t),p˙i=(LTp)i+pi(fi−fˉ),fˉ=∑jpjfj.(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)}

The subtraction of fˉ\bar f 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−ppT)f(\operatorname{diag} p - pp^{\mathsf{T}})f. It is a gradient of fˉ\bar{f} in the Fisher geometry on the interior simplex when ff is fixed. Its squared gradient norm is Var⁡p(f)\operatorname{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,

References

No references were extracted for this paper.

Paper details

vlnc:W_2npty4fy
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents