Introduction and summary of results
The reason it is difficult to find observational signatures that are truly stringy, rather than effects that can be mimicked by particle effective field theory, is simple: light strings look like particles. The situation can be much more interesting, however, in time-dependent backgrounds. In particular, when the string coupling grows with time, string theory also admits light strings that do not look at all like particles: Instant Folded Strings (IFSs) [1 ] . As their name suggests, these are closed folded strings nucleated at an instant. After creation, they expand, with their folds asymptotically approaching null trajectories. IFSs are light while remaining intrinsically non-particle-like. Their macroscopic spacetime extent, instantaneous creation, and folded structure have no natural counterpart in a particle effective field theory. They therefore raise the exciting possibility that time-dependent settings, and cosmology in particular, may exhibit genuinely stringy low-energy signatures.
The evidence for IFSs now goes well beyond the classical solution. The exact conformal field theory (CFT) description of the S L ( 2 , R ) / U ( 1 ) SL(2,\mathbb{R})/U(1) S L ( 2 , R ) / U ( 1 ) black hole was exploited in [2 , 3 ] to identify the worldsheet operator associated with an IFS and investigate its role in the Lorentzian version of FZZ duality [2 ] . A complementary perspective on this duality and the role of folded strings was developed in Ref. [4 ] , highlighting a possible connection to the ER=EPR proposal [5 ] . In a timelike linear-dilaton CFT, an exact worldsheet description was constructed in [6 ] , making the IFS production rate calculable. As shown in Ref. [7 ] , the resulting rate agrees with a calculation based on the timelike Liouville partition function of Ref. [8 ] . A second check comes from the S L ( 2 , R ) / U ( 1 ) SL(2,\mathbb{R})/U(1) S L ( 2 , R ) / U ( 1 ) black hole [9 , 10 , 11 , 12 ] : in its future-increasing-coupling interior, the number of IFSs encountered by an infalling observer scales parametrically with the Bekenstein–Hawking entropy [6 ] . In parallel, the collective backreaction of IFSs in homogeneous time-dependent dilaton-gravity backgrounds has been developed in Refs. [13 , 14 ] .
Despite this progress, IFSs present a basic consistency question. The same stress that underlies the proposed applications to black-hole interiors and cosmology also gives a macroscopic violation of the ANEC [15 ] . To illustrate this, we begin in 1 + 1 1+1 1 + 1 dimensions, where the energy-momentum tensor of an IFS nucleated at ( t , x ) = ( 0 , 0 ) (t,x)=(0,0) ( t , x ) = ( 0 , 0 ) is1
T u u = − μ v Θ ( v ) δ ( u ) , T v v = − μ u Θ ( u ) δ ( v ) , T u v = μ Θ ( u ) Θ ( v ) , (1) \begin{aligned}
T_{uu} &= -\mu v \Theta(v)\delta(u),\\
T_{vv} &= -\mu u \Theta(u)\delta(v),\\
T_{uv} &= \mu\Theta(u)\Theta(v), \tag*{(1)}
\end{aligned} T uu T v v T uv = − μv Θ ( v ) δ ( u ) , = − μu Θ ( u ) δ ( v ) , = μ Θ ( u ) Θ ( v ) , (1)
where μ = 1 / ( 2 π α ′ ) \mu=1/(2\pi\alpha') μ = 1/ ( 2 π α ′ ) , u = t − x u=t-x u = t − x , and v = t + x v=t+x v = t + x .
The first two terms are localized on the folds and carry negative null energy, whose magnitude grows with time. The third term fills the interior of the folded string and carries its traceful bulk stress. Although these contributions are individually nonzero, they cancel in the total Minkowski energy and momentum:
E t o t ( t ) ≡ ∫ d x T 00 ( t , x ) = 0 , P x t o t ( t ) ≡ ∫ d x T 01 ( t , x ) = 0. (2) E^{\mathrm{tot}}(t) \equiv\int dx\,T_{00}(t,x)=0,\qquad P_x^{\mathrm{tot}}(t) \equiv\int dx\,T^{01}(t,x)=0. \tag*{(2)} E tot ( t ) ≡ ∫ d x T 00 ( t , x ) = 0 , P x tot ( t ) ≡ ∫ d x T 01 ( t , x ) = 0. (2)
They do not, however, cancel in the integrated pressure:
p t o t ( t ) ≡ ∫ d x T 11 ( t , x ) = − 8 μ t Θ ( t ) , (3) p^{\mathrm{tot}}(t) \equiv\int dx\,T_{11}(t,x)=-8\mu t\Theta(t), \tag*{(3)} p tot ( t ) ≡ ∫ d x T 11 ( t , x ) = − 8 μ t Θ ( t ) , (3)
which is negative for t > 0 t>0 t > 0 and grows linearly in magnitude. This unusual combination of vanishing total energy and negative pressure is the origin of the possible importance of IFSs in cosmology and black hole interiors. At the same time, Eq. (1) implies a violation of the ANEC,
A ( v ) ≡ ∫ − ∞ ∞ d u T u u ( u , v ) = − μ Θ ( v ) v , (4) \mathcal{A}(v) \equiv\int_{-\infty}^{\infty} du\,T_{uu}(u,v)=-\mu\Theta(v)v, \tag*{(4)} A ( v ) ≡ ∫ − ∞ ∞ d u T uu ( u , v ) = − μ Θ ( v ) v , (4)
with an analogous result under u ↔ v u\leftrightarrow v u ↔ v .
This behavior is markedly different from that of an ordinary QFT in 1 + 1 1+1 1 + 1 dimensions, where any state with vanishing total energy and momentum, normally the vacuum, must saturate the ANEC [16 ] . To see this, note that energy-momentum conservation implies
d A ( v ) d v = − [ T u v ( u , v ) ] u = − ∞ u = + ∞ . (5) \frac{d\mathcal{A}(v)}{dv}=-\left[T_{uv}(u,v)\right]_{u=-\infty}^{u=+\infty}. \tag*{(5)} d v d A ( v ) = − [ T uv ( u , v ) ] u = − ∞ u = + ∞ . (5)
For conventional QFT states, the traceful component T u v T_{uv} T uv vanishes at the asymptotic endpoints of the null ray, and so A ( v ) \mathcal{A}(v) A ( v ) is independent of v v v . If the total energy and momentum vanish, this constant is necessarily zero, and the ANEC is therefore saturated.
An IFS evades this conclusion since its stress tensor retains traceful support at null infinity, so T u v T_{uv} T uv does not vanish at the endpoint of the null ray. As a result, A ( v ) \mathcal{A}(v) A ( v ) need not be constant, and the IFS can violate the ANEC despite carrying vanishing total energy and momentum.
In flat-space quantum field theory, the ANEC is closely tied to causality [17 ] . It can also be derived using monotonicity of relative entropy [18 ] , within the relative-entropy framework emphasized in Ref. [19 ] . The tension is particularly sharp because IFS production requires only a locally increasing string coupling: it can occur at arbitrarily weak coupling and arbitrarily small curvature, precisely in the regime where one would expect flat-space positivity arguments to remain valid. Moreover, the violation in (1.4) is macroscopic. Its magnitude is controlled by the extent of the folded string along the null ray, rather than by the local curvature scale.
For some time, we took a conservative view of this puzzle. Although we did not understand how it was resolved, the accumulating evidence indicated that IFSs are genuine objects in string theory. If so, the resolution should arise intrinsically within string theory, rather than through an additional constraint imposed from outside — for example, by simply declaring that IFSs are not part of the theory. Surprisingly, the resolution of this tension is straightforward: at finite string coupling, an IFS, like any other string, can split. At first sight, it is hard to imagine how IFS splitting can be relevant for the ANEC puzzle. The local splitting probability,
d P s p l i t = Γ s p l i t d u d v , Γ s p l i t ∼ g s 2 ℓ s 2 , (6) dP_{\mathrm{split}} = \Gamma_{\mathrm{split}}\,du\,dv,\qquad\Gamma_{\mathrm{split}} \sim\frac{g_s^2}{\ell_s^2}, \tag*{(6)} d P split = Γ split d u d v , Γ split ∼ ℓ s 2 g s 2 , (6)
vanishes in the weak-coupling limit. One might therefore expect splitting to give only a small perturbative correction to the classical IFS, far too small to repair the macroscopic ANEC violation in (4) . Examining a single split seems to support this. Suppose that an IFS breaks at ( v 0 , u 0 ) (v_0,u_0) ( v 0 , u 0 ) with v 0 , u 0 > 0 v_0,u_0>0 v 0 , u 0 > 0 . The split produces two segments moving in opposite directions at the speed of light (see Figure 1 ). Each has a new inner fold carrying positive null energy that grows with time. This affects the negative pressure and the ANEC violation in a way that seems correlated:
Figure 1. A single split configuration. The solid lines illustrate why the violation of the ANEC stops growing after the split. The growth of the negative contribution from the red dot is balanced by that of the green dot. The dashed segments indicate the portions of the strings contributing to the negative integrated pressure; their total length remains constant after the split.
(i) The IFS pressure stops growing for t > t 0 = ( v 0 + u 0 ) / 2 t>t_0=(v_0+u_0)/2 t > t 0 = ( v 0 + u 0 ) /2 ,
p t o t a l = { 0 , t < 0 , − 8 μ t , 0 < t < t 0 , − 8 μ t 0 , t 0 < t . (7) p^{\mathrm{total}} =
\begin{cases}
0, & t<0,\\
-8\mu t, & 0<t<t_0,\\
-8\mu t_0, & t_0<t.
\end{cases} \tag*{(7)} p total = ⎩ ⎨ ⎧ 0 , − 8 μ t , − 8 μ t 0 , t < 0 , 0 < t < t 0 , t 0 < t . (7)
The reason is that the size of the two segments is fixed after the splitting, and the growth of the negative null energy at the external fold is canceled by the growth of the positive null energy in the internal fold (the green arrows in Figure 1 ).
(ii) ANEC violation stops growing as well for v > v 0 v>v_0 v > v 0 ,
A ( v ) = { 0 , v < 0 , − μ v , 0 < v < v 0 , − μ v 0 , v 0 < v . (8) \mathcal{A}(v) =
\begin{cases}
0, & v<0,\\
-\mu v, & 0<v<v_0,\\
-\mu v_0, & v_0<v.
\end{cases} \tag*{(8)} A ( v ) = ⎩ ⎨ ⎧ 0 , − μv , − μ v 0 , v < 0 , 0 < v < v 0 , v 0 < v . (8)
as follows from the cancellation between the contribution of the external fold (the red dot) and the internal fold (the green dot).
This suggests a simple interpretation: the negative pressure ( i ) (i) ( i ) and the ANEC violation ( i i ) (ii) ( ii ) are two manifestations of the same physics viewed from two different perspectives: in Minkowski time as negative pressure, and in light-cone time as ANEC violation. It is therefore natural to expect that subsequent splittings modify the late-time evolution of the negative pressure and the ANEC violation, while preserving the qualitative link between them.
Despite its plausibility, the expectation that negative pressure and ANEC violation would continue to go hand in hand is incorrect. The distinction becomes clear when we consider the full decay. At any finite Minkowski time, only finitely many splittings have occurred, whereas a complete null ray samples contributions from arbitrarily many splits extending to arbitrarily late times.
The perpetual nature of the decay is reflected in the statistical boost symmetry,
u → λ u , v → λ − 1 v , (9) u \to\lambda u,\qquad v \to\lambda^{-1}v, \tag*{(9)} u → λ u , v → λ − 1 v , (9)
of both the stress tensor (1) and the splitting rate (1.6). At fixed u v uv uv , a boost can move a splitting event to arbitrarily late Minkowski time, t = ( u + v ) / 2 t=(u+v)/2 t = ( u + v ) /2 . Thus, a single split, or any finite sequence of splits, can describe part of the decay, but cannot capture the complete boost-invariant decay process. Put differently, boost invariance implies that a complete decay of an IFS must take an infinite amount of time, and must involve an infinite number of decay products. The decay products are yo-yos [20 , 21 ] that are detached from the remnant of the original IFS (see Figure 2 ). These yo-yos can split again, and their daughters can do the same. Their fragmentation ultimately reaches the string scale, where the classical yo-yo description gives way to the light-string modes of the terminal sector.
Figure 2. Successive emissions build the completed null cancellation. Positive folds (green) compensate the original negative fold (red). Later yo-yos, not shown, account for the remaining deficit marked by v 1 v_1 v 1 .
Our main claim is that, once all descendants and the terminal radiation sector are included, the complete decay cascade of an IFS saturates the ANEC:
A ( v ) = ∫ − ∞ ∞ d u T u u ( u , v ) = 0 , (10) \mathcal{A}(v) = \int_{-\infty}^{\infty} du\, T_{uu}(u,v) = 0, \tag*{(10)} A ( v ) = ∫ − ∞ ∞ d u T uu ( u , v ) = 0 , (10)
for every v v v (and similarly with u ↔ v u \leftrightarrow v u ↔ v ).
There are two complementary ways to prove this claim. The first is a direct fold-by-fold construction, illustrated schematically in Figure 2 . Each split creates positive-energy null folds. As the remnant continues to split, the contributions from the emitted yo-yos form a sum that compensates the negative null energy carried by the external fold of the original IFS.
The second is a conservation argument, illustrated in Figure 3 . Since the stress vanishes before nucleation, ∂ v T u u + ∂ u T u v = 0 \partial_v T_{uu} + \partial_u T_{uv} = 0 ∂ v T uu + ∂ u T uv = 0 relates the green null cut A \mathcal{A} A to the red segment B \mathcal{B} B :
Figure 3. The three integration cuts A \mathcal{A} A , B \mathcal{B} B and C \mathcal{C} C , used in the conservation and pressure arguments, are shown superimposed over a numerical simulation of the complete decay cascade. The dashed hyperbola illustrates the stochastic boost invariance of the cascade. Green A \mathcal{A} A is the ray v = v 0 v=v_0 v = v 0 , continued into the vacuum at u < 0 u<0 u < 0 and truncated at u = U u=U u = U . Red B \mathcal{B} B lies at u = U u=U u = U ; its integral runs from v = 0 v=0 v = 0 to v = v 0 v=v_0 v = v 0 , opposite to the red arrow. Blue C \mathcal{C} C is a constant-Minkowski-time slice.
A ( v 0 ; U ) ≡ ∫ A d u T u u = − ∫ B d v T u v . (11) \mathcal{A}(v_0;U) \equiv\int_{\mathcal{A}} du\,T_{uu} = -\int_{\mathcal{B}} dv\,T_{uv}. \tag*{(11)} A ( v 0 ; U ) ≡ ∫ A d u T uu = − ∫ B d v T uv . (11)
Statistical boost invariance gives ⟨ T u v ( u , v ) ⟩ = f ( u v ) ≥ 0 \langle T_{uv}(u,v)\rangle=f(uv)\geq0 ⟨ T uv ( u , v )⟩ = f ( uv ) ≥ 0 . Writing I ( z ) ≡ ∫ 0 z d ζ f ( ζ ) I(z)\equiv\int_0^z d\zeta\,f(\zeta) I ( z ) ≡ ∫ 0 z d ζ f ( ζ ) , we obtain
⟨ A ( v 0 ; U ) ⟩ = − ∫ 0 v 0 d v f ( U v ) = − I ( U v 0 ) U . (12) \langle\mathcal{A}(v_0;U)\rangle=-\int_0^{v_0}dv\,f(Uv)=-\frac{I(Uv_0)}{U}. \tag*{(12)} ⟨ A ( v 0 ; U )⟩ = − ∫ 0 v 0 d v f ( U v ) = − U I ( U v 0 ) . (12)
Saturation along each fixed null ray therefore requires only I ( z ) / z → 0 I(z)/z\to0 I ( z ) / z → 0 .
The physical reason this condition holds is that splitting prevents the integrated trace from growing without bound. Let W t o t ( t ) ≡ − ∫ C d x T μ μ W_{\mathrm{tot}}(t)\equiv-\int_{\mathcal{C}}dx\,T^\mu_{\ \mu} W tot ( t ) ≡ − ∫ C d x T μ μ denote the trace weight on the blue slice C \mathcal{C} C . A single split already halts its linear growth. More strongly, if the first split occurs at t s t_s t s , all subsequent splittings and yo-yo oscillations obey W t o t ( t ) ≤ 8 μ t s W_{\mathrm{tot}}(t)\leq8\mu t_s W tot ( t ) ≤ 8 μ t s .2 Since ⟨ t s ⟩ = O ( Γ s p l i t − 1 / 2 ) \langle t_s\rangle=O(\Gamma_{\mathrm{split}}^{-1/2}) ⟨ t s ⟩ = O ( Γ split − 1/2 ) , this gives ⟨ W t o t ( t ) ⟩ ≲ μ / Γ s p l i t \langle W_{\mathrm{tot}}(t)\rangle\lesssim\mu/\sqrt{\Gamma_{\mathrm{split}}} ⟨ W tot ( t )⟩ ≲ μ / Γ split without any assumption about the subsequent decay law.
Positivity and boost invariance translate this bounded trace into geometric dilution:
I ( z ) z ≤ ⟨ W t o t ( z ) ⟩ 4 z , (13) \frac{I(z)}{z}\leq\frac{\langle W_{\mathrm{tot}}(\sqrt{z})\rangle}{4\sqrt{z}}, \tag*{(13)} z I ( z ) ≤ 4 z ⟨ W tot ( z )⟩ , (13)
where changing variables from x x x to ζ = t 2 − x 2 \zeta= t^2 - x^2 ζ = t 2 − x 2 along C C C gives ⟨ W t o t ( t ) ⟩ ≥ 4 I ( t 2 ) / t \langle W_{\mathrm{tot}}(t)\rangle\ge4I(t^2)/t ⟨ W tot ( t )⟩ ≥ 4 I ( t 2 ) / t . Because the numerator is bounded, I ( z ) / z I(z)/z I ( z ) / z falls at least as 1 / z = 1 / u v 1/\sqrt{z} = 1/\sqrt{uv} 1/ z = 1/ uv , even if the trace weight itself does not decay. This geometric dilution suffices for ANEC saturation without a particular decay law or terminal conversion. The stronger terminal result I ( ∞ ) < ∞ I(\infty) < \infty I ( ∞ ) < ∞ improves the convergence to O ( U − 1 ) O(U^{-1}) O ( U − 1 ) .
The same figure explains why the integrated pressure can nevertheless remain negative. Since W t o t = E t o t − p t o t W_{\mathrm{tot}} = E_{\mathrm{tot}} - p_{\mathrm{tot}} W tot = E tot − p tot and the complete state has E t o t = 0 E_{\mathrm{tot}} = 0 E tot = 0 ,
⟨ p t o t ( t ) ⟩ = − ⟨ W t o t ( t ) ⟩ = − 4 ∫ C d x f ( t 2 − x 2 ) < 0 , t > 0. (14) \langle p_{\mathrm{tot}}(t)\rangle= -\langle W_{\mathrm{tot}}(t)\rangle= -4 \int_C dx\, f(t^2-x^2) < 0, \qquad t > 0. \tag*{(14)} ⟨ p tot ( t )⟩ = − ⟨ W tot ( t )⟩ = − 4 ∫ C d x f ( t 2 − x 2 ) < 0 , t > 0. (14)
The null folds carry no trace. A completed null ray collects their compensating flux, while a finite spacelike slice still intersects the traceful remnant and live descendants.
In summary, the complete IFS decay cascade exhibits a remarkable combination of properties:
It saturates the ANEC and has vanishing total energy and momentum, yet retains negative integrated pressure at every finite time.
This raises many questions about the IFS decay cascade and its possible applications. For example:
(i) How fast is the ANEC saturated? Saturation of the complete null integral does not tell us how rapidly the negative flux disappears as the cutoff is increased. A benchmark is provided by the geometric bound (1.13), which gives
⟨ A ( v 0 ; U ) ⟩ ≳ − μ Γ s p l i t v 0 U . (15) \langle A(v_0;U)\rangle\gtrsim-\frac{\mu}{\sqrt{\Gamma_{\mathrm{split}}}}\sqrt{\frac{v_0}{U}}. \tag*{(15)} ⟨ A ( v 0 ; U )⟩ ≳ − Γ split μ U v 0 . (15)
For every fixed v 0 v_0 v 0 , the residual approaches zero as U → ∞ U \to\infty U → ∞ , but the convergence is not uniform in v 0 v_0 v 0 . This is particularly important for homogeneous cosmological settings, where coarse-graining involves an unbounded family of translated null rays. Consequently, completing each null ray and taking the homogeneous limit need not commute. This motivates investigating how subsequent yo-yo splittings modify the rate of ANEC saturation and its dependence on v 0 v_0 v 0 .
Another motivation comes from the gravitational time advance generated by the IFS decay cascade, which is proportional to ⟨ A ( v 0 ; U ) ⟩ \langle A(v_0;U)\rangle ⟨ A ( v 0 ; U )⟩ . Since the effective cutoff depends on the wavelength of the probing matter, U ∼ λ U \sim\lambda U ∼ λ , the time advance may be strongly wavelength-dependent. This could play an important role in determining how different infalling observers experience the interior of the S L ( 2 ) / U ( 1 ) SL(2)/U(1) S L ( 2 ) / U ( 1 ) black hole, where IFSs are expected to be abundant.
(ii) What is the total integrated pressure as a function of time? The identity p t o t = − W t o t p_{\mathrm{tot}}=-W_{\mathrm{tot}} p tot = − W tot leaves open how rapidly fragmentation removes the surviving tension. We determine the time and coupling dependence of p t o t p_{\mathrm{tot}} p tot and its crossover to the string-scale terminal regime. This one-event response is also the input to the homogeneous IFS cosmologies of Refs. [13 , 14 ] .
(iii) What are the spatial moments of the stress tensor of an IFS cascade? For the homogeneous evolution, the pressure is enough, but perturbations require more information: how localized is the traceful stress, and how do its spatial moments grow? These moments determine the leading nonlocal response of an IFS cascade and are therefore relevant for cosmological perturbations and possible observational imprints.
(iv) What happens to the ANEC saturation in higher dimensions? In D > 2 D>2 D > 2 the situation appears to be more complicated. For example, the light ray associated with the ANEC and the IFS need not be aligned, and similarly the compensating radiation can pass through neighboring rays.
(v) Does ANEC saturation persist in cosmological settings? In particular, does the fact that the IFS decay cascade saturates the ANEC in flat spacetime imply that, at finite coupling, IFSs cannot mediate a cosmological bounce?
To address these questions, it is useful to introduce logarithmic Milne coordinates in the future wedge, in which boosts act as translations:
η = 1 2 ln ( u v ) , ϕ = 1 2 ln ( u / v ) , d s 2 = e 2 η ( − d η 2 + d ϕ 2 ) , (16) \eta=\frac{1}{2}\ln(uv),\qquad\phi=\frac{1}{2}\ln(u/v),\qquad ds^2=e^{2\eta}(-d\eta^2+d\phi^2), \tag*{(16)} η = 2 1 ln ( uv ) , ϕ = 2 1 ln ( u / v ) , d s 2 = e 2 η ( − d η 2 + d ϕ 2 ) , (16)
with
e η = u v (17) e^\eta=\sqrt{uv} \tag*{(17)} e η = uv (17)
being the invariant Milne scale of the cascade, while a boost acts as a translation of ϕ \phi ϕ .
Figure 4 summarizes our findings. The horizontal direction is rapidity and the vertical direction is logarithmic Milne time. A particular decay history contains random split points, but the ensemble is statistically translation invariant in ϕ \phi ϕ . The physically meaningful crossovers are therefore horizontal bands, set by e η = u v e^\eta=\sqrt{uv} e η = uv . The figure displays three regimes. At early Milne time,
Figure 4. Finite-coupling IFS decay cascade in logarithmic future-wedge coordinates η = 1 2 [ ln u + ln v ] \eta= \frac{1}{2}[\ln u + \ln v] η = 2 1 [ ln u + ln v ] , ϕ = 1 2 [ ln u − ln v ] \phi= \frac{1}{2}[\ln u - \ln v] ϕ = 2 1 [ ln u − ln v ] . Translations in ϕ \phi ϕ are boosts, so boost invariance of the local decay process appears as statistical translation invariance in rapidity. The lower light-blue shaded region is the traceful original IFS bulk/remnant, the orange structures are finite classical yo-yo-like descendants, the black crosses mark splitting and terminal events, and the blue upper region consists of light string modes. The upper band depicts terminal conversion, not a sharp surface beyond which every history has zero trace. The ensemble-averaged trace decreases toward the future null endpoint; the integrated condition for ANEC saturation is given by Eq. (12) .
e η ≲ Γ s p l i t − 1 / 2 ∼ ℓ s g s , (18) e^\eta\lesssim\Gamma_{\mathrm{split}}^{-1/2}\sim\frac{\ell_s}{g_s}, \tag*{(18)} e η ≲ Γ split − 1/2 ∼ g s ℓ s , (18)
the expected number of splitting events in a causal patch is still of order one. The configuration is close to the unbroken classical IFS. The first active splits form a causal staircase: moving from left to right, successive split points have increasing u u u and decreasing v v v . This staircase cuts the original traceful bulk into remnant pieces and finite yo-yos. It supplies the initial data for the later cascade, but it does not eliminate the traceful stress.
The next regime is the semiclassical yo-yo cascade,
ℓ s g s ≪ e η ≪ ℓ s g s 2 (19) \frac{\ell_s}{g_s} \ll e^\eta\ll\frac{\ell_s}{g_s^2} \tag*{(19)} g s ℓ s ≪ e η ≪ g s 2 ℓ s (19)
where the finite strings are large compared with the string scale and are well described as classical yo-yos. Each yo-yo contains traceful bulk stress, oscillates periodically, and can split into smaller yo-yos. Since there is no additional scale in this window, the decay becomes self-similar: trace weight is redistributed among progressively smaller descendants and gradually depleted. The orange network in Figure 4 is the visual form of this self-similar trace transport.3
Finally, for
e η ≳ ℓ s g s 2 , (20) e^\eta\gtrsim\frac{\ell_s}{g_s^2}, \tag*{(20)} e η ≳ g s 2 ℓ s , (20)
finite descendants have typically reached the string scale and decay into light string modes. A string-scale yo-yo has period O ( ℓ s ) O(\ell_s) O ( ℓ s ) and splits or decays with probability O ( g s 2 ) O(g_s^2) O ( g s 2 ) per oscillation, so its terminal lifetime is O ( ℓ s / g s 2 ) O(\ell_s/g_s^2) O ( ℓ s / g s 2 ) . Above this scale the finite traceful descendants have mostly become traceless light string modes, T u v = 0 T_{uv}=0 T uv = 0 , shown as the blue region in the figure. This is an asymptotic conversion regime, rather than a simultaneous end to all traceful evolution. The persistent remnant continues to seed new trees near the outgoing folds. These regimes determine the approach to ANEC saturation and the decay of the surviving negative pressure, quantified below.
The Milne picture also suggests a useful analogy between a single IFS and a microscopic 1 + 1 1+1 1 + 1 -dimensional big bang. Defining a Milne time τ M ≡ e η = u v \tau_M \equiv e^\eta= \sqrt{uv} τ M ≡ e η = uv , we see that a single IFS is nucleated at the tip τ M = 0 \tau_M=0 τ M = 0 of an open Milne universe, while boosts act as translations of the comoving coordinate ϕ \phi ϕ . Statistical boost invariance is therefore spatial homogeneity in this cosmological description, and the three horizontal bands in Figure 4 may be viewed as successive epochs. The growing multiplicity of decay products also suggests coarse-grained entropy production analogous to reheating. The first finite yo-yos are emitted when Γ s p l i t u v = O ( 1 ) \Gamma_{\mathrm{split}}^{uv}=O(1) Γ split uv = O ( 1 ) , so their mass is of order m s / g s m_s/g_s m s / g s . Each subsequent split produces daughters of smaller invariant mass, so along each branch the cascade flows from m ∼ m s / g s m\sim m_s/g_s m ∼ m s / g s , through progressively lighter extended strings, to m = O ( m s ) m=O(m_s) m = O ( m s ) , where the semiclassical yo-yo description crosses over to massless, traceless light-string excitations. In this sense, a single IFS realizes a microscopic 1 + 1 1+1 1 + 1 -dimensional big-bang history: a Milne origin, a first generation of extended strings with invariant mass m s / g s m_s/g_s m s / g s , a self-similar fragmentation era, and a final radiative phase.
In the semiclassical yo-yo regime, the answers to questions ( i ) (i) ( i ) –( i i i ) (iii) ( iii ) are controlled by a single exponent,
q I F S = 3 − 2 2 ≃ 0.17157. (21) q_{\mathrm{IFS}}=3-2\sqrt{2}\simeq0.17157. \tag*{(21)} q IFS = 3 − 2 2 ≃ 0.17157. (21)
Equivalently, the self-similar branching has Malthusian exponent
κ I F S = 1 − q I F S ≃ 0.82843. (22) \kappa_{\mathrm{IFS}}=1-q_{\mathrm{IFS}}\simeq0.82843. \tag*{(22)} κ IFS = 1 − q IFS ≃ 0.82843. (22)
This exponent characterizes the fractal-like part of the cascade: it is the power governing the growth of the number of terminal leaves produced by a large finite yo-yo. The answer to question ( i ) (i) ( i ) is
⟨ A ( v ; U ) ⟩ ∝ − U α v β , with α = 2 − 2 , β = 2 − 1. (23) \langle\mathcal{A}(v;U)\rangle\propto-U^{\alpha}v^{\beta}, \qquad\text{with}\quad\alpha=\sqrt{2}-2,\qquad\beta=\sqrt{2}-1. \tag*{(23)} ⟨ A ( v ; U )⟩ ∝ − U α v β , with α = 2 − 2 , β = 2 − 1. (23)
Conservation and boost invariance fix β − α = 1 \beta-\alpha=1 β − α = 1 , while
β + α = − q I F S , (24) \beta+\alpha=-q_{\mathrm{IFS}}, \tag*{(24)} β + α = − q IFS , (24)
relates the dynamical decay of the surviving tension to the approach to null cancellation, as derived in § 5 . The answer to question ( i i ) (ii) ( ii ) is that the pressure, again in the yo-yo regime, behaves as
p t o t ( t ) ∼ − μ Γ s p l i t ( Γ s p l i t t ) − q I F S , 1 ≪ Γ s p l i t t ≪ g s − 1 . (25) p_{\mathrm{tot}}(t)\sim-\frac{\mu}{\sqrt{\Gamma_{\mathrm{split}}}}\left(\sqrt{\Gamma_{\mathrm{split}}}\,t\right)^{-q_{\mathrm{IFS}}},\qquad1\ll\sqrt{\Gamma_{\mathrm{split}}}\,t\ll g_s^{-1}. \tag*{(25)} p tot ( t ) ∼ − Γ split μ ( Γ split t ) − q IFS , 1 ≪ Γ split t ≪ g s − 1 . (25)
Using Γ s p l i t ∼ g s 2 / ℓ s 2 \Gamma_{\mathrm{split}}\sim g_s^2/\ell_s^2 Γ split ∼ g s 2 / ℓ s 2 , this may be written as
p t o t ( t ) ∼ 1 ℓ s g s − 1 − q I F S ( t / ℓ s ) − q I F S , ℓ s g s ≪ t ≪ ℓ s g s 2 , (26) p_{\mathrm{tot}}(t)\sim\frac{1}{\ell_s}g_s^{-1-q_{\mathrm{IFS}}}(t/\ell_s)^{-q_{\mathrm{IFS}}},\qquad\frac{\ell_s}{g_s}\ll t\ll\frac{\ell_s}{g_s^2}, \tag*{(26)} p tot ( t ) ∼ ℓ s 1 g s − 1 − q IFS ( t / ℓ s ) − q IFS , g s ℓ s ≪ t ≪ g s 2 ℓ s , (26)
which, interestingly, exhibits a non-integer power also of the string coupling, g s − 2 + κ I F S g_s^{-2+\kappa_{\mathrm{IFS}}} g s − 2 + κ IFS . This anomalous coupling dependence is another manifestation of the fractal cascade: it arises from resumming arbitrarily many splitting histories and therefore cannot be associated with any fixed order in the usual genus expansion. In a sense, the ordinary g s − 2 g_s^{-2} g s − 2 scaling is dressed by the Malthusian exponent of the self-similar decay cascade.
This is the intermediate semiclassical power law generated by self-similar trace-weight transport. Terminal conversion makes the invariant integrated trace profile finite,
I ∞ ≡ ∫ 0 ∞ d z f ( z ) < ∞ . (27) I_\infty\equiv\int_0^\infty dz\,f(z)<\infty. \tag*{(27)} I ∞ ≡ ∫ 0 ∞ d z f ( z ) < ∞. (27)
As derived in § 6 , this same quantity fixes both the terminal null residual and the surviving negative pressure:
⟨ A ( v ; U ) ⟩ ∼ − I ∞ U , U v ≫ ℓ s 2 g s 4 , (28) \langle\mathcal{A}(v;U)\rangle\sim-\frac{I_\infty}{U},\qquad Uv\gg\frac{\ell_s^2}{g_s^4}, \tag*{(28)} ⟨ A ( v ; U )⟩ ∼ − U I ∞ , U v ≫ g s 4 ℓ s 2 , (28)
⟨ p t o t ( t ) ⟩ ∼ − 4 I ∞ t , t ≫ ℓ s g s 2 . (29) \langle p_{\mathrm{tot}}(t)\rangle\sim-\frac{4I_\infty}{t},\qquad t\gg\frac{\ell_s}{g_s^2}. \tag*{(29)} ⟨ p tot ( t )⟩ ∼ − t 4 I ∞ , t ≫ g s 2 ℓ s . (29)
Thus the complete null integral vanishes while the pressure remains negative at every finite time. The terminal bound 0 ≤ − ⟨ A ( v ; U ) ⟩ ≤ I ∞ / U 0\le-\langle\mathcal{A}(v;U)\rangle\le I_\infty/U 0 ≤ − ⟨ A ( v ; U )⟩ ≤ I ∞ / U is uniform in v > 0 v>0 v > 0 at a fixed IFS origin. It does not, however, provide an integrable bound over an infinite distribution of source positions, which translate both null coordinates. Null completion and homogeneous coarse graining therefore need not commute. This distinction, and the retarded stress relevant to cosmology, are discussed in Sec. 8.
For question ( i i i ) (iii) ( iii ) , boost invariance fixes the relative coefficients of the spatial trace moments as well as their powers. Define
M n ( W ) ( t ) = ∫ d x x n ( T 00 − T 11 ) ( x , t ) . (30) M_n^{(W)}(t) = \int dx\, x^n (T_{00}-T_{11})(x,t). \tag*{(30)} M n ( W ) ( t ) = ∫ d x x n ( T 00 − T 11 ) ( x , t ) . (30)
In the semiclassical regime, the trace has a self-similar spatial profile spread across the light-cone scale ∣ x ∣ ∼ t |x| \sim t ∣ x ∣ ∼ t . Hence, for the left-right symmetric ensemble,
⟨ M 2 n ( W ) ( t ) ⟩ ∝ t 2 n ⟨ W t o t ( t ) ⟩ ∝ t 2 n − q I F S , ⟨ M 2 n + 1 ( W ) ( t ) ⟩ = 0. (31) \langle M_{2n}^{(W)}(t)\rangle\propto t^{2n}\langle W_{\mathrm{tot}}(t)\rangle\propto t^{2n-q_{\mathrm{IFS}}}, \qquad\langle M_{2n+1}^{(W)}(t)\rangle= 0. \tag*{(31)} ⟨ M 2 n ( W ) ( t )⟩ ∝ t 2 n ⟨ W tot ( t )⟩ ∝ t 2 n − q IFS , ⟨ M 2 n + 1 ( W ) ( t )⟩ = 0. (31)
The relative coefficients are generally different from unity in this regime. After terminal conversion, the surviving trace instead concentrates near the outgoing folds, so each even moment approaches t 2 n t^{2n} t 2 n times the total trace weight. This terminal localization underlies the narrow-band description used later for cosmology.
Energy moments obey different constraints. Define
E n ( t ) = ∫ d x x n T 00 ( x , t ) , (32) E_n(t) = \int dx\, x^n T_{00}(x,t), \tag*{(32)} E n ( t ) = ∫ d x x n T 00 ( x , t ) , (32)
then conservation gives
E 0 = 0 , E 1 = 0 (33) E_0 = 0,\qquad E_1 = 0 \tag*{(33)} E 0 = 0 , E 1 = 0 (33)
since the IFS is nucleated with zero total energy, momentum, and center-of-energy moment. The first nontrivial energy moment obeys
d 2 E 2 d t 2 = 2 p t o t ( t ) , (34) \frac{d^2 E_2}{dt^2} = 2p_{\mathrm{tot}}(t), \tag*{(34)} d t 2 d 2 E 2 = 2 p tot ( t ) , (34)
and therefore, in the semiclassical regime,
E 2 ( t ) ∝ − t 2 − q I F S . (35) E_2(t) \propto-t^{2-q_{\mathrm{IFS}}}. \tag*{(35)} E 2 ( t ) ∝ − t 2 − q IFS . (35)
The derivation and the higher moment generalizations are given in Appendix E. Question ( i v ) (iv) ( i v ) is addressed in Section 7, where we show that transverse momentum conservation implies that an IFS is nucleated in a state that saturates the ANEC for all light rays. We also discuss the amount of ANEC violation associated with other, more localized states. Question ( v ) (v) ( v ) is the subject of Section 8. There we derive the leading homogeneous transport equations for the highly boosted strings and the remnant source, and show that the IFS decay cascade can violate both the NEC and ANEC in a contracting universe. This violation, in turn, can mediate a cosmological bounce.
Geometry of the finite-coupling cascade
We construct the decay directly on the worldsheet, showing how local breaks divide the expanding IFS into a persistent remnant and detached strings. Throughout this section we work in the flat-space, locally constant-coupling approximation. The sole local input is a uniform splitting probability per unit active worldsheet area, where “active” denotes traceful parent bulk that has not been removed by an earlier split. A split removes the causal future of its point from the parent bulk and emits positive null folds. Only the causal-minimal splits affect the surviving parent stress; their records form a monotone staircase.
The construction separates two classes of objects: the unbounded remnant retains the original null arms, while crossings of neighboring positive folds emit finite yo-yos. Their invariant mass is set by the product of their null extents and their boost by the ratio. The sampler below determines these quantities without generating irrelevant candidate splits.
Fixed-background decay model. We condition on nucleation in a growing-dilaton background and evolve the leading small-gradient stress at fixed g s g_s g s , neglecting gravitational backreaction. Complete-null limits refer to this fixed-background decay model. In a varying background it describes a finite history while the coupling and curvature change little. Writing Q = ∂ t ln g s Q = \partial_t \ln g_s Q = ∂ t ln g s , this requires Q t ≪ 1 Qt \ll1 Qt ≪ 1 , and in particular Q t × ≪ 1 Qt_\times\ll1 Q t × ≪ 1 to reach the terminal crossover within the approximation.
We work in null coordinates u = t − x u = t - x u = t − x , v = t + x v = t + x v = t + x . The physical starting point is the IFS splitting argument [7 ] : splitting is local and, away from the folds, does not distinguish one position along the string from another.
At locally constant coupling, we describe these equivalent bulk patches by a uniform splitting probability per unit worldsheet area. The closed-string splitting amplitude is of order g s g_s g s , giving a probability of order g s 2 g_s^2 g s 2 per string-scale area. Restoring dimensions gives
d P s p l i t = Γ s p l i t d u d v . (36) dP_{\mathrm{split}} = \Gamma_{\mathrm{split}}\,du\,dv. \tag*{(36)} d P split = Γ split d u d v . (36)
Here Γ s p l i t ∼ g s 2 / ℓ s 2 \Gamma_{\mathrm{split}} \sim g_s^2/\ell_s^2 Γ split ∼ g s 2 / ℓ s 2 ; its normalization absorbs the numerical coefficient and the area convention. We implement the local law by independent splitting events, so the splitting points form a homogeneous Poisson process on any active bulk region. Thus, for a region A \mathcal{A} A , the probability that no split occurs in A \mathcal{A} A is
P 0 ( A ) = exp [ − Γ s p l i t ∣ A ∣ ] . (37) P_0(\mathcal{A}) = \exp[-\Gamma_{\mathrm{split}}|\mathcal{A}|]. \tag*{(37)} P 0 ( A ) = exp [ − Γ split ∣ A ∣ ] . (37)
The growing IFS exposes an area of order t 2 t^2 t 2 , so the first split becomes likely when Γ s p l i t t 2 = O ( 1 ) \Gamma_{\mathrm{split}}t^2 = O(1) Γ split t 2 = O ( 1 ) . This recovers the scale t ∼ ℓ s / g s t \sim\ell_s/g_s t ∼ ℓ s / g s estimated in [7 ] . The construction below follows the subsequent splittings and their causal removal of active bulk.
For a parent worldsheet domain D \mathcal{D} D , a split at p = ( u p , v p ) p=(u_p,v_p) p = ( u p , v p ) creates two positive-energy null folds and removes the causal future
J + ( p ) = { ( u , v ) ∣ u > u p , v > v p } (38) J^{+}(p)=\{(u,v)\mid u>u_p,\ v>v_p\} \tag*{(38)} J + ( p ) = {( u , v ) ∣ u > u p , v > v p } (38)
from the bulk of the parent object. For several splits in D \mathcal{D} D , the surviving parent bulk is therefore
D p a r e n t = D ∖ ⋃ i J + ( p i ) . (39) \mathcal{D}_{\mathrm{parent}}=\mathcal{D}\setminus\bigcup_i J^{+}(p_i). \tag*{(39)} D parent = D ∖ i ⋃ J + ( p i ) . (39)
Equivalently, the sequential active splitting process may be coupled to a homogeneous Poisson cloud on the original parent domain D \mathcal{D} D . Points lying in the causal future of an earlier causal-minimal split are candidates in regions that have already been removed, and therefore do not affect the parent stress tensor. By the independent-increments property of the Poisson process, retaining only the causal-minimal points gives the same distribution for the physical parent history as sampling sequentially on the active region. Regions lying in the future of more than one split are removed only once. In the numerical construction, this is implemented as a mask on the parent contribution to T u v T_{uv} T uv .
The boundary left by these causal removals forms a monotone staircase. The positive folds emitted from neighboring staircase vertices intersect, producing finite yo-yos. Each yo-yo may then split by the same mechanism, generating the recursive cascade shown in Figure 5 .
Figure 5. The persistent IFS remnant (blue) and its decay cascade. Causal-minimal splits { ( u i , v i ) } \{(u_i,v_i)\} {( u i , v i )} leave a staircase attached to the original null arms. Neighboring positive folds cross to emit first-generation yo-yos (green), whose own splits produce later generations (yellow). A finite sampling window does not terminate the remnant.
Causal records and the common staircase construction
Consider the null rectangle D = [ u a , u b ] × [ v a , v b ] \mathcal{D}=[u_a,u_b]\times[v_a,v_b] D = [ u a , u b ] × [ v a , v b ] , a causal diamond in ( t , x ) (t,x) ( t , x ) . With p ≺ q p\prec q p ≺ q when u p < u q u_p<u_q u p < u q and v p < v q v_p<v_q v p < v q , the physical records are the Poisson points with no other point in their causal past. Ordering them as p i = ( u i , v i ) p_i=(u_i,v_i) p i = ( u i , v i ) , i = 1 , … , N i=1,\ldots,N i = 1 , … , N , by increasing u u u gives
u 1 < u 2 < ⋯ < u N , v 1 > v 2 > ⋯ > v N . (40) u_1<u_2<\cdots<u_N,\qquad v_1>v_2>\cdots>v_N. \tag*{(40)} u 1 < u 2 < ⋯ < u N , v 1 > v 2 > ⋯ > v N . (40)
Set ( u 0 , v 0 ) = ( u a , v b ) (u_0,v_0)=(u_a,v_b) ( u 0 , v 0 ) = ( u a , v b ) and u N + 1 = u b u_{N+1}=u_b u N + 1 = u b to complete the sampling boundary; these are not additional splitting events. The surviving parent bulk is the union of non-overlapping rectangles
S = ⋃ i = 0 N R i , R i = [ u i , u i + 1 ] × [ v a , v i ] . (41) \mathcal{S}=\bigcup_{i=0}^{N}R_i,\qquad R_i=[u_i,u_{i+1}]\times[v_a,v_i]. \tag*{(41)} S = i = 0 ⋃ N R i , R i = [ u i , u i + 1 ] × [ v a , v i ] . (41)
Thus T u v s t a i r = μ χ S T_{uv}^{\mathrm{stair}}=\mu\chi_{\mathcal{S}} T uv stair = μ χ S , where χ S \chi_{\mathcal{S}} χ S is the indicator of the surviving region. Overlapping removed futures are counted only once.
The same staircase also determines the detached yo-yos. Whenever consecutive vertices carry physical folds, the fold from p i p_i p i at fixed v i v_i v i meets the fold from p i + 1 p_{i+1} p i + 1 at fixed u i + 1 u_{i+1} u i + 1 and emits a yo-yo with null extents
Δ u i = u i + 1 − u i , Δ v i = v i − v i + 1 . (42) \Delta u_i = u_{i+1} - u_i,\qquad\Delta v_i = v_i - v_{i+1}. \tag*{(42)} Δ u i = u i + 1 − u i , Δ v i = v i − v i + 1 . (42)
Its birth event and opposite cross-corner are
b i = ( u i + 1 , v i ) , b ˉ i = ( u i , v i + 1 ) , (43) b_i = (u_{i+1},v_i),\qquad\bar{b}_i = (u_i,v_{i+1}), \tag*{(43)} b i = ( u i + 1 , v i ) , b ˉ i = ( u i , v i + 1 ) , (43)
which span an invariant rectangle of area A i Y = Δ u i Δ v i A_i^{Y} = \Delta u_i\Delta v_i A i Y = Δ u i Δ v i . For the remnant, this rule applies to adjacent physical records; artificial sampling edges do not emit daughters. For a finite yo-yo, the physical turning points also carry folds and close the staircase, as described below. The bulk construction is common to both; their boundary folds differ.
Direct sampling of the records. Suppose that the current vertex is ( u i , v i ) (u_i, v_i) ( u i , v i ) . The event u i + 1 > U u_{i+1} > U u i + 1 > U is equivalent to finding no split in [ u i , U ] × [ v a , v i ] [u_i, U] \times[v_a, v_i] [ u i , U ] × [ v a , v i ] , so Eq. (37) gives
P ( u i + 1 ≥ U ∣ u i , v i ) = exp [ − Γ s p l i t ( U − u i ) ( v i − v a ) ] . (44) P(u_{i+1} \ge U \mid u_i, v_i) = \exp[-\Gamma_{\mathrm{split}}(U-u_i)(v_i-v_a)] . \tag*{(44)} P ( u i + 1 ≥ U ∣ u i , v i ) = exp [ − Γ split ( U − u i ) ( v i − v a )] . (44)
The horizontal step is therefore exponential, and the next height is uniform over the surviving vertical interval:
Δ u i ∼ Exp ( Γ s p l i t ( v i − v a ) ) , (45) \Delta u_i \sim\operatorname{Exp}\bigl(\Gamma_{\mathrm{split}}(v_i-v_a)\bigr), \tag*{(45)} Δ u i ∼ Exp ( Γ split ( v i − v a ) ) , (45)
v i + 1 = v a + Y i ( v i − v a ) , Y i ∼ U ( 0 , 1 ) . (46) v_{i+1}=v_a+Y_i(v_i-v_a), \qquad Y_i\sim U(0,1). \tag*{(46)} v i + 1 = v a + Y i ( v i − v a ) , Y i ∼ U ( 0 , 1 ) . (46)
If the trial u i + Δ u i u_i+\Delta u_i u i + Δ u i exceeds u b u_b u b , there is no further record and the sampler stops at the final bulk rectangle. Otherwise the point is accepted and the rule is repeated. This one-dimensional Markov chain samples the causal-minimal records without generating the full Poisson cloud. Their number grows only logarithmically with the expected number of candidates in a large domain. Ordering by u u u is a sampling convention; the underlying process is symmetric under u ↔ v u\leftrightarrow v u ↔ v .
The persistent remnant and its emission scale
The original IFS retains its two negative-energy null folds while neighboring positive folds detach as first-generation yo-yos (Figure 5 ). Its full staircase is unbounded; the rectangle D \mathcal{D} D is only a sampling window, extended by continuing the same record construction. Within that window the bulk is given by Eq. (41) . The remnant is an emitting subsystem: at daughter births it transfers null momentum to the detached sector,
∂ v T u u ( 0 ) + ∂ u T u v ( 0 ) = − J u , ∂ u T v v ( 0 ) + ∂ v T u v ( 0 ) = − J v . (47) \partial_v T_{uu}^{(0)}+\partial_u T_{uv}^{(0)}=-J_u,\qquad\partial_u T_{vv}^{(0)}+\partial_v T_{uv}^{(0)}=-J_v. \tag*{(47)} ∂ v T uu ( 0 ) + ∂ u T uv ( 0 ) = − J u , ∂ u T v v ( 0 ) + ∂ v T uv ( 0 ) = − J v . (47)
The descendants have the opposite divergences, so the complete stress is conserved, including at emission. The fold stresses and vertex currents J u J_u J u , J v J_v J v are given in App. § 11 .§ 1 .
The source’s invariant scale follows from the record statistics. With the IFS origin at ( u a , v a ) = ( 0 , 0 ) (u_a,v_a)=(0,0) ( u a , v a ) = ( 0 , 0 ) , the continuing process defined by Eqs. (45) –(46) has independent marks
X i ≡ Γ s p l i t v i ( u i + 1 − u i ) ∼ Exp ( 1 ) , Y i ≡ v i + 1 v i ∼ U ( 0 , 1 ) , (48) X_i\equiv\Gamma_{\mathrm{split}}v_i(u_{i+1}-u_i)\sim\operatorname{Exp}(1),\qquad Y_i\equiv\frac{v_{i+1}}{v_i}\sim U(0,1), \tag*{(48)} X i ≡ Γ split v i ( u i + 1 − u i ) ∼ Exp ( 1 ) , Y i ≡ v i v i + 1 ∼ U ( 0 , 1 ) , (48)
also independent of the earlier staircase history. Here X i X_i X i is the newly swept area in units of Γ s p l i t − 1 \Gamma_{\mathrm{split}}^{-1} Γ split − 1 , and Y i Y_i Y i is the fraction of the previous height retained. The dimensionless record area obeys the affine recursion
A i ≡ Γ s p l i t u i v i , A i + 1 = Y i ( A i + X i ) . (49) A_i\equiv\Gamma_{\mathrm{split}}u_iv_i,\qquad A_{i+1}=Y_i(A_i+X_i). \tag*{(49)} A i ≡ Γ split u i v i , A i + 1 = Y i ( A i + X i ) . (49)
Fresh area replenishes the process, while the contraction erases its initial condition: two trajectories driven by the same marks have a separation proportional to ∏ i < N Y i \prod_{i<N} Y_i ∏ i < N Y i , which decays exponentially because ⟨ ln Y ⟩ = − 1 \langle\ln Y\rangle=-1 ⟨ ln Y ⟩ = − 1 . Here ⟨ ⋅ ⟩ \langle\cdot\rangle ⟨ ⋅ ⟩ denotes stochastic expectation.
For the stationary Laplace transform Φ A ( s ) = ⟨ e − s A ⟩ \Phi_A(s)=\langle e^{-sA}\rangle Φ A ( s ) = ⟨ e − s A ⟩ , independence gives
Φ A ( s ) = ∫ 0 1 Φ A ( s y ) 1 + s y d y = 1 s ∫ 0 s Φ A ( t ) 1 + t d t , \Phi_A(s)=\int_0^1\frac{\Phi_A(sy)}{1+sy}\,\mathrm{d}y=\frac{1}{s}\int_0^s\frac{\Phi_A(t)}{1+t}\,\mathrm{d}t, Φ A ( s ) = ∫ 0 1 1 + sy Φ A ( sy ) d y = s 1 ∫ 0 s 1 + t Φ A ( t ) d t ,
Φ A ′ ( s ) = − Φ A ( s ) 1 + s , Φ A ( 0 ) = 1 ⟹ Φ A ( s ) = 1 1 + s . (50) \Phi_A'(s)=-\frac{\Phi_A(s)}{1+s},\qquad\Phi_A(0)=1\quad\Longrightarrow\quad\Phi_A(s)=\frac{1}{1+s}. \tag*{(50)} Φ A ′ ( s ) = − 1 + s Φ A ( s ) , Φ A ( 0 ) = 1 ⟹ Φ A ( s ) = 1 + s 1 . (50)
Thus the unique stationary law and its area moments are
A = Γ s p l i t u v ∼ Exp ( 1 ) , ⟨ ( u v ) m ⟩ = m ! Γ s p l i t m , ⟨ u v ⟩ = σ ( u v ) = Γ s p l i t − 1 . (51) A=\Gamma_{\mathrm{split}}uv\sim\operatorname{Exp}(1),\qquad\left\langle(uv)^m\right\rangle=\frac{m!}{\Gamma_{\mathrm{split}}^m},\qquad\langle uv\rangle=\sigma(uv)=\Gamma_{\mathrm{split}}^{-1}. \tag*{(51)} A = Γ split uv ∼ Exp ( 1 ) , ⟨ ( uv ) m ⟩ = Γ split m m ! , ⟨ uv ⟩ = σ ( uv ) = Γ split − 1 . (51)
This law describes the continuing remnant staircase away from the edges of a finite sampling window. It has order-one relative area fluctuations and restores the u ↔ v u\leftrightarrow v u ↔ v symmetry hidden by the coordinate-ordered sampler.
For adjacent records in this regime, the daughter rule (42) yields
Γ s p l i t A i Y = X i ( 1 − Y i ) , ⟨ ( Γ s p l i t A i Y ) m ⟩ = m ! m + 1 . (52) \Gamma_{\mathrm{split}}A_i^Y=X_i(1-Y_i),\qquad\left\langle\left(\Gamma_{\mathrm{split}}A_i^Y\right)^m\right\rangle=\frac{m!}{m+1}. \tag*{(52)} Γ split A i Y = X i ( 1 − Y i ) , ⟨ ( Γ split A i Y ) m ⟩ = m + 1 m ! . (52)
In particular,
⟨ A i Y ⟩ = 1 2 Γ s p l i t ∼ ℓ s 2 g s 2 . (53) \left\langle A_i^Y\right\rangle=\frac{1}{2\Gamma_{\mathrm{split}}}\sim\frac{\ell_s^2}{g_s^2}. \tag*{(53)} ⟨ A i Y ⟩ = 2 Γ split 1 ∼ g s 2 ℓ s 2 . (53)
The mass–area relation (57) therefore gives a typical first-generation mass of order m s / g s m_s/g_s m s / g s . Splitting selects this invariant scale, rather than a last emission time: later records lie farther along the outgoing null arms and emit increasingly boosted yo-yos. This is the geometric origin of the continuing source in Figure 4 .
Finite yo-yos: physical endpoints and conservation
A finite yo-yo sweeps the bounded rectangle shown in Figure 6 , D Y = [ u a , u b ] × [ v a , v b ] \mathcal{D}_Y=[u_a,u_b]\times[v_a,v_b] D Y = [ u a , u b ] × [ v a , v b ] during one oscillation, with extents Δ u = u b − u a \Delta u=u_b-u_a Δ u = u b − u a and Δ v = v b − v a \Delta v=v_b-v_a Δ v = v b − v a . Its diagonal vertices c − = ( u a , v a ) c_-=(u_a,v_a) c − = ( u a , v a ) and c + = ( u b , v b ) c_+=(u_b,v_b) c + = ( u b , v b ) are consecutive fold-coincidence events, at which its spatial extent vanishes. The other two corners are its physical turning points. They close the staircase by supplying the endpoint vertices
Figure 6. Splitting during one yo-yo oscillation. The diagonal corners are fold-coincidence events; the other two corners are physical turning points. Together with the interior split vertices, the turning points close the staircase around the surviving blue bulk. The turning-point folds supply the end daughters; adjacent interior records supply the others. Thus N N N splits give N + 1 N+1 N + 1 outgoing yo-yos (green). These physical endpoints distinguish this construction from a truncated remnant window.
p 0 = ( u a , v b ) , p N + 1 = ( u b , v a ) . (54) p_0=(u_a,v_b),\qquad p_{N+1}=(u_b,v_a). \tag*{(54)} p 0 = ( u a , v b ) , p N + 1 = ( u b , v a ) . (54)
Unlike the edges of a remnant sampling window, these endpoints carry physical folds and participate in the daughter construction. They are not stochastic splits; only the N N N interior records are. The bounded and unbounded staircases refer to worldsheet domains; both the yo-yo and the IFS are closed strings.
The surviving bulk is again Eq. (41) , and the daughter extents and birth events follow from Eqs. (42) –(43) , now for i = 0 , … , N i=0,\ldots,N i = 0 , … , N . Thus N ≥ 1 N\geq1 N ≥ 1 physical splits produce N + 1 N+1 N + 1 daughters. For N = 0 N=0 N = 0 , the single outgoing piece is the same unsplit parent continuing its oscillation, not a new generation. The next oscillation sweeps a translated rectangle with the same extents.
The physical endpoints make conservation transparent: the daughter extents telescope to those of the parent,
∑ i = 0 N Δ u i = Δ u , ∑ i = 0 N Δ v i = Δ v . (55) \sum_{i=0}^{N}\Delta u_i=\Delta u,\qquad\sum_{i=0}^{N}\Delta v_i=\Delta v. \tag*{(55)} i = 0 ∑ N Δ u i = Δ u , i = 0 ∑ N Δ v i = Δ v . (55)
For any yo-yo, the stress tensor in App. § 11 .§ 2 gives
E Y = μ ( Δ u + Δ v ) , P Y = μ ( Δ v − Δ u ) , (56) E_Y=\mu(\Delta u+\Delta v),\qquad P_Y=\mu(\Delta v-\Delta u), \tag*{(56)} E Y = μ ( Δ u + Δ v ) , P Y = μ ( Δ v − Δ u ) , (56)
m Y = E Y 2 − P Y 2 = 2 μ Δ u Δ v . (57) m_Y=\sqrt{E_Y^2-P_Y^2}=2\mu\sqrt{\Delta u\Delta v}. \tag*{(57)} m Y = E Y 2 − P Y 2 = 2 μ Δ u Δ v . (57)
We use P = ∫ d x T 01 = ∫ d x ( T u u − T v v ) P=\int dx\,T^{01}=\int dx\,(T_{uu}-T_{vv}) P = ∫ d x T 01 = ∫ d x ( T uu − T v v ) ; with u = t − x u=t-x u = t − x , v = t + x v=t+x v = t + x , a right-moving null packet contributes to T u u T_{uu} T uu . Applying the same relations to each daughter immediately gives
∑ i = 0 N E i = E Y , ∑ i = 0 N P i = P Y . (58) \sum_{i=0}^{N} E_i = E_Y, \qquad\sum_{i=0}^{N} P_i = P_Y. \tag*{(58)} i = 0 ∑ N E i = E Y , i = 0 ∑ N P i = P Y . (58)
Energy and momentum conservation are therefore encoded in the staircase geometry, without a separate branching rule.
Recursive sampling and the dilute single-split limit
Applying the same construction to each daughter closes the recursive cascade. Every yo-yo samples Eqs. (45) –(46) in its own oscillation rectangle, of invariant area A Y = Δ u Δ v A_Y = \Delta u \Delta v A Y = Δ u Δ v . Deep in the cascade,
Λ Y ≡ Γ s p l i t A Y ≪ 1. (59) \Lambda_Y \equiv\Gamma_{\mathrm{split}} A_Y \ll1. \tag*{(59)} Λ Y ≡ Γ split A Y ≪ 1. (59)
Most oscillations then contain no split. A more efficient implementation can therefore skip empty oscillations analytically and invoke the staircase sampler only in the first occupied oscillation. This is the cycle-skipping rule used in the numerical implementation.
For the analytic discussion below, a key simplification comes from the dilute single-split limit. The probability that one oscillation contains no split is
P 0 ( Y ) = e − Γ s p l i t A Y . (60) P_0(Y) = e^{-\Gamma_{\mathrm{split}} A_Y}. \tag*{(60)} P 0 ( Y ) = e − Γ split A Y . (60)
If K K K is the number of complete unsplit oscillations before the first occupied one, then
P ( K = k ) = e − k Λ Y ( 1 − e − Λ Y ) , k = 0 , 1 , 2 , … (61) P(K = k) = e^{-k\Lambda_Y}\left(1-e^{-\Lambda_Y}\right), \qquad k = 0,1,2,\ldots \tag*{(61)} P ( K = k ) = e − k Λ Y ( 1 − e − Λ Y ) , k = 0 , 1 , 2 , … (61)
Let N Y ∼ Poisson ( Λ Y ) N_Y \sim\operatorname{Poisson}(\Lambda_Y) N Y ∼ Poisson ( Λ Y ) be the number of Poisson splitting points in one oscillation. For Λ Y ≪ 1 \Lambda_Y \ll1 Λ Y ≪ 1 , an occupied oscillation contains exactly one Poisson point, and hence one causal-minimal record, with probability 1 − O ( Λ Y ) 1-O(\Lambda_Y) 1 − O ( Λ Y ) , since
P ( N Y ≥ 2 ∣ N Y ≥ 1 ) = Λ Y 2 + O ( Λ Y 2 ) . P(N_Y \ge2 \mid N_Y \ge1) = \frac{\Lambda_Y}{2} + O(\Lambda_Y^2). P ( N Y ≥ 2 ∣ N Y ≥ 1 ) = 2 Λ Y + O ( Λ Y 2 ) .
Moreover, the position of the first split in an occupied oscillation is approximately uniform over the yo-yo worldsheet,
ρ ( u s , v s ∣ N Y ≥ 1 ) = 1 A Y [ 1 + O ( Λ Y ) ] . (62) \rho(u_s,v_s \mid N_Y \ge1) = \frac{1}{A_Y}\left[1+O(\Lambda_Y)\right]. \tag*{(62)} ρ ( u s , v s ∣ N Y ≥ 1 ) = A Y 1 [ 1 + O ( Λ Y ) ] . (62)
Conditioned on there being exactly one split, the position is uniformly distributed over D Y \mathcal{D}_Y D Y for any value of Λ Y \Lambda_Y Λ Y .
Thus the late semiclassical cascade is dominated by long sequences of empty oscillations, separated by isolated splitting periods containing a single, nearly uniform split. This is the single-split regime used in the renewal analysis below, while the numerical construction retains the full staircase whenever more than one causal-minimal split occurs.
Kinematics and trace-weight flow in the decay cascade
Section 2 gives the stochastic geometry of the finite-coupling cascade. The pressure is governed by a smaller set of kinematic variables, which we introduce here. Each finite yo-yo conserves energy and momentum when it splits. The pressure is controlled by the trace weight
W = − ∫ d x T μ μ , (63) W=-\int dx\,T^\mu_{\ \mu}, \tag*{(63)} W = − ∫ d x T μ μ , (63)
which, for the complete zero-energy IFS decay state, is equal to minus the integrated pressure. The remnant injects trace weight into highly boosted first-generation yo-yos; later splittings transport this weight through the hierarchy and deplete it when the descendants reach finite boost.
Let T μ ν ( 0 ) T^{(0)}_{\mu\nu} T μν ( 0 ) denote the remnant bulk and folds of § 2.2 , with each positive boundary fold ending at its first daughter-birth vertex. Its directly emitted yo-yos form generation Y 1 \mathcal{Y}_1 Y 1 ; each subsequent split advances the generation by one. The semiclassical stress is
T μ ν t o t = T μ ν ( 0 ) + ∑ g = 1 ∞ ∑ Y ∈ Y g l i v e T μ ν Y . (64) T^{\mathrm{tot}}_{\mu\nu}=T^{(0)}_{\mu\nu}+\sum_{g=1}^{\infty}\sum_{Y\in\mathcal{Y}^{\mathrm{live}}_g}T^Y_{\mu\nu}. \tag*{(64)} T μν tot = T μν ( 0 ) + g = 1 ∑ ∞ Y ∈ Y g live ∑ T μν Y . (64)
The sum includes only yo-yos whose live intervals intersect the time slice under consideration. Contributions belonging to distinct objects are added even when their spacetime supports overlap. Terminal string-scale decay products will be included separately when the semiclassical cascade is cut off.
The equations below distinguish exact instantaneous quantities from the slowly varying, coarse-grained ones used in the renewal analysis. The identity p t o t = − W t o t p_{\mathrm{tot}}=-W_{\mathrm{tot}} p tot = − W tot is exact for the complete zero-energy cascade. The renewal calculation is first formulated for ensemble- and period-averaged yo-yo weights; Appendix § 13 .§ 5 then shows that the exact instantaneous ensemble pressure has the same leading semiclassical asymptotic. The boost-spectrum calculation retains the period-averaged weight unless stated otherwise.
Finite-yo-yo kinematics
The null extents introduced above determine both the area exposed to splitting and the elapsed Minkowski time:
A Y = Δ u Δ v , T Y = Δ u + Δ v 2 . (65) A_Y=\Delta u\Delta v,\qquad T_Y=\frac{\Delta u+\Delta v}{2}. \tag*{(65)} A Y = Δ u Δ v , T Y = 2 Δ u + Δ v . (65)
Here T Y T_Y T Y is the time between consecutive fold-coincidence events. One such interval sweeps one rectangle with splitting intensity Γ s p l i t d u d v \Gamma_{\mathrm{split}}\,du\,dv Γ split d u d v . Following labeled folds for two rectangles doubles both the time and the area, leaving the splitting rate unchanged.
Using the energy, momentum and mass in Eqs. (56) –(57) , the boost and velocity are
γ Y = E Y m Y = Δ u + Δ v 2 Δ u Δ v , β Y = P Y E Y = Δ v − Δ u Δ u + Δ v . (66) \gamma_Y = \frac{E_Y}{m_Y} = \frac{\Delta u + \Delta v}{2\sqrt{\Delta u\Delta v}}, \qquad\beta_Y = \frac{P_Y}{E_Y} = \frac{\Delta v - \Delta u}{\Delta u + \Delta v}. \tag*{(66)} γ Y = m Y E Y = 2 Δ u Δ v Δ u + Δ v , β Y = E Y P Y = Δ u + Δ v Δ v − Δ u . (66)
The sign agrees with the center-of-mass displacement between consecutive fold coincidences: Δ x = ( Δ v − Δ u ) / 2 \Delta x = (\Delta v - \Delta u)/2 Δ x = ( Δ v − Δ u ) /2 and Δ t = ( Δ u + Δ v ) / 2 \Delta t = (\Delta u + \Delta v)/2 Δ t = ( Δ u + Δ v ) /2 . The invariant mass is fixed by the area Δ u Δ v \Delta u\Delta v Δ u Δ v , whereas the boost is fixed by the aspect ratio Δ u / Δ v \Delta u/\Delta v Δ u /Δ v . This is the reason the first-generation source can emit yo-yos with an approximately fixed invariant scale while their boosts grow with the emission time.
Trace weight and pressure
Energy moves between a yo-yo’s bulk and folds during each oscillation, so its instantaneous pressure oscillates. The cascade calculation follows the period average. Integrating the stress over one oscillation gives the particle-like relation
p ˉ Y = γ Y m Y − m Y γ Y = γ Y m Y β Y 2 = P Y 2 E Y . (67) \bar{p}_Y = \gamma_Y m_Y - \frac{m_Y}{\gamma_Y} = \gamma_Y m_Y \beta_Y^2 = \frac{P_Y^2}{E_Y}. \tag*{(67)} p ˉ Y = γ Y m Y − γ Y m Y = γ Y m Y β Y 2 = E Y P Y 2 . (67)
Origin of the negative integrated pressure. Thus an isolated positive-energy yo-yo has non-negative period-averaged pressure, p ˉ Y = P Y 2 / E Y ≥ 0 \bar{p}_Y = P_Y^2/E_Y \ge0 p ˉ Y = P Y 2 / E Y ≥ 0 . The negative integrated pressure belongs to the combined zero-energy state of the remnant and its descendants.
For any contribution O \mathcal{O} O , define its trace weight
W O = − ∫ O d x T μ μ = ∫ O d x ( T 00 − T 11 ) = 4 ∫ O d x T u v . (68) W_{\mathcal{O}} = -\int_{\mathcal{O}} dx\,T^\mu{}_\mu= \int_{\mathcal{O}} dx\,(T_{00} - T_{11}) = 4\int_{\mathcal{O}} dx\,T_{uv}. \tag*{(68)} W O = − ∫ O d x T μ μ = ∫ O d x ( T 00 − T 11 ) = 4 ∫ O d x T uv . (68)
The null folds do not contribute to this combination. It measures the traceful bulk content carried by the remnant and the live finite yo-yos. For a yo-yo at oscillation phase τ \tau τ , the instantaneous weight is
W Y i n s t ( τ ) = E Y − p Y ( τ ) , (69) W_Y^{\mathrm{inst}}(\tau) = E_Y - p_Y(\tau), \tag*{(69)} W Y inst ( τ ) = E Y − p Y ( τ ) , (69)
and its period average is
W Y ≡ 1 T Y ∫ 0 T Y d τ W Y i n s t ( τ ) = E Y − p ˉ Y = m Y γ Y = 4 μ Δ u Δ v Δ u + Δ v . (70) W_Y \equiv\frac{1}{T_Y}\int_0^{T_Y} d\tau\,W_Y^{\mathrm{inst}}(\tau) = E_Y - \bar{p}_Y = \frac{m_Y}{\gamma_Y} = 4\mu\frac{\Delta u\Delta v}{\Delta u + \Delta v}. \tag*{(70)} W Y ≡ T Y 1 ∫ 0 T Y d τ W Y inst ( τ ) = E Y − p ˉ Y = γ Y m Y = 4 μ Δ u + Δ v Δ u Δ v . (70)
An unsplit oscillation has maximum bulk trace 4 μ min ( Δ u , Δ v ) 4\mu\min(\Delta u,\Delta v) 4 μ min ( Δ u , Δ v ) , so
0 ≤ W Y i n s t ( t ) ≤ 2 W Y . (71) 0 \le W_Y^{\mathrm{inst}}(t) \le2W_Y. \tag*{(71)} 0 ≤ W Y inst ( t ) ≤ 2 W Y . (71)
Removing bulk by fragmentation cannot increase this parent contribution. The branching process transports W Y = m Y / γ Y W_Y = m_Y/\gamma_Y W Y = m Y / γ Y . The exact instantaneous trace still oscillates within an individual yo-yo period; Appendix C.5 controls its ensemble asymptotic directly from physical split events, so no separate assumption about phase randomization is needed.
The corresponding weight of the zeroth-generation remnant is
W 0 ( t ) ≡ ∫ d x ( T 00 ( 0 ) − T 11 ( 0 ) ) . (72) W_0(t) \equiv\int dx \left(T_{00}^{(0)}-T_{11}^{(0)}\right). \tag*{(72)} W 0 ( t ) ≡ ∫ d x ( T 00 ( 0 ) − T 11 ( 0 ) ) . (72)
and the exact total weight is
W t o t ( t ) = W 0 ( t ) + ∑ g = 1 ∞ ∑ Y ∈ Y g l i v e W Y i n s t ( t ) . (73) W_{\mathrm{tot}}(t) = W_0(t) + \sum_{g=1}^{\infty}\sum_{Y\in\mathcal{Y}^{\mathrm{live}}_g} W_Y^{\mathrm{inst}}(t). \tag*{(73)} W tot ( t ) = W 0 ( t ) + g = 1 ∑ ∞ Y ∈ Y g live ∑ W Y inst ( t ) . (73)
The complete IFS decay state has vanishing total energy. Therefore
p t o t ( t ) ≡ ∫ d x T 11 ( t , x ) = E t o t ( t ) − W t o t ( t ) = − W t o t ( t ) . (74) p_{\mathrm{tot}}(t) \equiv\int dx\,T_{11}(t,x) = E_{\mathrm{tot}}(t)-W_{\mathrm{tot}}(t) = -W_{\mathrm{tot}}(t). \tag*{(74)} p tot ( t ) ≡ ∫ d x T 11 ( t , x ) = E tot ( t ) − W tot ( t ) = − W tot ( t ) . (74)
This identity is the origin of the negative integrated pressure: the positive energy carried by finite yo-yos is cancelled by the remnant and fold sector, while the trace weight remains positive.
The trace weight W W W therefore measures the tension that remains after cancellation of the large positive and negative energy contributions. The decay problem is reduced to determining how rapidly fragmentation removes this traceful content.
The renewal calculation uses the period-averaged weights W Y W_Y W Y ; the exact time-slice identity above uses W Y i n s t W_Y^{\mathrm{inst}} W Y inst .
The remnant and its trace-weight source
The zeroth-generation remnant contributes directly to the pressure and continuously supplies the first-generation yo-yos which seed the branching cascade. A point ( u , v ) (u,v) ( u , v ) remains in the original IFS bulk if and only if its causal past rectangle contains no splitting point. For u , v > 0 u,v>0 u , v > 0 , the Poisson void probability derived in Sec. 2 therefore gives
⟨ T u v ( 0 ) ( u , v ) ⟩ = μ e − Γ s p l i t u v . (75) \left\langle T_{uv}^{(0)}(u,v)\right\rangle= \mu e^{-\Gamma_{\mathrm{split}}uv}. \tag*{(75)} ⟨ T uv ( 0 ) ( u , v ) ⟩ = μ e − Γ split uv . (75)
On a constant-time slice, u = t − x u=t-x u = t − x , v = t + x v=t+x v = t + x , and hence u v = t 2 − x 2 uv=t^2-x^2 uv = t 2 − x 2 . At late times the corresponding remnant weight is
⟨ W 0 ( t ) ⟩ = 4 μ Γ s p l i t t [ 1 + O ( 1 Γ s p l i t t 2 ) ] . (76) \left\langle W_0(t)\right\rangle= \frac{4\mu}{\Gamma_{\mathrm{split}}t}\left[1+\mathcal{O}\left(\frac{1}{\Gamma_{\mathrm{split}}t^2}\right)\right]. \tag*{(76)} ⟨ W 0 ( t ) ⟩ = Γ split t 4 μ [ 1 + O ( Γ split t 2 1 ) ] . (76)
The direct remnant contribution therefore decays as t − 1 t^{-1} t − 1 . Since the yo-yo cascade will decay more slowly in the semiclassical scaling regime, the remnant is subleading as a direct pressure contribution. Its more important role is as a persistent source.
At late emission time Γ s p l i t t 0 2 ≫ 1 \Gamma_{\mathrm{split}}t_0^2 \gg1 Γ split t 0 2 ≫ 1 , the two null arms supply asymptotically two first-generation yo-yos per unit ln t 0 \ln t_0 ln t 0 :
N ˙ 1 ( t 0 ) ∼ 2 t 0 . (77) \dot{N}_1(t_0) \sim\frac{2}{t_0}. \tag*{(77)} N ˙ 1 ( t 0 ) ∼ t 0 2 . (77)
Their typical invariant mass remains of order μ / Γ s p l i t \mu/\sqrt{\Gamma_{\mathrm{split}}} μ / Γ split , as established in § 2.2 , while their boost grows as γ 1 = O ( Γ s p l i t t 0 ) \gamma_1 = O(\sqrt{\Gamma_{\mathrm{split}}}t_0) γ 1 = O ( Γ split t 0 ) . Later emissions have increasingly elongated null rectangles at the same invariant scale, suppressing the trace W 1 = m 1 / γ 1 W_1 = m_1/\gamma_1 W 1 = m 1 / γ 1 carried by each yo-yo. The fixed-emission-time source measure derived in App. § 12 .§ 1 gives
⟨ W 1 ∣ t 0 ⟩ ∼ 2 μ Γ s p l i t t 0 . (78) \langle W_1 \mid t_0\rangle\sim\frac{2\mu}{\Gamma_{\mathrm{split}}t_0}. \tag*{(78)} ⟨ W 1 ∣ t 0 ⟩ ∼ Γ split t 0 2 μ . (78)
This boost suppression supplies a second inverse power of time to the trace-weight injection rate:
W ˙ 1 s r c ( t 0 ) = N ˙ 1 ( t 0 ) ⟨ W 1 ∣ t 0 ⟩ ∼ 4 μ Γ s p l i t t 0 2 . (79) \dot{W}^{\mathrm{src}}_1(t_0) = \dot{N}_1(t_0)\langle W_1 \mid t_0\rangle\sim\frac{4\mu}{\Gamma_{\mathrm{split}}t_0^2}. \tag*{(79)} W ˙ 1 src ( t 0 ) = N ˙ 1 ( t 0 ) ⟨ W 1 ∣ t 0 ⟩ ∼ Γ split t 0 2 4 μ . (79)
Appendix B.1 gives the full birth distribution and the subleading corrections to these source laws.
These are direct remnant emissions. Their first splittings can be non-dilute, so the population entering the later binary cascade has different mass, boost and birth-time statistics. The entrance construction in App. § 12 .§ 3 controls this intermediate fragmentation; the renewal calculation keeps that entrance population distinct from the first-generation source.
Transfer of trace weight between later generations
The later generations redistribute W = m / γ W = m/\gamma W = m / γ among descendants. In the semiclassical dilute window
ℓ s 2 ≪ Δ u Δ v ≪ Γ s p l i t − 1 , (80) \ell_s^2 \ll\Delta u\Delta v \ll\Gamma_{\mathrm{split}}^{-1}, \tag*{(80)} ℓ s 2 ≪ Δ u Δ v ≪ Γ split − 1 , (80)
§ 2.4 gives a single, uniformly distributed split ( u s , v s ) (u_s,v_s) ( u s , v s ) in an occupied oscillation to leading order in Λ Y = Γ s p l i t Δ u Δ v \Lambda_Y = \Gamma_{\mathrm{split}}\Delta u\Delta v Λ Y = Γ split Δ u Δ v . We write
α U ≡ u s − u a Δ u , α V ≡ u b − u s Δ v , α U , α V ∼ U ( 0 , 1 ) , (81) \alpha_U \equiv\frac{u_s-u_a}{\Delta u}, \qquad\alpha_V \equiv\frac{u_b-u_s}{\Delta v}, \qquad\alpha_U,\alpha_V \sim\mathcal{U}(0,1), \tag*{(81)} α U ≡ Δ u u s − u a , α V ≡ Δ v u b − u s , α U , α V ∼ U ( 0 , 1 ) , (81)
with α U \alpha_U α U and α V \alpha_V α V independent. The daughter extents are
( Δ u 1 , Δ v 1 ) = ( α U Δ u , α V Δ v ) , ( Δ u 2 , Δ v 2 ) = ( ( 1 − α U ) Δ u , ( 1 − α V ) Δ v ) . (82) (\Delta u_1,\Delta v_1) = (\alpha_U\Delta u,\alpha_V\Delta v), \qquad(\Delta u_2,\Delta v_2) = ((1-\alpha_U)\Delta u,(1-\alpha_V)\Delta v). \tag*{(82)} ( Δ u 1 , Δ v 1 ) = ( α U Δ u , α V Δ v ) , ( Δ u 2 , Δ v 2 ) = (( 1 − α U ) Δ u , ( 1 − α V ) Δ v ) . (82)
This is the binary limit of the exact staircase construction derived in § 2.3 .
The daughter mass fractions are
m 1 m = α U α V , m 2 m = ( 1 − α U ) ( 1 − α V ) (83) \frac{m_1}{m}=\sqrt{\alpha_U\alpha_V},\qquad\frac{m_2}{m}=\sqrt{(1-\alpha_U)(1-\alpha_V)} \tag*{(83)} m m 1 = α U α V , m m 2 = ( 1 − α U ) ( 1 − α V ) (83)
Define q A = W A / W q_A=W_A/W q A = W A / W . The component expressions for the boosts and trace fractions are collected in Appendix A.3; their physical content follows directly from conservation.
The trace lost in a split measures the relative motion of the daughters. Write E = ∑ A E A E=\sum_A E_A E = ∑ A E A , P = ∑ A P A P=\sum_A P_A P = ∑ A P A , v A = P A / E A v_A=P_A/E_A v A = P A / E A , and v ˉ = P / E \bar{v}=P/E v ˉ = P / E . Since W = m 2 / E = E − P 2 / E W=m^2/E=E-P^2/E W = m 2 / E = E − P 2 / E , energy and momentum conservation give
W − ∑ A W A = ∑ A P A 2 E A − P 2 E = ∑ A E A ( v A − v ˉ ) 2 ≥ 0. (84) W-\sum_A W_A=\sum_A\frac{P_A^2}{E_A}-\frac{P^2}{E}=\sum_A E_A(v_A-\bar{v})^2\ge0. \tag*{(84)} W − A ∑ W A = A ∑ E A P A 2 − E P 2 = A ∑ E A ( v A − v ˉ ) 2 ≥ 0. (84)
For two daughters the loss is E 1 E 2 ( v 1 − v 2 ) 2 / E E_1E_2(v_1-v_2)^2/E E 1 E 2 ( v 1 − v 2 ) 2 / E . Equal fractions leave both daughters at the parent’s velocity and preserve the trace. Unequal fractions convert part of the trace weight into kinematic pressure while conserving energy and momentum.
By reflection take Δ u ≥ Δ v \Delta u\ge\Delta v Δ u ≥ Δ v . For a highly boosted parent, γ ≫ 1 \gamma\gg1 γ ≫ 1 ,
q 1 ⟶ α V , q 2 ⟶ 1 − α V , q 1 + q 2 ⟶ 1 , (85) q_1\longrightarrow\alpha_V,\qquad q_2\longrightarrow1-\alpha_V,\qquad q_1+q_2\longrightarrow1, \tag*{(85)} q 1 ⟶ α V , q 2 ⟶ 1 − α V , q 1 + q 2 ⟶ 1 , (85)
up to corrections that vanish as γ → ∞ \gamma\to\infty γ → ∞ . Highly boosted splittings therefore transport the total trace weight almost conservatively. The motion of this nearly conserved weight is most transparent in logarithmic boost. Define
Δ ln γ A ≡ ln γ A γ . (86) \Delta\ln\gamma_A\equiv\ln\frac{\gamma_A}{\gamma}. \tag*{(86)} Δ ln γ A ≡ ln γ γ A . (86)
At large boost,
Δ ln γ 1 = 1 2 ln α U α V + O ( γ − 2 ) , Δ ln γ 2 = 1 2 ln 1 − α U 1 − α V + O ( γ − 2 ) . (87) \Delta\ln\gamma_1=\frac{1}{2}\ln\frac{\alpha_U}{\alpha_V}+\mathcal{O}(\gamma^{-2}),\qquad\Delta\ln\gamma_2=\frac{1}{2}\ln\frac{1-\alpha_U}{1-\alpha_V}+\mathcal{O}(\gamma^{-2}). \tag*{(87)} Δ ln γ 1 = 2 1 ln α V α U + O ( γ − 2 ) , Δ ln γ 2 = 2 1 ln 1 − α V 1 − α U + O ( γ − 2 ) . (87)
Mass weighting has no preferred sign for these jumps, by α U ↔ α V \alpha_U\leftrightarrow\alpha_V α U ↔ α V symmetry. Trace weighting favors the less boosted daughter because W = m / γ W=m/\gamma W = m / γ .
Using Eq. (85) , the trace-weighted drift is
⟨ Δ ln γ ⟩ W ≡ ⟨ ∑ A q A Δ ln γ A ⟩ = 1 2 ⟨ α V ln α U α V + ( 1 − α V ) ln 1 − α U 1 − α V ⟩ = − 1 4 . (88) \langle\Delta\ln\gamma\rangle_W\equiv\left\langle\sum_A q_A\Delta\ln\gamma_A\right\rangle=\frac{1}{2}\left\langle\alpha_V\ln\frac{\alpha_U}{\alpha_V}+(1-\alpha_V)\ln\frac{1-\alpha_U}{1-\alpha_V}\right\rangle=-\frac{1}{4}. \tag*{(88)} ⟨ Δ ln γ ⟩ W ≡ ⟨ A ∑ q A Δ ln γ A ⟩ = 2 1 ⟨ α V ln α V α U + ( 1 − α V ) ln 1 − α V 1 − α U ⟩ = − 4 1 . (88)
Thus highly boosted splits transport trace almost conservatively, with mean drift toward smaller boost. At finite boost, Eq. (84) gives the depletion of this transported weight.
The binary branching rule is
( ln γ , W ) ⟶ ( ln γ 1 , q 1 W ) + ( ln γ 2 , q 2 W ) . (89) (\ln\gamma,W)\longrightarrow(\ln\gamma_1,q_1W)+(\ln\gamma_2,q_2W). \tag*{(89)} ( ln γ , W ) ⟶ ( ln γ 1 , q 1 W ) + ( ln γ 2 , q 2 W ) . (89)
Trace transport and the semiclassical scaling regime
The slow trace decay follows from the two effects identified in Sec. 3. The remnant feeds highly boosted strings into the cascade. Their splittings transport trace almost conservatively toward lower boost, where relative daughter motion depletes it. But a daughter with less trace also splits more slowly. The late pressure reflects the balance between this slowing clock and the multiplication of branches.
We use the period-averaged yo-yo weight W = m / γ W = m/\gamma W = m / γ to expose the branching clock in a covariant dilute binary reference process. For the entrance population constructed in Appendix B, its large-time limit describes the intermediate window 1 ≪ Γ s p l i t t ≪ g s − 1 1 \ll\sqrt{\Gamma_{\mathrm{split}}t} \ll g_s^{-1} 1 ≪ Γ split t ≪ g s − 1 . Appendix C.5 then restores the full geometric multi-split dynamics and proves the same leading result for the instantaneous ensemble pressure through the physical split-event measure. This section studies the unabsorbed semiclassical cascade; Appendix G estimates the gravitational losses neglected in the model.
Branch multiplication and the slowing clock
A yo-yo sweeps worldsheet area A Y A_Y A Y in period T Y T_Y T Y . Since W Y = 2 μ A Y / T Y W_Y = 2\mu A_Y/T_Y W Y = 2 μ A Y / T Y , its splitting rate and mean laboratory lifetime are
λ Y = Γ s p l i t A Y T Y = Γ s p l i t 2 μ W Y , τ ( W Y ) = 2 μ Γ s p l i t W Y . (90) \lambda_Y = \frac{\Gamma_{\mathrm{split}} A_Y}{T_Y} = \frac{\Gamma_{\mathrm{split}}}{2\mu}W_Y,\qquad\tau(W_Y) = \frac{2\mu}{\Gamma_{\mathrm{split}}W_Y}. \tag*{(90)} λ Y = T Y Γ split A Y = 2 μ Γ split W Y , τ ( W Y ) = Γ split W Y 2 μ . (90)
Thus the weight that measures the surviving trace also sets the clock. Each daughter inherits no more than the parent’s trace and has a correspondingly longer mean lifetime.
First set aside these waits and count the available mass hierarchy. The independent uniform null fractions give daughter mass fractions
z 1 = α U α V , z 2 = ( 1 − α U ) ( 1 − α V ) . (91) z_1 = \sqrt{\alpha_U\alpha_V},\qquad z_2 = \sqrt{(1-\alpha_U)(1-\alpha_V)}. \tag*{(91)} z 1 = α U α V , z 2 = ( 1 − α U ) ( 1 − α V ) . (91)
Their branching moment is
M ( p ) = ⟨ z 1 p + z 2 p ⟩ = 2 ( 1 + p / 2 ) 2 . (92) M(p) = \left\langle z_1^p + z_2^p\right\rangle= \frac{2}{(1+p/2)^2}. \tag*{(92)} M ( p ) = ⟨ z 1 p + z 2 p ⟩ = ( 1 + p /2 ) 2 2 . (92)
A split doubles the branch count, M ( 0 ) = 2 M(0) = 2 M ( 0 ) = 2 , while the expected sum of rest masses decreases, M ( 1 ) = 8 / 9 M(1) = 8/9 M ( 1 ) = 8/9 . Energy is conserved; relative motion accounts for the difference. The intermediate power which balances multiplication and shrinking masses satisfies
M ( κ I F S ) = 1 , κ I F S = 2 ( 2 − 1 ) , ⟨ m 1 κ I F S + m 2 κ I F S ∣ m ⟩ = m κ I F S . (93) M(\kappa_{\mathrm{IFS}}) = 1,\qquad\kappa_{\mathrm{IFS}} = 2(\sqrt{2}-1),\qquad\left\langle m_1^{\kappa_{\mathrm{IFS}}} + m_2^{\kappa_{\mathrm{IFS}}}\mid m\right\rangle= m^{\kappa_{\mathrm{IFS}}}. \tag*{(93)} M ( κ IFS ) = 1 , κ IFS = 2 ( 2 − 1 ) , ⟨ m 1 κ IFS + m 2 κ IFS ∣ m ⟩ = m κ IFS . (93)
The expected κ I F S \kappa_{\mathrm{IFS}} κ IFS -weighted mass is therefore conserved across generations. Since M ( p ) M(p) M ( p ) decreases strictly, its positive balancing exponent is unique. For a hierarchy stopped at a lower mass cutoff, with initial mass-to-cutoff ratio C C C , Appendix § 12 .§ 2 then gives
N ( C ) ∼ C N C κ I F S . (94) \mathcal{N}(C) \sim C_N C^{\kappa_{\mathrm{IFS}}}. \tag*{(94)} N ( C ) ∼ C N C κ IFS . (94)
The first source-scale strings can undergo several splits in one oscillation. As fragmentation reduces their mass, the worldsheet area per period decreases and such events become rare. Deep generations therefore approach the dilute binary law. The earlier evolution supplies the strings entering this regime and fixes their overall contribution; the repeated binary law determines the exponent and normalized limiting boost profile. Appendix § 12 derives this entrance population, and Appendix § 13 .§ 5 establishes its matching to the full geometric cascade.
From the branching hierarchy to pressure
The physical argument has four steps. The mass hierarchy supplies the power κ I F S \kappa_{\mathrm{IFS}} κ IFS . The splitting clock turns that hierarchy into an accumulated number of events growing as t κ I F S t^{\kappa_{\mathrm{IFS}}} t κ IFS . Averaging over source boosts preserves the same power. Finally, the instantaneous splitting rate measures the surviving trace, giving W l i v e ∼ t κ I F S − 1 W_{\mathrm{live}} \sim t^{\kappa_{\mathrm{IFS}}-1} W live ∼ t κ IFS − 1 . A late daughter’s birth time contains the correlated waits of all its ancestors. Retaining those waits determines the normalization as well as the scaling law.
Split count and surviving trace. Each live yo-yo splits at rate Γ s p l i t W Y / ( 2 μ ) \Gamma_{\mathrm{split}}W_Y/(2\mu) Γ split W Y / ( 2 μ ) . Summing these rates gives, within the period-averaged binary model,
d d t ⟨ N s p l i t ( t ) ⟩ = Γ s p l i t 2 μ W l i v e ( t ) , (95) \frac{d}{dt}\langle N_{\mathrm{split}}(t)\rangle= \frac{\Gamma_{\mathrm{split}}}{2\mu}W_{\mathrm{live}}(t), \tag*{(95)} d t d ⟨ N split ( t )⟩ = 2 μ Γ split W live ( t ) , (95)
where N s p l i t ( t ) N_{\mathrm{split}}(t) N split ( t ) counts descendant splitting events before t t t . More surviving tension gives a larger mean splitting rate. Moreover, every dilute node contributes the same expected time-integrated trace,
W Y τ Y = 2 μ Γ s p l i t , (96) W_Y\tau_Y = \frac{2\mu}{\Gamma_{\mathrm{split}}}, \tag*{(96)} W Y τ Y = Γ split 2 μ , (96)
independently of its mass and boost. (96) links branch counting directly to the surviving trace.
The inherited splitting clock. The inherited waiting times simplify when each node is measured in its own lifetime units. For a node of mass m m m and signed velocity rapidity y y y , define
U = Γ s p l i t m 2 μ e y u , V = Γ s p l i t m 2 μ e − y v . (97) U = \frac{\Gamma_{\mathrm{split}}m}{2\mu}e^{y}u,\qquad V = \frac{\Gamma_{\mathrm{split}}m}{2\mu}e^{-y}v. \tag*{(97)} U = 2 μ Γ split m e y u , V = 2 μ Γ split m e − y v . (97)
If the parent waits a dimensionless proper time ϵ ∼ Exp ( 1 ) \epsilon\sim\operatorname{Exp}(1) ϵ ∼ Exp ( 1 ) and a selected daughter receives null momentum fractions a , b a,b a , b , so that its mass fraction is z = a b z = \sqrt{ab} z = ab , then its inherited birth coordinates obey the exact affine recursion
( U , V ) ⟼ ( a ( U + ϵ ) , b ( V + ϵ ) ) . (98) (U,V) \longmapsto\left(a(U+\epsilon), b(V+\epsilon)\right). \tag*{(98)} ( U , V ) ⟼ ( a ( U + ϵ ) , b ( V + ϵ ) ) . (98)
Thus each fresh waiting time enters at order unity in the daughter’s own lifetime units, while the accumulated timing information inherited from earlier generations is multiplied by shrinking daughter fractions. Appendix C.2 makes this loss of memory precise and proves that the rescaled birth-coordinate process approaches a stationary law with the inverse-radius moments needed below. The ancestral waits therefore renormalize the coefficient, but introduce no new leading power or logarithm.
Boost averaging and the scaling exponent. An event at invariant radius ρ = u v \rho= \sqrt{uv} ρ = uv lies on the boost orbit u = ρ e φ u = \rho e^{\varphi} u = ρ e φ , v = ρ e − φ v = \rho e^{-\varphi} v = ρ e − φ , with laboratory time t = ρ cosh φ t = \rho\cosh\varphi t = ρ cosh φ . The contribution of the complete boost orbit to a laboratory-time event rate is exactly
K ρ ( t ) ≡ ∫ − ∞ ∞ d φ δ ( t − ρ cosh φ ) = 2 Θ ( t − ρ ) t 2 − ρ 2 , (99) K_{\rho}(t) \equiv\int_{-\infty}^{\infty} d\varphi\,\delta(t-\rho\cosh\varphi) = \frac{2\Theta(t-\rho)}{\sqrt{t^2-\rho^2}}, \tag*{(99)} K ρ ( t ) ≡ ∫ − ∞ ∞ d φ δ ( t − ρ cosh φ ) = t 2 − ρ 2 2Θ ( t − ρ ) , (99)
and its Mellin transform is
∫ ρ ∞ d t t − ζ K ρ ( t ) = B ( ζ 2 , 1 2 ) ρ − ζ . (100) \int_{\rho}^{\infty} dt\,t^{-\zeta}K_{\rho}(t) = \mathrm{B}\left(\frac{\zeta}{2},\frac{1}{2}\right)\rho^{-\zeta}. \tag*{(100)} ∫ ρ ∞ d t t − ζ K ρ ( t ) = B ( 2 ζ , 2 1 ) ρ − ζ . (100)
The beta function is finite and analytic at ζ = κ I F S \zeta= \kappa_{\mathrm{IFS}} ζ = κ IFS . Hence boost averaging changes the normalization but not the selected power. The same is true of the stationary chronological clock: the only nonanalytic factor at the leading exponent comes from repeated branching.
The branching pole and the resulting pressure. Set t ∗ = Γ s p l i t − 1 / 2 t_* = \Gamma_{\mathrm{split}}^{-1/2} t ∗ = Γ split − 1/2 and T = t / t ∗ T = t/t_* T = t / t ∗ , and define the dimensionless live-trace response W ( T ) = Γ s p l i t W l i v e ( t ∗ T ) / ( 2 μ ) \mathcal{W}(T) = \sqrt{\Gamma_{\mathrm{split}}}\,W_{\mathrm{live}}(t_*T)/(2\mu) W ( T ) = Γ split W live ( t ∗ T ) / ( 2 μ ) . Appendix C shows that its Mellin transform, W ^ ( ζ ) ≡ ∫ 0 ∞ d T T − ζ W ( T ) \widehat{\mathcal{W}}(\zeta) \equiv\int_{0}^{\infty} dT\,T^{-\zeta}\mathcal{W}(T) W ( ζ ) ≡ ∫ 0 ∞ d T T − ζ W ( T ) , has the form
W ^ ( ζ ) = B ( ζ 2 , 1 2 ) [ C ( ζ ) 1 − M ( ζ ) + H ( ζ ) ] , C ( κ I F S ) > 0 , (101) \widehat{\mathcal{W}}(\zeta) = \mathrm{B}\left(\frac{\zeta}{2},\frac{1}{2}\right)\left[\frac{C(\zeta)}{1-M(\zeta)}+H(\zeta)\right],\qquad C(\kappa_{\mathrm{IFS}})>0, \tag*{(101)} W ( ζ ) = B ( 2 ζ , 2 1 ) [ 1 − M ( ζ ) C ( ζ ) + H ( ζ ) ] , C ( κ IFS ) > 0 , (101)
where C C C and H H H are analytic at the Malthusian line. Thus the only leading nonanalyticity is the simple branching pole selected by M ( κ I F S ) = 1 M(\kappa_{\mathrm{IFS}})=1 M ( κ IFS ) = 1 ; the waiting-time and boost integrals change its residue but cannot generate a leading logarithm. Equivalently,
⟨ N s p l i t ( t ∗ T ) ⟩ ∼ D κ I F S T κ I F S , W ( T ) ∼ D T κ I F S − 1 . (102) \langle N_{\mathrm{split}}(t_*T)\rangle\sim\frac{D}{\kappa_{\mathrm{IFS}}}T^{\kappa_{\mathrm{IFS}}},\qquad\mathcal{W}(T)\sim DT^{\kappa_{\mathrm{IFS}}-1}. \tag*{(102)} ⟨ N split ( t ∗ T )⟩ ∼ κ IFS D T κ IFS , W ( T ) ∼ D T κ IFS − 1 . (102)
Hence
q I F S = 1 − κ I F S = 3 − 2 2 , W l i v e ( t ) = C P μ Γ s p l i t − 1 / 2 ( Γ s p l i t t ) − q I F S [ 1 + o ( 1 ) ] . (103) q_{\mathrm{IFS}} = 1-\kappa_{\mathrm{IFS}} = 3-2\sqrt{2},\qquad W_{\mathrm{live}}(t)=C_P\mu\Gamma_{\mathrm{split}}^{-1/2}\left(\sqrt{\Gamma_{\mathrm{split}}}\,t\right)^{-q_{\mathrm{IFS}}}[1+o(1)]. \tag*{(103)} q IFS = 1 − κ IFS = 3 − 2 2 , W live ( t ) = C P μ Γ split − 1/2 ( Γ split t ) − q IFS [ 1 + o ( 1 )] . (103)
The amplitude depends on the entrance population, while the power does not. For a fixed mass-threshold entrance, the direct remnant and the pre-entrance trace are both O ( t − 1 ) O(t^{-1}) O ( t − 1 ) and are subleading.
Appendix C.5 proves that the same simple pole survives the exact geometric cascade. In particular, the local physical split intensity is proportional to the instantaneous ⟨ T u v ⟩ \langle T_{uv}\rangle ⟨ T uv ⟩ ; an exact Abel relation between the radial split-event measure and a constant-time slice then gives the same leading coefficient for the instantaneous ensemble pressure.
Boost distribution of the surviving trace
The total trace fixes the surviving weight, while its boost distribution identifies the strings that carry it. Write ρ W ( t , γ ) \rho_W(t,\gamma) ρ W ( t , γ ) for the trace density per unit ln γ \ln\gamma ln γ , so that ρ W ( t , γ ) d ln γ \rho_W(t,\gamma)\,d\ln\gamma ρ W ( t , γ ) d ln γ is the trace in a logarithmic boost bin. Each string contributes m / γ m/\gamma m / γ ; its birth history and waiting time determine whether it is still present on the slice.
Let y y y denote a string’s signed velocity rapidity, γ = cosh y \gamma=\cosh y γ = cosh y . It need not equal the spacetime rapidity of its birth event. Consequently a boost-invariant entrance source does not give a flat velocity histogram on a Minkowski-time slice. Appendix C.4 derives a parameter-free occupation factor F κ o c c ( w ) \mathcal{F}^{\mathrm{occ}}_{\kappa}(w) F κ occ ( w ) from the stationary birth history and final wait, with κ = κ I F S \kappa=\kappa_{\mathrm{IFS}} κ = κ IFS ; its explicit definition is Eq. (C.36). The normalized trace profile is
R κ ( γ ) = 2 Z κ γ − κ 1 − γ − 2 F κ o c c ( w γ ) , w γ = 1 + 1 − γ − 2 2 . (104) \mathcal{R}_{\kappa}(\gamma)=\frac{2}{Z_{\kappa}}\frac{\gamma^{-\kappa}}{\sqrt{1-\gamma^{-2}}}\mathcal{F}^{\mathrm{occ}}_{\kappa}(w_{\gamma}),\qquad w_{\gamma}=\frac{1+\sqrt{1-\gamma^{-2}}}{2}. \tag*{(104)} R κ ( γ ) = Z κ 2 1 − γ − 2 γ − κ F κ occ ( w γ ) , w γ = 2 1 + 1 − γ − 2 . (104)
Here w γ = ( 1 + ∣ β ∣ ) / 2 w_{\gamma}=(1+|\beta|)/2 w γ = ( 1 + ∣ β ∣ ) /2 depends only on speed. The constant Z κ Z_{\kappa} Z κ fixes ∫ 1 ∞ R κ ( γ ) d ln γ = 1 \int_{1}^{\infty}\mathcal{R}_{\kappa}(\gamma)\,d\ln\gamma=1 ∫ 1 ∞ R κ ( γ ) d ln γ = 1 . The result is
ρ W ( t , γ ) d ln γ W l i v e ( t ) ⟹ R κ ( γ ) d ln γ . (105) \frac{\rho_W(t,\gamma)\,d\ln\gamma}{W_{\mathrm{live}}(t)}\Longrightarrow\mathcal{R}_{\kappa}(\gamma)\,d\ln\gamma. \tag*{(105)} W live ( t ) ρ W ( t , γ ) d ln γ ⟹ R κ ( γ ) d ln γ . (105)
Because m = γ W m=\gamma W m = γ W , the corresponding sum-of-rest-masses spectrum is
d M d ln γ ( t , γ ) ∼ W l i v e ( t ) γ R κ ( γ ) . (106) \frac{dM}{d\ln\gamma}(t,\gamma)\sim W_{\mathrm{live}}(t)\gamma\mathcal{R}_{\kappa}(\gamma). \tag*{(106)} d ln γ d M ( t , γ ) ∼ W live ( t ) γ R κ ( γ ) . (106)
The three factors have distinct meanings. The branching weight gives γ − κ \gamma^{-\kappa} γ − κ , and conversion from trace to mass changes this to γ 1 − κ = γ q I F S \gamma^{1-\kappa}=\gamma^{q_{\mathrm{IFS}}} γ 1 − κ = γ q IFS . The denominator is the Jacobian from signed rapidity to logarithmic boost. The occupation factor accounts for the correlated birth histories and lifetimes; it approaches a finite positive constant at large boost. Normalizing the profile by the total trace removes the entrance strength.
At small boost, the rapidity Jacobian gives the integrable enhancement R κ ( γ ) ∝ ( γ − 1 ) − 1 / 2 R_{\kappa}(\gamma) \propto(\gamma-1)^{-1/2} R κ ( γ ) ∝ ( γ − 1 ) − 1/2 . Equation (105) fixes the distribution on each fixed boost interval. The total mass also samples the growing source region at γ s r c ∼ Γ s p l i t t \gamma_{\mathrm{src}} \sim\sqrt{\Gamma_{\mathrm{split}}t} γ src ∼ Γ split t ; Appendix § 14 treats its contribution.
Physical consequences and numerical tests
Energy, summed rest mass, and trace probe different parts of the population: one string contributes γ m \gamma m γ m , m m m , and m / γ m/\gamma m / γ , respectively. A large boost enhances its energy but suppresses its trace relative to its rest mass. Here M i n v = ∑ Y m Y M_{\mathrm{inv}} = \sum_Y m_Y M inv = ∑ Y m Y denotes the sum of live descendant rest masses, not the invariant mass of their total four-momentum.
The live boost distribution fixes the mass spectrum through m = γ W m=\gamma W m = γ W . With 1 − κ I F S = q I F S 1-\kappa_{\mathrm{IFS}}=q_{\mathrm{IFS}} 1 − κ IFS = q IFS , the limiting profile has the large-boost scaling
ρ W ( t , γ ) ∝ t − q I F S γ − κ I F S , d M d ln γ = γ ρ W ( t , γ ) ∝ t − q I F S γ q I F S . (107) \rho_W(t,\gamma) \propto t^{-q_{\mathrm{IFS}}}\gamma^{-\kappa_{\mathrm{IFS}}}, \qquad\frac{\mathrm{d}M}{\mathrm{d}\ln\gamma}=\gamma\rho_W(t,\gamma)\propto t^{-q_{\mathrm{IFS}}}\gamma^{q_{\mathrm{IFS}}}. \tag*{(107)} ρ W ( t , γ ) ∝ t − q IFS γ − κ IFS , d ln γ d M = γ ρ W ( t , γ ) ∝ t − q IFS γ q IFS . (107)
The temporal pressure exponent and the boost-space mass-spectrum slope are therefore the same. The occupation factor changes the finite-boost shape while preserving this exponent relation.
Figure 7 compares the predicted mass spectrum with 1,000 unabsorbed cascade histories at dimensionless time t ^ = Γ s p l i t t = 10 6 \widehat{t}=\sqrt{\Gamma_{\mathrm{split}}}t=10^6 t = Γ split t = 1 0 6 . With only the overall amplitude fitted, the calculated shape describes both the low-boost enhancement and the rising spectrum across five decades in boost. The upper edge, γ = 10 5 \gamma=10^5 γ = 1 0 5 , lies a factor of ten below the characteristic source scale. Appendix § 20 .§ 3 gives the correlated fit and its sensitivity checks. An independent binary-cascade simulation also tests the chronological profile and trace amplitude without a fitted normalization (App. § 20 .§ 2 ).
Figure 7. Sum-of-rest-masses spectrum at Γ s p l i t t = 10 6 \sqrt{\Gamma_{\rm split}t}=10^6 Γ split t = 1 0 6 , averaged over 1,000 histories without terminal absorption, in 50 logarithmic bins over 1 ≤ γ ≤ 10 5 1 \le\gamma\le10^5 1 ≤ γ ≤ 1 0 5 . Here M ^ = Γ s p l i t M / ( 2 μ ) \widehat{M}=\sqrt{\Gamma_{\rm split}}M/(2\mu) M = Γ split M / ( 2 μ ) . The blue curve is the predicted density, the orange staircase its bin averages, and the black crosses the data. The lower panel gives percentage residuals from the bin prediction. Error bars are correlated standard errors of the mean. The fit gives p b o o t = 0.486 p_{\rm boot}=0.486 p boot = 0.486 ; its template and statistical procedure are given in App. § 20 .§ 3 .
The complete population also extends to the moving source scale
γ s r c ( t ) ∼ Γ s p l i t t (108) \gamma_{\mathrm{src}}(t)\sim\sqrt{\Gamma_{\mathrm{split}}t} \tag*{(108)} γ src ( t ) ∼ Γ split t (108)
Energy and total invariant mass sample this growing boost range. Their integrated behavior is constrained separately by conservation and the full source history.
Energy, summed rest mass, and pressure. The different weights allow descendant energy to grow while trace decays. The binary calculation also predicts a plateau in summed rest mass under the source conditions of App. § 14 :
E y o y o ( t ) = 2 μ t − 2 μ Γ s p l i t t + O ( t − 3 ) , (109) E_{\rm yoyo}(t)=2\mu t-\frac{2\mu}{\Gamma_{\rm split}t}+O(t^{-3}), \tag*{(109)} E yoyo ( t ) = 2 μ t − Γ split t 2 μ + O ( t − 3 ) , (109)
M i n v ( t ) = M ∞ − B t ^ − q I F S + o ( t ^ − q I F S ) , (110) M_{\rm inv}(t)=M_{\infty}-B\hat{t}^{-q_{\rm IFS}}+o(\hat{t}^{-q_{\rm IFS}}), \tag*{(110)} M inv ( t ) = M ∞ − B t ^ − q IFS + o ( t ^ − q IFS ) , (110)
∣ p t o t ( t ) ∣ = C W ( Γ s p l i t t ) − q I F S [ 1 + o ( 1 ) ] . (111) |p_{\rm tot}(t)|=C_{\rm W}\left(\sqrt{\Gamma_{\rm split}t}\right)^{-q_{\rm IFS}}\left[1+o(1)\right]. \tag*{(111)} ∣ p tot ( t ) ∣ = C W ( Γ split t ) − q IFS [ 1 + o ( 1 ) ] . (111)
The first relation is an energy sum rule: later splittings redistribute energy among descendants but do not change the energy injected by the first-generation source. The remnant and fold sector balances this positive descendant energy, preserving E t o t = 0 E_{\rm tot}=0 E tot = 0 for the complete state.
In the chronological binary model, Appendix § 14 proves the plateau and its leading approach from below under its additional mass-moment and source-tail conditions. In that model the surviving tension and remaining mass deficit share an exponent, and Eq. (D.24) predicts B / A ≃ 3.08 B/A \simeq3.08 B / A ≃ 3.08 when W l i v e ∼ A t ^ − q I F S W_{\mathrm{live}} \sim A\hat{t}^{-q_{\mathrm{IFS}}} W live ∼ A t ^ − q IFS . The geometric simulations below test these integrated-mass predictions. An analytic extension of the mass coefficient to the geometric process requires matching its mass-loss weight and controlling the pre-entrance mass remainder, as specified in that appendix. Since q I F S < 1 q_{\mathrm{IFS}} < 1 q IFS < 1 , the live cascade dominates the pressure tail over the direct remnant contribution of order t − 1 t^{-1} t − 1 .
Table 1. Direct tests with q I F S = 3 − 2 2 q_{\mathrm{IFS}} = 3 - 2\sqrt{2} q IFS = 3 − 2 2 fixed, using 1,000 histories and 20 logarithmic time intervals. The joint row imposes B / A = 3.0774 B/A = 3.0774 B / A = 3.0774 and fits A , C , M ∞ A,C,M_\infty A , C , M ∞ . The probabilities are calibrated by whole-history resampling.
Direct tests of the time dependence. We test these predictions with q = q I F S q = q_{\mathrm{IFS}} q = q IFS fixed, fitting
W l i v e = A t ^ − q + C / t ^ , M i n v = M ∞ − B t ^ − q . (112) W_{\mathrm{live}} = A\hat{t}^{-q} + C/\hat{t}, \qquad M_{\mathrm{inv}} = M_\infty- B\hat{t}^{-q}. \tag*{(112)} W live = A t ^ − q + C / t ^ , M inv = M ∞ − B t ^ − q . (112)
The comparison uses 1,000 histories over 10 3 ≤ t ^ ≤ 10 6 10^3 \le\hat{t} \le10^6 1 0 3 ≤ t ^ ≤ 1 0 6 , with results in Table 1 and Figure 8 . Appendix J.4 gives the interval averaging, correlated fits, and amplitudes.
Figure 8. Live-trace decay and invariant-mass deficit in 1,000 histories, with q = 3 − 2 2 q = 3 - 2\sqrt{2} q = 3 − 2 2 fixed and μ = Γ s p l i t = 1 \mu= \Gamma_{\mathrm{split}} = 1 μ = Γ split = 1 . The horizontal coordinate is t ^ = Γ s p l i t t \hat{t} = \sqrt{\Gamma_{\mathrm{split}}}t t ^ = Γ split t ; the vertical coordinates show W W W and M M M in these physical units. The left panel fits W l i v e = A t ^ − q + C / t ^ W_{\mathrm{live}} = A\hat{t}^{-q} + C/\hat{t} W live = A t ^ − q + C / t ^ ; the right shows the mass deficit from the fitted plateau. Blue curves are fitted predictions, orange staircases their time-bin averages, and black crosses the data with correlated standard errors. The lower panels show percentage residuals. Fit statistics appear in Table 1, with the averaging and error analysis in App. § 20 .§ 4 . The live trace excludes the direct remnant contribution.
The increased precision resolves a small negative finite-time correction to the trace. Including it gives agreement with the predicted leading exponent and mass-to-trace amplitude ratio. The fitted C / t ^ C/\hat{t} C / t ^ term describes the observed window; the asymptotic subleading power remains undetermined. Spatial moments also resolve small departures from their leading ratios (App. J.6).
Fluctuations in the early remnant leave a persistent imprint on the late cascade: histories with weaker early emission retain a smaller trace at later times. Most of the trace variance is a common amplitude fluctuation, so changes of shape are measured more precisely than individual levels. Appendix J.4 quantifies this memory of the source.
These simulations omit terminal absorption and test the semiclassical predictions in the intermediate string regime 1 ≪ Γ s p l i t t ^ ≪ g s − 1 1 \ll\sqrt{\Gamma_{\mathrm{split}}}\hat{t} \ll g_s^{-1} 1 ≪ Γ split t ^ ≪ g s − 1 .
The cascade has now determined how quickly the surviving tension decreases on a time slice. Conservation relates this same tension to the positive flux still missing from a truncated null ray, so the pressure law also fixes the approach to null cancellation.
Cutoff ANEC convergence
A truncated null ray has crossed the original negative fold but has not yet collected all the compensating positive flux. Its residual measures how much cancellation is still to come. We determine that residual from the surviving trace, using the pressure law of § 4 . The numerical diagnostics are ensemble averages; the related equal-time moment tests are collected in App. § 15 .
From a time slice to a null segment
The pressure and the null deficit sample the same positive trace profile on different cuts. The cutoff observable is
A ( v ; U ) = ∫ − ∞ U d u T u u ( u , v ) . (113) \mathcal{A}(v; U) = \int_{-\infty}^{U} du\, T_{uu}(u,v). \tag*{(113)} A ( v ; U ) = ∫ − ∞ U d u T uu ( u , v ) . (113)
Here U U U advances along a null ray and v v v labels which ray is measured. Conservation gives
∂ v A ( v ; U ) = − T u v ( U , v ) . (114) \partial_v \mathcal{A}(v;U)=-T_{uv}(U,v). \tag*{(114)} ∂ v A ( v ; U ) = − T uv ( U , v ) . (114)
Since the stress vanishes before nucleation and T u v ≥ 0 T_{uv}\ge0 T uv ≥ 0 , A ( v ; U ) ≤ 0 \mathcal{A}(v;U)\le0 A ( v ; U ) ≤ 0 history by history. Thus the numerical absolute-value diagnostic satisfies ⟨ ∣ A ∣ ⟩ = − ⟨ A ⟩ \langle|\mathcal{A}|\rangle=-\langle\mathcal{A}\rangle ⟨ ∣ A ∣ ⟩ = − ⟨ A ⟩ .
Write ⟨ T u v ( u , v ) ⟩ = f ( u v ) \langle T_{uv}(u,v)\rangle=f(uv) ⟨ T uv ( u , v )⟩ = f ( uv ) and I ( S ) = ∫ 0 S f ( z ) d z I(S)=\int_0^S f(z)\,dz I ( S ) = ∫ 0 S f ( z ) d z . Conservation fixes the null residual, ⟨ A ( v ; U ) ⟩ = − I ( U v ) / U \langle\mathcal{A}(v;U)\rangle=-I(Uv)/U ⟨ A ( v ; U )⟩ = − I ( U v ) / U . The pressure measures this same profile on z = t 2 − x 2 z=t^2-x^2 z = t 2 − x 2 . Changing variables on that time slice gives
W ( t ) = − ⟨ p t o t ( t ) ⟩ = 4 ∫ 0 t 2 f ( z ) t 2 − z d z . (115) W(t)=-\langle p_{\mathrm{tot}}(t)\rangle=4\int_0^{t^2}\frac{f(z)}{\sqrt{t^2-z}}\,dz. \tag*{(115)} W ( t ) = − ⟨ p tot ( t )⟩ = 4 ∫ 0 t 2 t 2 − z f ( z ) d z . (115)
The square-root factor is the spatial length occupied by each interval of invariant age. Applying the same kernel a second time removes that geometric weight:
I ( S ) = 1 4 π ∫ 0 S W ( r ) S − r d r , S > 0. (116) I(S)=\frac{1}{4\pi}\int_0^S\frac{W(\sqrt{r})}{\sqrt{S-r}}\,dr,\qquad S>0. \tag*{(116)} I ( S ) = 4 π 1 ∫ 0 S S − r W ( r ) d r , S > 0. (116)
Indeed, exchanging the two nonnegative integrals leaves ∫ z S [ ( r − z ) ( S − r ) ] − 1 / 2 d r = π \int_z^S[(r-z)(S-r)]^{-1/2}\,dr=\pi ∫ z S [( r − z ) ( S − r ) ] − 1/2 d r = π . Pressure therefore determines the accumulated trace and the null deficit without requiring reconstruction of the local density f f f .
Put q = q I F S q=q_{\mathrm{IFS}} q = q IFS . The pressure law is W ( t ) ∼ A W t − q W(t)\sim A_Wt^{-q} W ( t ) ∼ A W t − q , with A W = C P μ Γ s p l i t − ( 1 + q ) / 2 > 0 A_W=C_P\mu\Gamma_{\mathrm{split}}^{-(1+q)/2}>0 A W = C P μ Γ split − ( 1 + q ) /2 > 0 . Substituting r = S a r=Sa r = S a gives
I ( S ) ∼ A W 4 π B ( 1 − q / 2 , 1 / 2 ) S ( 1 − q ) / 2 . (117) I(S)\sim\frac{A_W}{4\pi}B(1-q/2,1/2)S^{(1-q)/2}. \tag*{(117)} I ( S ) ∼ 4 π A W B ( 1 − q /2 , 1/2 ) S ( 1 − q ) /2 . (117)
The limit follows directly by separating a fixed early interval, whose contribution is O ( S − 1 / 2 ) O(S^{-1/2}) O ( S − 1/2 ) , from the late tail. On the tail the pressure asymptotic is uniform, and a − q / 2 ( 1 − a ) − 1 / 2 a^{-q/2}(1-a)^{-1/2} a − q /2 ( 1 − a ) − 1/2 is integrable at both endpoints. Consequently
⟨ A ( v ; U ) ⟩ ∼ − C A U − ( 1 + q ) / 2 v ( 1 − q ) / 2 , C A = A W 4 π B ( 1 − q / 2 , 1 / 2 ) . (118) \langle\mathcal{A}(v;U)\rangle\sim-C_AU^{-(1+q)/2}v^{(1-q)/2},\qquad C_A=\frac{A_W}{4\pi}B(1-q/2,1/2). \tag*{(118)} ⟨ A ( v ; U )⟩ ∼ − C A U − ( 1 + q ) /2 v ( 1 − q ) /2 , C A = 4 π A W B ( 1 − q /2 , 1/2 ) . (118)
The limit is U v → ∞ Uv\to\infty U v → ∞ in the unabsorbed boost-invariant model. For finite-coupling strings, it describes the intermediate invariant-age window and inherits the pressure theorem’s entrance and averaging assumptions.
For comparison with the two-variable numerical fit, define
⟨ A ( v ; U ) ⟩ ∼ − C A U α v β , α = 2 − 2 , β = 2 − 1. (119) \langle\mathcal{A}(v;U)\rangle\sim-C_AU^\alpha v^\beta,\qquad\alpha=\sqrt{2}-2,\qquad\beta=\sqrt{2}-1. \tag*{(119)} ⟨ A ( v ; U )⟩ ∼ − C A U α v β , α = 2 − 2 , β = 2 − 1. (119)
Their difference is fixed by the boost weight and their sum by the decay of trace:
β − α = 1 , α + β = − q I F S . (120) \beta-\alpha=1,\qquad\alpha+\beta=-q_{\mathrm{IFS}}. \tag*{(120)} β − α = 1 , α + β = − q IFS . (120)
The same branching exponent also explains β = κ I F S / 2 \beta= \kappa_{\mathrm{IFS}}/2 β = κ IFS /2 : the invariant radius is U v \sqrt{Uv} U v . Growth with the ray label v v v and decay with the advancing cutoff U U U are therefore compatible. Equations (118) and (119) fix both the scaling and its normalization for the numerical test.
The cancellation also holds history by history. After the original negative fold is crossed, each subsequent positive contribution reduces the deficit. Its vanishing ensemble limit therefore implies saturation with probability one. Appendix § 16 .§ 3 gives the argument simultaneously for all longitudinal rays.
Numerical test of the ANEC prediction
We test Eq. (118) on 1,000 independent cascade histories evolved without terminal absorption, using μ = Γ s p l i t = 1 \mu= \Gamma_{\mathrm{split}} = 1 μ = Γ split = 1 . The fiducial fit averages the data into five logarithmic v v v bands over 0.5 ≤ v ≤ 100 0.5 \le v \le100 0.5 ≤ v ≤ 100 and six logarithmic U U U intervals over 10 3 ≤ U ≤ 10 6 10^3 \le U \le10^6 1 0 3 ≤ U ≤ 1 0 6 . We apply the same averaging to the prediction, fix q = 3 − 2 2 q = 3 - 2\sqrt{2} q = 3 − 2 2 , and fit only C A C_A C A using the full covariance of the 30 averages. App. § 20 .§ 5 reports the amplitude and fit statistics; the fiducial whole-history fit gives p b o o t = 0.797 p_{\mathrm{boot}} = 0.797 p boot = 0.797 . Figure 9 shows a finer binning of the same histories. All tested windows and binnings remain compatible with the prediction; the procedure and checks are given in App. § 20 .§ 5 .
Figure 9. Cutoff ANEC convergence in the unabsorbed semiclassical cascade, with μ = Γ s p l i t = 1 \mu= \Gamma_{\mathrm{split}} = 1 μ = Γ split = 1 . The 1,000-history ensemble is averaged over six logarithmic v v v bands and eight logarithmic U U U intervals. Black crosses mark bin means and their marginal standard errors; the short horizontal strokes locate the means and do not denote horizontal uncertainty. Blue curves show C A U − ( 1 + q ) / 2 v ( 1 − q ) / 2 C_A U^{-(1+q)/2}v^{(1-q)/2} C A U − ( 1 + q ) /2 v ( 1 − q ) /2 averaged over each v v v band, with q = 3 − 2 2 q = 3 - 2\sqrt{2} q = 3 − 2 2 fixed and one common fitted amplitude C A = 0.72978 C_A = 0.72978 C A = 0.72978 . Each mean is plotted at the effective cutoff where the continuous curve equals its bin-averaged prediction; data values and errors are not rescaled. Both this finer 48-cell fit and the fiducial 30-cell comparison use the full covariance; their statistics appear in App. § 20 .§ 5 .
Allowing q q q to vary while retaining the boost relation β − α = 1 \beta- \alpha= 1 β − α = 1 gives q = 0.17135 q = 0.17135 q = 0.17135 , with a 95% bootstrap interval [ 0.17084 , 0.17183 ] [0.17084, 0.17183] [ 0.17084 , 0.17183 ] , containing the predicted 0.17157. The improvement is only Δ χ 2 = 0.802 \Delta\chi^2 = 0.802 Δ χ 2 = 0.802 , with nested-bootstrap p Δ = 0.392 p_{\Delta} = 0.392 p Δ = 0.392 . Thus these data support the predicted exponent without requiring an additional fitted slope.
The pressure fit supplies a further test of the normalization. Using A W = 4.30203 A_W = 4.30203 A W = 4.30203 from the corrected trace fit in App. § 20 .§ 4 , Eq. (118) predicts C A ( W ) = 0.72372 C_A^{(W)} = 0.72372 C A ( W ) = 0.72372 . The fiducial ANEC fit gives
C A C A ( W ) = 1.010 , 95 % paired-bootstrap interval: [ 0.991 , 1.026 ] . (121) \frac{C_A}{C_A^{(W)}} = 1.010,\qquad95\% \text{ paired-bootstrap interval: } [0.991, 1.026]. \tag*{(121)} C A ( W ) C A = 1.010 , 95% paired-bootstrap interval: [ 0.991 , 1.026 ] . (121)
The paired amplitudes agree within their uncertainty. They probe different age ranges: the ANEC endpoints have invariant ages U v ≤ 10 4 \sqrt{Uv} \le10^4 U v ≤ 1 0 4 , whereas the pressure reference is fitted over 10 3 ≤ t ≤ 10 6 10^3 \le t \le10^6 1 0 3 ≤ t ≤ 1 0 6 .
Terminal decay in 1 + 1 dimensions
The semiclassical cascade eventually reaches masses m = O ( m s ) m = O(m_s) m = O ( m s ) , with m s ∼ α ′ − 1 / 2 m_s \sim\alpha'^{-1/2} m s ∼ α ′ − 1/2 , where microscopic light-string dynamics takes over. We allow a finite-width crossover localized near this scale. The 1 / t 1/t 1/ t pressure result applies to terminal completions for which each finite tree—one emitted yo-yo together with all its descendants—has a finite integrated trace response and the final products are traceless.
The remnant continues to emit new trees at arbitrarily late times, even after earlier trees have decayed, supplying an order-one number per logarithmic interval of time. Under the scale-covariance and source-tail conditions below, convolving this continuing source with the finite response of its trees gives a 1 / t 1/t 1/ t pressure tail. The terminal dynamics determines its coefficient and the properties of the emitted radiation.
Stable free massive products instead belong to the broader completed-flux class treated in § 7 : they retain trace m j 2 / E j m_j^2/E_j m j 2 / E j indefinitely and therefore do not satisfy the finite-response assumption used for the pressure law. The analysis proceeds from the response of one tree to its convolution with the continuing source and finally to the crossover in a sharp absorbing reference model. Unless stated otherwise, tree responses and late-time pressure laws are ensemble-averaged over cascade histories and period-averaged over yo-yo oscillations; we suppress ensemble brackets on asymptotic pressure laws.
Node counting and finite-tree response
The splitting clock (90) gives each dilute yo-yo the same expected integrated trace:
W τ ( W ) = 2 μ Γ s p l i t . (122) W_{\tau}(W) = \frac{2\mu}{\Gamma_{\mathrm{split}}}. \tag*{(122)} W τ ( W ) = Γ split 2 μ . (122)
A node with less trace weight waits proportionally longer before splitting. The two effects cancel in its time integral: every node in the dilute self-similar part of the cascade contributes the same expected integrated trace weight, independently of its mass and boost. Summing a complete tree’s response therefore amounts to counting its splitting nodes. If the early stages of each tree supply a finite, nonzero κ I F S \kappa_{\mathrm{IFS}} κ IFS -weighted entrance mass, independent of the terminal hierarchy, they modify the coefficient of the complete-tree response.
Section 4.2 used waiting times to determine which descendants are reached by a given observation time. Here we integrate each tree over its full history. The cutoff is instead the terminal mass: we count the semiclassical nodes produced before the string-scale crossover. Write ξ = m / m t e r m \xi= m/m_{\mathrm{term}} ξ = m / m term , with m t e r m = O ( m s ) m_{\mathrm{term}} = O(m_{\mathrm{s}}) m term = O ( m s ) . The complete-tree response, with T T T measured from the root’s birth, is
R t r e e ( ξ ) ≡ ∫ 0 ∞ d T ⟨ W t r e e ( T ; ξ ) ⟩ . (123) \mathcal{R}_{\mathrm{tree}}(\xi) \equiv\int_{0}^{\infty} dT\,\langle W_{\mathrm{tree}}(T;\xi)\rangle. \tag*{(123)} R tree ( ξ ) ≡ ∫ 0 ∞ d T ⟨ W tree ( T ; ξ )⟩ . (123)
Here W t r e e W_{\mathrm{tree}} W tree is the spatially integrated trace of that one tree at age T T T , including the trace carried during terminal evolution before conversion. Thus R t r e e \mathcal{R}_{\mathrm{tree}} R tree measures its response accumulated over its entire lifetime. In the sharp absorbing model only live semiclassical nodes contribute, and the expected number N i n t ( ξ ) \mathcal{N}_{\mathrm{int}}(\xi) N int ( ξ ) of above-threshold splitting nodes is the stopped-tree count of App. B.2. Its recurrence and Laplace transform are derived there; the result is
N i n t ( ξ ) ∼ C N t e r m ξ κ I F S , κ I F S = 2 ( 2 − 1 ) = 1 − q I F S . (124) \mathcal{N}_{\mathrm{int}}(\xi) \sim C_{N}^{\mathrm{term}}\xi^{\kappa_{\mathrm{IFS}}}, \qquad\kappa_{\mathrm{IFS}} = 2(\sqrt{2}-1) = 1-q_{\mathrm{IFS}}. \tag*{(124)} N int ( ξ ) ∼ C N term ξ κ IFS , κ IFS = 2 ( 2 − 1 ) = 1 − q IFS . (124)
Multiplying this stopped-tree count by the response per node gives
R t r e e s h a r p ( ξ ) = 2 μ Γ s p l i t N i n t ( ξ ) , R t r e e ( ξ ) ∼ 2 μ Γ s p l i t C t e r m ξ κ I F S . (125) \mathcal{R}_{\mathrm{tree}}^{\mathrm{sharp}}(\xi) = \frac{2\mu}{\Gamma_{\mathrm{split}}}\mathcal{N}_{\mathrm{int}}(\xi), \qquad\mathcal{R}_{\mathrm{tree}}(\xi) \sim\frac{2\mu}{\Gamma_{\mathrm{split}}}C_{\mathrm{term}}\xi^{\kappa_{\mathrm{IFS}}}. \tag*{(125)} R tree sharp ( ξ ) = Γ split 2 μ N int ( ξ ) , R tree ( ξ ) ∼ Γ split 2 μ C term ξ κ IFS . (125)
The smooth-crossover form assumes the finite entrance weight above and a terminal law localized near ξ = O ( 1 ) \xi= O(1) ξ = O ( 1 ) , with no additional parametric lifetime hierarchy. Terminal dwell times then change C t e r m C_{\mathrm{term}} C term while preserving the hierarchy exponent. Appendix H.1 calculates their contribution in a finite-rate example and gives the renewal equation for competing rates.
Persistent source and the universal 1 / t 1/t 1/ t tail
Summing these tree responses over successive emissions of one IFS remnant gives the complete descendant response. Time t t t is measured from the IFS nucleation, while t 0 t_0 t 0 is the birth time of an emitted yo-yo. At late t 0 t_0 t 0 , the remnant produces first-generation trees at the asymptotic rate
N ˙ 1 ( t 0 ) = 2 t 0 [ 1 + o ( 1 ) ] . (126) \dot{N}_1(t_0) = \frac{2}{t_0}\left[1+o(1)\right]. \tag*{(126)} N ˙ 1 ( t 0 ) = t 0 2 [ 1 + o ( 1 ) ] . (126)
Their invariant-mass distribution approaches a time-independent form, while their trace weight scales as t 0 − 1 t_0^{-1} t 0 − 1 . Their complete response is therefore scale covariant: increasing the emission time by a factor λ \lambda λ stretches the complete tree history by the same factor and reduces its instantaneous trace weight by λ − 1 \lambda^{-1} λ − 1 . After averaging over the masses and boosts supplied by this first-generation source, the response of one emitted tree takes the form
⟨ W t r e e ( t ; t 0 ) ⟩ s r c = 1 t 0 F ( t − t 0 t 0 ) , ∫ 0 ∞ d y F ( y ) = ⟨ R t r e e ⟩ s r c . (127) \left\langle W_{\mathrm{tree}}(t;t_0)\right\rangle_{\mathrm{src}} = \frac{1}{t_0}\mathcal{F}\left(\frac{t-t_0}{t_0}\right), \qquad\int_0^\infty dy\,\mathcal{F}(y) = \left\langle R_{\mathrm{tree}}\right\rangle_{\mathrm{src}}. \tag*{(127)} ⟨ W tree ( t ; t 0 ) ⟩ src = t 0 1 F ( t 0 t − t 0 ) , ∫ 0 ∞ d y F ( y ) = ⟨ R tree ⟩ src . (127)
The variable y = ( t − t 0 ) / t 0 y=(t-t_0)/t_0 y = ( t − t 0 ) / t 0 measures a tree’s age relative to its emission time. Summing over all emission times gives W d e s c ( t ) W_{\mathrm{desc}}(t) W desc ( t ) , the total descendant trace of this single IFS. Changing the integration variable to y y y gives d t 0 / t 0 2 = − d y / t dt_0/t_0^2=-dy/t d t 0 / t 0 2 = − d y / t , so the scale-covariant late-source contribution depends only on the integrated tree response:
W d e s c ( t ) ≃ ∫ t d t 0 N ˙ 1 ( t 0 ) ⟨ W t r e e ( t ; t 0 ) ⟩ s r c = 2 t ∫ 0 ∞ F ( y ) d y + o ( t − 1 ) = 2 t ⟨ R t r e e ⟩ s r c + o ( t − 1 ) . (128) \begin{aligned}
W_{\mathrm{desc}}(t) &\simeq\int^{t} dt_0\,\dot{N}_1(t_0)\left\langle W_{\mathrm{tree}}(t;t_0)\right\rangle_{\mathrm{src}} \\
&= \frac{2}{t}\int_0^\infty\mathcal{F}(y)\,dy + o(t^{-1}) \\
&= \frac{2}{t}\left\langle R_{\mathrm{tree}}\right\rangle_{\mathrm{src}} + o(t^{-1}). \tag*{(128)}
\end{aligned} W desc ( t ) ≃ ∫ t d t 0 N ˙ 1 ( t 0 ) ⟨ W tree ( t ; t 0 ) ⟩ src = t 2 ∫ 0 ∞ F ( y ) d y + o ( t − 1 ) = t 2 ⟨ R tree ⟩ src + o ( t − 1 ) . (128)
Under the stated finite-response and source-tail assumptions, the continuing source fixes the 1 / t 1/t 1/ t power; the integrated response of its trees fixes the coefficient.
The same 1 / t 1/t 1/ t behavior follows directly from the late-time spatial geometry. A finite invariant-profile integral I ∞ I_\infty I ∞ places most of the surviving trace at bounded invariant age z = t 2 − x 2 z=t^2-x^2 z = t 2 − x 2 . On a late time slice these regions lie near the outgoing folds, at distance t − ∣ x ∣ = z / ( t + ∣ x ∣ ) t-|x|=z/(t+|x|) t − ∣ x ∣ = z / ( t + ∣ x ∣ ) ; their widths shrink as 1 / t 1/t 1/ t . The pressure integral then gives
W ( t ) ∼ 4 I ∞ t . (129) W(t)\sim\frac{4I_\infty}{t}. \tag*{(129)} W ( t ) ∼ t 4 I ∞ . (129)
Appendix H.1 bounds the contribution from large invariant ages and establishes this limit. The tree convolution and the invariant-profile argument are two representations of the same asymptotic coefficient: the former separates remnant emission from descendant response, while the latter combines them in I ∞ I_\infty I ∞ .
The characteristic first-generation mass hierarchy is
ξ ∗ ≡ 2 μ m t e r m Γ s p l i t ∼ g s − 1 . (130) \xi_* \equiv\frac{2\mu}{m_{\mathrm{term}}\sqrt{\Gamma_{\mathrm{split}}}} \sim g_s^{-1}. \tag*{(130)} ξ ∗ ≡ m term Γ split 2 μ ∼ g s − 1 . (130)
Averaging Eq. (125) over the broad but time-independent source distribution gives
⟨ R t r e e ( ξ ∗ ) ⟩ s r c ∼ 2 μ Γ s p l i t C s r c ξ ∗ κ I F S , ξ ∗ → ∞ . (131) \left\langle\mathcal{R}_{\mathrm{tree}}(\xi_*) \right\rangle_{\mathrm{src}} \sim\frac{2\mu}{\Gamma_{\mathrm{split}}} C_{\mathrm{src}} \xi_*^{\kappa_{\mathrm{IFS}}}, \qquad\xi_* \to\infty. \tag*{(131)} ⟨ R tree ( ξ ∗ ) ⟩ src ∼ Γ split 2 μ C src ξ ∗ κ IFS , ξ ∗ → ∞. (131)
The order-one coefficient C s r c C_{\mathrm{src}} C src includes the full first-generation mass distribution and the terminal response.
Including the direct remnant, the complete zero-energy state has, at fixed hierarchy ξ ∗ \xi_* ξ ∗ ,
p t o t ( t ) = − 1 t [ 4 μ Γ s p l i t + 2 ⟨ R t r e e ( ξ ∗ ) ⟩ s r c ] + o ( t − 1 ) . (132) p_{\mathrm{tot}}(t)=-\frac{1}{t}\left[\frac{4\mu}{\Gamma_{\mathrm{split}}}+2\left\langle\mathcal{R}_{\mathrm{tree}}(\xi_*)\right\rangle_{\mathrm{src}}\right]+o(t^{-1}). \tag*{(132)} p tot ( t ) = − t 1 [ Γ split 4 μ + 2 ⟨ R tree ( ξ ∗ ) ⟩ src ] + o ( t − 1 ) . (132)
At weak coupling the descendant term dominates. Using Γ s p l i t α ′ ∝ g s 2 \Gamma_{\mathrm{split}}\alpha' \propto g_s^2 Γ split α ′ ∝ g s 2 and κ I F S = 2 2 − 2 \kappa_{\mathrm{IFS}}=2\sqrt{2}-2 κ IFS = 2 2 − 2 ,
∣ p t o t ( t ) ∣ ∝ g s 2 − 2 2 t . (133) \left|p_{\mathrm{tot}}(t)\right|\propto\frac{g_s^{2-2\sqrt{2}}}{t}. \tag*{(133)} ∣ p tot ( t ) ∣ ∝ t g s 2 − 2 2 . (133)
The noninteger coupling power counts the extra generations exposed at weak coupling. The source starts at mass O ( m s / g s ) O(m_s/g_s) O ( m s / g s ) , and a tree produces O ( g s − κ I F S ) O(g_s^{-\kappa_{\mathrm{IFS}}}) O ( g s − κ IFS ) semiclassical nodes before reaching the string scale. Each contributes an integrated trace of order g s − 2 g_s^{-2} g s − 2 . Their product gives g s − 2 − κ I F S = g s − 2 2 g_s^{-2-\kappa_{\mathrm{IFS}}}=g_s^{-2\sqrt{2}} g s − 2 − κ IFS = g s − 2 2 . The coupling exponent assumes that the large-hierarchy binary fragmentation law remains valid down to a scale-local terminal crossover and that the terminal dynamics introduces no new power of g s g_s g s .
The transition from the intermediate t − q I F S t^{-q_{\mathrm{IFS}}} t − q IFS law is resolved in § 6.3 . The marginal terminal power q = 1 q=1 q = 1 has two immediate consequences. Stress-tensor conservation gives M 2 ( E ) , M 2 ( p ) ∝ − t log ( t / t × ) M_2^{(E)},M_2^{(p)}\propto-t\log(t/t_\times) M 2 ( E ) , M 2 ( p ) ∝ − t log ( t / t × ) at leading order. The even trace-weight moments have a common terminal coefficient: M 2 k ( W ) / ( t 2 k M 0 ( W ) ) → 1 M_{2k}^{(W)}/(t^{2k}M_0^{(W)})\to1 M 2 k ( W ) / ( t 2 k M 0 ( W ) ) → 1 for each fixed k k k , as derived in App. § 15 . This expresses the concentration of the remaining trace near the original folds on the scale ∣ x ∣ / t |x|/t ∣ x ∣/ t .
In the collinear massless completion, the terminal products carry the conserved right- and left-moving null charges of the absorbed branch and have T u v r a d = 0 T_{uv}^{\mathrm{rad}}=0 T uv rad = 0 . With the momentum convention of § 2.3 , their total energies are E R = ( E + P ) / 2 = μ Δ v E_R=(E+P)/2=\mu\Delta v E R = ( E + P ) /2 = μ Δ v and E L = ( E − P ) / 2 = μ Δ u E_L=(E-P)/2=\mu\Delta u E L = ( E − P ) /2 = μ Δ u . The endpoint is controlled by the integrated trace of the entire continuing cascade, rather than by termination of each finite tree separately. Boost invariance and conservation give
⟨ A ( v ; U ) ⟩ = − I ( U v ) U = − I ∞ U + o ( U − 1 ) ⟶ 0 , (134) \langle\mathcal{A}(v;U)\rangle=-\frac{I(Uv)}{U}=-\frac{I_{\infty}}{U}+o(U^{-1})\longrightarrow0, \tag*{(134)} ⟨ A ( v ; U )⟩ = − U I ( U v ) = − U I ∞ + o ( U − 1 ) ⟶ 0 , (134)
at fixed v > 0 v>0 v > 0 . Here I I I includes the remnant and all live descendants, and I ∞ = ∫ 0 ∞ f ( z ) d z < ∞ I_{\infty}=\int_{0}^{\infty}f(z)\,\mathrm{d}z<\infty I ∞ = ∫ 0 ∞ f ( z ) d z < ∞ under the finite-response assumptions above, as quantified in App. F.6. In particular, 0 ≤ − ⟨ A ( v ; U ) ⟩ ≤ I ∞ / U 0\leq-\langle\mathcal{A}(v;U)\rangle\leq I_{\infty}/U 0 ≤ − ⟨ A ( v ; U )⟩ ≤ I ∞ / U . This controls the persistent source as well as its finite trees and establishes the terminal 1 + 1 1+1 1 + 1 -dimensional null balance.
The terminal crossover
The source argument fixes the late-time power. Its connection to the semiclassical regime can also be seen directly in the boost spectrum. At fixed time, the string-scale cutoff removes the lower-mass, lower-boost part of the population. In the sharp absorbing binary reference, the stationary clock of App. C.4 gives, at large boost,
( d W / d ln γ ) s h a r p ( d W / d ln γ ) s c ≃ S α ( T / γ ) , S α ( x ) = Γ ( α , x ) Γ ( α ) , α = 3 − 2 , (135) \frac{(dW/d\ln\gamma)_{\mathrm{sharp}}}{(dW/d\ln\gamma)_{\mathrm{sc}}}\simeq S_{\alpha}(T/\gamma),\qquad S_{\alpha}(x)=\frac{\Gamma(\alpha,x)}{\Gamma(\alpha)},\qquad\alpha=3-\sqrt{2}, \tag*{(135)} ( d W / d ln γ ) sc ( d W / d ln γ ) sharp ≃ S α ( T / γ ) , S α ( x ) = Γ ( α ) Γ ( α , x ) , α = 3 − 2 , (135)
where Γ ( α , x ) \Gamma(\alpha,x) Γ ( α , x ) is the upper incomplete gamma function and T = t / t × T=t/t_{\times} T = t / t × . The crossover scale is
t × = 2 μ Γ s p l i t m t e r m , Γ s p l i t t × = ξ ∗ , t × ∼ ℓ s g s 2 . (136) t_{\times}=\frac{2\mu}{\Gamma_{\mathrm{split}}m_{\mathrm{term}}},\qquad\sqrt{\Gamma_{\mathrm{split}}}\,t_{\times}=\xi_{*},\qquad t_{\times}\sim\frac{\ell_{s}}{g_{s}^{2}}. \tag*{(136)} t × = Γ split m term 2 μ , Γ split t × = ξ ∗ , t × ∼ g s 2 ℓ s . (136)
For T ≫ 1 T\gg1 T ≫ 1 , the surviving trace lies at boosts of order T T T and above. Integrating the semiclassical spectrum d W / d ln γ ∝ t κ I F S − 1 γ − κ I F S dW/d\ln\gamma\propto t^{\kappa_{\mathrm{IFS}}-1}\gamma^{-\kappa_{\mathrm{IFS}}} d W / d ln γ ∝ t κ IFS − 1 γ − κ IFS against this suppression factor supplies T − κ I F S T^{-\kappa_{\mathrm{IFS}}} T − κ IFS , giving W ∝ t − 1 W\propto t^{-1} W ∝ t − 1 . Terminal decay therefore removes the slow descendants while increasingly boosted strings continue to carry the trace. The cutoff moves through the spectrum with a fixed width in ln γ \ln\gamma ln γ ; weak coupling separates it from the source edge without making this width grow.
To follow the same IFS through the full crossover, we integrate its total descendant trace up to the observation time. The joint stationary clock and boost projection include moderate boosts as well as the tail. Appendix H.1 constructs a positive reference function Ψ s h a r p \Psi_{\mathrm{sharp}} Ψ sharp such that, in the large-hierarchy limit at fixed T T T ,
R d e s c s h a r p ( t ) ≡ ∫ 0 t W d e s c s h a r p ( s ) d s ∼ A t e r m Ψ s h a r p ( T ) , \mathcal{R}_{\mathrm{desc}}^{\mathrm{sharp}}(t)\equiv\int_{0}^{t}W_{\mathrm{desc}}^{\mathrm{sharp}}(s)\,\mathrm{d}s\sim\mathcal{A}_{\mathrm{term}}\Psi_{\mathrm{sharp}}(T), R desc sharp ( t ) ≡ ∫ 0 t W desc sharp ( s ) d s ∼ A term Ψ sharp ( T ) ,
Ψ s h a r p ( T ) = { C e a r l y T κ I F S [ 1 + o ( 1 ) ] , T ≪ 1 , 2 ln T + C s h a r p + o ( 1 ) , T ≫ 1. (137) \Psi_{\mathrm{sharp}}(T)=
\begin{cases}
C_{\mathrm{early}}T^{\kappa_{\mathrm{IFS}}}[1+o(1)], & T\ll1,\\
2\ln T+C_{\mathrm{sharp}}+o(1), & T\gg1.
\end{cases} \tag*{(137)} Ψ sharp ( T ) = { C early T κ IFS [ 1 + o ( 1 )] , 2 ln T + C sharp + o ( 1 ) , T ≪ 1 , T ≫ 1. (137)
The logarithmic growth counts the continuing arrival of new trees; each tree still has a finite lifetime-integrated response.
Here A t e r m \mathcal{A}_{\mathrm{term}} A term is fixed by the entrance mass moment, as in Eq. (H.6); the early limit is taken after entrance relaxation. Figure 10 shows this calculated interpolation. The persistent remnant contributes its additional response from § 6.2 .
Figure 10. Sharp-cutoff binary reference in the large-hierarchy limit. Left: the accumulated response function Ψ s h a r p ( T ) \Psi_{\mathrm{sharp}}(T) Ψ sharp ( T ) , with its early power and late logarithmic asymptotes. Right: suppression of the high-boost trace spectrum by the mass cutoff, Eq. (135) . The surviving trace moves to γ ∼ T \gamma\sim T γ ∼ T with a fixed logarithmic transition width. The curve on the left is evaluated from the stationary joint clock in App. § 18 .§ 1 ; its normalization is separate from the entrance-dependent amplitude A t e r m \mathcal{A}_{\mathrm{term}} A term .
A finite-width terminal layer changes the crossover shape and its normalization. With finite response and the delay moments specified in Appendix § 18 .§ 1 , the semiclassical power and late logarithmic accumulation persist. The same appendix gives the convergence estimates for the sharp reference and more general terminal laws.
Higher-dimensional extension
In higher dimensions, the situation is more complicated. The null ray entering the ANEC need not lie in the plane of the IFS. Moreover, even when it does, terminal radiation can be emitted in arbitrary directions and need not remain in that plane. In either case, the positive flux that compensates the negative-energy fold can miss the null ray, leading to ANEC violation. Thus a transversely localized IFS need not saturate the ANEC on every null ray.
However, IFSs are not nucleated in generic states. For an unconditioned state produced from the vacuum by spatially translation-invariant dynamics, momentum conservation places the complete IFS state, including all descendants, in the exact zero-spatial-momentum sector. The state is therefore spatially translation invariant. In the next subsection we combine this symmetry with the complete cascade’s flux balance to establish ANEC saturation on every null ray. We then turn to a transversely localized IFS and quantify the residual after smearing over a finite transverse scale.
Transverse zero modes and ANEC
We first establish the flux balance integrated over the transverse screen, then use translation invariance to determine the light-ray expectation on each generator.
Let k ^ = ( 1 , n ^ ) \hat{k}=(1,\hat{\mathbf{n}}) k ^ = ( 1 , n ^ ) , with n ^ 2 = 1 \hat{\mathbf{n}}^2=1 n ^ 2 = 1 , be a future-directed null vector. A displacement b = ( 0 , b ) \mathbf{b}=(0,\mathbf{b}) b = ( 0 , b ) with b ⋅ n ^ = 0 \mathbf{b}\cdot\hat{\mathbf{n}}=0 b ⋅ n ^ = 0 labels neighboring generators. For a reference point x 0 x_0 x 0 , their complete light-ray expectation is
A k ^ ( b ) ≡ ∫ − ∞ ∞ d λ ⟨ T μ ν ( x 0 + λ k ^ + b ) ⟩ k ^ μ k ^ ν . (138) \mathcal{A}_{\hat{k}}(\mathbf{b}) \equiv\int_{-\infty}^{\infty} d\lambda\,\langle T_{\mu\nu}(x_0+\lambda\hat{k}+\mathbf{b})\rangle\hat{k}^{\mu}\hat{k}^{\nu}. \tag*{(138)} A k ^ ( b ) ≡ ∫ − ∞ ∞ d λ ⟨ T μν ( x 0 + λ k ^ + b )⟩ k ^ μ k ^ ν . (138)
For a localized IFS this can depend on b \mathbf{b} b and be negative. The conserved current J k ^ μ = − T μ ν k ^ ν J^\mu_{\hat{k}}=-T^{\mu\nu}\hat{k}_{\nu} J k ^ μ = − T μν k ^ ν has Cauchy-surface flux
Q k ^ = ∫ d D − 1 x ⟨ J k ^ 0 ⟩ = ⟨ P 0 − P ⋅ n ^ ⟩ , P μ = ∫ d D − 1 x T 0 μ . (139) Q_{\hat{k}}=\int d^{D-1}x\,\langle J^0_{\hat{k}}\rangle=\langle P^0-\mathbf{P}\cdot\hat{\mathbf{n}}\rangle,\qquad P^\mu=\int d^{D-1}x\,T^{0\mu}. \tag*{(139)} Q k ^ = ∫ d D − 1 x ⟨ J k ^ 0 ⟩ = ⟨ P 0 − P ⋅ n ^ ⟩ , P μ = ∫ d D − 1 x T 0 μ . (139)
Evaluating this current on t − n ^ ⋅ x = s > 0 t-\hat{\mathbf{n}}\cdot\mathbf{x}=s>0 t − n ^ ⋅ x = s > 0 relates it to the flux integrated over the complete screen, provided all the compensating flux is collected.
It remains to establish that no finite amount of compensating flux is permanently delayed beyond the cutoff, since the cascade continues to emit at arbitrarily late times. For directions not parallel to the original folds, both negative folds cross the null plane in finite time. The region still ahead of the plane then contains only positive stress. Conservation and causal support bound its remaining null charge by a quantity that tends to zero as the cutoff is removed. For longitudinal directions, the first-split bound applied to the transversely integrated longitudinal stress gives the same conclusion. Appendix F.3 establishes both limits directly for the full cascade. Consequently,
∫ d D − 2 b A k ^ ( b ) = ⟨ P 0 − P ⋅ n ^ ⟩ = 0 , (140) \int d^{D-2}b\,\mathcal{A}_{\hat{k}}(\mathbf{b})=\langle P^0-\mathbf{P}\cdot\hat{\mathbf{n}}\rangle=0, \tag*{(140)} ∫ d D − 2 b A k ^ ( b ) = ⟨ P 0 − P ⋅ n ^ ⟩ = 0 , (140)
using the vanishing total energy and momentum of the complete state. This balance does not require finite integrated trace or a specified terminal lifetime.
To implement a normalized translation average, retain this completed flux balance in a periodic transverse regulator of volume V ⊥ V_{\perp} V ⊥ and average uniformly over translations of the localized response. Translation invariance makes the averaged response independent of b \mathbf{b} b , so
V ⊥ A k ^ = ∫ V ⊥ d D − 2 b A k ^ ( b ) = 0. (141) V_{\perp}\mathcal{A}_{\hat{k}}=\int_{V_{\perp}}d^{D-2}b\,\mathcal{A}_{\hat{k}}(\mathbf{b})=0. \tag*{(141)} V ⊥ A k ^ = ∫ V ⊥ d D − 2 b A k ^ ( b ) = 0. (141)
Thus the ANEC is saturated on every generator in this completed, translation-invariant construction.
The same reasoning applies directly to a translation-invariant quantum state. Let P ⊥ \mathbf{P}_{\perp} P ⊥ denote the momentum conjugate to translations within the screen. An exact zero-momentum state obeys
P ⊥ ∣ Ψ I F S ⟩ = 0 , e − i a ⊥ ⋅ P ⊥ ∣ Ψ I F S ⟩ = ∣ Ψ I F S ⟩ . (142) \mathbf{P}_{\perp}|\Psi_{\mathrm{IFS}}\rangle= 0,\qquad e^{-i\mathbf{a}_{\perp}\cdot\mathbf{P}_{\perp}}|\Psi_{\mathrm{IFS}}\rangle= |\Psi_{\mathrm{IFS}}\rangle. \tag*{(142)} P ⊥ ∣ Ψ IFS ⟩ = 0 , e − i a ⊥ ⋅ P ⊥ ∣ Ψ IFS ⟩ = ∣ Ψ IFS ⟩ . (142)
Its stress expectation, and hence its complete light-ray expectation, is independent of the transverse position. Combined with the completed flux balance, this gives a vanishing integral on each generator. A normalized positional mixture provides the corresponding semiclassical realization of the translation average.
This average keeps the total IFS number fixed. A homogeneous population at fixed number density contains more events as the volume grows and can retain negative averaged null energy. The stronger finite-number estimate uniform in the longitudinal cutoff is given in Appendix § 16 .§ 3 ; the cosmological population is discussed in § 8 .
Finite transverse smearing
For a localized IFS, with collinear preterminal evolution, smear the complete light-ray integral over the screen of a fixed null plane s = − k ⋅ x > 0 s=-k\cdot x>0 s = − k ⋅ x > 0 :
A k , L ( s , b 0 ) = ∫ d D − 2 b K L ( b 0 − b ) A k ( s , b ) , K L ( b ) = L − ( D − 2 ) K ( b / L ) . (143) A_{k,L}(s,\mathbf{b}_0)=\int d^{D-2}b\,K_L(\mathbf{b}_0-\mathbf{b})A_k(s,\mathbf{b}),\qquad K_L(\mathbf{b})=L^{-(D-2)}K(\mathbf{b}/L). \tag*{(143)} A k , L ( s , b 0 ) = ∫ d D − 2 b K L ( b 0 − b ) A k ( s , b ) , K L ( b ) = L − ( D − 2 ) K ( b / L ) . (143)
The normalized kernel K K K and its first two derivatives are bounded. Positive and negative flux at nearby screen positions then receive nearly equal weights. The residual is controlled by their separation, weighted by the compensating flux.
The first-split trace bound controls the mean distance of the collinear positive energy from the folds. Terminal radiation adds angular spreading, but its mass budget is bounded by that of the first-generation yo-yos. Appendix F combines these observations to give the sufficient smearing scale
L s m e a r ≡ Γ s p l i t − 1 / 2 ∼ ℓ s g s . (144) L_{\mathrm{smear}}\equiv\Gamma_{\mathrm{split}}^{-1/2}\sim\frac{\ell_s}{g_s}. \tag*{(144)} L smear ≡ Γ split − 1/2 ∼ g s ℓ s . (144)
For the boost-invariant cascade with local terminal conversion into free products of positive energy, the bound is
sup b 0 ∣ A k , L ( s , b 0 ) ∣ ≤ C K F k ( s ) L D − 2 L s m e a r L , s > 0 , every normalized k . (145) \sup_{\mathbf{b}_0}|A_{k,L}(s,\mathbf{b}_0)|\le C_K\frac{\mathcal{F}_k(s)}{L^{D-2}}\frac{L_{\mathrm{smear}}}{L},\qquad s>0,\quad\text{every normalized }k. \tag*{(145)} b 0 sup ∣ A k , L ( s , b 0 ) ∣ ≤ C K L D − 2 F k ( s ) L L smear , s > 0 , every normalized k . (145)
Here F k ( s ) \mathcal{F}_k(s) F k ( s ) is the magnitude of the negative fold flux, equal to the total compensating positive flux. The constant is independent of the null direction, including grazing directions: nearly parallel products can travel far but carry little test flux. A fractional residual at most ε \varepsilon ε therefore requires only L ≳ C K L s m e a r / ε L \gtrsim C_K L_{\mathrm{smear}}/\varepsilon L ≳ C K L smear / ε . Its sign at finite resolution can be either positive or negative. No terminal node count or finite I ∞ I_{\infty} I ∞ is needed for this estimate.
Cosmological implications
The cosmological interest in instant folded strings (IFSs) rests on an unusual property of their energy-momentum tensor. An isolated IFS has vanishing total energy in Minkowski space but a large negative integrated pressure. A homogeneous population can consequently violate the null energy condition (NEC), ρ + p ≥ 0 \rho+ p \ge0 ρ + p ≥ 0 , and make H ˙ \dot{H} H ˙ positive in a contracting universe. On the expanding branch, IFS production can oppose the rolling dilaton and sustain a dark-energy-like stress [13 , 14 ] . In the cyclic construction, the dilaton crosses an enhanced symmetry point (ESP), where nonadiabatically produced particles grow heavier as the field moves away. Their restoring force turns the dilaton back through the moduli-trapping mechanism of [22 ] . IFS production then begins while the universe is still contracting, and its negative pressure can turn the geometry around.
The finite-coupling decay raises an apparent obstacle to this picture. The averaged null energy condition (ANEC) constrains the stress integrated along a complete null ray and has a firmer basis in Minkowski spacetime than the pointwise NEC [16 , 17 , 18 , 23 ] . The preceding sections show that the IFS cascade saturates ANEC along complete longitudinal rays in 1 + 1 1+1 1 + 1 dimensions: positive folds and terminal radiation eventually compensate the negative null flux. § 7 identifies the corresponding completed transverse zero mode in higher dimensions. Yet a regular, nonstatic, two-sided null-complete flat-FRW cosmology has negative complete affine ANEC under the assumptions of § 8.5 . In a cyclic universe, the affine integral is already negative over each full cycle bounded by equivalent zeros of H H H , even when a a a grows from cycle to cycle. How can the same decay process restore the Minkowski null-energy balance and still support a bounce?
The finite-lifetime approximation used in Refs. [13 , 14 ] already illustrates this compatibility. In Minkowski space, let the traceful string bulk convert locally into collinear null products within a finite time4 . The complete longitudinal ANEC integral then vanishes already in this approximation: by Eq. (11) , it equals the flux of − T u v -T_{uv} − T uv through the closing segment, which vanishes once that segment lies entirely after conversion. For the translation-invariant single-event state considered in § 7.1 , the completed null-flux balance extends this saturation to every null ray.
The effect of contraction or expansion follows directly from energy conservation. To isolate this effect, hold the dilaton fixed and follow a homogeneous, isotropically averaged cohort of IFSs and their decay products. In a comoving volume of physical size V ∝ a 3 V \propto a^3 V ∝ a 3 , its energy E = ρ V E = \rho V E = ρ V obeys d E = − p d V dE = -p\,dV d E = − p d V . While the pressure is negative, contraction lowers the energy of a cohort born with E = 0 E = 0 E = 0 , whereas expansion raises it. After conversion to null products, the cohort therefore retains a negative or positive energy balance, respectively.5
A rolling dilaton also exchanges energy with the traceful strings. The cosmological equations of Refs. [13 , 14 ] include this work and explicitly admit bouncing solutions. The finite-lifetime description thus already illustrates how completed Minkowski compensation can coexist with a cosmological bounce.
In the full cascade, the persistent remnant continually produces traceful descendants, while completed branches deposit compensating positive null energy over a broad range of delays. The trace controls the force on the dilaton. When the response is strong enough to balance the driving force with little field motion, that balance fixes the trace required by the background. The IFS energy density and pressure then follow from conservation and the initial stress. We first derive the bounce and expanding solutions from force balance and conservation, then use the retarded cascade response to test the bounce and estimate the rolling rate on the expanding branch.
Force balance and the cosmological stress
The quantity followed through the decay cascade is also the one that opposes the dilaton’s motion. For a homogeneous, isotropically averaged population, let W = ρ I F S − 3 p I F S W = \rho_{\mathrm{IFS}} - 3p_{\mathrm{IFS}} W = ρ IFS − 3 p IFS , and write g s = g 0 e ϕ / ( 2 M P l ) g_s = g_0e^{\phi/(\sqrt{2}M_{\mathrm{Pl}})} g s = g 0 e ϕ / ( 2 M Pl ) and Π = ϕ ˙ \Pi= \dot{\phi} Π = ϕ ˙ . Orientation averaging preserves the event trace, since 3 p = p ∥ 3p = p_{\parallel} 3 p = p ∥ ; the accumulated event responses therefore determine W W W . Traceless terminal products contribute to the energy and pressure, but the dilaton force is set by the remaining trace:
Π ¨ + 3 H Π ˙ + V , ϕ = − W 2 M P l . (146) \ddot{\Pi} + 3H\dot{\Pi} + V_{,\phi} = -\frac{W}{\sqrt{2}M_{\mathrm{Pl}}}. \tag*{(146)} Π ¨ + 3 H Π ˙ + V , ϕ = − 2 M Pl W . (146)
For V , ϕ < 0 V_{,\phi} < 0 V , ϕ < 0 , the potential drives the field toward stronger coupling, where positive rolling produces IFSs. Increasing production increases the opposing force. If this response is sufficiently strong that scalar acceleration and Hubble drag are small compared with the two forces, the required trace is
W ≃ W V , b a l ≡ − 2 M P l V , ϕ . (147) W \simeq W_{V,\mathrm{bal}} \equiv-\sqrt{2}M_{\mathrm{Pl}}V_{,\phi}. \tag*{(147)} W ≃ W V , bal ≡ − 2 M Pl V , ϕ . (147)
The potential slope thus fixes the required trace; the cascade determines the rolling rate needed to supply it. When ESP particles are present, their force replaces V , ϕ V_{,\phi} V , ϕ by V , ϕ + ∂ ϕ ρ χ V_{,\phi}+\partial_\phi\rho_\chi V , ϕ + ∂ ϕ ρ χ . Whether the delayed response can maintain this balance is a dynamical question, tested below for the homogeneous bounce.
The trace fixes the dilaton force, while the geometry also depends on the IFS energy density. In spatially flat Einstein-frame FLRW, d s 2 = − d t 2 + a 2 ( t ) d x 2 ds^2=-dt^2+a^2(t)d\mathbf{x}^2 d s 2 = − d t 2 + a 2 ( t ) d x 2 ,
3 M P l 2 H 2 = ρ , H ˙ = − ρ + p 2 M P l 2 , (148) 3M_{\mathrm{Pl}}^2H^2=\rho,\qquad\dot{H}=-\frac{\rho+p}{2M_{\mathrm{Pl}}^2}, \tag*{(148)} 3 M Pl 2 H 2 = ρ , H ˙ = − 2 M Pl 2 ρ + p , (148)
where ρ \rho ρ and p p p include all sectors. Applied to this total stress, the Minkowski values ρ = 0 \rho=0 ρ = 0 , p < 0 p<0 p < 0 describe the instant of a bounce: H = 0 H=0 H = 0 and H ˙ > 0 \dot{H}>0 H ˙ > 0 .6 The singular ratio p / ρ p/\rho p / ρ conceals a regular cancellation between positive and negative contributions whose subsequent evolution differs.
The trace and null contributions can be kept separately as
ρ I F S = W 4 + ρ N , p I F S = − W 4 + ρ N 3 , ρ I F S + p I F S = 4 3 ρ N . (149) \rho_{\mathrm{IFS}}=\frac{W}{4}+\rho_N,\qquad p_{\mathrm{IFS}}=-\frac{W}{4}+\frac{\rho_N}{3},\qquad\rho_{\mathrm{IFS}}+p_{\mathrm{IFS}}=\frac{4}{3}\rho_N. \tag*{(149)} ρ IFS = 4 W + ρ N , p IFS = − 4 W + 3 ρ N , ρ IFS + p IFS = 3 4 ρ N . (149)
The first component, − W g μ ν / 4 -Wg_{\mu\nu}/4 − W g μν /4 , carries the full trace and has zero null contraction. The signed, traceless remainder ρ N \rho_N ρ N carries all the homogeneous IFS null energy. In the Minkowski state, ρ N = − W / 4 \rho_N=-W/4 ρ N = − W /4 : the energy densities cancel while the pressures add to − W / 3 -W/3 − W /3 . This split stays regular through zero total density. Both components describe the combined stress of strings, folds, and decay products; W / 4 W/4 W /4 is not separately the live-string energy.
Energy conservation, including the work exchanged with the dilaton, gives
ρ ˙ I F S + 4 H ρ I F S = ( H + Π 2 M P l ) W , \dot{\rho}_{\mathrm{IFS}}+4H\rho_{\mathrm{IFS}}=\left(H+\frac{\Pi}{\sqrt{2}M_{\mathrm{Pl}}}\right)W, ρ ˙ IFS + 4 H ρ IFS = ( H + 2 M Pl Π ) W ,
ρ ˙ N + 4 H ρ N = Π W 2 M P l − W ˙ 4 . (150) \dot{\rho}_N+4H\rho_N=\frac{\Pi W}{\sqrt{2}M_{\mathrm{Pl}}}-\frac{\dot{W}}{4}. \tag*{(150)} ρ ˙ N + 4 H ρ N = 2 M Pl Π W − 4 W ˙ . (150)
For nearly constant W W W and negligible dilaton work, the trace component is vacuum-like and ρ N \rho_N ρ N scales freely as a − 4 a^{-4} a − 4 . Expansion dilutes an initially negative null component, leaving positive vacuum-like stress; contraction amplifies it. A nearly constant potential slope can therefore sustain an approximately constant IFS density through Eq. (147) , even though each individual event’s trace decays.
A changing potential slope can be described along the trajectory by
ϵ V ′ ≡ − 1 2 d ln ∣ V , ϕ ∣ d ln a , V , ϕ [ ϕ ( a ) ] ≃ V , ϕ , ∗ ( a a ∗ ) − 2 ϵ V ′ . (151) \epsilon_{V'}\equiv-\frac{1}{2}\frac{d\ln\left|V_{,\phi}\right|}{d\ln a},\qquad V_{,\phi}[\phi(a)]\simeq V_{,\phi,*}\left(\frac{a}{a_*}\right)^{-2\epsilon_{V'}}. \tag*{(151)} ϵ V ′ ≡ − 2 1 d ln a d ln ∣ V , ϕ ∣ , V , ϕ [ ϕ ( a )] ≃ V , ϕ , ∗ ( a ∗ a ) − 2 ϵ V ′ . (151)
The power law applies when ϵ V ′ \epsilon_{V'} ϵ V ′ is approximately constant; this is a running of the slope along the trajectory, distinct from ϵ V = M P l 2 ( V , ϕ / V ) 2 / 2 \epsilon_V = M_{\mathrm{Pl}}^2(V_{,\phi}/V)^2/2 ϵ V = M Pl 2 ( V , ϕ / V ) 2 /2 . In force balance, W W W follows the same power law. Neglecting dilaton work, integration of Eq. (150) gives, for ϵ V ′ ≠ 2 \epsilon_{V'} \ne2 ϵ V ′ = 2 , the dual-fluid prescription ,
ρ I F S = ρ I D E + ρ R , p I F S = p I D E + ρ R 3 , ρ I D E = W 4 − 2 ϵ V ′ , p I D E = ( − 1 + 2 ϵ V ′ 3 ) ρ I D E . (152) \begin{aligned}
\rho_{\mathrm{IFS}} &= \rho_{\mathrm{IDE}}+\rho_R,\qquad p_{\mathrm{IFS}} = p_{\mathrm{IDE}}+\frac{\rho_R}{3},\\
\rho_{\mathrm{IDE}} &= \frac{W}{4-2\epsilon_{V'}},\qquad p_{\mathrm{IDE}} = \left(-1+\frac{2\epsilon_{V'}}{3}\right)\rho_{\mathrm{IDE}}. \tag*{(152)}
\end{aligned} ρ IFS ρ IDE = ρ IDE + ρ R , p IFS = p IDE + 3 ρ R , = 4 − 2 ϵ V ′ W , p IDE = ( − 1 + 3 2 ϵ V ′ ) ρ IDE . (152)
Here ρ R = C R ( a ∗ / a ) 4 \rho_R = C_R(a_*/a)^4 ρ R = C R ( a ∗ / a ) 4 , with C R C_R C R fixed by the initial stress. The label IDE refers to instant dark energy ; this component is vacuum-like only when ϵ V ′ = 0 \epsilon_{V'} = 0 ϵ V ′ = 0 . For a changing slope, it carries the stress that follows the driving force. The freely scaling radiation-like term is ρ R \rho_R ρ R ; the pure null remainder is ρ N = ρ R + ϵ V ′ ρ I D E / 2 \rho_N = \rho_R + \epsilon_{V'}\rho_{\mathrm{IDE}}/2 ρ N = ρ R + ϵ V ′ ρ IDE /2 . For example, a force proportional to a − 3 a^{-3} a − 3 gives ρ I D E = W \rho_{\mathrm{IDE}} = W ρ IDE = W and p I D E = 0 p_{\mathrm{IDE}} = 0 p IDE = 0 . This is the matter-like IFS tracking contribution relevant to the ESP bounce.
Appendix I.4 includes dilaton work through a string-frame scaling prescription and derives the transfer between the effective components when its exponent varies. At the onset of production, where a scaling exponent need not be defined, the trace/null equations above remain regular.
ESP turnaround and the IFS-mediated bounce
The cyclic scenario requires two turnarounds: first the dilaton, then the geometry. Initially the universe contracts with Π < 0 \Pi< 0 Π < 0 , so no new IFSs are produced. Passage through the ESP creates χ \chi χ particles with m χ = λ χ ∣ ϕ − ϕ E S P ∣ m_\chi= \lambda_\chi|\phi-\phi_{\mathrm{ESP}}| m χ = λ χ ∣ ϕ − ϕ ESP ∣ and conserved comoving number density. Away from the production interval they become nonrelativistic, and their increasing mass supplies a restoring force toward the ESP [14 , 22 ] . We use this nonrelativistic description, take the particles to be stable over the bounce, and neglect the ordinary dilaton potential. On the weak-coupling side, define the constant comoving force F χ ≡ λ χ n χ , ∗ > 0 F_\chi\equiv\lambda_\chi n_{\chi,*} > 0 F χ ≡ λ χ n χ , ∗ > 0 and P = a 3 Π P = a^3\Pi P = a 3 Π . Then
ρ χ = F χ a − 3 ( ϕ E S P − ϕ ) , P ˙ = F χ − a 3 W 2 M P l . (153) \rho_\chi= F_\chi a^{-3}(\phi_{\mathrm{ESP}}-\phi),\qquad\dot{P} = F_\chi-\frac{a^3W}{\sqrt{2}M_{\mathrm{Pl}}}. \tag*{(153)} ρ χ = F χ a − 3 ( ϕ ESP − ϕ ) , P ˙ = F χ − 2 M Pl a 3 W . (153)
With no inherited IFSs, W = 0 W = 0 W = 0 while Π < 0 \Pi< 0 Π < 0 , and P ˙ = F χ > 0 \dot{P} = F_\chi> 0 P ˙ = F χ > 0 brings the initially negative momentum to zero during contraction.7
Once Π > 0 \Pi> 0 Π > 0 , IFS production begins and its trace opposes the ESP force. Instantaneous balance would require
W χ , b a l = 2 M P l F χ a − 3 . (154) W_{\chi,\mathrm{bal}} = \sqrt{2}M_{\mathrm{Pl}}F_\chi a^{-3}. \tag*{(154)} W χ , bal = 2 M Pl F χ a − 3 . (154)
The enhanced cascade response allows slow positive rolling to generate a large opposing force: faster motion produces more IFSs, which push back on the dilaton. The response is delayed, however, and contraction amplifies residual kinetic motion. The dynamical question is whether the accumulated response keeps the field close to its first turn long enough for the geometry to bounce.
The a − 3 a^{-3} a − 3 dependence of the ESP force makes its balancing IFS stress matter-like. In the nearly fixed-dilaton limit, the dual-fluid prescription gives
ρ I D E ≃ W χ , b a l = 2 M P l F χ a 3 , p I D E ≃ 0 , ρ R = − c r a 4 . \rho_{\mathrm{IDE}} \simeq W_{\chi,\mathrm{bal}} = \frac{\sqrt{2}M_{\mathrm{Pl}}F_{\chi}}{a^{3}}, \qquad p_{\mathrm{IDE}} \simeq0, \qquad\rho_{R} = -\frac{c_{r}}{a^{4}}. ρ IDE ≃ W χ , bal = a 3 2 M Pl F χ , p IDE ≃ 0 , ρ R = − a 4 c r .
The trace and its associated energy thus grow together, while the negative radiation term grows faster during contraction. Let t m t_m t m be the first dilaton turn, with d m = ϕ E S P − ϕ m d_m = \phi_{\mathrm{ESP}} - \phi_m d m = ϕ ESP − ϕ m . Holding the field near ϕ m \phi_m ϕ m also gives ρ χ ≃ F χ d m / a 3 \rho_{\chi} \simeq F_{\chi}d_m/a^3 ρ χ ≃ F χ d m / a 3 . Neglecting the small dilaton kinetic energy, the total density therefore gives the reference trajectory
3 M P l 2 H 2 ≃ c m a 3 − c r a 4 , c m = F χ ( d m + 2 M P l ) , c r = c m a m − 3 M P l 2 a m 4 H m 2 . (155) 3M_{\mathrm{Pl}}^{2}H^{2} \simeq\frac{c_m}{a^{3}}-\frac{c_r}{a^{4}}, \qquad c_m = F_{\chi}(d_m+\sqrt{2}M_{\mathrm{Pl}}), \qquad c_r = c_m a_m - 3M_{\mathrm{Pl}}^{2}a_m^{4}H_m^{2}. \tag*{(155)} 3 M Pl 2 H 2 ≃ a 3 c m − a 4 c r , c m = F χ ( d m + 2 M Pl ) , c r = c m a m − 3 M Pl 2 a m 4 H m 2 . (155)
At the first turn Π m = 0 \Pi_m=0 Π m = 0 and no IFSs have yet been produced. With no additional radiation, matching to the zero IFS entrance density fixes c r = 2 M P l F χ a m > 0 c_r=\sqrt{2}M_{\mathrm{Pl}}F_{\chi}a_m>0 c r = 2 M Pl F χ a m > 0 . The reference’s two IFS contributions therefore cancel initially. As contraction amplifies the negative radiation term relative to the positive matter-like contributions, the total density reaches zero at a b = c r / c m a_b=c_r/c_m a b = c r / c m . In the fiducial example below, a b / a m ≃ 0.956 a_b/a_m\simeq0.956 a b / a m ≃ 0.956 : only about 4 % 4\% 4% further contraction is needed after the dilaton turns.
The trace starts from zero at t m t_m t m , so establishing the balancing force takes time. The reference remains predictive if this adjustment produces little field motion. By Eq. (153) , the momentum P P P measures the force mismatch accumulated since the first turn, so small momentum permits brief departures from instantaneous balance. Combining the scalar and geometric equations eliminates the explicit trace term: for J = H + Π / ( 3 2 M P l ) J=H+\Pi/(3\sqrt{2}M_{\mathrm{Pl}}) J = H + Π/ ( 3 2 M Pl ) , the W W W dependence cancels from J ˙ \dot{J} J ˙ (Appendix § 19 .§ 3 ). Small velocity and field displacement can therefore keep the geometry close to the reference even during the response’s initial adjustment.
At an exact zero of H H H , the Friedmann and Raychaudhuri equations give
6 M P l 2 H ˙ = ρ χ + W − Π 2 . (156) 6M_{\mathrm{Pl}}^{2}\dot{H}=\rho_{\chi}+W-\Pi^{2}. \tag*{(156)} 6 M Pl 2 H ˙ = ρ χ + W − Π 2 . (156)
Since W ≥ 0 W\geq0 W ≥ 0 , the condition Π 2 < ρ χ \Pi^{2}<\rho_{\chi} Π 2 < ρ χ ensures a strict bounce from contraction to expansion.
The full retarded evolution in the sharp-cutoff binary model keeps the dilaton close enough to rest that the reference fixed at the first turn predicts the bounce without a subsequent fit (Figure 11 ). On the settled comparison interval, the field moves by only about 0.18% of its distance to the ESP at the interval’s start, and the Hubble histories differ by less than 4 × 10 − 4 H 0 4 \times10^{-4}H_0 4 × 1 0 − 4 H 0 , where H 0 = − H ( t E S P ) H_0 = -H(t_{\mathrm{ESP}}) H 0 = − H ( t ESP ) . This calculation includes the persistent remnant and descendant response through the terminal crossover, with the full optical history and traceless terminal products.
Figure 11. ESP-mediated bounce in the sharp-cutoff binary reference, including the persistent remnant and descendant response. Left: the retarded Hubble evolution (solid) and nearly fixed-dilaton reference (dashed), matched at the first dilaton turn without a later fit. Right: kinetic fraction and field drift on the settled comparison interval, with the drift measured from ϕ c = ϕ ( x = 0.051 ) \phi_c = \phi(x = 0.051) ϕ c = ϕ ( x = 0.051 ) . Numerical definitions, convergence tests, and parameter variations are given in Appendix § 19 .§ 3 .
Appendix § 19 .§ 3 gives the matching, convergence checks, and residual-motion diagnostics. The homogeneous bounce also survives the tested variations of coupling, terminal cutoff, and descendant normalization. These tests use the sharp-cutoff binary response; non-dilute entrance and finite-hierarchy corrections remain outside this benchmark.
Expansion and a slowly changing potential slope
During expansion, the same force balance can sustain an IFS contribution to dark energy [13 , 14 ] . Consider a negative potential slope with negligible ESP force and sufficiently slow positive rolling that Eq. (147) holds and dilaton work is small. A nearly constant slope then maintains nearly constant W W W . Expansion dilutes the signed radiation component, leaving ρ I F S ≃ − p I F S ≃ W V b a l / 4 \rho_{\mathrm{IFS}} \simeq-p_{\mathrm{IFS}} \simeq W V_{\mathrm{bal}}/4 ρ IFS ≃ − p IFS ≃ W V bal /4 . The negative component that can drive the contracting geometry toward a bounce becomes less important on the expanding branch.
When the slope changes, the dual-fluid prescription describes how the IFS stress follows it. For approximately constant ϵ V ′ < 2 \epsilon_{V'} < 2 ϵ V ′ < 2 , the radiation term decays relative to the IDE contribution. Once it is negligible,
ρ I F S ≃ ρ I D E ≃ W V , b a l 4 − 2 ϵ V ′ , p I F S ρ I F S ≃ − 1 + 2 ϵ V ′ 3 . (157) \rho_{\mathrm{IFS}} \simeq\rho_{\mathrm{IDE}} \simeq\frac{W_{\mathrm{V,bal}}}{4-2\epsilon_{V'}} , \qquad\frac{p_{\mathrm{IFS}}}{\rho_{\mathrm{IFS}}} \simeq-1+\frac{2\epsilon_{V'}}{3}. \tag*{(157)} ρ IFS ≃ ρ IDE ≃ 4 − 2 ϵ V ′ W V , bal , ρ IFS p IFS ≃ − 1 + 3 2 ϵ V ′ . (157)
A decreasing slope magnitude gives positive IFS null energy; an increasing magnitude gives negative null energy, even after the freely redshifting term has diluted. Thus the departure from vacuum-like stress measures the evolution of the force being balanced. Neglecting dilaton kinetic energy, the combined density is
ρ D E ≃ V + W V , b a l 4 − 2 ϵ V ′ = V − M P l 2 ( 2 − ϵ V ′ ) V , ϕ . (158) \rho_{\mathrm{DE}} \simeq V+\frac{W_{\mathrm{V,bal}}}{4-2\epsilon_{V'}} = V-\frac{M_{\mathrm{Pl}}}{\sqrt{2}(2-\epsilon_{V'})}V_{,\phi}. \tag*{(158)} ρ DE ≃ V + 4 − 2 ϵ V ′ W V , bal = V − 2 ( 2 − ϵ V ′ ) M Pl V , ϕ . (158)
For a nearly constant slope this becomes V − M P l V , ϕ / ( 2 2 ) V-M_{\mathrm{Pl}}V_{,\phi}/(2\sqrt{2}) V − M Pl V , ϕ / ( 2 2 ) . The extra positive contribution is controlled by the slope, so the combined density can be positive even where V < 0 V<0 V < 0 [13 ] . These expressions retain the leading slope variation while neglecting work corrections of relative order ∣ Π ∣ / ( M P l H ) |\Pi|/(M_{\mathrm{Pl}}H) ∣Π∣/ ( M Pl H ) . When dilaton work is comparable to the slope-variation correction, the string-frame formulation in Appendix § 19 .§ 4 gives the corresponding stress and energy transfer. A varying exponent also sources ρ R \rho_R ρ R through Eq. (I.(422) ; its accumulated contribution must then be kept in the total stress.
§ 8.4 estimates the rolling rate required to sustain this branch. Whether the branch attracts solutions of the full retarded system remains to be established.
How the cascade supplies the balancing trace
The bounce and expanding branch require a large trace with little dilaton motion. The cascade can provide this response because an IFS continues to contribute long after its first splits. In the stationary problem of § 6 , the persistent remnant emits an order-one number of descendant trees per logarithmic age interval. Each tree has a finite time-integrated trace, giving the late event response w e v , s ( t ) ≃ c s / ( t − s ) w_{\mathrm{ev},s}(t)\simeq c_s/(t-s) w ev , s ( t ) ≃ c s / ( t − s ) for birth time s s s . Each semiclassical node contributes an integrated response of order g s − 2 g_s^{-2} g s − 2 , and the hierarchy contains O ( g s − κ I F S ) O(g_s^{-\kappa_{\mathrm{IFS}}}) O ( g s − κ IFS ) nodes. Their combined response therefore scales as C d e s c ∝ g s − 2 − κ I F S = g s − 2 2 C_{\mathrm{desc}}\propto g_s^{-2-\kappa_{\mathrm{IFS}}}=g_s^{-2\sqrt{2}} C desc ∝ g s − 2 − κ IFS = g s − 2 2 . The cosmological response follows by transporting these events in the evolving background and summing over their production history.
The worldsheet production calculation [6 ] , in the four-dimensional normalization of Ref. [14 ] , gives β Π + 2 \beta\Pi_+^2 β Π + 2 births per unit physical volume and time, with Π + = max ( Π , 0 ) \Pi_+=\max(\Pi,0) Π + = max ( Π , 0 ) and β = 1 / ( 32 π 6 ) \beta=1/(32\pi^6) β = 1/ ( 32 π 6 ) . If w e v , s w_{\mathrm{ev},s} w ev , s is the spatially integrated trace of one event, including its remnant and descendants, the population trace is
W ( t ) = ∫ − ∞ t d s [ a ( s ) a ( t ) ] 3 β Π + 2 ( s ) w e v , s ( t ) . (159) W(t)=\int_{-\infty}^{t} ds\left[\frac{a(s)}{a(t)}\right]^3\beta\Pi_+^2(s)w_{\mathrm{ev},s}(t). \tag*{(159)} W ( t ) = ∫ − ∞ t d s [ a ( t ) a ( s ) ] 3 β Π + 2 ( s ) w ev , s ( t ) . (159)
The integral covers the full production history and is assumed to converge; for a finite onset it starts at that onset. The volume ratio converts the event density at birth to its value at observation, while w e v , s w_{\mathrm{ev},s} w ev , s retains each event’s decay history. Together with the energy boundary condition and the other matter sources, this response closes the homogeneous equations for independently evolving cascades.
The event response can be derived from the common time-dilation factor in a yo-yo’s trace and splitting rate. Let m m m be its constant string-frame mass between splits. Oscillation averaging gives an integrated trace m / γ m/\gamma m / γ and a splitting rate λ s = Γ t m / ( 2 μ γ ) \lambda_s = \Gamma_t m/(2\mu\gamma) λ s = Γ t m / ( 2 μ γ ) , where Γ t = Γ s p l i t ( s ) ( t ) \Gamma_t = \Gamma_{\mathrm{split}}^{(s)}(t) Γ t = Γ split ( s ) ( t ) . With a s = Ω a a_s = \Omega a a s = Ω a , d t s = Ω d t \mathrm{d}t_s = \Omega\,\mathrm{d}t d t s = Ω d t , and Ω = g s / g 0 \Omega= g_s/g_0 Ω = g s / g 0 , the Einstein-frame trace is w E = Ω m / γ w_E = \Omega m/\gamma w E = Ω m / γ . Spatial homogeneity conserves k = a s p s k = a_s p_s k = a s p s , so at high boost γ ≃ k / ( a s m ) \gamma\simeq k/(a_s m) γ ≃ k / ( a s m ) and
w E ≃ a Ω 2 m 2 k , λ s d t s ≃ m 2 2 μ k a Ω 2 Γ t d t . (160) w_E \simeq a\Omega^2\frac{m^2}{k}, \qquad\lambda_s\mathrm{d}t_s \simeq\frac{m^2}{2\mu k}a\Omega^2\Gamma_t\mathrm{d}t. \tag*{(160)} w E ≃ a Ω 2 k m 2 , λ s d t s ≃ 2 μ k m 2 a Ω 2 Γ t d t . (160)
The two expressions share the factor a Ω 2 a\Omega^2 a Ω 2 . Absorbing this and the splitting rate into an optical age gives
Q s ( t ) ≡ Q ˉ ( t ) − Q ˉ ( s ) = ∫ s t d u a ( u ) Ω 2 ( u ) Γ u . (161) Q_s(t) \equiv\bar{Q}(t)-\bar{Q}(s)=\int_s^t \mathrm{d}u\,a(u)\Omega^2(u)\Gamma_u. \tag*{(161)} Q s ( t ) ≡ Q ˉ ( t ) − Q ˉ ( s ) = ∫ s t d u a ( u ) Ω 2 ( u ) Γ u . (161)
Per unit Q ˉ \bar{Q} Q ˉ , the splitting rate is m 2 / ( 2 μ k ) m^2/(2\mu k) m 2 / ( 2 μ k ) , with no explicit background dependence. The mass and momentum distribution therefore evolves with the same leading high-boost branching law, while each live string’s trace carries the observation factor Q ˙ s / Γ t \dot{Q}_s/\Gamma_t Q ˙ s / Γ t . The local small-string conditions and the finite-boost corrections are given in Appendix § 19 .§ 1 .
To follow the response through terminal conversion, we transport the stationary sharp-cutoff crossover of § 6.3 in this clock. Matching the remnant’s emissions to their stationary source requires little coupling variation during the formation intervals that supply the observed trace; subsequent background evolution is retained in Q s Q_s Q s . Let F s ( Q ) \mathcal{F}_s(Q) F s ( Q ) denote the accumulated event response, including the remnant and descendants, normalized at birth. The instantaneous Einstein-frame trace is
w e v , s ( t ) ≃ Γ s Γ t d d t F s ( Q s ( t ) ) = Γ s Γ t Q ˙ s ( t ) F s ′ ( Q s ( t ) ) . (162) w_{\mathrm{ev},s}(t) \simeq\frac{\Gamma_s}{\Gamma_t}\frac{\mathrm{d}}{\mathrm{d}t}\mathcal{F}_s(Q_s(t))=\frac{\Gamma_s}{\Gamma_t}\dot{Q}_s(t)\mathcal{F}'_s(Q_s(t)). \tag*{(162)} w ev , s ( t ) ≃ Γ t Γ s d t d F s ( Q s ( t )) = Γ t Γ s Q ˙ s ( t ) F s ′ ( Q s ( t )) . (162)
Here Γ s \Gamma_s Γ s is the splitting rate at the IFS birth. The derivative acts on F s ( Q s ( t ) ) \mathcal{F}_s(Q_s(t)) F s ( Q s ( t )) ; the rate ratio converts its birth normalization to the trace at observation. The full function F s \mathcal{F}_s F s retains the buildup and crossover of the response. For the sharp-cutoff reference used in the bounce calculation,
F s s h a r p ( Q ) = A s Ψ s h a r p ( Q Q × , s ) + 4 μ Γ s Ξ ( Q Q ∗ , s ) . (163) \mathcal{F}_s^{\mathrm{sharp}}(Q)=A_s\Psi_{\mathrm{sharp}}\left(\frac{Q}{Q_{\times,s}}\right)+\frac{4\mu}{\Gamma_s}\Xi\left(\frac{Q}{Q_{*,s}}\right). \tag*{(163)} F s sharp ( Q ) = A s Ψ sharp ( Q × , s Q ) + Γ s 4 μ Ξ ( Q ∗ , s Q ) . (163)
The two terms are the accumulated descendant and remnant responses. The scales Q × , s Q_{\times,s} Q × , s and Q ∗ , s Q_{*,s} Q ∗ , s mark terminal conversion and first splitting, respectively. Their normalization and the remnant function Ξ \Xi Ξ are specified in Appendix § 19 .§ 3 ; the youngest-cohort continuation is tested in Appendix § 20 .§ 7 .
At late optical age, the logarithmic accumulation derived in § 6 gives
F s ( Q ) = C s ln ( Q / Q r e f , s ) + O ( 1 ) , (164) \mathcal{F}_s(Q)=C_s\ln(Q/Q_{\mathrm{ref},s})+O(1), \tag*{(164)} F s ( Q ) = C s ln ( Q / Q ref , s ) + O ( 1 ) , (164)
with Q r e f , s Q_{\mathrm{ref},s} Q ref , s a fixed optical scale. The corresponding tail F s ′ ( Q ) ∼ C s / Q \mathcal{F}'_s(Q)\sim C_s/Q F s ′ ( Q ) ∼ C s / Q yields
w e v , s l a t e ( t ) ≃ C s Γ s Q ˙ s ( t ) Γ t Q s ( t ) = C s Γ s a ( t ) Ω 2 ( t ) ∫ s t d u a ( u ) Ω 2 ( u ) Γ u . (165) w_{\mathrm{ev},s}^{\mathrm{late}}(t)\simeq C_s\frac{\Gamma_s\dot{Q}_s(t)}{\Gamma_t Q_s(t)}=\frac{C_s\Gamma_s a(t)\Omega^2(t)}{\int_s^t du\,a(u)\Omega^2(u)\Gamma_u}. \tag*{(165)} w ev , s late ( t ) ≃ C s Γ t Q s ( t ) Γ s Q ˙ s ( t ) = ∫ s t d u a ( u ) Ω 2 ( u ) Γ u C s Γ s a ( t ) Ω 2 ( t ) . (165)
In the sharp reference, C s = 2 A s + 4 μ / Γ s C_s=2\mathcal{A}_s+4\mu/\Gamma_s C s = 2 A s + 4 μ / Γ s ; more generally the descendant term is twice the source-averaged integrated tree response. At constant coupling, the expression reduces to C s a ( t ) / ∫ s t a ( u ) d u C_s a(t)/\int_s^t a(u)du C s a ( t ) / ∫ s t a ( u ) d u , recovering C s / ( t − s ) C_s/(t-s) C s / ( t − s ) in Minkowski space. The background affects both the rate at which a cascade ages and its trace at observation. The late expression applies after the crossover; the full kernel (8.17) is needed during the buildup of production.
For a slowly varying expanding branch, the convolution gives a simple estimate of the rolling rate. With negligible inherited trace and nearly constant production over the contributing history,
W ≃ A c a s Π + 2 , Π ∗ 2 ≃ W V , b a l A c a s , A c a s > 0. (166) W\simeq A_{\mathrm{cas}}\Pi_+^2,\qquad\Pi_*^2\simeq\frac{W_{V,\mathrm{bal}}}{A_{\mathrm{cas}}},\qquad A_{\mathrm{cas}}>0. \tag*{(166)} W ≃ A cas Π + 2 , Π ∗ 2 ≃ A cas W V , bal , A cas > 0. (166)
For nearly constant H > 0 H>0 H > 0 and coupling, the event tail becomes w e v , s l a t e ≃ C s H / [ 1 − e − H ( t − s ) ] w_{\mathrm{ev},s}^{\mathrm{late}}\simeq C_sH/[1-e^{-H(t-s)}] w ev , s late ≃ C s H / [ 1 − e − H ( t − s ) ] . For t × ≪ t − s ≪ H − 1 t_\times\ll t-s\ll H^{-1} t × ≪ t − s ≪ H − 1 , successive logarithmic age intervals contribute equally to the population trace. Expansion suppresses older cohorts through the volume factor in Eq. (159) . When H t × ≪ 1 Ht_\times\ll1 H t × ≪ 1 , this gives
A c a s ≃ β C ( g s ) [ ln 1 H t × + O ( 1 ) ] , C ( g s ) ∝ g s − 2 2 . (167) A_{\mathrm{cas}}\simeq\beta C(g_s)\left[\ln\frac{1}{Ht_\times}+O(1)\right],\qquad C(g_s)\propto g_s^{-2\sqrt{2}}. \tag*{(167)} A cas ≃ β C ( g s ) [ ln H t × 1 + O ( 1 ) ] , C ( g s ) ∝ g s − 2 2 . (167)
Here t × = 2 μ / ( Ω Γ t m t e r m ) t_\times=2\mu/(\Omega\Gamma_t m_{\mathrm{term}}) t × = 2 μ / ( Ω Γ t m term ) , with string-frame splitting rate Γ t \Gamma_t Γ t ; the integral is evaluated in Appendix § 19 .§ 4 . At fixed balancing trace, Π ∗ ∝ g s 2 / ln [ 1 / ( H t × ) ] \Pi_*\propto g_s^{\sqrt{2}}/\sqrt{\ln[1/(Ht_\times)]} Π ∗ ∝ g s 2 / ln [ 1/ ( H t × )] . In this regime, the cascade enhancement suppresses the rolling rate required to sustain the balancing trace. The estimate applies while H t × ≪ 1 Ht_\times\ll1 H t × ≪ 1 and production and coupling vary slowly over a Hubble time. The string-frame crossover time t × ( s ) = Ω t × t_\times^{(s)}=\Omega t_\times t × ( s ) = Ω t × scales as g s − 2 g_s^{-2} g s − 2 at fixed string-frame scales. Thus the weak-coupling limit is not uniform at fixed H / Ω H/\Omega H /Ω : the condition H t × ≪ 1 Ht_\times\ll1 H t × ≪ 1 must still hold, and the full convolution is required outside this regime.
Once production ceases, an inherited population can lose its influence on the dilaton despite the persistence of individual cascades. At approximately constant coupling, sufficiently late in a radiation era, its trace scales as W o l d ∝ a − 3 t − 1 ∝ a − 5 W_{\mathrm{old}}\propto a^{-3}t^{-1}\propto a^{-5} W old ∝ a − 3 t − 1 ∝ a − 5 (Appendix § 19 .§ 4 ). Population dilution and decay of the event response both contribute. The traceful force then fades faster than radiation, while the signed traceless stress can retain an a − 4 a^{-4} a − 4 mode. Thus the surviving remnants need not maintain an appreciable dilaton force, even when an a − 4 a^{-4} a − 4 energy component remains.
The complete cosmological null-energy balance
At the homogeneous level, the affine null-energy balance is fixed by the geometry. Along a radial null ray with d t / d λ = 1 / a dt/d\lambda= 1/a d t / d λ = 1/ a ,
I γ ≡ ∫ γ d λ T μ ν k μ k ν = ∫ γ ρ + p a d t , k μ = d x μ d λ , (168) I_\gamma\equiv\int_\gamma d\lambda\,T_{\mu\nu}k^\mu k^\nu= \int_\gamma\frac{\rho+p}{a}\,dt,\qquad k^\mu= \frac{dx^\mu}{d\lambda}, \tag*{(168)} I γ ≡ ∫ γ d λ T μν k μ k ν = ∫ γ a ρ + p d t , k μ = d λ d x μ , (168)
where ρ \rho ρ and p p p include the dilaton, ESP particles, and the full IFS cascade. ANEC requires I γ ≥ 0 I_\gamma\ge0 I γ ≥ 0 on a complete ray. Using ρ + p = − 2 M P l 2 H ˙ \rho+p=-2M_{\mathrm{Pl}}^2\dot{H} ρ + p = − 2 M Pl 2 H ˙ gives
∫ t i t f ρ + p a d t = − 2 M P l 2 [ H a ] t i t f − 2 M P l 2 ∫ t i t f H 2 a d t , \int_{t_i}^{t_f}\frac{\rho+p}{a}\,dt = -2M_{\mathrm{Pl}}^2\left[\frac{H}{a}\right]_{t_i}^{t_f}-2M_{\mathrm{Pl}}^2\int_{t_i}^{t_f}\frac{H^2}{a}\,dt, ∫ t i t f a ρ + p d t = − 2 M Pl 2 [ a H ] t i t f − 2 M Pl 2 ∫ t i t f a H 2 d t ,
I n = − 2 M P l 2 ∫ t n t n + 1 H 2 a d t < 0 , H ( t n ) = H ( t n + 1 ) = 0 , H ≢ 0. (169) I_n=-2M_{\mathrm{Pl}}^2\int_{t_n}^{t_{n+1}}\frac{H^2}{a}\,dt<0,\qquad H(t_n)=H(t_{n+1})=0,\qquad H\not\equiv0. \tag*{(169)} I n = − 2 M Pl 2 ∫ t n t n + 1 a H 2 d t < 0 , H ( t n ) = H ( t n + 1 ) = 0 , H ≡ 0. (169)
Between successive bounces, the unweighted integral ∫ ( ρ + p ) d t \int(\rho+p)\,dt ∫ ( ρ + p ) d t vanishes, but the affine weight 1 / a 1/a 1/ a leaves a net deficit. Thus, if the cascade-supported bounce is embedded in the cyclic cosmology of Ref. [14 ] , every cycle contributes negative affine null energy, including its positive-NEC phases. Growth of the scale factor from one cycle to the next does not change this sign.
For a regular, nonstatic, two-sided null-complete flat-FRW history satisfying the Einstein equations, with finite limits of H / a H/a H / a at both ends, the complete affine integral is negative, possibly − ∞ -\infty − ∞ [31 ] . Since complete radial rays in spatially flat FLRW with R 3 \mathbb{R}^3 R 3 spatial sections are achronal, a self-consistent semiclassical realization of such a history would violate the proposed achronal ANEC [32 ] . The net-expanding cyclic solution of Ref. [14 ] , however, is past null-incomplete when extrapolated through infinitely many cycles, so its negative cycle integrals do not constitute a counterexample to this conjecture. The earlier bounce construction of Ref. [13 ] uses an approximately constant effective potential over the bounce and does not establish the global completeness of the full evolution.
ANEC violations on prescribed curved backgrounds [33 , 34 ] do not by themselves establish violations by the total stress sourcing a self-consistent geometry. Positive curved-space results include the massless-scalar theorem in two dimensions [35 ] and weak-curvature scalar-field results assuming null convergence [36 , 37 ] . The latter assumption fails at the bounce. Wall derives achronal ANEC from the generalized second law for minimally coupled quantum fields, at leading order in ℏ \hbar ℏ about a classical NEC-respecting background [38 ] . This links the consistency question to horizon entropy as well as to the stress tensor. For a review of these distinctions, see Ref. [39 ] .
Discussion
The main result of this paper is that IFS decay never terminates. The original remnant persists indefinitely, continuously seeding a self-similar fragmentation cascade that produces infinitely many strings. Remarkably, the complete cascade simultaneously:
The coexistence of these features may have important implications for several proposed applications of IFSs. Having discussed some of the cosmological implications in the previous section, we now turn briefly to two other applications.
(i) Black holes. IFSs have been proposed to play an important role in black-hole interiors [1 , 2 ] . In particular, for near-extremal black fivebranes, Ref. [40 ] proposed that their backreaction stabilizes the dilaton, giving rise to a constant-dilaton A d S 2 \mathrm{AdS}_2 AdS 2 interior and a horizon shock that obstructs infalling matter.
The decay cascade studied here opens a possible route toward a microscopic understanding of the black-hole information puzzle. The negative-energy folds can strongly affect the propagation of the interior partners of Hawking particles, while the cascade continuously produces correlated descendants and compensating positive-energy flux. Interactions between Hawking partners and this evolving string population, together with the resulting backreaction, could modify the entanglement structure of the black-hole state and potentially provide a mechanism for information to be encoded in the outgoing Hawking radiation.
A natural first step is to determine the backreaction of the complete cascade in the black-hole geometry. In 1 + 1 1+1 1 + 1 dimensions, the stress-tensor trace and dilaton source, together with appropriate state and boundary data, determine the remaining stress-tensor components through the diffeomorphism Ward identities. Understanding the subsequent interactions of Hawking partners with the cascade could then bring us closer to a microscopic derivation of the Page curve [41 ] .
(ii) Wavefunction collapse. In Ref. [42 ] , it was argued that, if the present cosmic acceleration is driven by IFSs, their decay could give rise to a Diósi–Penrose-like mechanism for objective wavefunction collapse [43 , 44 ] . The much more detailed understanding of the IFS decay cascade developed here provides an opportunity to revisit this proposal, sharpen its predictions, and explore possible experimental tests.
Acknowledgments
This work was supported in part by the ISF through Grant No. 256/22. U.P. was supported in part by the John Templeton Foundation under grant number 63749. The opinions expressed in this publication are those of the author(s) and do not necessarily reflect the views of the John Templeton Foundation. We thank Paul Steinhardt, Alek Bedroya and Nicolas Patino for helpful discussions.
Staircase stress tensors and yo-yo kinematics
This appendix gives the explicit distributional stress tensors associated with the staircase constructions used in Sec. 2. The formulas are local to the displayed sampling rectangle. For the original IFS remnant, additional records outside the finite rectangle complete the full infinite remnant.
Unbounded staircase of the original IFS remnant
Let p i = ( u i , v i ) p_i=(u_i,v_i) p i = ( u i , v i ) , i = 1 , … , N i=1,\ldots,N i = 1 , … , N , be the causal-minimal records, ordered by u 1 < ⋯ < u N u_1<\cdots<u_N u 1 < ⋯ < u N and v 1 > ⋯ > v N v_1>\cdots>v_N v 1 > ⋯ > v N . Set ( u 0 , v 0 ) = ( u a , v b ) (u_0,v_0)=(u_a,v_b) ( u 0 , v 0 ) = ( u a , v b ) and u N + 1 = u b u_{N+1}=u_b u N + 1 = u b . The surviving bulk is
S 0 = ⋃ i = 0 N [ u i , u i + 1 ] × [ v a , v i ] . (170) S_0=\bigcup_{i=0}^{N}[u_i,u_{i+1}]\times[v_a,v_i]. \tag*{(170)} S 0 = i = 0 ⋃ N [ u i , u i + 1 ] × [ v a , v i ] . (170)
Writing χ [ a , b ] ( z ) = Θ ( z − a ) Θ ( b − z ) \chi_{[a,b]}(z)=\Theta(z-a)\Theta(b-z) χ [ a , b ] ( z ) = Θ ( z − a ) Θ ( b − z ) , its stress and the positive folds on the stepped boundary are
T u v s t a i r ( u , v ) = μ ∑ i = 0 N χ [ u i , u i + 1 ] ( u ) χ [ v a , v i ] ( v ) , (171) T^{\mathrm{stair}}_{uv}(u,v)=\mu\sum_{i=0}^{N}\chi_{[u_i,u_{i+1}]}(u)\chi_{[v_a,v_i]}(v), \tag*{(171)} T uv stair ( u , v ) = μ i = 0 ∑ N χ [ u i , u i + 1 ] ( u ) χ [ v a , v i ] ( v ) , (171)
T u u s t a i r ( u , v ) = μ ∑ i = 1 N ( v − v i ) χ [ v i , v i − 1 ] ( v ) δ ( u − u i ) , (172) T^{\mathrm{stair}}_{uu}(u,v)=\mu\sum_{i=1}^{N}(v-v_i)\chi_{[v_i,v_{i-1}]}(v)\delta(u-u_i), \tag*{(172)} T uu stair ( u , v ) = μ i = 1 ∑ N ( v − v i ) χ [ v i , v i − 1 ] ( v ) δ ( u − u i ) , (172)
T v v s t a i r ( u , v ) = μ ∑ i = 1 N ( u − u i ) χ [ u i , u i + 1 ] ( u ) δ ( v − v i ) . (173) T^{\mathrm{stair}}_{vv}(u,v)=\mu\sum_{i=1}^{N}(u-u_i)\chi_{[u_i,u_{i+1}]}(u)\delta(v-v_i). \tag*{(173)} T v v stair ( u , v ) = μ i = 1 ∑ N ( u − u i ) χ [ u i , u i + 1 ] ( u ) δ ( v − v i ) . (173)
Each fold starts with zero null energy at its splitting point and grows as it removes bulk from the parent IFS. The original negative-energy folds remain attached to the two persistent null arms:
T u u I F S = − μ ( v − v a ) Θ ( v − v a ) δ ( u − u a ) , (174) T^{\mathrm{IFS}}_{uu}=-\mu(v-v_a)\Theta(v-v_a)\delta(u-u_a), \tag*{(174)} T uu IFS = − μ ( v − v a ) Θ ( v − v a ) δ ( u − u a ) , (174)
T v v I F S = − μ ( u − u a ) Θ ( u − u a ) δ ( v − v a ) . (175) T^{\mathrm{IFS}}_{vv}=-\mu(u-u_a)\Theta(u-u_a)\delta(v-v_a). \tag*{(175)} T v v IFS = − μ ( u − u a ) Θ ( u − u a ) δ ( v − v a ) . (175)
The full zeroth-generation remnant stress tensor is therefore
T u v ( 0 ) = T u v s t a i r , T u u ( 0 ) = T u u s t a i r + T u u I F S , T v v ( 0 ) = T v v s t a i r + T v v I F S . (176) \begin{aligned}
T^{(0)}_{uv}&=T^{\mathrm{stair}}_{uv}, \qquad T^{(0)}_{uu}=T^{\mathrm{stair}}_{uu}+T^{\mathrm{IFS}}_{uu},\\
T^{(0)}_{vv}&=T^{\mathrm{stair}}_{vv}+T^{\mathrm{IFS}}_{vv}. \tag*{(176)}
\end{aligned} T uv ( 0 ) T v v ( 0 ) = T uv stair , T uu ( 0 ) = T uu stair + T uu IFS , = T v v stair + T v v IFS . (176)
The remnant is an emitting subsystem. Inside the sampling patch its distributional divergence is
∂ v T u u ( 0 ) + ∂ u T u v ( 0 ) = − J u , ∂ u T v v ( 0 ) + ∂ v T u v ( 0 ) = − J v , (177) \partial_v T^{(0)}_{uu}+\partial_u T^{(0)}_{uv}=-J_u,\qquad
\partial_u T^{(0)}_{vv}+\partial_v T^{(0)}_{uv}=-J_v, \tag*{(177)} ∂ v T uu ( 0 ) + ∂ u T uv ( 0 ) = − J u , ∂ u T v v ( 0 ) + ∂ v T uv ( 0 ) = − J v , (177)
with positive null-momentum transfer currents
J u ( u , v ) = μ ∑ i = 1 N − 1 Δ v i δ ( u − u i + 1 ) δ ( v − v i ) , (178) J_u(u,v)=\mu\sum_{i=1}^{N-1}\Delta v_i\,\delta(u-u_{i+1})\delta(v-v_i), \tag*{(178)} J u ( u , v ) = μ i = 1 ∑ N − 1 Δ v i δ ( u − u i + 1 ) δ ( v − v i ) , (178)
J v ( u , v ) = μ ∑ i = 1 N − 1 Δ u i δ ( u − u i + 1 ) δ ( v − v i ) . (179) J_v(u,v)=\mu\sum_{i=1}^{N-1}\Delta u_i\,\delta(u-u_{i+1})\delta(v-v_i). \tag*{(179)} J v ( u , v ) = μ i = 1 ∑ N − 1 Δ u i δ ( u − u i + 1 ) δ ( v − v i ) . (179)
Here Δ u i = u i + 1 − u i \Delta u_i=u_{i+1}-u_i Δ u i = u i + 1 − u i and Δ v i = v i − v i + 1 \Delta v_i=v_i-v_{i+1} Δ v i = v i − v i + 1 . These identities are understood against test functions supported in the interior of the sampling rectangle; its artificial boundaries are supplied by continuation of the staircase, not by additional physical sources.
For example, differentiating the upper endpoint of the positive T u u T_{uu} T uu fold at u = u i + 1 u=u_{i+1} u = u i + 1 gives − μ Δ v i δ ( u − u i + 1 ) δ ( v − v i ) -\mu\Delta v_i\delta(u-u_{i+1})\delta(v-v_i) − μ Δ v i δ ( u − u i + 1 ) δ ( v − v i ) . The corresponding endpoint of T v v T_{vv} T v v gives the − μ Δ u i -\mu\Delta u_i − μ Δ u i term. Along each null arm the bulk and fold derivatives cancel; at the original split the fold starts with zero energy. At b i = ( u i + 1 , v i ) b_i=(u_{i+1},v_i) b i = ( u i + 1 , v i ) , the daughter inherits the two incoming fold momenta. Consequently the stress tensor of all detached descendants, including their later decay products, obeys
∂ v T u u d e s c + ∂ u T u v d e s c = J u , ∂ u T v v d e s c + ∂ v T u v d e s c = J v . (180) \partial_v T^{\mathrm{desc}}_{uu}+\partial_u T^{\mathrm{desc}}_{uv}=J_u,\qquad
\partial_u T^{\mathrm{desc}}_{vv}+\partial_v T^{\mathrm{desc}}_{uv}=J_v. \tag*{(180)} ∂ v T uu desc + ∂ u T uv desc = J u , ∂ u T v v desc + ∂ v T uv desc = J v . (180)
Transfers between later generations cancel within this sector. Adding the remnant therefore restores the source-free conservation equations for the complete cascade, including every birth vertex.
Bounded staircase within a yo-yo oscillation
For a parent yo-yo with oscillation rectangle D Y = [ u a , u b ] × [ v a , v b ] \mathcal{D}_Y=[u_a,u_b]\times[v_a,v_b] D Y = [ u a , u b ] × [ v a , v b ] , include both turning points as boundary vertices: p 0 = ( u a , v b ) p_0=(u_a,v_b) p 0 = ( u a , v b ) , p N + 1 = ( u b , v a ) p_{N+1}=(u_b,v_a) p N + 1 = ( u b , v a ) . The bulk T u v Y , s t a i r T^{Y,\mathrm{stair}}_{uv} T uv Y , stair is given by Eq. (A.2). The outgoing components T u u Y , s t a i r T^{Y,\mathrm{stair}}_{uu} T uu Y , stair and T v v Y , s t a i r T^{Y,\mathrm{stair}}_{vv} T v v Y , stair have the forms (A.3) and (A.4), with the summation ranges extended to i = 1 , … , N + 1 i=1,\ldots,N+1 i = 1 , … , N + 1 and i = 0 , … , N i=0,\ldots,N i = 0 , … , N , respectively. The added endpoint terms are the outgoing folds from the two turning points.
The incoming folds from ( u a , v a ) (u_a,v_a) ( u a , v a ) carry positive null energy, which decreases as the string bulk expands:
T u u Y , b o t t o m ( u , v ) = μ ( v b − v ) χ [ v a , v b ] ( v ) δ ( u − u a ) , (181) T^{Y,\mathrm{bottom}}_{uu}(u,v)=\mu(v_b-v)\chi_{[v_a,v_b]}(v)\delta(u-u_a), \tag*{(181)} T uu Y , bottom ( u , v ) = μ ( v b − v ) χ [ v a , v b ] ( v ) δ ( u − u a ) , (181)
T v v Y , b o t t o m ( u , v ) = μ ( u b − u ) χ [ u a , u b ] ( u ) δ ( v − v a ) . (182) T^{Y,\mathrm{bottom}}_{vv}(u,v)=\mu(u_b-u)\chi_{[u_a,u_b]}(u)\delta(v-v_a). \tag*{(182)} T v v Y , bottom ( u , v ) = μ ( u b − u ) χ [ u a , u b ] ( u ) δ ( v − v a ) . (182)
The complete stress tensor of the parent during this oscillation is
T u v Y = T u v Y , s t a i r , T u u Y = T u u Y , s t a i r + T u u Y , b o t t o m , T v v Y = T v v Y , s t a i r + T v v Y , b o t t o m . (183) \begin{aligned}
T^Y_{uv}&=T^{Y,\mathrm{stair}}_{uv},\qquad T^Y_{uu}=T^{Y,\mathrm{stair}}_{uu}+T^{Y,\mathrm{bottom}}_{uu},\\
T^Y_{vv}&=T^{Y,\mathrm{stair}}_{vv}+T^{Y,\mathrm{bottom}}_{vv}. \tag*{(183)}
\end{aligned} T uv Y T v v Y = T uv Y , stair , T uu Y = T uu Y , stair + T uu Y , bottom , = T v v Y , stair + T v v Y , bottom . (183)
For N = 0 N=0 N = 0 , the formulas reduce to the unsplit yo-yo stress tensor,
T u v Y = μ χ [ u a , u b ] ( u ) χ [ v a , v b ] ( v ) , (184) T^Y_{uv}=\mu\chi_{[u_a,u_b]}(u)\chi_{[v_a,v_b]}(v), \tag*{(184)} T uv Y = μ χ [ u a , u b ] ( u ) χ [ v a , v b ] ( v ) , (184)
T u u Y = μ [ ( v b − v ) δ ( u − u a ) + ( v − v a ) δ ( u − u b ) ] χ [ v a , v b ] ( v ) , (185) T^Y_{uu}=\mu\left[(v_b-v)\delta(u-u_a)+(v-v_a)\delta(u-u_b)\right]\chi_{[v_a,v_b]}(v), \tag*{(185)} T uu Y = μ [ ( v b − v ) δ ( u − u a ) + ( v − v a ) δ ( u − u b ) ] χ [ v a , v b ] ( v ) , (185)
T v v Y = μ [ ( u b − u ) δ ( v − v a ) + ( u − u a ) δ ( v − v b ) ] χ [ u a , u b ] ( u ) . (186) T^Y_{vv}=\mu\left[(u_b-u)\delta(v-v_a)+(u-u_a)\delta(v-v_b)\right]\chi_{[u_a,u_b]}(u). \tag*{(186)} T v v Y = μ [ ( u b − u ) δ ( v − v a ) + ( u − u a ) δ ( v − v b ) ] χ [ u a , u b ] ( u ) . (186)
Component form of the binary transfer map
These expressions implement the conservation-based trace transfer used in § 3.4 .
Using the u ↔ v u\leftrightarrow v u ↔ v symmetry, take Δ u ≥ Δ v \Delta u\ge\Delta v Δ u ≥ Δ v and define
r = Δ u Δ v = ( γ + γ 2 − 1 ) 2 ≥ 1. (187) r=\frac{\Delta u}{\Delta v}=\left(\gamma+\sqrt{\gamma^2-1}\right)^2\ge1. \tag*{(187)} r = Δ v Δ u = ( γ + γ 2 − 1 ) 2 ≥ 1. (187)
The daughter boosts are
γ 1 = r α U + α V 2 r α U α V , γ 2 = r ( 1 − α U ) + ( 1 − α V ) 2 r ( 1 − α U ) ( 1 − α V ) . (188) \gamma_1=\frac{r\alpha_U+\alpha_V}{2\sqrt{r\alpha_U\alpha_V}},\qquad
\gamma_2=\frac{r(1-\alpha_U)+(1-\alpha_V)}{2\sqrt{r(1-\alpha_U)(1-\alpha_V)}}. \tag*{(188)} γ 1 = 2 r α U α V r α U + α V , γ 2 = 2 r ( 1 − α U ) ( 1 − α V ) r ( 1 − α U ) + ( 1 − α V ) . (188)
We denote ℓ γ , A = ln γ A \ell_{\gamma,A}=\ln\gamma_A ℓ γ , A = ln γ A , A = 1 , 2 A=1,2 A = 1 , 2 .
The fractions of the parent trace weight inherited by the daughters are
q 1 ( γ ; α U , α V ) ≡ W 1 W = ( r + 1 ) α U α V r α U + α V , q 2 ( γ ; α U , α V ) ≡ W 2 W = ( r + 1 ) ( 1 − α U ) ( 1 − α V ) r ( 1 − α U ) + ( 1 − α V ) . (189) \begin{aligned}
q_1(\gamma;\alpha_U,\alpha_V)&\equiv\frac{W_1}{W}=\frac{(r+1)\alpha_U\alpha_V}{r\alpha_U+\alpha_V},\\
q_2(\gamma;\alpha_U,\alpha_V)&\equiv\frac{W_2}{W}=\frac{(r+1)(1-\alpha_U)(1-\alpha_V)}{r(1-\alpha_U)+(1-\alpha_V)}. \tag*{(189)}
\end{aligned} q 1 ( γ ; α U , α V ) q 2 ( γ ; α U , α V ) ≡ W W 1 = r α U + α V ( r + 1 ) α U α V , ≡ W W 2 = r ( 1 − α U ) + ( 1 − α V ) ( r + 1 ) ( 1 − α U ) ( 1 − α V ) . (189)
The equivalent fractional trace deficit is
1 − q 1 − q 2 = r ( α U − α V ) 2 ( r α U + α V ) [ r ( 1 − α U ) + ( 1 − α V ) ] ≥ 0. (190) 1-q_1-q_2=\frac{r(\alpha_U-\alpha_V)^2}{(r\alpha_U+\alpha_V)[r(1-\alpha_U)+(1-\alpha_V)]}\ge0. \tag*{(190)} 1 − q 1 − q 2 = ( r α U + α V ) [ r ( 1 − α U ) + ( 1 − α V )] r ( α U − α V ) 2 ≥ 0. (190)
From the geometric cascade to the dilute entrance population
The remnant supplies the cascade with strings of fixed characteristic mass and increasing boost. We derive their emission statistics and follow them through the initial, non-dilute splittings to a fixed small mass threshold. The population crossing that threshold supplies the source for the late-time calculation in Appendix C.
The first-generation source
The first-generation yo-yos are emitted where positive folds from consecutive records p i = ( u i , v i ) p_i=(u_i,v_i) p i = ( u i , v i ) and p i + 1 = ( u i + 1 , v i + 1 ) p_{i+1}=(u_{i+1},v_{i+1}) p i + 1 = ( u i + 1 , v i + 1 ) meet. The emission vertex, time, and null extents are
b i = ( u i + 1 , v i ) , t 0 = u i + 1 + v i 2 , (191) b_i=(u_{i+1},v_i), \qquad t_0=\frac{u_{i+1}+v_i}{2}, \tag*{(191)} b i = ( u i + 1 , v i ) , t 0 = 2 u i + 1 + v i , (191)
Δ u i = u i + 1 − u i , Δ v i = v i − v i + 1 . (192) \Delta u_i=u_{i+1}-u_i,\qquad\Delta v_i=v_i-v_{i+1}. \tag*{(192)} Δ u i = u i + 1 − u i , Δ v i = v i − v i + 1 . (192)
The stationary record variables
A i = Γ s p l i t u i v i , X i = Γ s p l i t v i ( u i + 1 − u i ) , Y i = v i + 1 v i (193) A_i=\Gamma_{\mathrm{split}}u_iv_i,\qquad X_i=\Gamma_{\mathrm{split}}v_i(u_{i+1}-u_i),\qquad Y_i=\frac{v_{i+1}}{v_i} \tag*{(193)} A i = Γ split u i v i , X i = Γ split v i ( u i + 1 − u i ) , Y i = v i v i + 1 (193)
are independent, with A i , X i ∼ Exp ( 1 ) A_i,X_i\sim\operatorname{Exp}(1) A i , X i ∼ Exp ( 1 ) and Y i ∼ U ( 0 , 1 ) Y_i\sim\mathcal{U}(0,1) Y i ∼ U ( 0 , 1 ) . Set
B = A i + X i = Γ s p l i t u i + 1 v i , Q = X i B , S = 1 − Y i . (194) B=A_i+X_i=\Gamma_{\mathrm{split}}u_{i+1}v_i,\qquad Q=\frac{X_i}{B},\qquad S=1-Y_i. \tag*{(194)} B = A i + X i = Γ split u i + 1 v i , Q = B X i , S = 1 − Y i . (194)
The variable B B B fixes the invariant position of the emission vertex, while Q Q Q and S S S determine the horizontal and vertical sizes of the emitted yo-yo. The sum and ratio of the two independent exponentials give
ρ B ( B ) = B e − B , B > 0 , Q , S ∼ U ( 0 , 1 ) (195) \rho_B(B)=Be^{-B},\qquad B>0,\qquad Q,S\sim\mathcal{U}(0,1) \tag*{(195)} ρ B ( B ) = B e − B , B > 0 , Q , S ∼ U ( 0 , 1 ) (195)
independently of B B B .
At fixed B , t 0 B,t_0 B , t 0 , the product u i + 1 v i = B / Γ s p l i t u_{i+1}v_i=B/\Gamma_{\mathrm{split}} u i + 1 v i = B / Γ split and sum u i + 1 + v i = 2 t 0 u_{i+1}+v_i=2t_0 u i + 1 + v i = 2 t 0 give the two reflection-related branches
v σ = t 0 + σ t 0 2 − B Γ s p l i t , u σ = B Γ s p l i t v σ , σ = ± . (196) v_\sigma=t_0+\sigma\sqrt{t_0^2-\frac{B}{\Gamma_{\mathrm{split}}}},\qquad u_\sigma=\frac{B}{\Gamma_{\mathrm{split}}v_\sigma},\qquad\sigma=\pm. \tag*{(196)} v σ = t 0 + σ t 0 2 − Γ split B , u σ = Γ split v σ B , σ = ± . (196)
Since − log Y i ∼ Exp ( 1 ) -\log Y_i\sim\operatorname{Exp}(1) − log Y i ∼ Exp ( 1 ) , the stationary record process has unit mean density per d log v d\log v d log v . Changing variables from ( B , log v ) (B,\log v) ( B , log v ) to ( B , t 0 ) (B,t_0) ( B , t 0 ) gives the source measure on each branch:
d N ˙ 1 , σ = B e − B t 0 2 − B / Γ s p l i t d B d Q d S , 0 < B < Γ s p l i t t 0 2 . (197) d\dot{N}_{1,\sigma}=\frac{Be^{-B}}{\sqrt{t_0^2-B/\Gamma_{\mathrm{split}}}}\,dB\,dQ\,dS,\qquad0<B<\Gamma_{\mathrm{split}}t_0^2. \tag*{(197)} d N ˙ 1 , σ = t 0 2 − B / Γ split B e − B d B d Q d S , 0 < B < Γ split t 0 2 . (197)
This measure determines the birth statistics before any further fragmentation. Its conditional averages are
⟨ F ⟩ t 0 = 1 N ˙ 1 ( t 0 ) ∑ σ = ± ∫ d N ˙ 1 , σ F ( B , Q , S ; σ ) . (198) \langle F\rangle_{t_0}=\frac{1}{\dot{N}_1(t_0)}\sum_{\sigma=\pm}\int d\dot{N}_{1,\sigma}\,F(B,Q,S;\sigma). \tag*{(198)} ⟨ F ⟩ t 0 = N ˙ 1 ( t 0 ) 1 σ = ± ∑ ∫ d N ˙ 1 , σ F ( B , Q , S ; σ ) . (198)
The extents and kinematic quantities entering these averages are
Δ u σ = B Q Γ s p l i t v σ , Δ v σ = v σ S , (199) \Delta u_\sigma=\frac{BQ}{\Gamma_{\rm split}v_\sigma},\qquad\Delta v_\sigma=v_\sigma S, \tag*{(199)} Δ u σ = Γ split v σ B Q , Δ v σ = v σ S , (199)
A 1 = B Q S Γ s p l i t , m 1 = 2 μ Γ s p l i t B Q S , (200) A_1=\frac{BQS}{\Gamma_{\rm split}},\qquad m_1=\frac{2\mu}{\sqrt{\Gamma_{\rm split}}}\sqrt{BQS}, \tag*{(200)} A 1 = Γ split B QS , m 1 = Γ split 2 μ B QS , (200)
γ 1 = B Q + h σ S 2 h σ B Q S , h σ = Γ s p l i t v σ 2 . (201) \gamma_1=\frac{BQ+h_\sigma S}{2\sqrt{h_\sigma BQS}},\qquad h_\sigma=\Gamma_{\rm split}v_\sigma^2. \tag*{(201)} γ 1 = 2 h σ B QS B Q + h σ S , h σ = Γ split v σ 2 . (201)
Write z = Γ s p l i t t 0 2 z=\Gamma_{\rm split}t_0^2 z = Γ split t 0 2 and define
M ν ( z ) = ∫ 0 z d B B ν e − B ( 1 − B z ) − 1 / 2 . (202) \mathcal{M}_\nu(z)=\int_0^z dB\,B^\nu e^{-B}\left(1-\frac{B}{z}\right)^{-1/2}. \tag*{(202)} M ν ( z ) = ∫ 0 z d B B ν e − B ( 1 − z B ) − 1/2 . (202)
Integrating over Q Q Q , S S S and the two branches gives the rate and area moments,
N ˙ 1 ( t 0 ) = 2 t 0 M 1 ( z ) = 2 t 0 [ 1 + 1 z + O ( z − 2 ) ] , (203) \dot{N}_1(t_0)=\frac{2}{t_0}\mathcal{M}_1(z)=\frac{2}{t_0}\left[1+\frac{1}{z}+O(z^{-2})\right], \tag*{(203)} N ˙ 1 ( t 0 ) = t 0 2 M 1 ( z ) = t 0 2 [ 1 + z 1 + O ( z − 2 ) ] , (203)
⟨ A 1 p ∣ t 0 ⟩ = 1 Γ s p l i t p M p + 1 ( z ) ( p + 1 ) 2 M 1 ( z ) . (204) \langle A_1^p\mid t_0\rangle=\frac{1}{\Gamma_{\rm split}^p}\frac{\mathcal{M}_{p+1}(z)}{(p+1)2\mathcal{M}_1(z)}. \tag*{(204)} ⟨ A 1 p ∣ t 0 ⟩ = Γ split p 1 ( p + 1 ) 2 M 1 ( z ) M p + 1 ( z ) . (204)
For the mean boost, the intermediate average is ⟨ γ 1 ∣ B , t 0 ⟩ = 4 Γ s p l i t t 0 / ( 3 B ) \langle\gamma_1\mid B,t_0\rangle=4\sqrt{\Gamma_{\rm split}}t_0/(3\sqrt{B}) ⟨ γ 1 ∣ B , t 0 ⟩ = 4 Γ split t 0 / ( 3 B ) . The first moments and their late-time limits are therefore
⟨ A 1 ∣ t 0 ⟩ = M 2 ( z ) 4 Γ s p l i t M 1 ( z ) ⟶ 1 2 Γ s p l i t , (205) \langle A_1\mid t_0\rangle=\frac{\mathcal{M}_2(z)}{4\Gamma_{\rm split}\mathcal{M}_1(z)}\longrightarrow\frac{1}{2\Gamma_{\rm split}}, \tag*{(205)} ⟨ A 1 ∣ t 0 ⟩ = 4 Γ split M 1 ( z ) M 2 ( z ) ⟶ 2 Γ split 1 , (205)
⟨ m 1 ∣ t 0 ⟩ = 8 μ 9 Γ s p l i t M 3 / 2 ( z ) M 1 ( z ) ⟶ 2 π 3 μ Γ s p l i t , (206) \langle m_1\mid t_0\rangle=\frac{8\mu}{9\sqrt{\Gamma_{\rm split}}}\frac{\mathcal{M}_{3/2}(z)}{\mathcal{M}_1(z)}\longrightarrow\frac{2\sqrt{\pi}}{3}\frac{\mu}{\sqrt{\Gamma_{\rm split}}}, \tag*{(206)} ⟨ m 1 ∣ t 0 ⟩ = 9 Γ split 8 μ M 1 ( z ) M 3/2 ( z ) ⟶ 3 2 π Γ split μ , (206)
⟨ γ 1 ∣ t 0 ⟩ = 4 3 z M 1 / 2 ( z ) M 1 ( z ) = 2 π 3 z [ 1 − 1 4 z + O ( z − 2 ) ] . (207) \langle\gamma_1\mid t_0\rangle=\frac{4}{3}\sqrt{z}\frac{\mathcal{M}_{1/2}(z)}{\mathcal{M}_1(z)}=\frac{2\sqrt{\pi}}{3}\sqrt{z}\left[1-\frac{1}{4z}+O(z^{-2})\right]. \tag*{(207)} ⟨ γ 1 ∣ t 0 ⟩ = 3 4 z M 1 ( z ) M 1/2 ( z ) = 3 2 π z [ 1 − 4 z 1 + O ( z − 2 ) ] . (207)
Thus the remnant emits two yo-yos per logarithmic interval of late time, with a fixed invariant-mass scale and boosts growing linearly with t 0 t_0 t 0 .
The trace average has a logarithmic subleading term that is absent from the area and mean-boost expansions. On the branch u ≤ v u\le v u ≤ v , averaging W = 4 μ u v Q S / ( u Q + v S ) W=4\mu uvQS/(uQ+vS) W = 4 μuv QS / ( u Q + v S ) first over Q Q Q , S S S gives
⟨ W ∣ u , v ⟩ 4 μ = u 2 − u 2 v ∫ 0 1 Q 2 ln ( 1 + v u Q ) d Q . (208) \frac{\langle W\mid u,v\rangle}{4\mu}=\frac{u}{2}-\frac{u^2}{v}\int_0^1 Q^2\ln\left(1+\frac{v}{uQ}\right)dQ. \tag*{(208)} 4 μ ⟨ W ∣ u , v ⟩ = 2 u − v u 2 ∫ 0 1 Q 2 ln ( 1 + u Q v ) d Q . (208)
Expanding this integral and averaging with the conditional source measure (B.7), with z = Γ s p l i t t 0 2 z = \Gamma_{\mathrm{split}}t_0^2 z = Γ split t 0 2 , yields
⟨ W 1 ∣ t 0 ⟩ = 2 μ Γ s p l i t t 0 [ 1 + 2 − γ E / 2 − ln 2 − 1 2 ln z z + O ( ln z z 2 ) ] (209) \langle W_1 \mid t_0 \rangle= \frac{2\mu}{\Gamma_{\mathrm{split}}t_0}\left[1+\frac{2-\gamma_E/2-\ln2-\frac{1}{2}\ln z}{z}+O\left(\frac{\ln z}{z^2}\right)\right] \tag*{(209)} ⟨ W 1 ∣ t 0 ⟩ = Γ split t 0 2 μ [ 1 + z 2 − γ E /2 − ln 2 − 2 1 ln z + O ( z 2 ln z ) ] (209)
where γ E \gamma_E γ E is Euler’s constant. This gives the leading law (3.16) and its logarithmic relative correction. Multiplying by Eq. (B.13) gives Eq. (79) with the same O ( ln z / z ) O(\ln z/z) O ( ln z / z ) relative accuracy; there is no leading logarithm in these source laws.
Raw-source moments for chronological matching. The same variables verify the source assumptions in Appendix § 13 .§ 1 . With c = Γ s p l i t / ( 2 μ ) c = \Gamma_{\mathrm{split}}/(2\mu) c = Γ split / ( 2 μ ) , t ∗ = Γ s p l i t − 1 / 2 t_* = \Gamma_{\mathrm{split}}^{-1/2} t ∗ = Γ split − 1/2 , and the emitted yo-yo’s signed momentum rapidity y 0 y_0 y 0 , its marks are
s 0 = c m 0 t ∗ = B Q S , U 0 = c m 0 e y 0 u b = B S , V 0 = c m 0 e − y 0 v b = B Q . (210) s_0 = cm_0t_* = \sqrt{BQ\!S},\qquad U_0 = cm_0e^{y_0}u_b = BS,\qquad V_0 = cm_0e^{-y_0}v_b = BQ. \tag*{(210)} s 0 = c m 0 t ∗ = B Q S , U 0 = c m 0 e y 0 u b = B S , V 0 = c m 0 e − y 0 v b = B Q . (210)
Changing from the birth-event rapidity to y 0 y_0 y 0 only shifts rapidity by a function of the invariant marks, preserving its unit intensity. The raw mass moment per unit rapidity is therefore
S 0 ( p ) = ∫ s 0 p d ν 0 = Γ ( 2 + p / 2 ) ( 1 + p / 2 ) 2 , p > − 2. (211) S_0(p)=\int s_0^p\,d\nu_0=\frac{\Gamma(2+p/2)}{(1+p/2)^2},\qquad p>-2. \tag*{(211)} S 0 ( p ) = ∫ s 0 p d ν 0 = ( 1 + p /2 ) 2 Γ ( 2 + p /2 ) , p > − 2. (211)
Here Γ \Gamma Γ in the numerator is Euler’s gamma function. Positive polynomial birth-coordinate moments and the mass-weighted negative logarithmic moments near the selected exponent are finite. The dimensionless birth radius is U 0 V 0 / s 0 = B \sqrt{U_0V_0}/s_0=\sqrt{B} U 0 V 0 / s 0 = B , so its mass-weighted large-radius tail is exponentially suppressed. The following estimates propagate these source moments through the non-dilute stages to the mass-threshold entrance population.
Daughter mass moments and the branching hierarchy
Let s = c m t ∗ s = cmt_* s = c m t ∗ , with c = Γ s p l i t / ( 2 μ ) c = \Gamma_{\mathrm{split}}/(2\mu) c = Γ split / ( 2 μ ) and t ∗ = Γ s p l i t − 1 / 2 t_* = \Gamma_{\mathrm{split}}^{-1/2} t ∗ = Γ split − 1/2 . The exact staircase law controls the mass weight passed to later generations, even before the binary approximation applies.
A bound from the exact staircase daughter law. Condition on n ≥ 1 n \ge1 n ≥ 1 Poisson candidates in the occupied rectangle. Rescale it to the unit square, retain the causal-minimal records, and include the boundary vertices ( 0 , 1 ) (0,1) ( 0 , 1 ) and ( 1 , 0 ) (1,0) ( 1 , 0 ) . If consecutive vertices have gaps a i = Δ u i a_i=\Delta u_i a i = Δ u i and b i = − Δ v i b_i=-\Delta v_i b i = − Δ v i , the daughter mass fractions are z i = a i b i z_i=\sqrt{a_i b_i} z i = a i b i . Thus ∑ i a i = ∑ i b i = 1 \sum_i a_i=\sum_i b_i=1 ∑ i a i = ∑ i b i = 1 and ∑ i z i ≤ 1 \sum_i z_i\le1 ∑ i z i ≤ 1 in every realization. More precisely, with h = 1 + p / 2 h=1+p/2 h = 1 + p /2 , their conditional moment is
K n ( p ) ≡ E [ ∑ i z i p ∣ n ] = Γ ( n + 1 ) Γ ( h ) h Γ ( n + h ) [ 2 + ψ ( n + h ) − ψ ( h + 1 ) ] , p > 0 , (212) K_n(p)\equiv\mathbb{E}\left[\sum_i z_i^p\mid n\right]=\frac{\Gamma(n+1)\Gamma(h)}{h\Gamma(n+h)}\left[2+\psi(n+h)-\psi(h+1)\right],\qquad p>0, \tag*{(212)} K n ( p ) ≡ E [ i ∑ z i p ∣ n ] = h Γ ( n + h ) Γ ( n + 1 ) Γ ( h ) [ 2 + ψ ( n + h ) − ψ ( h + 1 ) ] , p > 0 , (212)
where ψ \psi ψ is the digamma function. To obtain this expression, the two boundary gaps contribute 2 n B ( h , n ) / h 2nB(h,n)/h 2 n B ( h , n ) / h . For an internal gap, let x x x be its right endpoint and y y y its upper endpoint. The other n − 2 n-2 n − 2 candidates must avoid the rectangle of area x y xy x y . Integrating the two gap lengths gives
n ( n − 1 ) h 2 ∫ 0 1 ( − ln z ) z h ( 1 − z ) n − 2 d z (213) \frac{n(n-1)}{h^2}\int_0^1(-\ln z)z^h(1-z)^{n-2}\,\mathrm{d}z \tag*{(213)} h 2 n ( n − 1 ) ∫ 0 1 ( − ln z ) z h ( 1 − z ) n − 2 d z (213)
with this contribution absent for n = 1 n=1 n = 1 ; adding the boundary terms gives Eq. (B.22).
For p ≥ κ I F S p \ge\kappa_{\mathrm{IFS}} p ≥ κ IFS , K n ( p ) K_n(p) K n ( p ) is nonincreasing in n n n . Indeed, setting L n = 2 + ψ ( n + h ) − ψ ( h + 1 ) L_n=2+\psi(n+h)-\psi(h+1) L n = 2 + ψ ( n + h ) − ψ ( h + 1 ) gives K n + 1 / K n = ( n + 1 ) [ 1 + 1 / ( ( n + h ) L n ) ] / ( n + h ) K_{n+1}/K_n=(n+1)[1+1/((n+h)L_n)]/(n+h) K n + 1 / K n = ( n + 1 ) [ 1 + 1/ (( n + h ) L n )] / ( n + h ) . The required inequality is ( h − 1 ) ( n + h ) L n ≥ n + 1 (h-1)(n+h)L_n\ge n+1 ( h − 1 ) ( n + h ) L n ≥ n + 1 . At n = 1 n=1 n = 1 it follows from h 2 ≥ 2 h^2\ge2 h 2 ≥ 2 ; its left-minus-right difference increases with n n n , since its increment is at least 3 h − 4 > 0 3h-4>0 3 h − 4 > 0 . Averaging over the conditional Poisson count therefore gives, at every Λ > 0 \Lambda>0 Λ > 0 ,
E [ ∑ i s i κ I F S ] ≤ s κ I F S , E [ ∑ i s i ] ≤ 8 9 s . (214) \mathbb{E}\left[\sum_i s_i^{\kappa_{\mathrm{IFS}}}\right]\le s^{\kappa_{\mathrm{IFS}}},\qquad
\mathbb{E}\left[\sum_i s_i\right]\le\frac{8}{9}s. \tag*{(214)} E [ i ∑ s i κ IFS ] ≤ s κ IFS , E [ i ∑ s i ] ≤ 9 8 s . (214)
The first two candidate counts have K 1 ( κ I F S ) = K 2 ( κ I F S ) = 1 K_1(\kappa_{\mathrm{IFS}})=K_2(\kappa_{\mathrm{IFS}})=1 K 1 ( κ IFS ) = K 2 ( κ IFS ) = 1 ; larger counts lower this moment. The exact geometric mass hierarchy therefore has the same nonincreasing mass weight as the binary model, although its full offspring distribution and clock differ.
The binary node-count benchmark. For comparison, the binary mass hierarchy can be solved explicitly. For a binary hierarchy stopped at mass m c u t m_{\mathrm{cut}} m cut , define C = m / m c u t C=m/m_{\mathrm{cut}} C = m / m cut . Each daughter reduces this ratio by its mass fraction:
C A = z A C , A = 1 , 2. (215) C_A=z_A C,\qquad A=1,2. \tag*{(215)} C A = z A C , A = 1 , 2. (215)
Let N ( C ) \mathcal{N}(C) N ( C ) be the expected number of nodes above the cutoff. Counting the parent and its two descendant hierarchies gives
N ( C ) = 1 + ⟨ N ( z 1 C ) + N ( z 2 C ) ⟩ , N ( C ) = 0 ( C ≤ 1 ) . (216) \mathcal{N}(C)=1+\left\langle\mathcal{N}(z_1C)+\mathcal{N}(z_2C)\right\rangle,\qquad
\mathcal{N}(C)=0\quad(C\le1). \tag*{(216)} N ( C ) = 1 + ⟨ N ( z 1 C ) + N ( z 2 C ) ⟩ , N ( C ) = 0 ( C ≤ 1 ) . (216)
Set r = ln C r=\ln C r = ln C , n ( r ) = N ( e r ) n(r)=\mathcal{N}(e^r) n ( r ) = N ( e r ) and ℓ A = − ln z A > 0 \ell_A=-\ln z_A>0 ℓ A = − ln z A > 0 . Then n ( r ) = 1 + ⟨ n ( r − ℓ 1 ) + n ( r − ℓ 2 ) ⟩ n(r)=1+\langle n(r-\ell_1)+n(r-\ell_2)\rangle n ( r ) = 1 + ⟨ n ( r − ℓ 1 ) + n ( r − ℓ 2 )⟩ , with n ( r ) = 0 n(r)=0 n ( r ) = 0 for r ≤ 0 r\le0 r ≤ 0 . A shift by ℓ A \ell_A ℓ A multiplies the Laplace transform by z A p z_A^p z A p , giving n ^ ( p ) = 1 / p + M ( p ) n ^ ( p ) \hat n(p)=1/p+M(p)\hat n(p) n ^ ( p ) = 1/ p + M ( p ) n ^ ( p ) and hence
n ^ ( p ) ≡ ∫ 0 ∞ e − p r n ( r ) d r = 1 p [ 1 − M ( p ) ] . (217) \hat n(p)\equiv\int_0^\infty e^{-pr}n(r)\,\mathrm{d}r=\frac{1}{p[1-M(p)]}. \tag*{(217)} n ^ ( p ) ≡ ∫ 0 ∞ e − p r n ( r ) d r = p [ 1 − M ( p )] 1 . (217)
Its unique leading pole is at M ( κ I F S ) = 1 M(\kappa_{\mathrm{IFS}})=1 M ( κ IFS ) = 1 , so n ( r ) ∼ C N e κ I F S r n(r)\sim C_Ne^{\kappa_{\mathrm{IFS}}r} n ( r ) ∼ C N e κ IFS r , yielding Eq. (94) . Scale-local changes to the stopping rule affect the coefficient but leave this pole unchanged. The exponent describes the ensemble-averaged, multiplicity-weighted cascade of § 4.1 . It also determines the expected active worldsheet area: every counted node sweeps mean area Γ s p l i t − 1 \Gamma_{\mathrm{split}}^{-1} Γ split − 1 before its next split.
Mass-threshold entrance into the dilute regime
The first-generation law can be carried through the non-dilute layer without specifying that layer’s full distribution. Fix a dimensionless mass s d > 0 s_d > 0 s d > 0 , and stop each branch when its mass s = c m t ∗ s = cm t_\ast s = c m t ∗ first satisfies s ≤ s d s \le s_d s ≤ s d . Here t ∗ = Γ s p l i t − 1 / 2 t_\ast= \Gamma_{\mathrm{split}}^{-1/2} t ∗ = Γ split − 1/2 and c = Γ s p l i t / ( 2 μ ) c = \Gamma_{\mathrm{split}}/(2\mu) c = Γ split / ( 2 μ ) , so the mean number of candidates in a full oscillation is Λ = s 2 \Lambda= s^2 Λ = s 2 . The threshold defines a stopping population using only the past evolution, so the subsequent splitting law is unchanged. We use complete oscillation rectangles of the unabsorbed geometric model.
Entrance masses and birth times. Starting from one raw root of mass s 0 s_0 s 0 , the expected sum of masses over all nodes is at most 9 s 0 9s_0 9 s 0 . Consequently the expected number of nodes before threshold entrance is at most 9 s 0 / s d 9s_0/s_d 9 s 0 / s d . Moreover, stopping the first inequality in Eq. (B.24) gives
E [ ∑ e ∈ E d s e κ I F S ] ≤ s 0 κ I F S . (218) \mathbb{E}\left[\sum_{e \in\mathcal{E}_d} s_e^{\kappa_{\mathrm{IFS}}}\right] \le s_0^{\kappa_{\mathrm{IFS}}}. \tag*{(218)} E [ e ∈ E d ∑ s e κ IFS ] ≤ s 0 κ IFS . (218)
Both bounds follow first for a finite number of generations and then by monotone convergence or Fatou’s lemma. The first-generation law has finite polynomial moments, so these bounds are integrable per unit source rapidity. Raw roots already below the threshold have positive intensity, so the entrance moment at κ I F S \kappa_{\mathrm{IFS}} κ IFS is strictly positive.
Timing requires a separate estimate. Use the invariant birth coordinates U = c m e y u U = cme^{y_u} U = c m e y u , V = c m e − y v V = cme^{-y_v} V = c m e − y v of Appendix § 13 .§ 1 , and set H = 1 + U + V H = 1 + U + V H = 1 + U + V . For a parent of mass s s s , couple the number of empty oscillations to ϵ ∼ Exp ( 1 ) \epsilon\sim\operatorname{Exp}(1) ϵ ∼ Exp ( 1 ) as K = ⌊ ϵ / s 2 ⌋ K = \lfloor\epsilon/s^2 \rfloor K = ⌊ ϵ / s 2 ⌋ . The occupied rectangle lies within a displacement s 2 ( K + 1 ) ≤ ϵ + s 2 s^2(K+1) \le\epsilon+s^2 s 2 ( K + 1 ) ≤ ϵ + s 2 in each parent clock coordinate. Every daughter therefore obeys H i ≤ H + 2 ( ϵ + s 0 2 ) H_i \le H+2(\epsilon+s_0^2) H i ≤ H + 2 ( ϵ + s 0 2 ) , since masses never increase. The empty-cycle count is independent of the occupied staircase. For any fixed r ≥ 1 r \ge1 r ≥ 1 , applying ( H + D ) r ≤ ( 1 + ϵ ) H r + C r , ϵ D r (H+D)^r \le(1+\epsilon)H^r+C_{r,\epsilon}D^r ( H + D ) r ≤ ( 1 + ϵ ) H r + C r , ϵ D r with ( 1 + ϵ ) 8 / 9 < 1 (1+\epsilon)8/9 < 1 ( 1 + ϵ ) 8/9 < 1 yields
E [ ∑ pre-entrance Y s Y H Y r ] ≤ C r s 0 [ H 0 r + ( 1 + s 0 2 ) r ] . (219) \mathbb{E}\left[\sum_{\text{pre-entrance }Y} s_Y H_Y^r\right] \le C_r s_0\left[H_0^r+(1+s_0^2)^r\right]. \tag*{(219)} E pre-entrance Y ∑ s Y H Y r ≤ C r s 0 [ H 0 r + ( 1 + s 0 2 ) r ] . (219)
At an occupied split the expected number of daughters is bounded by E [ n + 1 ∣ n ≥ 1 ] ≤ s 0 2 + 2 \mathbb{E}[n+1\mid n\ge1]\le s_0^2+2 E [ n + 1 ∣ n ≥ 1 ] ≤ s 0 2 + 2 . Using s Y > s d s_Y>s_d s Y > s d before entrance, the same bound controls the unweighted entrance count and its H r H^r H r moments, with constants depending on s d s_d s d . Together with the raw-source marks (B.20), this proves
∫ s σ ( 1 + U + V ) r d ν d < ∞ ( σ > 0 , r ≥ 0 ) (220) \int s^\sigma(1+U+V)^r\,d\nu_d < \infty\qquad(\sigma>0,\ r\ge0) \tag*{(220)} ∫ s σ ( 1 + U + V ) r d ν d < ∞ ( σ > 0 , r ≥ 0 ) (220)
for the threshold entrance measure d ν d d\nu_d d ν d . Mass-weighted negative logarithms follow by using a slightly smaller positive σ \sigma σ . For a daughter entering at a parent’s occupied rectangle, its birth radius obeys r e ≤ [ H Y + 2 ( ϵ + s 0 2 ) ] / ( 2 s d ) r_e \le[H_Y + 2(\epsilon+ s_0^2)]/(2s d) r e ≤ [ H Y + 2 ( ϵ + s 0 2 )] / ( 2 s d ) . The direct raw roots have r 0 = B r_0 = \sqrt{B} r 0 = B . Thus the entrance radial moments are finite as well, implying the injection-tail condition used in Eq. (D.10).
Late-time injection into the dilute cascade. For one boost orbit with marks ( s , U , V ) (s,U,V) ( s , U , V ) , its laboratory birth time is T b ( y ) = ( U e − y + V e y ) / ( 2 s ) T_b(y) = (Ue^{-y} + Ve^y)/(2s) T b ( y ) = ( U e − y + V e y ) / ( 2 s ) and its injected dimensionless trace is s / cosh y s/\cosh y s / cosh y . If T b > T T_b > T T b > T , at least one of U e − y > s T Ue^{-y} > sT U e − y > s T and V e y > s T Ve^y > sT V e y > s T holds. Integrating the corresponding two rapidity tails gives
J > ( T ) ≡ ∫ d y d ν d s cosh y 1 { T b ( y ) > T } ≤ 2 T ∫ ( U + V ) d ν d = O ( T − 1 ) . (221) J_{>}(T) \equiv\int dy\,d\nu_d\,\frac{s}{\cosh y}\,\mathbf{1}_{\{T_b(y)>T\}} \le\frac{2}{T}\int(U+V)d\nu_d = O(T^{-1}). \tag*{(221)} J > ( T ) ≡ ∫ d y d ν d cosh y s 1 { T b ( y ) > T } ≤ T 2 ∫ ( U + V ) d ν d = O ( T − 1 ) . (221)
Thus the trace still awaiting injection is subleading to the T κ I F S − 1 T^{\kappa_{\mathrm{IFS}}-1} T κ IFS − 1 cascade tail. The same estimate applied to the end of each pre-entrance occupied rectangle, using Eq. (B.29), gives c t ∗ W p r e ( t ∗ T ) = O ( T − 1 ) c t_* W_{\mathrm{pre}}(t_*T) = O(T^{-1}) c t ∗ W pre ( t ∗ T ) = O ( T − 1 ) . Within an oscillation the instantaneous trace is at most twice its period average, so this bound also controls the geometric pre-entrance stress.
Together these estimates supply the entrance hypotheses of Appendix § 13 : the stopped mass moment is finite and positive, Eq. (B.30) controls the inherited clock, and Eq. (B.31) makes delayed injection and pre-entrance trace subleading. Appendix C.5 then follows the geometric evolution with these inherited birth times. The mass asymptotics use the additional pre-entrance mass condition stated in App. D.
Chronological renewal equation and the large-age limit
Repeated fragmentation both multiplies the strings and lengthens their lifetimes. We show how their competition fixes the surviving trace, W ∝ T κ − 1 W \propto T^{\kappa-1} W ∝ T κ − 1 , and its normalized boost distribution. The source sets a common amplitude; the late generations determine the power and profile. Here T = t / t ∗ T = t/t_* T = t / t ∗ and q I F S = 1 − κ q_{\mathrm{IFS}} = 1-\kappa q IFS = 1 − κ .
The calculation starts with the dilute binary process, retaining the birth time inherited from every ancestor. Its mass-weighted clock loses memory of the entrance data and approaches a stationary distribution. Summing generations then produces the pole M ( κ ) = 1 M(\kappa)=1 M ( κ ) = 1 shared by the observables below. Monotonicity converts the accumulated response into the trace on a time slice, while an occupation integral resolves that trace in boost. Finally, App. C.5 restores the geometric splitting clock and multiple-daughter events. Their corrections are summable; conservation also bounds events close to the observation slice, establishing the same leading coefficient for the instantaneous ensemble pressure and the period-averaged live trace.
Chronological variables and source assumptions
For the dilute binary reference process, each node propagates along its center-of-mass worldline until an exponentially distributed split. Its proper-time rate is c m cm c m , with c = Γ s p l i t / ( 2 μ ) c = \Gamma_{\mathrm{split}}/(2\mu) c = Γ split / ( 2 μ ) , and the two daughter null fractions are uniform and independent. Set
t ∗ = Γ s p l i t − 1 / 2 , κ = κ I F S . (222) t_* = \Gamma_{\mathrm{split}}^{-1/2}, \qquad\kappa= \kappa_{\mathrm{IFS}}. \tag*{(222)} t ∗ = Γ split − 1/2 , κ = κ IFS . (222)
For a node of mass m m m , signed rapidity y y y , and birth event ( u , v ) (u,v) ( u , v ) , define the birth coordinates in its own lifetime units,
Z = ( U , V ) , U = c m e y u , V = c m e − y v . (223) Z = (U,V), \qquad U = c m e^y u, \qquad V = c m e^{-y} v. \tag*{(223)} Z = ( U , V ) , U = c m e y u , V = c m e − y v . (223)
After a dimensionless proper waiting time ϵ ∼ Exp ( 1 ) \epsilon\sim\operatorname{Exp}(1) ϵ ∼ Exp ( 1 ) , its split radius and one daughter’s coordinates are
R ≡ c m ρ = ( U + ϵ ) ( V + ϵ ) , ρ = u s p l i t v s p l i t , (224) R \equiv c m \rho= \sqrt{(U+\epsilon)(V+\epsilon)}, \qquad\rho= \sqrt{u_{\mathrm{split}}v_{\mathrm{split}}}, \tag*{(224)} R ≡ c m ρ = ( U + ϵ ) ( V + ϵ ) , ρ = u split v split , (224)
z = a b , Z ′ = ( a ( U + ϵ ) , b ( V + ϵ ) ) , (225) z = \sqrt{ab}, \qquad Z' = \left(a(U+\epsilon),b(V+\epsilon)\right), \tag*{(225)} z = ab , Z ′ = ( a ( U + ϵ ) , b ( V + ϵ ) ) , (225)
with the second daughter obtained from ( 1 − a , 1 − b ) (1-a,1-b) ( 1 − a , 1 − b ) . In these coordinates a proper wait ϵ / ( c m ) \epsilon/(cm) ϵ / ( c m ) adds ϵ \epsilon ϵ to both entries. The daughter’s mass and rapidity then rescale the two entries by a a a and b b b . Thus Z Z Z expresses the position inherited from all preceding waits and boost changes in the current node’s lifetime units.
Entrance source. The strings that first enter a fixed dilute threshold form a stopping population with boost-invariant expected measure
d y 0 d ν ( s , U 0 , V 0 ) , s = c m 0 t ∗ > 0. (226) dy_0\,d\nu(s,U_0,V_0), \qquad s = c m_0 t_* > 0. \tag*{(226)} d y 0 d ν ( s , U 0 , V 0 ) , s = c m 0 t ∗ > 0. (226)
and source Mellin factor
S ( ζ ) = ∫ s ζ d ν . (227) \mathcal{S}(\zeta) = \int s^\zeta d\nu. \tag*{(227)} S ( ζ ) = ∫ s ζ d ν . (227)
For some 0 < δ < 1 − κ 0 < \delta< 1-\kappa 0 < δ < 1 − κ and sufficiently small η > 0 \eta> 0 η > 0 , assume
0 < S ( κ ) < ∞ , ∫ s σ ( 1 + U 0 δ + V 0 δ ) d ν < ∞ , ∣ σ − κ ∣ ≤ η . (228) 0 < \mathcal{S}(\kappa) < \infty, \qquad\int s^\sigma(1+U_0^\delta+V_0^\delta)\,d\nu< \infty, \qquad\lvert\sigma-\kappa\rvert\leq\eta. \tag*{(228)} 0 < S ( κ ) < ∞ , ∫ s σ ( 1 + U 0 δ + V 0 δ ) d ν < ∞ , ∣ σ − κ ∣ ≤ η . (228)
with κ + η + δ < 1 \kappa+\eta+\delta< 1 κ + η + δ < 1 . The expected future injection satisfies J > ( T ) = O ( T − 1 ) J_{>}(T) = O(T^{-1}) J > ( T ) = O ( T − 1 ) for T = t / t ∗ T=t/t_* T = t / t ∗ . Appendix § 12 .§ 3 verifies these conditions, together with a subleading pre-entrance trace, for the geometric mass-threshold population.
A common branching resolvent
The mass fraction z z z is correlated with the inherited clock, so the generation sum must keep both. For real p p p near κ \kappa κ , selecting one daughter with the mass weight gives
d P p ( a , b ) = 2 ( a b ) p / 2 M ( p ) d a d b , 0 < a , b < 1 , (229) \mathrm{d}\mathbb{P}_p(a,b)=\frac{2(ab)^{p/2}}{M(p)}\,\mathrm{d}a\,\mathrm{d}b,\qquad0<a,b<1, \tag*{(229)} d P p ( a , b ) = M ( p ) 2 ( ab ) p /2 d a d b , 0 < a , b < 1 , (229)
and therefore
⟨ ∑ ∣ Y ∣ = n ( m Y m 0 ) p R Y − p ⟩ = M ( p ) n E p , Z 0 [ R n − p ] . (230) \left\langle\sum_{|Y|=n}\left(\frac{m_Y}{m_0}\right)^p R_Y^{-p}\right\rangle=M(p)^n\mathbb{E}_{p,Z_0}\left[R_n^{-p}\right]. \tag*{(230)} ⟨ ∣ Y ∣ = n ∑ ( m 0 m Y ) p R Y − p ⟩ = M ( p ) n E p , Z 0 [ R n − p ] . (230)
At p = κ p=\kappa p = κ , M ( p ) = 1 M(p)=1 M ( p ) = 1 , so this weighted generation sum is itself a probability expectation. The transfer operator advances one generation with this mass weight; h ζ h_\zeta h ζ averages the final split radius over the last waiting time:
h ζ ( U , V ) = ∫ 0 ∞ d ϵ e − ϵ [ ( U + ϵ ) ( V + ϵ ) ] − ζ / 2 , (231) h_\zeta(U,V)=\int_0^\infty\mathrm{d}\epsilon\,e^{-\epsilon}[(U+\epsilon)(V+\epsilon)]^{-\zeta/2}, \tag*{(231)} h ζ ( U , V ) = ∫ 0 ∞ d ϵ e − ϵ [( U + ϵ ) ( V + ϵ ) ] − ζ /2 , (231)
( T ζ f ) ( U , V ) = 2 ∫ 0 1 d a ∫ 0 1 d b ( a b ) ζ / 2 ∫ 0 ∞ d ϵ e − ϵ f ( a ( U + ϵ ) , b ( V + ϵ ) ) . (232) (\mathcal{T}_\zeta f)(U,V)=2\int_0^1\mathrm{d}a\int_0^1\mathrm{d}b\,(ab)^{\zeta/2}\int_0^\infty\mathrm{d}\epsilon\,e^{-\epsilon}f\bigl(a(U+\epsilon),b(V+\epsilon)\bigr). \tag*{(232)} ( T ζ f ) ( U , V ) = 2 ∫ 0 1 d a ∫ 0 1 d b ( ab ) ζ /2 ∫ 0 ∞ d ϵ e − ϵ f ( a ( U + ϵ ) , b ( V + ϵ ) ) . (232)
Then T ζ 1 = M ( ζ ) \mathcal{T}_\zeta1=M(\zeta) T ζ 1 = M ( ζ ) . Define
d δ ( Z , Z ′ ) = ∣ U − U ′ ∣ δ + ∣ V − V ′ ∣ δ , [ f ] δ = sup Z ≠ Z ′ ∣ f ( Z ) − f ( Z ′ ) ∣ d δ ( Z , Z ′ ) , w ( Z ) = U δ + V δ . d_\delta(Z,Z')=|U-U'|^\delta+|V-V'|^\delta,\qquad[f]_\delta=\sup_{Z\ne Z'}\frac{|f(Z)-f(Z')|}{d_\delta(Z,Z')},\qquad w(Z)=U^\delta+V^\delta. d δ ( Z , Z ′ ) = ∣ U − U ′ ∣ δ + ∣ V − V ′ ∣ δ , [ f ] δ = Z = Z ′ sup d δ ( Z , Z ′ ) ∣ f ( Z ) − f ( Z ′ ) ∣ , w ( Z ) = U δ + V δ .
Short waiting times make h ζ h_\zeta h ζ non-Lipschitz at the origin, but for σ = Re ζ \sigma=\operatorname{Re}\zeta σ = Re ζ near κ \kappa κ it is Hölder:
∣ h ζ ( Z ) ∣ ≤ Γ ( 1 − σ ) , ∣ h ζ ( Z ) − h ζ ( Z ′ ) ∣ ≤ C ζ Γ ( 1 − σ − δ ) ( ∣ U − U ′ ∣ δ + ∣ V − V ′ ∣ δ ) . (233) |h_\zeta(Z)|\le\Gamma(1-\sigma),\qquad|h_\zeta(Z)-h_\zeta(Z')|\le C_\zeta\Gamma(1-\sigma-\delta)(|U-U'|^\delta+|V-V'|^\delta). \tag*{(233)} ∣ h ζ ( Z ) ∣ ≤ Γ ( 1 − σ ) , ∣ h ζ ( Z ) − h ζ ( Z ′ ) ∣ ≤ C ζ Γ ( 1 − σ − δ ) ( ∣ U − U ′ ∣ δ + ∣ V − V ′ ∣ δ ) . (233)
Evolving two copies with the same a a a , b b b , ϵ \epsilon ϵ multiplies their coordinate differences by a a a , b b b , giving
[ T ζ f ] δ ≤ ϑ δ ( σ ) [ f ] δ , ϑ δ ( σ ) = 2 ( 1 + σ / 2 ) ( 1 + σ / 2 + δ ) < 1 (234) [\mathcal{T}_\zeta f]_\delta\le\vartheta_\delta(\sigma)[f]_\delta,\qquad\vartheta_\delta(\sigma)=\frac{2}{(1+\sigma/2)(1+\sigma/2+\delta)}<1 \tag*{(234)} [ T ζ f ] δ ≤ ϑ δ ( σ ) [ f ] δ , ϑ δ ( σ ) = ( 1 + σ /2 ) ( 1 + σ /2 + δ ) 2 < 1 (234)
in a sufficiently narrow strip around κ \kappa κ .
The same contraction applies to any analytic observable weight b ζ ( Z ) b_\zeta(Z) b ζ ( Z ) with these boundedness and Hölder estimates. With f n = T ζ n b ζ f_n=\mathcal{T}_\zeta^n b_\zeta f n = T ζ n b ζ , separate f n = a n + g n f_n=a_n+g_n f n = a n + g n with a n = f n ( 0 , 0 ) a_n=f_n(0,0) a n = f n ( 0 , 0 ) . The coordinate-dependent part contracts geometrically,
∣ g n ( Z ) ∣ ≤ C ζ θ n w ( Z ) , ∣ d n ( ζ ) ∣ ≤ C ′ θ n , d n = ( T ζ g n ) ( 0 , 0 ) , θ < 1 , (235) |g_n(Z)|\le C_\zeta\theta^n w(Z),\qquad|d_n(\zeta)|\le C'\theta^n,\qquad d_n=(\mathcal{T}_\zeta g_n)(0,0),\qquad\theta<1, \tag*{(235)} ∣ g n ( Z ) ∣ ≤ C ζ θ n w ( Z ) , ∣ d n ( ζ ) ∣ ≤ C ′ θ n , d n = ( T ζ g n ) ( 0 , 0 ) , θ < 1 , (235)
and the scalar recurrence a n + 1 = M ( ζ ) a n + d n a_{n+1}=M(\zeta)a_n+d_n a n + 1 = M ( ζ ) a n + d n gives
∑ n ≥ 0 ( T ζ n b ζ ) ( Z ) = A b ( ζ ) 1 − M ( ζ ) + R b , ζ ( Z ) , (236) \sum_{n\ge0}(\mathcal{T}_\zeta^n b_\zeta)(Z)=\frac{\mathcal{A}_b(\zeta)}{1-M(\zeta)}+\mathcal{R}_{b,\zeta}(Z), \tag*{(236)} n ≥ 0 ∑ ( T ζ n b ζ ) ( Z ) = 1 − M ( ζ ) A b ( ζ ) + R b , ζ ( Z ) , (236)
where A b \mathcal{A}_b A b and R b , ζ \mathcal{R}_{b,\zeta} R b , ζ are analytic across the Malthusian line after source integration. Equivalently, the analytic pieces may be written as
A b ( ζ ) = b ζ ( 0 , 0 ) + ∑ n ≥ 0 d n ( ζ ) , R b , ζ ( Z ) = ∑ n ≥ 0 g n ( Z ) . (237) \mathcal{A}_b(\zeta)=b_\zeta(0,0)+\sum_{n\ge0}d_n(\zeta),\qquad\mathcal{R}_{b,\zeta}(Z)=\sum_{n\ge0}g_n(Z). \tag*{(237)} A b ( ζ ) = b ζ ( 0 , 0 ) + n ≥ 0 ∑ d n ( ζ ) , R b , ζ ( Z ) = n ≥ 0 ∑ g n ( Z ) . (237)
For b = h b=h b = h , the κ \kappa κ -weighted recursion is a Markov chain. Its birth coordinates approach a stationary law π \pi π with finite first moments, and
A h ( κ ) = π ( h κ ) , 0 < π ( h κ ) ≤ Γ ( 1 − κ ) < ∞ . (238) \mathcal{A}_h(\kappa)=\pi(h_\kappa),\qquad0<\pi(h_\kappa)\le\Gamma(1-\kappa)<\infty. \tag*{(238)} A h ( κ ) = π ( h κ ) , 0 < π ( h κ ) ≤ Γ ( 1 − κ ) < ∞. (238)
Thus the radial split-event transform
J ( ζ ) ≡ ∫ d ν s ζ ∑ n ≥ 0 ( T ζ n h ζ ) ( Z 0 ) = S ( ζ ) A h ( ζ ) 1 − M ( ζ ) + H s p l i t ( ζ ) (239) \mathcal{J}(\zeta)\equiv\int d\nu\,s^\zeta\sum_{n\ge0}(\mathcal{T}_\zeta^n h_\zeta)(Z_0)=\frac{\mathcal{S}(\zeta)\mathcal{A}_h(\zeta)}{1-M(\zeta)}+\mathcal{H}_{\mathrm{split}}(\zeta) \tag*{(239)} J ( ζ ) ≡ ∫ d ν s ζ n ≥ 0 ∑ ( T ζ n h ζ ) ( Z 0 ) = 1 − M ( ζ ) S ( ζ ) A h ( ζ ) + H split ( ζ ) (239)
has one simple pole on Re ζ = κ \operatorname{Re}\zeta=\kappa Re ζ = κ , with positive radial residue
R ∗ ≡ Res ζ = κ J ( ζ ) = S ( κ ) π ( h κ ) − M ′ ( κ ) > 0. (240) R_\ast\equiv\operatorname*{Res}_{\zeta=\kappa}\mathcal{J}(\zeta)=\frac{\mathcal{S}(\kappa)\pi(h_\kappa)}{-M'(\kappa)}>0. \tag*{(240)} R ∗ ≡ ζ = κ Res J ( ζ ) = − M ′ ( κ ) S ( κ ) π ( h κ ) > 0. (240)
The corresponding laboratory-time residue is
D ≡ B ( κ 2 , 1 2 ) R ∗ > 0. (241) D\equiv\mathcal{B}\left(\frac{\kappa}{2},\frac{1}{2}\right)R_\ast>0. \tag*{(241)} D ≡ B ( 2 κ , 2 1 ) R ∗ > 0. (241)
The second zero of 1 − M ( ζ ) 1-M(\zeta) 1 − M ( ζ ) lies to the left, so subtracting the pole leaves a function analytic across the whole boundary line.
From the accumulated response to a time slice. The pole fixes the response integrated up to time T T T . To recover the live trace, we use its decrease under splitting: after including future injection, the value on a time slice is bounded by averages over adjacent time intervals. For a nonnegative live observable Q ( T ) Q(T) Q ( T ) whose transform has residue D Q D_Q D Q and for which adding the future source tail makes Q ( T ) + J > ( T ) Q(T)+J_{>}(T) Q ( T ) + J > ( T ) nonincreasing, Wiener–Ikehara applied to its primitive gives8
∫ 1 T Q ( s ) d s ∼ D Q κ T κ . (242) \int_1^T Q(s)\,ds\sim\frac{D_Q}{\kappa}T^\kappa. \tag*{(242)} ∫ 1 T Q ( s ) d s ∼ κ D Q T κ . (242)
The compensated quantity
Q c ( T ) = Q ( T ) + J > ( T ) (243) Q_{\mathrm{c}}(T)=Q(T)+J_{>}(T) \tag*{(243)} Q c ( T ) = Q ( T ) + J > ( T ) (243)
has the same primitive to leading order because J > = O ( T − 1 ) J_{>} = O(T^{-1}) J > = O ( T − 1 ) . Writing G ( T ) = ∫ 1 T Q c ( s ) d s G(T) = \int_{1}^{T} Q_c(s)\,ds G ( T ) = ∫ 1 T Q c ( s ) d s , for any fixed λ > 1 \lambda> 1 λ > 1 monotonicity gives
G ( λ T ) − G ( T ) ( λ − 1 ) T ≤ Q c ( T ) ≤ G ( T ) − G ( T / λ ) ( 1 − 1 / λ ) T , (244) \frac{G(\lambda T)-G(T)}{(\lambda-1)T} \le Q_c(T) \le\frac{G(T)-G(T/\lambda)}{(1-1/\lambda)T}, \tag*{(244)} ( λ − 1 ) T G ( λ T ) − G ( T ) ≤ Q c ( T ) ≤ ( 1 − 1/ λ ) T G ( T ) − G ( T / λ ) , (244)
so Q ( T ) ∼ D Q T κ − 1 Q(T) \sim D_Q T^{\kappa-1} Q ( T ) ∼ D Q T κ − 1 . We use this corollary for occupation observables, including the unmarked binary trace. The exact geometric instantaneous stress will instead use the positive split measure together with the endpoint bound in Eq. (C.64).
Radial split measure and instantaneous pressure
Let N ( d r ) \mathcal{N}(dr) N ( d r ) be the expected binary split-event measure per unit source rapidity, expressed in the dimensionless invariant radius r = ρ / t ∗ r = \rho/t_* r = ρ / t ∗ . By construction its Mellin transform is J ( ζ ) \mathcal{J}(\zeta) J ( ζ ) . The positive-measure Wiener–Ikehara theorem therefore gives
N ( ( 0 , R ] ) ∼ R ∗ κ R κ . (245) \mathcal{N}((0,R]) \sim\frac{R_*}{\kappa}R^\kappa. \tag*{(245)} N (( 0 , R ]) ∼ κ R ∗ R κ . (245)
This is the chronological version of the Malthusian node count: the clock changes the residue, but not the power.
To convert radial events to a Minkowski-time slice, write u = ρ e φ u = \rho e^\varphi u = ρ e φ , v = ρ e − φ v = \rho e^{-\varphi} v = ρ e − φ , so t = ρ cosh φ t = \rho\cosh\varphi t = ρ cosh φ . The boost orbit contributes
K ρ ( t ) = ∫ R d φ δ ( t − ρ cosh φ ) = 2 Θ ( t − ρ ) t 2 − ρ 2 , ∫ r ∞ 2 T − ζ d T T 2 − r 2 = B ( ζ 2 , 1 2 ) r − ζ . (246) K_\rho(t) = \int_{\mathbb{R}} d\varphi\,\delta(t-\rho\cosh\varphi) = \frac{2\Theta(t-\rho)}{\sqrt{t^2-\rho^2}}, \qquad\int_r^\infty\frac{2T^{-\zeta}\,dT}{\sqrt{T^2-r^2}} = \mathrm{B}\left(\frac{\zeta}{2},\frac{1}{2}\right)r^{-\zeta}. \tag*{(246)} K ρ ( t ) = ∫ R d φ δ ( t − ρ cosh φ ) = t 2 − ρ 2 2Θ ( t − ρ ) , ∫ r ∞ T 2 − r 2 2 T − ζ d T = B ( 2 ζ , 2 1 ) r − ζ . (246)
The binary splitting-rate identity then gives the exact Abel relation
W ( T ) = 2 ∫ ( 0 , T ] N ( d r ) T 2 − r 2 , W ( T ) = c t ∗ W l i v e ( t ∗ T ) . (247) \mathcal{W}(T) = 2\int_{(0,T]} \frac{\mathcal{N}(dr)}{\sqrt{T^2-r^2}}, \qquad\mathcal{W}(T) = ct_* W_{\mathrm{live}}(t_*T). \tag*{(247)} W ( T ) = 2 ∫ ( 0 , T ] T 2 − r 2 N ( d r ) , W ( T ) = c t ∗ W live ( t ∗ T ) . (247)
The boost projection enhances events near r = T r = T r = T . In the binary model, monotonicity controls their contribution: energy and momentum conservation give ∑ i m i 2 / E i ≤ m 2 / E \sum_i m_i^2/E_i \le m^2/E ∑ i m i 2 / E i ≤ m 2 / E at a split, so W + J > \mathcal{W}+J_> W + J > is nonincreasing. Its Mellin transform is B ( ζ / 2 , 1 / 2 ) J ( ζ ) \mathrm{B}(\zeta/2,1/2)\mathcal{J}(\zeta) B ( ζ /2 , 1/2 ) J ( ζ ) , with residue D D D . Equations (C.21)–(C.23) therefore give
W ( T ) ∼ D T κ − 1 , W l i v e ( t ) ∼ 2 D μ Γ s p l i t − 1 / 2 ( Γ s p l i t t ) κ − 1 . (248) \mathcal{W}(T) \sim DT^{\kappa-1}, \qquad W_{\mathrm{live}}(t) \sim2D\mu\Gamma_{\mathrm{split}}^{-1/2}\left(\sqrt{\Gamma_{\mathrm{split}}t}\right)^{\kappa-1}. \tag*{(248)} W ( T ) ∼ D T κ − 1 , W live ( t ) ∼ 2 D μ Γ split − 1/2 ( Γ split t ) κ − 1 . (248)
Thus the binary pressure has the pure power in Eq. (103) , with coefficient fixed by the stationary clock and the entrance mass moment.
Boost distribution at fixed time
The boost profile uses the same resolvent with a different reward. Let
W ( T , y ) = c t ∗ d W l i v e ( t ∗ T ) d y \mathcal{W}(T,y)=ct_{\ast}\frac{dW_{\mathrm{live}}(t_{\ast}T)}{dy} W ( T , y ) = c t ∗ d y d W live ( t ∗ T )
for signed velocity rapidity y y y . For
L y ( Z , ϵ ) = ( U + ϵ ) e − y + ( V + ϵ ) e y 2 , (249) L_y(Z,\epsilon)=\frac{(U+\epsilon)e^{-y}+(V+\epsilon)e^y}{2}, \tag*{(249)} L y ( Z , ϵ ) = 2 ( U + ϵ ) e − y + ( V + ϵ ) e y , (249)
and a bounded velocity test ψ \psi ψ , define the occupation reward
h ζ , ψ ( Z ) = ∫ R d y ψ ( tanh y ) ∫ 0 ∞ d ϵ e − ϵ L y ( Z , ϵ ) − ζ . (250) h_{\zeta,\psi}(Z)=\int_{\mathbb{R}}dy\,\psi(\tanh y)\int_0^\infty d\epsilon\,e^{-\epsilon}L_y(Z,\epsilon)^{-\zeta}. \tag*{(250)} h ζ , ψ ( Z ) = ∫ R d y ψ ( tanh y ) ∫ 0 ∞ d ϵ e − ϵ L y ( Z , ϵ ) − ζ . (250)
The factor e − ϵ e^{-\epsilon} e − ϵ here is the probability of surviving to proper age ϵ \epsilon ϵ : averaging an integral over a node’s lifetime gives the displayed integral over its survival probability. The reward obeys the same Hölder estimates as h ζ h_\zeta h ζ , with the additional finite factor ∫ d y ( cosh y ) − σ \int dy(\cosh y)^{-\sigma} ∫ d y ( cosh y ) − σ . The common resolvent therefore gives
∫ 0 ∞ d T T − ζ ∫ R d y ψ ( tanh y ) W ( T , y ) = S ( ζ ) A ψ ( ζ ) 1 − M ( ζ ) + H ψ ( ζ ) , (251) \int_0^\infty dT\,T^{-\zeta}\int_{\mathbb{R}}dy\,\psi(\tanh y)\mathcal{W}(T,y)=\frac{\mathcal{S}(\zeta)\mathcal{A}_\psi(\zeta)}{1-\mathcal{M}(\zeta)}+\mathcal{H}_\psi(\zeta), \tag*{(251)} ∫ 0 ∞ d T T − ζ ∫ R d y ψ ( tanh y ) W ( T , y ) = 1 − M ( ζ ) S ( ζ ) A ψ ( ζ ) + H ψ ( ζ ) , (251)
with residue
D ψ = S ( κ ) − M ′ ( κ ) ∫ R d y ψ ( tanh y ) P κ ( y ) , P κ ( y ) = E Z ∼ π , ϵ [ L y ( Z , ϵ ) − κ ] . (252) D_\psi=\frac{\mathcal{S}(\kappa)}{-\mathcal{M}'(\kappa)}\int_{\mathbb{R}}dy\,\psi(\tanh y)\mathcal{P}_\kappa(y),\qquad\mathcal{P}_\kappa(y)=\mathbb{E}_{Z\sim\pi,\epsilon}[L_y(Z,\epsilon)^{-\kappa}]. \tag*{(252)} D ψ = − M ′ ( κ ) S ( κ ) ∫ R d y ψ ( tanh y ) P κ ( y ) , P κ ( y ) = E Z ∼ π , ϵ [ L y ( Z , ϵ ) − κ ] . (252)
Integrating over y y y first gives
Z κ ≡ ∫ R P κ ( y ) d y = B ( κ / 2 , 1 / 2 ) π ( h κ ) , D 1 = D . (253) Z_\kappa\equiv\int_{\mathbb{R}}\mathcal{P}_\kappa(y)dy=B(\kappa/2,1/2)\pi(h_\kappa),\qquad D_1=D. \tag*{(253)} Z κ ≡ ∫ R P κ ( y ) d y = B ( κ /2 , 1/2 ) π ( h κ ) , D 1 = D . (253)
To recover a fixed-time distribution, use
ψ b ( v ) = 1 1 − b v , W b ( t ) = ∑ Y m Y 2 E Y − b P Y , ∣ b ∣ < 1. (254) \psi_b(v)=\frac{1}{1-bv},\qquad W_b(t)=\sum_Y\frac{m_Y^2}{E_Y-bP_Y},\qquad|b|<1. \tag*{(254)} ψ b ( v ) = 1 − b v 1 , W b ( t ) = Y ∑ E Y − b P Y m Y 2 , ∣ b ∣ < 1. (254)
Each split decreases W b W_b W b by the same Cauchy–Schwarz argument used for the unmarked trace. Since its future-injection tail is still O ( T − 1 ) O(T^{-1}) O ( T − 1 ) , the fixed-time corollary above gives
T 1 − κ ∫ d y ψ b ( tanh y ) W ( T , y ) ⟶ D ψ b . (255) T^{1-\kappa}\int dy\,\psi_b(\tanh y)\mathcal{W}(T,y)\longrightarrow D_{\psi_b}. \tag*{(255)} T 1 − κ ∫ d y ψ b ( tanh y ) W ( T , y ) ⟶ D ψ b . (255)
On the compact velocity interval [ − 1 , 1 ] [-1,1] [ − 1 , 1 ] , any subsequential limit has the displayed integrals against every ψ b \psi_b ψ b . The expansion ψ b ( v ) = ∑ n ≥ 0 b n v n \psi_b(v)=\sum_{n\geq0}b^nv^n ψ b ( v ) = ∑ n ≥ 0 b n v n , uniformly convergent for ∣ b ∣ < 1 |b|<1 ∣ b ∣ < 1 , then fixes all its moments and hence the measure. The candidate limit has no mass at v = ± 1 v=\pm1 v = ± 1 , so this also gives weak convergence of the normalized rapidity measure. In terms of γ = cosh y \gamma=\cosh y γ = cosh y ,
ρ W ( t , γ ) d ln γ W l i v e ( t ) ⟹ R κ ( γ ) d ln γ , R κ ( γ ) = 2 P κ ( y ) Z κ tanh y , y = arcosh γ . (256) \frac{\rho_W(t,\gamma)d\ln\gamma}{W_{\mathrm{live}}(t)} \Longrightarrow\mathcal{R}_{\kappa}(\gamma)d\ln\gamma,\qquad\mathcal{R}_{\kappa}(\gamma)=\frac{2\mathcal{P}_{\kappa}(y)}{\mathcal{Z}_{\kappa}\tanh y},\qquad y=\operatorname{arcosh}\gamma. \tag*{(256)} W live ( t ) ρ W ( t , γ ) d ln γ ⟹ R κ ( γ ) d ln γ , R κ ( γ ) = Z κ tanh y 2 P κ ( y ) , y = arcosh γ . (256)
For the finite-boost shape, let A = U + ϵ A=U+\epsilon A = U + ϵ , B = V + ϵ B=V+\epsilon B = V + ϵ , and w = ( 1 + tanh y ) / 2 w=(1+\tanh y)/2 w = ( 1 + tanh y ) /2 . Then
P κ ( y ) = ( cosh y ) − κ F κ o c c ( w ) , F κ o c c ( w ) = E [ ( ( 1 − w ) A + w B ) − κ ] . (257) \mathcal{P}_{\kappa}(y)=(\cosh y)^{-\kappa}\mathcal{F}_{\kappa}^{\mathrm{occ}}(w),\qquad\mathcal{F}_{\kappa}^{\mathrm{occ}}(w)=\mathbb{E}\left[((1-w)A+wB)^{-\kappa}\right]. \tag*{(257)} P κ ( y ) = ( cosh y ) − κ F κ occ ( w ) , F κ occ ( w ) = E [ (( 1 − w ) A + w B ) − κ ] . (257)
Exchange symmetry gives F κ o c c ( w ) = F κ o c c ( 1 − w ) \mathcal{F}_{\kappa}^{\mathrm{occ}}(w)=\mathcal{F}_{\kappa}^{\mathrm{occ}}(1-w) F κ occ ( w ) = F κ occ ( 1 − w ) . Its endpoint value follows from the stationary Gamma(2 \sqrt{2} 2 , 1 1 1 ) marginal,
F κ o c c ( 1 ) = Γ ( 2 + 1 − κ ) Γ ( 2 + 1 ) . (258) \mathcal{F}_{\kappa}^{\mathrm{occ}}(1)=\frac{\Gamma(\sqrt{2}+1-\kappa)}{\Gamma(\sqrt{2}+1)}. \tag*{(258)} F κ occ ( 1 ) = Γ ( 2 + 1 ) Γ ( 2 + 1 − κ ) . (258)
Consequently
R κ ( γ ) ∼ 2 F κ o c c ( 1 ) Z κ γ − κ , γ → ∞ , (259) \mathcal{R}_{\kappa}(\gamma)\sim\frac{2\mathcal{F}_{\kappa}^{\mathrm{occ}}(1)}{\mathcal{Z}_{\kappa}}\gamma^{-\kappa},\qquad\gamma\to\infty, \tag*{(259)} R κ ( γ ) ∼ Z κ 2 F κ occ ( 1 ) γ − κ , γ → ∞ , (259)
R κ ( γ ) ∼ 2 F κ o c c ( 1 / 2 ) Z κ γ − 1 , γ → 1. (260) \mathcal{R}_{\kappa}(\gamma)\sim\frac{\sqrt{2}\mathcal{F}_{\kappa}^{\mathrm{occ}}(1/2)}{\mathcal{Z}_{\kappa}\sqrt{\gamma-1}},\qquad\gamma\to1. \tag*{(260)} R κ ( γ ) ∼ Z κ γ − 1 2 F κ occ ( 1/2 ) , γ → 1. (260)
A fixed finite boost bin therefore has a well-defined asymptotic mass; no uniform approximation up to the moving source edge is assumed.
Matching the binary and geometric cascades
As a string fragments, its worldsheet area per oscillation decreases. Both the error in the exponential waiting-time approximation and the probability of multiple candidates in one oscillation therefore become small. Their mass-weighted sum over the tree is finite, preserving the branching exponent and normalized boost profile.
The argument first establishes this summability and then applies it to the live population and to the physical split-event measure. The latter is proportional to the instantaneous bulk trace and gives the pressure directly. For the occupation calculation a parent is removed at its first physical split; daughters enter at their actual birth events. The occupied-period bound below also covers the snapshot convention used in the simulations.
A common geometric perturbation bound. Write s = c m t ∗ s=cm t_* s = c m t ∗ , Z = ( U , V ) Z=(U,V) Z = ( U , V ) , H = 1 + U + V H=1+U+V H = 1 + U + V , and Λ = s 2 \Lambda=s^2 Λ = s 2 . The exact generation operator is
( G s , ζ f ) ( Z ) = E [ ∑ i ζ i s f ( Z i G ) ∣ s , Z ] . (261) (\mathcal{G}_{s,\zeta}f)(Z)=\mathbb{E}\left[\sum_i \zeta_i^s f(Z_i^{G})\mid s,Z\right]. \tag*{(261)} ( G s , ζ f ) ( Z ) = E [ i ∑ ζ i s f ( Z i G ) ∣ s , Z ] . (261)
Couple the number of empty oscillations to a unit exponential by K = ⌊ ϵ / Λ ⌋ K=\lfloor\epsilon/\Lambda\rfloor K = ⌊ ϵ /Λ ⌋ . Every event in the occupied rectangle lies within one period of the exponential clock,
∣ Λ ( K + θ ) − ϵ ∣ ≤ Λ , 0 ≤ θ ≤ 1. (262) \lvert\Lambda(K+\theta)-\epsilon\rvert\le\Lambda,\qquad0\le\theta\le1. \tag*{(262)} ∣ Λ ( K + θ ) − ϵ ∣ ≤ Λ , 0 ≤ θ ≤ 1. (262)
Conditional on one candidate, the daughter fractions are exactly the binary ones and the inherited-coordinate discrepancy is O ( Λ ) O(\Lambda) O ( Λ ) ; the probability of more than one candidate is O ( Λ ) O(\Lambda) O ( Λ ) . Choose a a a and a narrow strip about κ \kappa κ such that 0 < a < 1 − κ − η 0<a<1-\kappa-\eta 0 < a < 1 − κ − η and 2 a > η 2a>\eta 2 a > η . With the anchored Hölder norm
∥ f ∥ a , ∗ = ∣ f ( 0 , 0 ) ∣ + [ f ] a , ∣ f ( Z ) ∣ ≤ ∥ f ∥ a , ∗ ( 1 + U a + V a ) , (263) \lVert f\rVert_{a,*}=\lvert f(0,0)\rvert+[f]_{a},\qquad\lvert f(Z)\rvert\le\lVert f\rVert_{a,*}(1+U^{a}+V^{a}), \tag*{(263)} ∥ f ∥ a , ∗ = ∣ f ( 0 , 0 )∣ + [ f ] a , ∣ f ( Z )∣ ≤ ∥ f ∥ a , ∗ ( 1 + U a + V a ) , (263)
these facts give
∣ ( G s , ζ − T ζ ) f ( Z ) ∣ ≤ C s 2 a H a ∥ f ∥ a , ∗ . (264) \lvert(\mathcal{G}_{s,\zeta}-\mathcal{T}_{\zeta})f(Z)\rvert\le Cs^{2a}H^{a}\lVert f\rVert_{a,*}. \tag*{(264)} ∣( G s , ζ − T ζ ) f ( Z )∣ ≤ C s 2 a H a ∥ f ∥ a , ∗ . (264)
The clock weight can also be controlled generation by generation. Fix s 0 ≤ s d s_0\le s_d s 0 ≤ s d . The exact staircase law gives E [ ∑ i z i p ] ≤ M ( p ) < 1 \mathbb{E}\left[\sum_i z_i^p\right]\le M(p)<1 E [ ∑ i z i p ] ≤ M ( p ) < 1 for p > κ p>\kappa p > κ . Every daughter obeys H i ≤ H + 2 ( ϵ + s 2 ) H_i\le H+2(\epsilon+s^2) H i ≤ H + 2 ( ϵ + s 2 ) , as in App. B.3. The empty-cycle clock is independent of the occupied staircase. For 0 < r ≤ 1 0<r\le1 0 < r ≤ 1 , subadditivity therefore bounds the conditional sum of z i p H i r z_i^pH_i^r z i p H i r by M ( p ) ( H r + C r ) M(p)(H^r+C_r) M ( p ) ( H r + C r ) . Iterating this inequality, while the unweighted p p p -mass decreases as M ( p ) g M(p)^g M ( p ) g , gives
E [ ∑ ∣ Y ∣ = g ( s Y s 0 ) p H Y r ] ≤ C p , r ( g + 1 ) M ( p ) g H 0 r , 0 ≤ r ≤ 1. (265) \mathbb{E}\left[\sum_{\lvert Y\rvert=g}\left(\frac{s_Y}{s_0}\right)^pH_Y^r\right]\le C_{p,r}(g+1)M(p)^gH_0^r,\qquad0\le r\le1. \tag*{(265)} E ∣ Y ∣ = g ∑ ( s 0 s Y ) p H Y r ≤ C p , r ( g + 1 ) M ( p ) g H 0 r , 0 ≤ r ≤ 1. (265)
Define the weighted exact generation operator by
( P g G f ) ( Z 0 ) = E s 0 , Z 0 [ ∑ ∣ Y ∣ = g ( s Y / s 0 ) ζ f ( Z Y G ) ] , P 0 G = Id . (266) (\mathcal{P}^{\mathrm{G}}_g f)(Z_0)=\mathbb{E}_{s_0,Z_0}\left[\sum_{\lvert Y\rvert=g}(s_Y/s_0)^\zeta f(Z_Y^{\mathrm{G}})\right],\qquad\mathcal{P}^{\mathrm{G}}_0=\operatorname{Id}. \tag*{(266)} ( P g G f ) ( Z 0 ) = E s 0 , Z 0 ∣ Y ∣ = g ∑ ( s Y / s 0 ) ζ f ( Z Y G ) , P 0 G = Id . (266)
Hence the generation error E g = P g + 1 G − P g G T ζ \mathcal{E}_g=\mathcal{P}^{\mathrm{G}}_{g+1}-\mathcal{P}^{\mathrm{G}}_g\mathcal{T}_\zeta E g = P g + 1 G − P g G T ζ obeys
∣ ( E g f ) ( Z 0 ) ∣ ≤ C s 0 2 a ( g + 1 ) M ( σ + 2 a ) g H 0 a ∥ f ∥ a , ∗ , (267) \lvert(\mathcal{E}_g f)(Z_0)\rvert\le Cs_0^{2a}(g+1)M(\sigma+2a)^gH_0^a\lVert f\rVert_{a,*}, \tag*{(267)} ∣( E g f ) ( Z 0 )∣ ≤ C s 0 2 a ( g + 1 ) M ( σ + 2 a ) g H 0 a ∥ f ∥ a , ∗ , (267)
which is summable, including after entrance-source integration.
Now let b ζ b_\zeta b ζ be any binary observable weight covered by the common resolvent, and b s , ζ G b^G_{s,\zeta} b s , ζ G its geometric counterpart. If
∣ b s , ζ G ( Z ) − b ζ ( Z ) ∣ ≤ C s 2 a H a , (268) \lvert b^G_{s,\zeta}(Z)-b_\zeta(Z)\rvert\le Cs^{2a}H^a, \tag*{(268)} ∣ b s , ζ G ( Z ) − b ζ ( Z )∣ ≤ C s 2 a H a , (268)
then the Duhamel expansion
P n G = T ζ n + ∑ j = 0 n − 1 E j T ζ n − 1 − j (269) \mathcal{P}^{\mathrm{G}}_n=\mathcal{T}_\zeta^n+\sum_{j=0}^{n-1}\mathcal{E}_j\mathcal{T}_\zeta^{n-1-j} \tag*{(269)} P n G = T ζ n + j = 0 ∑ n − 1 E j T ζ n − 1 − j (269)
and the bounds above give
O ~ G ( ζ ) = S e f f ( ζ ) A b ( ζ ) 1 − M ( ζ ) + H G , b ( ζ ) , (270) \widetilde{\mathcal{O}}_G(\zeta)=\frac{\mathcal{S}_{\mathrm{eff}}(\zeta)\mathcal{A}_b(\zeta)}{1-M(\zeta)}+\mathcal{H}_{G,b}(\zeta), \tag*{(270)} O G ( ζ ) = 1 − M ( ζ ) S eff ( ζ ) A b ( ζ ) + H G , b ( ζ ) , (270)
with analytic remainder. To see why the source factor is common, sum the Duhamel identity over n n n and insert Eq. (C.15). Its singular part is constant in Z Z Z , so each generation error acts on 1 1 1 , independently of the chosen reward:
S e f f ( ζ ) = ∫ d v s 0 ζ [ 1 + ∑ j ≥ 0 ( E j 1 ) ( Z 0 ) ] . (271) \mathcal{S}_{\mathrm{eff}}(\zeta)=\int dv\,s_0^\zeta\left[1+\sum_{j\geq0}(\mathcal{E}_j1)(Z_0)\right]. \tag*{(271)} S eff ( ζ ) = ∫ d v s 0 ζ [ 1 + j ≥ 0 ∑ ( E j 1 ) ( Z 0 ) ] . (271)
The coordinate-dependent resolvent remainder and the direct reward errors have summable analytic contributions. For the constant reward the clock drops out: the O ( s 2 ) O(s^2) O ( s 2 ) probability of extra candidates gives ∣ E j 1 ∣ ≤ C s 0 2 M ( σ + 2 ) j |\mathcal{E}_j1|\leq Cs_0^2M(\sigma+2)^j ∣ E j 1∣ ≤ C s 0 2 M ( σ + 2 ) j . At ζ = κ \zeta=\kappa ζ = κ , this implies S e f f ( κ ) = S ( κ ) [ 1 + O ( s d 2 ) ] > 0 \mathcal{S}_{\mathrm{eff}}(\kappa)=\mathcal{S}(\kappa)[1+O(s_d^2)]>0 S eff ( κ ) = S ( κ ) [ 1 + O ( s d 2 )] > 0 for sufficiently dilute entrance. The residue is
D G , b = S e f f ( κ ) − M ′ ( κ ) π ( b κ ) . (272) D_{G,b}=\frac{\mathcal{S}_{\mathrm{eff}}(\kappa)}{-M'(\kappa)}\pi(b_\kappa). \tag*{(272)} D G , b = − M ′ ( κ ) S eff ( κ ) π ( b κ ) . (272)
Thus the early geometric evolution enters these observables through a common amplitude. Their pole and normalized residue ratios are those of the binary process.
String occupation and the boost profile. For a node of rapidity y y y , put B y = ( U e − y + V e y ) / 2 B_y=(Ue^{-y}+Ve^y)/2 B y = ( U e − y + V e y ) /2 and L y ( Z , r ) = B y + r cosh y L_y(Z,r)=B_y+r\cosh y L y ( Z , r ) = B y + r cosh y . If ℓ s ( y ) \ell_s(y) ℓ s ( y ) is its exact lifetime to the first physical split, the geometric occupation reward is
b s , ζ , ψ G ( Z ) = ∫ R d y ψ ( tanh y ) E [ ∫ 0 ℓ s ( y ) L y ( Z , r ) − ζ d r ] . (273) b^G_{s,\zeta,\psi}(Z)=\int_{\mathbb{R}}dy\,\psi(\tanh y)\mathbb{E}\left[\int_0^{\ell_s(y)}L_y(Z,r)^{-\zeta}\,dr\right]. \tag*{(273)} b s , ζ , ψ G ( Z ) = ∫ R d y ψ ( tanh y ) E [ ∫ 0 ℓ s ( y ) L y ( Z , r ) − ζ d r ] . (273)
Since ∣ ℓ s ( y ) − ϵ ∣ ≤ Λ |\ell_s(y)-\epsilon|\leq\Lambda ∣ ℓ s ( y ) − ϵ ∣ ≤ Λ , the endpoint discrepancy satisfies
∣ b s , ζ , ψ G − h ζ , ψ ∣ ≤ C ∥ ψ ∥ ∞ Λ , (274) \left|b^G_{s,\zeta,\psi}-h_{\zeta,\psi}\right|\leq C\|\psi\|_\infty\Lambda, \tag*{(274)} b s , ζ , ψ G − h ζ , ψ ≤ C ∥ ψ ∥ ∞ Λ , (274)
so Eq. (C.47) applies. The normalized residues therefore give exactly the binary profile R κ \mathcal{R}_\kappa R κ . The finite interval between a split and daughter formation contributes a subleading trace. Count a daughter as pending between its parent’s first split and its actual birth. The probability that a fixed time lies in the occupied period is
( 1 − e − Λ ) e − Λ ⌊ r / Λ ⌋ ≤ C Λ e − r , (275) (1-e^{-\Lambda})e^{-\Lambda\lfloor r/\Lambda\rfloor}\leq C\Lambda e^{-r}, \tag*{(275)} ( 1 − e − Λ ) e − Λ ⌊ r /Λ ⌋ ≤ C Λ e − r , (275)
and the same boost envelope as in the binary source bound gives
∣ P ψ ( T ) ∣ ≤ C ∥ ψ ∥ ∞ T ∫ d v d E [ ∑ Y s Y 2 H Y ] = O ( T − 1 ) . (276) |\mathcal{P}_\psi(T)|\leq\frac{C\|\psi\|_\infty}{T}\int dv_d\,\mathbb{E}\left[\sum_Y s_Y^2H_Y\right]=O(T^{-1}). \tag*{(276)} ∣ P ψ ( T ) ∣ ≤ T C ∥ ψ ∥ ∞ ∫ d v d E [ Y ∑ s Y 2 H Y ] = O ( T − 1 ) . (276)
Adding this pending term and the O ( T − 1 ) O(T^{-1}) O ( T − 1 ) future source makes the monotone test rewards W b W_b W b nonincreasing, so the fixed-time corollary of Eq. (C.23) yields
W G , ψ ( T ) ∼ D G , ψ T κ − 1 ( ψ = 1 or ψ b ) . (277) \mathcal{W}_{G,\psi}(T) \sim D_{G,\psi}T^{\kappa-1} \qquad(\psi= 1\ \text{or}\ \psi_b). \tag*{(277)} W G , ψ ( T ) ∼ D G , ψ T κ − 1 ( ψ = 1 or ψ b ) . (277)
The trace retained by the numerical parent convention differs only in the same occupied period and is bounded by the unsplit parent’s weight. Equation (C.55) therefore also bounds the difference between the two snapshot conventions by O ( T − 1 ) O(T^{-1}) O ( T − 1 ) . For physical bulk stress use the factor-two bound (71) . This proves the same normalized fixed-bin boost profile for both geometric occupation conventions.
Physical split events and the instantaneous pressure. The same active area carries string tension and admits splitting. One sheet has T u v = μ T_{uv}=\mu T uv = μ and physical split intensity Γ s p l i t d u d v \Gamma_{\mathrm{split}}\,du\,dv Γ split d u d v . Thus, for the developed descendant cascade,
d ⟨ N p h y s ⟩ = Γ s p l i t μ ⟨ T u v ⟩ d u d v . (278) d\langle N_{\mathrm{phys}}\rangle= \frac{\Gamma_{\mathrm{split}}}{\mu}\langle T_{uv}\rangle\,du\,dv. \tag*{(278)} d ⟨ N phys ⟩ = μ Γ split ⟨ T uv ⟩ d u d v . (278)
With u = t ∗ r e φ u=t_*re^\varphi u = t ∗ r e φ , v = t ∗ r e − φ v=t_*re^{-\varphi} v = t ∗ r e − φ define the radial physical split measure per unit d φ d\varphi d φ ,
n G ( d r ) = 2 r μ ⟨ T u v ( t ∗ 2 r 2 ) ⟩ d r . (279) n_G(dr)=\frac{2r}{\mu}\langle T_{uv}(t_*^2r^2)\rangle\,dr. \tag*{(279)} n G ( d r ) = μ 2 r ⟨ T uv ( t ∗ 2 r 2 )⟩ d r . (279)
The exact instantaneous trace on a Minkowski-time slice is therefore
W i n s t , G ( T ) = 2 ∫ ( 0 , T ] n G ( d r ) T 2 − r 2 . (280) \mathcal{W}_{\mathrm{inst},G}(T)=2\int_{(0,T]} \frac{n_G(dr)}{\sqrt{T^2-r^2}}. \tag*{(280)} W inst , G ( T ) = 2 ∫ ( 0 , T ] T 2 − r 2 n G ( d r ) . (280)
To match the physical split measure, let R a G = c m ρ a R_a^G=cm\rho_a R a G = c m ρ a be the dimensionless invariant radii of the causal-minimal records in the first occupied rectangle, in the parent’s lifetime units, and
ϵ s , ζ G ( Z ) = E [ ∑ a ( R a G ) − ζ ∣ s , Z ] . (281) \epsilon_{s,\zeta}^G(Z)=\mathbb{E}\left[\left.\sum_a(R_a^G)^{-\zeta}\right|s,Z\right]. \tag*{(281)} ϵ s , ζ G ( Z ) = E [ a ∑ ( R a G ) − ζ s , Z ] . (281)
For a one-candidate rectangle, the coupling above moves each rescaled null coordinate by at most Λ \Lambda Λ ; the Hölder estimate (C.12) therefore gives an O ( Λ a ) O(\Lambda^a) O ( Λ a ) difference from h ζ h_\zeta h ζ . Conditional on occupation, the expected number of additional candidates is O ( Λ ) O(\Lambda) O ( Λ ) . For occupied-period index K ≥ 1 K\geq1 K ≥ 1 their inverse-radius moment is bounded by C ( K Λ ) − σ C(K\Lambda)^{-\sigma} C ( K Λ ) − σ , while the exceptional first period contributes O ( Λ 2 − σ ) O(\Lambda^{2-\sigma}) O ( Λ 2 − σ ) . Summing over Pr ( K = k ) = ( 1 − e − Λ ) e − k Λ \Pr(K=k)=(1-e^{-\Lambda})e^{-k\Lambda} Pr ( K = k ) = ( 1 − e − Λ ) e − k Λ gives
∣ ϵ s , ζ G ( Z ) − h ζ ( Z ) ∣ ≤ C ζ Λ a = C ζ s 2 a , (282) \left|\epsilon_{s,\zeta}^G(Z)-h_\zeta(Z)\right|\leq C_\zeta\Lambda^a=C_\zeta s^{2a}, \tag*{(282)} ϵ s , ζ G ( Z ) − h ζ ( Z ) ≤ C ζ Λ a = C ζ s 2 a , (282)
and logarithmically weighted variants obey the same locally uniform bound. Thus the common geometric reward theorem applies to the actual split measure. Its Mellin transform has the same pole,
J G ( ζ ) = S e f f ( ζ ) A h ( ζ ) 1 − M ( ζ ) + R G s p l i t ( ζ ) , (283) \mathcal{J}_G(\zeta)=\frac{S_{\mathrm{eff}}(\zeta)\mathcal{A}_h(\zeta)}{1-M(\zeta)}+\mathcal{R}_G^{\mathrm{split}}(\zeta), \tag*{(283)} J G ( ζ ) = 1 − M ( ζ ) S eff ( ζ ) A h ( ζ ) + R G split ( ζ ) , (283)
so the positive-measure theorem gives
n G ( ( 0 , R ] ) ∼ R G , ∗ κ R κ , R G , ∗ = S e f f ( κ ) π ( h κ ) − M ′ ( κ ) . n_G((0,R]) \sim\frac{R_{G,*}}{\kappa}R^\kappa,\qquad R_{G,*}=\frac{S_{\mathrm{eff}}(\kappa)\pi(h_\kappa)}{-M'(\kappa)}. n G (( 0 , R ]) ∼ κ R G , ∗ R κ , R G , ∗ = − M ′ ( κ ) S eff ( κ ) π ( h κ ) .
The boost kernel enhances events close to the observation slice, r = T r=T r = T . Conservation and positivity bound this local contribution, allowing the radial asymptotic to pass through the Abel integral.
Include the remnant and pre-entrance stages and denote their full radial measure by n f u l l n_{\mathrm{full}} n full . Their expected event count per unit rapidity is finite by the raw-source law and the pre-entrance node and candidate bounds in App. B.3; hence their cumulative contribution is O ( 1 ) O(1) O ( 1 ) . Thus F ( R ) = n f u l l ( ( 0 , R ] ) ∼ ( R G , ∗ / κ ) R κ F(R)=n_{\mathrm{full}}((0,R])\sim(R_{G,*}/\kappa)R^\kappa F ( R ) = n full (( 0 , R ]) ∼ ( R G , ∗ / κ ) R κ . For the full conserved geometric stress, put f ( z ) = ⟨ T u v ⟩ f(z)=\langle T_{uv}\rangle f ( z ) = ⟨ T uv ⟩ and I ( z ) = ∫ 0 z f ( w ) d w I(z)=\int_0^z f(w)\,dw I ( z ) = ∫ 0 z f ( w ) d w . Positivity of its regular null stress gives 0 ≤ z f ( z ) ≤ I ( z ) 0\le zf(z)\le I(z) 0 ≤ z f ( z ) ≤ I ( z ) , by the conservation identity (F.9). At every finite radius, this implies
F ( R ) = Γ s p l i t μ I ( t ∗ 2 R 2 ) , 0 ≤ F ′ ( R ) ≤ 2 F ( R ) R . (284) F(R)=\frac{\Gamma_{\mathrm{split}}}{\mu}I(t_*^2R^2),\qquad0\le F'(R)\le\frac{2F(R)}{R}. \tag*{(284)} F ( R ) = μ Γ split I ( t ∗ 2 R 2 ) , 0 ≤ F ′ ( R ) ≤ R 2 F ( R ) . (284)
For 0 < ε < 1 / 2 0<\varepsilon<1/2 0 < ε < 1/2 and sufficiently large T T T , this implies
2 ∫ ( 1 − ε ) T T n G ( d r ) T 2 − r 2 ≤ 2 ∫ ( 1 − ε ) T T d F ( r ) T 2 − r 2 ≤ C T κ − 1 ε . (285) 2\int_{(1-\varepsilon)T}^{T}\frac{n_G(dr)}{\sqrt{T^2-r^2}}\le2\int_{(1-\varepsilon)T}^{T}\frac{dF(r)}{\sqrt{T^2-r^2}}\le CT^{\kappa-1}\sqrt{\varepsilon}. \tag*{(285)} 2 ∫ ( 1 − ε ) T T T 2 − r 2 n G ( d r ) ≤ 2 ∫ ( 1 − ε ) T T T 2 − r 2 d F ( r ) ≤ C T κ − 1 ε . (285)
On [ 0 , ( 1 − ε ) T ] [0,(1-\varepsilon)T] [ 0 , ( 1 − ε ) T ] , the rescaled Abel kernel is bounded and cumulative regular variation gives the limiting integral. Taking first T → ∞ T\to\infty T → ∞ , then ε → 0 \varepsilon\to0 ε → 0 , gives 2 R G , ∗ ∫ 0 1 x κ − 1 ( 1 − x 2 ) − 1 / 2 d x = B ( κ / 2 , 1 / 2 ) R G , ∗ 2R_{G,*}\int_0^1 x^{\kappa-1}(1-x^2)^{-1/2}\,dx=B(\kappa/2,1/2)R_{G,*} 2 R G , ∗ ∫ 0 1 x κ − 1 ( 1 − x 2 ) − 1/2 d x = B ( κ /2 , 1/2 ) R G , ∗ . Therefore
W i n s t , G ( T ) ∼ D G , 1 T κ − 1 . (286) \mathcal{W}_{\mathrm{inst},G}(T)\sim D_{G,1}T^{\kappa-1}. \tag*{(286)} W inst , G ( T ) ∼ D G , 1 T κ − 1 . (286)
The equality of the coefficient with the live-weight residue follows from
π ( h κ , 1 ) = B ( κ / 2 , 1 / 2 ) π ( h κ ) . \pi(h_{\kappa,1})=B(\kappa/2,1/2)\pi(h_\kappa). π ( h κ , 1 ) = B ( κ /2 , 1/2 ) π ( h κ ) .
The exact instantaneous ensemble pressure and the period-averaged live trace therefore have the same leading asymptotic. The remnant and pre-entrance contributions are O ( T − 1 ) O(T^{-1}) O ( T − 1 ) and remain subleading.
Energy and summed-rest-mass asymptotics
The detached strings carry an energy that grows with time, while their summed invariant mass approaches a plateau. The difference reflects the increasing boosts of newly emitted strings. We derive the exact energy sum rule and then calculate the plateau, its leading deficit, and its relation to the surviving trace. The mass calculation is a prediction of the binary process with the source moments specified below. Applying its leading deficit to the complete geometric population additionally requires the pre-entrance mass condition (D.25) and matching of the mass-loss observable. The comparison with geometric simulations is given in App. § 20 .§ 4 .
The energy normalization follows directly from conservation. A first-generation yo-yo emitted at time t 0 t_0 t 0 has
⟨ E 1 ∣ t 0 ⟩ = μ t 0 , (287) \langle E_1 \mid t_0\rangle= \mu t_0, \tag*{(287)} ⟨ E 1 ∣ t 0 ⟩ = μ t 0 , (287)
while the persistent remnant emits first-generation yo-yos at the rate N ˙ 1 ( t 0 ) = 2 / t 0 + o ( t 0 − 1 ) \dot{N}_1(t_0)=2/t_0+o(t_0^{-1}) N ˙ 1 ( t 0 ) = 2/ t 0 + o ( t 0 − 1 ) . Subsequent splittings only redistribute this energy. Consequently,
d d T ⟨ E y o y o ( T ) ⟩ = μ T N ˙ 1 ( T ) , (288) \frac{\mathrm{d}}{\mathrm{d}T}\langle E_{\mathrm{yoyo}}(T)\rangle=\mu T\dot{N}_1(T), \tag*{(288)} d T d ⟨ E yoyo ( T )⟩ = μ T N ˙ 1 ( T ) , (288)
and the exact source measure of App. § 12 gives
⟨ E y o y o ( T ) ⟩ = 2 μ T − 2 μ Γ s p l i t T + O ( T − 3 ) . (289) \langle E_{\mathrm{yoyo}}(T)\rangle=2\mu T-\frac{2\mu}{\Gamma_{\mathrm{split}}T}+\mathcal{O}(T^{-3}). \tag*{(289)} ⟨ E yoyo ( T )⟩ = 2 μ T − Γ split T 2 μ + O ( T − 3 ) . (289)
The energy therefore approaches 2 μ T 2\mu T 2 μ T with relative correction of order ( Γ s p l i t T ) − 2 (\sqrt{\Gamma_{\mathrm{split}}T})^{-2} ( Γ split T ) − 2 . Its leading value includes the moving high-boost source as well as the finite-boost population.
The invariant-mass plateau. For the entrance population, assume U 0 , V 0 > 0 U_0,V_0>0 U 0 , V 0 > 0 almost everywhere and the mass-weighted bound
∫ d ν s [ 1 + U 0 + V 0 + ( − ln U 0 ) + + ( − ln V 0 ) + ] < ∞ , s = c m 0 t ∗ . (290) \int\mathrm{d}\nu\,s\left[1+U_0+V_0+(-\ln U_0)_++(-\ln V_0)_+\right]<\infty,\qquad s=cm_0t_*. \tag*{(290)} ∫ d ν s [ 1 + U 0 + V 0 + ( − ln U 0 ) + + ( − ln V 0 ) + ] < ∞ , s = c m 0 t ∗ . (290)
Here ( z ) + = max ( z , 0 ) (z)_+=\max(z,0) ( z ) + = max ( z , 0 ) . Until the final extension below, M i n v M_{\mathrm{inv}} M inv denotes the mass of this entrance population and its descendants. For a fixed node history, vary the common boost of the source and ask which copies are alive on the observation slice. Its mass contribution is m m m times the total length of that range of source rapidities. With c = Γ s p l i t / ( 2 μ ) c=\Gamma_{\mathrm{split}}/(2\mu) c = Γ split / ( 2 μ ) , the rescaled birth and split radii are r b = U V r_b=\sqrt{UV} r b = U V , r s = ( U + ϵ ) ( V + ϵ ) r_s=\sqrt{(U+\epsilon)(V+\epsilon)} r s = ( U + ϵ ) ( V + ϵ ) . An event of rescaled radius r r r lies before time t t t on a rapidity interval of length 2 a + ( c m t / r ) 2a_+(cmt/r) 2 a + ( c m t / r ) , where a + ( z ) = Θ ( z − 1 ) arcosh z a_+(z)=\Theta(z-1)\operatorname{arcosh}z a + ( z ) = Θ ( z − 1 ) arcosh z . Subtracting copies that have already split from those already born gives
M n o d e ( t ) = 2 m [ a + ( c m t r b ) − a + ( c m t r s ) ] . (291) M_{\mathrm{node}}(t)=2m\left[a_+\left(\frac{cmt}{r_b}\right)-a_+\left(\frac{cmt}{r_s}\right)\right]. \tag*{(291)} M node ( t ) = 2 m [ a + ( r b c m t ) − a + ( r s c m t ) ] . (291)
Although each individual copy eventually splits, increasingly boosted copies are delayed by time dilation. Their net rapidity length approaches a constant. For each node,
M n o d e ( t ) ⟶ 2 m log ( r s / r b ) , M_{\mathrm{node}}(t)\longrightarrow2m\log(r_s/r_b), M node ( t ) ⟶ 2 m log ( r s / r b ) ,
0 ≤ M n o d e ( t ) ≤ 2 m log ( 2 r s / r b ) . (292) 0\leq M_{\mathrm{node}}(t)\leq2m\log(2r_s/r_b). \tag*{(292)} 0 ≤ M node ( t ) ≤ 2 m log ( 2 r s / r b ) . (292)
It remains to establish that the infinitely many nodes have a finite total contribution. The expected sum of daughter invariant masses is only M ( 1 ) = 8 / 9 M(1) = 8/9 M ( 1 ) = 8/9 of the parent mass, so successive generations carry a geometrically decreasing mass weight. We must also bound the logarithmic occupation factor. The normalized operator T 1 / M ( 1 ) \mathcal{T}_1/M(1) T 1 / M ( 1 ) has independent Beta( 3 / 2 , 1 ) (3/2,1) ( 3/2 , 1 ) multipliers and unit exponential waiting times. The coordinate means are uniformly bounded along this chain. Its negative logarithmic moments are also uniformly bounded after the first step, since U ′ = a ( U + ϵ ) ≥ a ϵ U' = a(U + \epsilon) \ge a\epsilon U ′ = a ( U + ϵ ) ≥ a ϵ , and similarly for V ′ V' V ′ . Finally, log ( 1 + ϵ / U ) ≤ log ( 1 + ϵ ) + ( − log U ) + \log(1 + \epsilon/U) \le\log(1 + \epsilon) + (-\log U)_+ log ( 1 + ϵ / U ) ≤ log ( 1 + ϵ ) + ( − log U ) + . Thus the expected logarithmic factor in Eq. (D.6) is uniformly bounded by an integrable function of the root marks. The remaining generation sum is geometric, ∑ n ( 8 / 9 ) n \sum_n(8/9)^n ∑ n ( 8/9 ) n ; the root integral is finite by Eq. (D.4). Dominated convergence yields
M i n v ( t ) ⟶ M ∞ = ∫ d ν m 0 ∑ n ≥ 0 ( J 1 n ℓ ) ( Z 0 ) , ℓ ( U , V ) = E ϵ [ log ( U + ϵ ) ( V + ϵ ) U V ] (293) M_{\mathrm{inv}}(t) \longrightarrow M_\infty= \int d\nu\,m_0 \sum_{n\ge0}(\mathcal{J}_1^n\ell)(Z_0), \qquad\ell(U,V) = \mathbb{E}_{\epsilon}\left[\log\frac{(U+\epsilon)(V+\epsilon)}{UV}\right] \tag*{(293)} M inv ( t ) ⟶ M ∞ = ∫ d ν m 0 n ≥ 0 ∑ ( J 1 n ℓ ) ( Z 0 ) , ℓ ( U , V ) = E ϵ [ log U V ( U + ϵ ) ( V + ϵ ) ] (293)
with 0 < M ∞ < ∞ 0 < M_\infty< \infty 0 < M ∞ < ∞ . This proves existence of the plateau in the unabsorbed binary model. The following radial mass balance determines its leading approach.
The leading mass deficit and its relation to the trace
The plateau measures the mass integrated over complete boost orbits. Its approach is controlled by the mass still awaiting fragmentation at large invariant radius. We express that remaining mass directly in terms of the chronological branching process. The resulting boost integral determines both the correction exponent and the sign of the deficit. Put T = t ^ T = \hat{t} T = t ^ , κ = κ I F S \kappa= \kappa_{\mathrm{IFS}} κ = κ IFS , q = 1 − κ q = 1 - \kappa q = 1 − κ , m ˉ = c t ∗ m \bar{m} = c_{t_*}m m ˉ = c t ∗ m , and M ( T ) = c t ∗ M i n v ( t ∗ T ) \mathcal{M}(T) = c_{t_*}M_{\mathrm{inv}}(t_*T) M ( T ) = c t ∗ M inv ( t ∗ T ) . Here c t ∗ = Γ s p l i t / ( 2 μ ) c_{t_*} = \sqrt{\Gamma_{\mathrm{split}}/(2\mu)} c t ∗ = Γ split / ( 2 μ ) , as in the trace normalization.
Mass remaining at invariant radius. A cut at fixed invariant radius intersects the live nodes of the tree. Denote their expected remaining invariant mass, in the dimensionless units above and per unit source rapidity, by F ( r ) \mathcal{F}(r) F ( r ) . Its balance has two terms: root births add mass, and fragmentation reduces the sum of descendant invariant masses. Let I ( d r ) \mathcal{I}(dr) I ( d r ) be the root-mass injection measure and D ( d r ) \mathcal{D}(dr) D ( d r ) the expected invariant-mass loss at descendant splits, per unit source rapidity. For a general entrance root, its dimensionless birth radius is r 0 = U 0 V 0 / s r_0 = \sqrt{U_0V_0}/s r 0 = U 0 V 0 / s . The measures are
I ( d r ) = ∫ d ν s δ r 0 ( d r ) , (294) \mathcal{I}(dr) = \int d\nu\,s\,\delta_{r_0}(dr), \tag*{(294)} I ( d r ) = ∫ d ν s δ r 0 ( d r ) , (294)
D ( d r ) = d m ∫ d ν E [ ∑ Y m ˉ Y δ r Y ( d r ) ] , d m = 1 − M ( 1 ) = 1 9 . (295) \mathcal{D}(dr) = d_m\int d\nu\,\mathbb{E}\left[\sum_Y \bar{m}_Y\delta_{r_Y}(dr)\right], \qquad d_m = 1 - M(1) = \frac{1}{9}. \tag*{(295)} D ( d r ) = d m ∫ d ν E [ Y ∑ m ˉ Y δ r Y ( d r ) ] , d m = 1 − M ( 1 ) = 9 1 . (295)
In addition to Eq. (D.4), assume, for σ \sigma σ near κ \kappa κ and some ε > 0 \varepsilon> 0 ε > 0 ,
∫ d ν s σ ( 1 + U 0 + V 0 ) 1 + δ < ∞ , G ( ( r , ∞ ) ) = O ( r − q − ε ) ( r → ∞ ) . (296) \int d\nu\,s^\sigma(1+U_0+V_0)^{1+\delta}<\infty,\qquad\mathcal{G}((r,\infty))=O(r^{-q-\varepsilon})\qquad(r\to\infty). \tag*{(296)} ∫ d ν s σ ( 1 + U 0 + V 0 ) 1 + δ < ∞ , G (( r , ∞ )) = O ( r − q − ε ) ( r → ∞ ) . (296)
These bounds control mass-weighted occupations and delayed injection; they are additional to the assumptions for the normalized trace profile. The sum includes the root’s split and all subsequent descendant splits; r Y = ρ Y / t ∗ r_Y=\rho_Y/t_\ast r Y = ρ Y / t ∗ is the event’s invariant radius. The mean loss at a split is d m m ‾ Y d_m\overline{m}_Y d m m Y , since daughter fractions are independent of the parent’s previous history and final waiting time. The geometric generation sum gives D ( ( 0 , ∞ ) ) = d m ∫ d ν m ‾ 0 ∑ n ≥ 0 M ( 1 ) n = G ( ( 0 , ∞ ) ) \mathcal{D}((0,\infty))=d_m\int d\nu\,\overline{m}_0\sum_{n\geq0}M(1)^n=\mathcal{G}((0,\infty)) D (( 0 , ∞ )) = d m ∫ d ν m 0 ∑ n ≥ 0 M ( 1 ) n = G (( 0 , ∞ )) . Thus the mass on the cut is the injected mass minus the loss already incurred, or equivalently the future loss minus future injection:
F ( r ) = G ( ( 0 , r ] ) − D ( ( 0 , r ] ) = D ( ( r , ∞ ) ) − G ( ( r , ∞ ) ) ≥ 0. (297) \mathcal{F}(r)=\mathcal{G}((0,r])-\mathcal{D}((0,r])=\mathcal{D}((r,\infty))-\mathcal{G}((r,\infty))\geq0. \tag*{(297)} F ( r ) = G (( 0 , r ]) − D (( 0 , r ]) = D (( r , ∞ )) − G (( r , ∞ )) ≥ 0. (297)
Positivity follows because F ( r ) \mathcal{F}(r) F ( r ) counts the mass of live strings on the cut.
An event at radius r r r has occurred by time T T T on a rapidity interval of length 2 arcosh ( T / r ) 2\operatorname{arcosh}(T/r) 2 arcosh ( T / r ) . Grouping each parent’s removal with its daughters’ births in the occupation sum, and then integrating by parts, yields the exact identity
M ( T ) = 2 ∫ 0 T arcosh ( T / r ) [ G − D ] ( d r ) = 2 ∫ 0 T T F ( r ) r T 2 − r 2 d r . (298) \mathcal{M}(T)=2\int_0^T\operatorname{arcosh}(T/r)[\mathcal{G}-\mathcal{D}](dr)=2\int_0^T\frac{T\mathcal{F}(r)}{r\sqrt{T^2-r^2}}\,dr. \tag*{(298)} M ( T ) = 2 ∫ 0 T arcosh ( T / r ) [ G − D ] ( d r ) = 2 ∫ 0 T r T 2 − r 2 T F ( r ) d r . (298)
Thus the remaining radial mass determines the time dependence through the boost projection.
The branching pole fixes the radial tail. For birth coordinates Z = ( U , V ) Z=(U,V) Z = ( U , V ) and an independent unit exponential waiting time ϵ \epsilon ϵ , write R = ( U + ϵ ) ( V + ϵ ) R=\sqrt{(U+\epsilon)(V+\epsilon)} R = ( U + ϵ ) ( V + ϵ ) , so that r = R / m ‾ r=R/\overline{m} r = R / m . The transform of the positive loss measure is
H M ( ζ ) ≡ ∫ 0 ∞ r 1 − ζ D ( d r ) = d m ∫ d ν m ‾ 0 ζ ∑ n ≥ 0 ( T ζ n k ζ ) ( Z 0 ) , k ζ ( Z ) = E ϵ [ R 1 − ζ ] . (299) H_M(\zeta)\equiv\int_0^\infty r^{1-\zeta}\mathcal{D}(dr)=d_m\int d\nu\,\overline{m}_0^\zeta\sum_{n\geq0}(\mathcal{T}_\zeta^n k_\zeta)(Z_0),\qquad k_\zeta(Z)=\mathbb{E}_{\epsilon}[R^{1-\zeta}]. \tag*{(299)} H M ( ζ ) ≡ ∫ 0 ∞ r 1 − ζ D ( d r ) = d m ∫ d ν m 0 ζ n ≥ 0 ∑ ( T ζ n k ζ ) ( Z 0 ) , k ζ ( Z ) = E ϵ [ R 1 − ζ ] . (299)
The identity m ‾ r 1 − ζ = m ‾ ζ R 1 − ζ \overline{m}r^{1-\zeta}=\overline{m}^{\zeta}R^{1-\zeta} m r 1 − ζ = m ζ R 1 − ζ gives this representation in its convergence strip just to the right of κ \kappa κ . We apply Eq. (C.15) with b ζ = k ζ b_\zeta=k_\zeta b ζ = k ζ . Unlike the pressure kernel, this positive moment grows with the birth coordinates, so its contraction requires the following separate check.
To extend the contraction argument to this observable, let σ = Re ζ \sigma=\operatorname{Re}\zeta σ = Re ζ and choose 0 < δ < ( 1 − σ ) / 2 0<\delta<(1-\sigma)/2 0 < δ < ( 1 − σ ) /2 in a narrow strip about κ \kappa κ . The power in k ζ k_\zeta k ζ obeys the weighted Hölder bound
∣ k ζ ( Z ) − k ζ ( Z ′ ) ∣ ≤ C ζ d δ ( Z , Z ′ ) ( 1 + U + V + U ′ + V ′ ) . (300) |k_\zeta(Z)-k_\zeta(Z')|\leq C_\zeta d_\delta(Z,Z')(1+U+V+U'+V'). \tag*{(300)} ∣ k ζ ( Z ) − k ζ ( Z ′ ) ∣ ≤ C ζ d δ ( Z , Z ′ ) ( 1 + U + V + U ′ + V ′ ) . (300)
where δ σ \delta_{\sigma} δ σ is defined in App. § 13 .§ 2 . Under T σ / M ( σ ) \mathcal{T}_{\sigma}/M(\sigma) T σ / M ( σ ) , the coordinate multipliers are independent Beta( 1 + σ / 2 , 1 ) (1+\sigma/2,1) ( 1 + σ /2 , 1 ) variables. Their contracting products and the exponential waiting times give uniformly bounded coordinate second moments, with polynomial dependence on the initial coordinates. Coupling to the chain started at zero and applying Cauchy–Schwarz therefore gives
∣ T ζ n κ ζ ( Z ) − T ζ n κ ζ ( 0 ) ∣ ≤ C θ δ ( σ ) n ( 1 + U + V ) 1 + δ , (301) \left|\mathcal{T}_{\zeta}^{n} \kappa_{\zeta}(Z)-\mathcal{T}_{\zeta}^{n} \kappa_{\zeta}(0)\right| \le C\theta_{\delta}(\sigma)^{n}(1+U+V)^{1+\delta}, \tag*{(301)} T ζ n κ ζ ( Z ) − T ζ n κ ζ ( 0 ) ≤ C θ δ ( σ ) n ( 1 + U + V ) 1 + δ , (301)
θ δ ( σ ) = M ( σ ) 1 + σ / 2 1 + σ / 2 + 2 δ < 1. (302) \theta_{\delta}(\sigma)=M(\sigma)\sqrt{\frac{1+\sigma/2}{1+\sigma/2+2\delta}}<1. \tag*{(302)} θ δ ( σ ) = M ( σ ) 1 + σ /2 + 2 δ 1 + σ /2 < 1. (302)
The strict inequality holds throughout a sufficiently narrow strip. Equation (D.10) supplies the required weighted polynomial moments of the entrance source. The first-step moments also bound d n d_n d n , so the hypotheses of Eq. (C.15) hold. Its pole has residue
A M ≡ Res ζ = κ H M ( ζ ) = d m S ( κ ) E π , ϵ [ R q ] − M ′ ( κ ) > 0. (303) A_M\equiv\underset{\zeta=\kappa}{\operatorname{Res}}\,H_M(\zeta)=\frac{d_m\mathcal{S}(\kappa)\mathbb{E}_{\pi,\epsilon}[R^q]}{-M'(\kappa)}>0. \tag*{(303)} A M ≡ ζ = κ Res H M ( ζ ) = − M ′ ( κ ) d m S ( κ ) E π , ϵ [ R q ] > 0. (303)
Here π \pi π is the same stationary birth-coordinate law as in the pressure calculation. The expectation includes the final waiting time.
Apply the positive-measure Wiener–Ikehara theorem in the form used in App. § 13 .§ 3 to S D ( R ) = ∫ ( 1 , R ] r D ( d r ) S_D(R)=\int_{(1,R]}r\mathcal{D}(dr) S D ( R ) = ∫ ( 1 , R ] r D ( d r ) . Removing r < 1 r<1 r < 1 changes the transform by an analytic function near the Malthusian line, since D \mathcal{D} D is finite. The theorem gives S D ( R ) ∼ ( A M / κ ) R κ S_D(R)\sim(A_M/\kappa)R^\kappa S D ( R ) ∼ ( A M / κ ) R κ . Stieltjes integration by parts then gives D ( ( r , ∞ ) ) ∼ ( A M / q ) r − q \mathcal{D}((r,\infty))\sim(A_M/q)r^{-q} D (( r , ∞ )) ∼ ( A M / q ) r − q ; the injection tail is o ( r − q ) o(r^{-q}) o ( r − q ) by Eq. (D.10), so
F ( r ) ∼ K M r − q , K M = A M / q > 0. (304) \mathcal{F}(r)\sim K_Mr^{-q},\qquad K_M=A_M/q>0. \tag*{(304)} F ( r ) ∼ K M r − q , K M = A M / q > 0. (304)
The boost integral fixes the sign of the deficit. Using the plateau and Eq. (D.12),
M ∞ − M ( T ) = 2 ∫ T ∞ F ( r ) r d r (305) \mathcal{M}_{\infty}-\mathcal{M}(T)=2\int_T^{\infty}\frac{\mathcal{F}(r)}{r}\,dr \tag*{(305)} M ∞ − M ( T ) = 2 ∫ T ∞ r F ( r ) d r (305)
M ∞ − M ( T ) + 2 ∫ 0 T F ( r ) r ( 1 − T T 2 − r 2 ) d r . (306) \mathrel{\phantom{\mathcal{M}_{\infty}-\mathcal{M}(T)}}+2\int_0^T\frac{\mathcal{F}(r)}{r}\left(1-\frac{T}{\sqrt{T^2-r^2}}\right)\,dr. \tag*{(306)} M ∞ − M ( T ) + 2 ∫ 0 T r F ( r ) ( 1 − T 2 − r 2 T ) d r . (306)
The first term is positive: radii beyond T T T have not yet appeared anywhere on the time slice, so their plateau contribution is missing. The second is negative: radii just inside the slice receive an enhanced rapidity weight from the boost Jacobian. The deficit is the balance of these two geometric effects; discarding either would give the wrong coefficient. Substituting the regularly varying tail (D.18) into the combined expression yields
M ∞ − M ( T ) ∼ K M T − q [ 2 q + 2 ∫ 0 1 u − q − 1 ( 1 − 1 1 − u 2 ) d u ] (307) \mathcal{M}_{\infty}-\mathcal{M}(T)\sim K_MT^{-q}\left[\frac{2}{q}+2\int_0^1u^{-q-1}\left(1-\frac{1}{\sqrt{1-u^2}}\right)\,du\right] \tag*{(307)} M ∞ − M ( T ) ∼ K M T − q [ q 2 + 2 ∫ 0 1 u − q − 1 ( 1 − 1 − u 2 1 ) d u ] (307)
= − K M B ( − q / 2 , 1 / 2 ) T − q . (308) =-K_MB(-q/2,1/2)T^{-q}. \tag*{(308)} = − K M B ( − q /2 , 1/2 ) T − q . (308)
The subtracted kernel is O ( u 2 ) O(u^2) O ( u 2 ) at zero and has an integrable singularity at one; the bounded-radius contribution is O ( T − 2 ) O(T^{-2}) O ( T − 2 ) . These facts justify the substitution. Since 0 < q < 1 0 < q < 1 0 < q < 1 , − B ( − q / 2 , 1 / 2 ) > 0 -\mathrm{B}(-q/2,1/2) > 0 − B ( − q /2 , 1/2 ) > 0 . In physical mass units,
M i n v ( t ) = M ∞ − B t ^ − q + o ( t ^ − q ) , B = − K M B ( − q / 2 , 1 / 2 ) c t ∗ > 0. (309) M_{\mathrm{inv}}(t) = M_{\infty} - B\hat{t}^{-q} + o(\hat{t}^{-q}), \qquad B = \frac{-K_M\mathrm{B}(-q/2,1/2)}{ct_*} > 0. \tag*{(309)} M inv ( t ) = M ∞ − B t ^ − q + o ( t ^ − q ) , B = c t ∗ − K M B ( − q /2 , 1/2 ) > 0. (309)
The positive coefficient fixes the leading approach from below.
A common coefficient relation. The mass left to lose and the surviving tension probe the same late generations, with different occupation weights. Their residues contain different moments of the stationary clock, but the same source strength and branching factor. The pressure residue in Eq. (C.20) contains the same S ( κ ) \mathcal{S}(\kappa) S ( κ ) and branching derivative; dividing the two amplitudes therefore cancels both:
a κ ≡ lim t → ∞ M ∞ − M i n v ( t ) W l i v e ( t ) = − B ( − q / 2 , 1 / 2 ) 9 q B ( κ / 2 , 1 / 2 ) E π , ϵ [ R q ] E π , ϵ [ R − κ ] ≃ 3.08. (310) a_{\kappa} \equiv\lim_{t\to\infty} \frac{M_{\infty}-M_{\mathrm{inv}}(t)}{W_{\mathrm{live}}(t)} = \frac{-\mathrm{B}(-q/2,1/2)}{9q\,\mathrm{B}(\kappa/2,1/2)}\frac{\mathbb{E}_{\pi,\epsilon}[R^q]}{\mathbb{E}_{\pi,\epsilon}[R^{-\kappa}]} \simeq3.08. \tag*{(310)} a κ ≡ t → ∞ lim W live ( t ) M ∞ − M inv ( t ) = 9 q B ( κ /2 , 1/2 ) − B ( − q /2 , 1/2 ) E π , ϵ [ R − κ ] E π , ϵ [ R q ] ≃ 3.08. (310)
The value a κ ≃ 3.08 a_{\kappa} \simeq3.08 a κ ≃ 3.08 follows from the stationary binary recursion, independently of the timeseries fit. Both clock moments are finite; the source factor cancels for entrance laws obeying the stated bounds. Extending this coefficient to the geometric cascade requires matching the growing mass-loss weight in the comparison of App. § 13 .§ 5 .
For the complete detached population, suppose also that the mass still in pre-entrance stages satisfies
M p r e ( t ∗ T ) = M p r e , ∞ + o ( T − q ) . (311) M_{\mathrm{pre}}(t_*T) = M_{\mathrm{pre},\infty} + o(T^{-q}). \tag*{(311)} M pre ( t ∗ T ) = M pre , ∞ + o ( T − q ) . (311)
Its plateau then adds to M ∞ M_{\infty} M ∞ , leaving the leading deficit and amplitude ratio unchanged once the geometric mass-loss coefficient is matched. This is a separate condition on mass: a highly boosted string carries mass γ W \gamma W γ W even when its trace W W W is small.
Stress-tensor moment diagnostics
Equal-time moments diagnose the spatial distribution of the surviving tension and the location of the negative energy relative to the compensating positive energy. Boost invariance relates the trace moments to the total pressure, and conservation then fixes the energy and pressure moments. Their powers and coefficient ratios test the same physical picture as the null integral, using different measurements of the cascade.
The semiclassical spatial distribution
Define the ensemble-averaged complete stress moments by
M n ( p ) ( t ) = ∫ d x x n ⟨ T 11 ( t , x ) ⟩ , (312) M^{(p)}_n(t) = \int dx\,x^n\langle T_{11}(t,x)\rangle, \tag*{(312)} M n ( p ) ( t ) = ∫ d x x n ⟨ T 11 ( t , x )⟩ , (312)
M n ( E ) ( t ) = ∫ d x x n ⟨ T 00 ( t , x ) ⟩ , (313) M^{(E)}_n(t) = \int dx\,x^n\langle T_{00}(t,x)\rangle, \tag*{(313)} M n ( E ) ( t ) = ∫ d x x n ⟨ T 00 ( t , x )⟩ , (313)
M n ( W ) ( t ) = M n ( E ) ( t ) − M n ( p ) ( t ) . (314) M^{(W)}_n(t) = M^{(E)}_n(t) - M^{(p)}_n(t). \tag*{(314)} M n ( W ) ( t ) = M n ( E ) ( t ) − M n ( p ) ( t ) . (314)
Here x x x is the position on a constant-time slice, and M 0 ( W ) M^{(W)}_0 M 0 ( W ) includes the remnant and all live descendants.
The trace defines a positive spatial measure. Its normalized moments describe localization. The complete energy density is signed, so its moments diagnose cancellation and cannot be read as moments of a probability distribution for size.
For the complete IFS decay state, the total energy and the first energy moment vanish,
M 0 ( E ) ( t ) = 0 , M 1 ( E ) ( t ) = 0. (315) M^{(E)}_0(t) = 0,\qquad M^{(E)}_1(t) = 0. \tag*{(315)} M 0 ( E ) ( t ) = 0 , M 1 ( E ) ( t ) = 0. (315)
The first identity is total energy conservation. The second is conservation of the center-of-energy moment, with the origin chosen at the center of the initial IFS and total momentum zero. More generally, stress-tensor conservation implies
d 2 d t 2 M n ( E ) ( t ) = n ( n − 1 ) M n − 2 ( p ) ( t ) , (316) \frac{d^2}{dt^2}M^{(E)}_n(t) = n(n-1)M^{(p)}_{n-2}(t), \tag*{(316)} d t 2 d 2 M n ( E ) ( t ) = n ( n − 1 ) M n − 2 ( p ) ( t ) , (316)
assuming that the boundary terms vanish. In particular, M ¨ 2 ( E ) = 2 p t o t \ddot{M}^{(E)}_2 = 2p_{\mathrm{tot}} M ¨ 2 ( E ) = 2 p tot . The pressure theorem fixes M 0 ( W ) = − p t o t M^{(W)}_0 = -p_{\mathrm{tot}} M 0 ( W ) = − p tot ; boost invariance then determines the higher trace moments, and conservation determines their energy and pressure counterparts.
Boost invariance and trace-moment coefficients. Boost invariance fixes the relative coefficients as well as the powers. For example, M 2 ( W ) / ( t 2 M 0 ( W ) ) M^{(W)}_2/(t^2M^{(W)}_0) M 2 ( W ) / ( t 2 M 0 ( W ) ) measures the mean of ( x / t ) 2 (x/t)^2 ( x / t ) 2 in the positive trace distribution. Its semiclassical limit 1 / ( 2 − q I F S ) ≃ 0.547 1/(2-q_{\mathrm{IFS}}) \simeq0.547 1/ ( 2 − q IFS ) ≃ 0.547 resolves a profile spread across the wedge. The terminal limit of one derived below means that the surviving weight is instead concentrated near x = ± t x=\pm t x = ± t .
All moments in the following identities are ensemble averages, with brackets suppressed. In the boost-invariant flat-space model, write ⟨ T u v ( u , v ) ⟩ = f ( u v ) \langle T_{uv}(u,v)\rangle= f(uv) ⟨ T uv ( u , v )⟩ = f ( uv ) , so the complete traceful density inside the future wedge is ⟨ T 00 − T 11 ⟩ = 4 f ( t 2 − x 2 ) \langle T_{00}-T_{11}\rangle= 4f(t^2-x^2) ⟨ T 00 − T 11 ⟩ = 4 f ( t 2 − x 2 ) . The original null folds and traceless terminal products do not contribute to this density. Changing variables to z = t 2 − x 2 z=t^2-x^2 z = t 2 − x 2 gives
M 2 k ( W ) ( t ) = 4 ∫ 0 t 2 ( t 2 − z ) k − 1 / 2 f ( z ) d z , k ≥ 0. (317) M^{(W)}_{2k}(t) = 4\int_0^{t^2}(t^2-z)^{k-1/2}f(z)\,dz,\qquad k\geq0. \tag*{(317)} M 2 k ( W ) ( t ) = 4 ∫ 0 t 2 ( t 2 − z ) k − 1/2 f ( z ) d z , k ≥ 0. (317)
For k ≥ 1 k \ge1 k ≥ 1 , the upper-endpoint term vanishes, and differentiation gives the exact relation
d d t M 2 k ( W ) ( t ) = ( 2 k − 1 ) t M 2 k − 2 ( W ) ( t ) . (318) \frac{d}{dt}M^{(W)}_{2k}(t)=(2k-1)tM^{(W)}_{2k-2}(t). \tag*{(318)} d t d M 2 k ( W ) ( t ) = ( 2 k − 1 ) t M 2 k − 2 ( W ) ( t ) . (318)
Local boundedness of the trace profile makes the derivative integrable.
Define the dimensionless semiclassical coefficients by
M 2 k ( W ) ( t ) ∼ C 2 k ( W ) μ Γ s p l i t − k − 1 / 2 ( Γ s p l i t t ) 2 k − q I F S . (319) M^{(W)}_{2k}(t)\sim C^{(W)}_{2k}\,\mu\Gamma_{\mathrm{split}}^{-k-1/2}\left(\sqrt{\Gamma_{\mathrm{split}}}t\right)^{2k-q_{\mathrm{IFS}}}. \tag*{(319)} M 2 k ( W ) ( t ) ∼ C 2 k ( W ) μ Γ split − k − 1/2 ( Γ split t ) 2 k − q IFS . (319)
Starting from the established zeroth-moment tail and integrating Eq. (E.7) successively yields
C 2 k ( W ) = 2 k − 1 2 k − q I F S C 2 k − 2 ( W ) , k ≥ 1. (320) C^{(W)}_{2k}=\frac{2k-1}{2k-q_{\mathrm{IFS}}}C^{(W)}_{2k-2},\qquad k\ge1. \tag*{(320)} C 2 k ( W ) = 2 k − q IFS 2 k − 1 C 2 k − 2 ( W ) , k ≥ 1. (320)
Finite-time integration constants are subleading. The zeroth coefficient is C 0 ( W ) = C P C^{(W)}_0=C_P C 0 ( W ) = C P ; the zeroth- and second-moment predictions are collected in Eq. (E.14).
Energy and pressure coefficients. The same inputs fix the leading coefficients of all even energy and pressure moments. Define their signed coefficients using the normalization of Eq. (E.8),
M 2 k ( A ) ( t ) ∼ C 2 k ( A ) μ Γ s p l i t − k − 1 / 2 ( Γ s p l i t t ) 2 k − q I F S , A = E , p , k ≥ 1. (321) M^{(A)}_{2k}(t)\sim C^{(A)}_{2k}\,\mu\Gamma_{\mathrm{split}}^{-k-1/2}\left(\sqrt{\Gamma_{\mathrm{split}}}t\right)^{2k-q_{\mathrm{IFS}}},\qquad A=E,p,\quad k\ge1. \tag*{(321)} M 2 k ( A ) ( t ) ∼ C 2 k ( A ) μ Γ split − k − 1/2 ( Γ split t ) 2 k − q IFS , A = E , p , k ≥ 1. (321)
Combining Eqs. (E.5) and (E.3) gives the exact hierarchy
d 2 d t 2 M 2 k ( E ) = 2 k ( 2 k − 1 ) ( M 2 k − 2 ( E ) − M 2 k − 2 ( W ) ) . (322) \frac{d^2}{dt^2}M^{(E)}_{2k}=2k(2k-1)\left(M^{(E)}_{2k-2}-M^{(W)}_{2k-2}\right). \tag*{(322)} d t 2 d 2 M 2 k ( E ) = 2 k ( 2 k − 1 ) ( M 2 k − 2 ( E ) − M 2 k − 2 ( W ) ) . (322)
Starting from M 0 ( E ) = 0 M^{(E)}_0=0 M 0 ( E ) = 0 , integrate this hierarchy twice at each step and use Eq. (E.9). The leading power t 2 k − q I F S t^{2k-q_{\mathrm{IFS}}} t 2 k − q IFS dominates the constant and linear integration terms because 0 < q I F S < 1 0<q_{\mathrm{IFS}}<1 0 < q IFS < 1 . Induction gives
C 2 k ( E ) = − 2 k 1 − q I F S C 2 k ( W ) = − k ( 1 + 2 ) C 2 k ( W ) , C 2 k ( p ) = − 2 k + 1 − q I F S 1 − q I F S C 2 k ( W ) . (323) \begin{aligned}
C^{(E)}_{2k}&=-\frac{2k}{1-q_{\mathrm{IFS}}}C^{(W)}_{2k}=-k(1+\sqrt{2})C^{(W)}_{2k},\\
C^{(p)}_{2k}&=-\frac{2k+1-q_{\mathrm{IFS}}}{1-q_{\mathrm{IFS}}}C^{(W)}_{2k}. \tag*{(323)}
\end{aligned} C 2 k ( E ) C 2 k ( p ) = − 1 − q IFS 2 k C 2 k ( W ) = − k ( 1 + 2 ) C 2 k ( W ) , = − 1 − q IFS 2 k + 1 − q IFS C 2 k ( W ) . (323)
Thus C 0 ( W ) = C P C^{(W)}_0=C_P C 0 ( W ) = C P fixes every leading even-moment coefficient. The signed moments include the original negative-energy folds, and conservation closes their hierarchy. Odd moments vanish by ensemble symmetry. The numerical comparison uses orders zero and two, with μ = Γ s p l i t = 1 \mu=\Gamma_{\mathrm{split}}=1 μ = Γ split = 1 .
For the numerical test, set q = q I F S q=q_{\mathrm{IFS}} q = q IFS and remove the predicted time powers:
X 0 = ( Γ s p l i t t ) q M 0 ( W ) μ Γ s p l i t − 1 / 2 , X 2 ( A ) = ( Γ s p l i t t ) q − 2 M 2 ( A ) μ Γ s p l i t − 3 / 2 , A = W , E . (324) X_0=\frac{\left(\sqrt{\Gamma_{\mathrm{split}}}t\right)^qM^{(W)}_0}{\mu\Gamma_{\mathrm{split}}^{-1/2}},\qquad
X^{(A)}_2=\frac{\left(\sqrt{\Gamma_{\mathrm{split}}}t\right)^{q-2}M^{(A)}_2}{\mu\Gamma_{\mathrm{split}}^{-3/2}},\qquad A=W,E. \tag*{(324)} X 0 = μ Γ split − 1/2 ( Γ split t ) q M 0 ( W ) , X 2 ( A ) = μ Γ split − 3/2 ( Γ split t ) q − 2 M 2 ( A ) , A = W , E . (324)
The three fitted observables have the limits
( X 0 , X 2 ( W ) , X 2 ( E ) ) ⟶ C P ( 1 , 1 2 − q , − 2 ( 1 − q ) ( 2 − q ) ) . (325) \left(X_0,X_2^{(\mathrm{W})},X_2^{(\mathrm{E})}\right)\longrightarrow C_{\mathrm{P}}\left(1,\frac{1}{2-q},-\frac{2}{(1-q)(2-q)}\right). \tag*{(325)} ( X 0 , X 2 ( W ) , X 2 ( E ) ) ⟶ C P ( 1 , 2 − q 1 , − ( 1 − q ) ( 2 − q ) 2 ) . (325)
Pressure supplies a conservation check but no independent coordinate, since M ( p ) = M ( E ) − M ( W ) M^{(\mathrm{p})}=M^{(\mathrm{E})}-M^{(\mathrm{W})} M ( p ) = M ( E ) − M ( W ) . Coefficient ratios are formed from ensemble means of signed moments. The higher even moments follow from the same hierarchy.
Comparison with the geometric cascade. Figure 12 compares the zeroth and second moments in 1,000 geometric histories with their predicted limits. The ratios lie close to the asymptotic values, but their correlated departures are statistically resolved. A correction proportional to 1 / t 1/t 1/ t in the total trace improves the joint fit while leaving further finite-time structure. Appendix § 20 .§ 6 gives the fits, their uncertainties, and the conservation checks.
Figure 12. Complete signed spatial moments from 1,000 unabsorbed histories. Top: rescaling each observable to the same predicted limiting amplitude; the dashed line is the fiducial fit. Bottom: departures of ratios of signed ensemble means from the predicted ratios. Error bars are marginal standard errors; ratio errors use linear error propagation with the within-history pairing retained. Shading marks times before the fiducial fitting window. These errors do not display the full covariance used in the joint tests.
Localization after terminal decay
The same exact moment identity also fixes the terminal limit. Under the finite-response assumptions of § 6 , let
M 0 ( W ) ( t ) ∼ C t e r m t , C t e r m = 4 I ∞ > 0. (326) M_0^{(\mathrm{W})}(t)\sim\frac{C_{\mathrm{term}}}{t},\qquad C_{\mathrm{term}}=4I_{\infty}>0. \tag*{(326)} M 0 ( W ) ( t ) ∼ t C term , C term = 4 I ∞ > 0. (326)
Successive integration of Eq. (E.7) gives, for every fixed k ≥ 0 k\ge0 k ≥ 0 ,
M 2 k ( W ) ( t ) ∼ C t e r m t 2 k − 1 , M 2 k ( W ) ( t ) t 2 k M 0 ( W ) ( t ) ⟶ 1. (327) M_{2k}^{(\mathrm{W})}(t)\sim C_{\mathrm{term}}t^{2k-1},\qquad\frac{M_{2k}^{(\mathrm{W})}(t)}{t^{2k}M_0^{(\mathrm{W})}(t)}\longrightarrow1. \tag*{(327)} M 2 k ( W ) ( t ) ∼ C term t 2 k − 1 , t 2 k M 0 ( W ) ( t ) M 2 k ( W ) ( t ) ⟶ 1. (327)
In the dimensionless convention of Eq. (E.8), with terminal power 2 k − 1 2k-1 2 k − 1 , this means C 2 k ( W ) , t e r m = Γ s p l i t C t e r m / μ C_{2k}^{(\mathrm{W}),\mathrm{term}}=\Gamma_{\mathrm{split}}C_{\mathrm{term}}/\mu C 2 k ( W ) , term = Γ split C term / μ for every fixed k k k . Positivity of the trace turns the second-moment limit into a physical localization statement: its normalized weight outside any fixed neighborhood of x / t = ± 1 x/t=\pm1 x / t = ± 1 vanishes. This describes localization relative to the light-cone size.
The corresponding energy and pressure moments have a logarithmic factor. Starting from M ¨ 0 ( E ) = − 2 M 0 ( W ) ∼ − 2 C t e r m / t \ddot{M}_0^{(\mathrm{E})}=-2M_0^{(\mathrm{W})}\sim-2C_{\mathrm{term}}/t M ¨ 0 ( E ) = − 2 M 0 ( W ) ∼ − 2 C term / t and integrating Eq. (E.11) successively yields, for each fixed k ≥ 1 k\ge1 k ≥ 1 ,
M 2 k ( E ) ( t ) ∼ − 2 k C t e r m t 2 k − 1 log ( t / t r e f ) , M 2 k ( p ) ( t ) ∼ − 2 k C t e r m t 2 k − 1 log ( t / t r e f ) . (328) \begin{aligned}
M_{2k}^{(\mathrm{E})}(t)&\sim-2k\,C_{\mathrm{term}}\,t^{2k-1}\log(t/t_{\mathrm{ref}}),\\
M_{2k}^{(\mathrm{p})}(t)&\sim-2k\,C_{\mathrm{term}}\,t^{2k-1}\log(t/t_{\mathrm{ref}}). \tag*{(328)}
\end{aligned} M 2 k ( E ) ( t ) M 2 k ( p ) ( t ) ∼ − 2 k C term t 2 k − 1 log ( t / t ref ) , ∼ − 2 k C term t 2 k − 1 log ( t / t ref ) . (328)
The logarithm comes from twice integrating the terminal pressure − C t e r m / t -C_{\mathrm{term}}/t − C term / t ; the hierarchy fixes the higher coefficients. The reference time t r e f > 0 t_{\mathrm{ref}}>0 t ref > 0 affects only subleading terms.
Finite smearing for a general null generator
We bound the residual after transverse smearing by separating the distance of the longitudinal cascade from the original folds from the additional spreading caused by terminal decay. At each terminal conversion, replace the outgoing products by two collinear null packets with the same total four-momentum. This reference remains in the original ( t , x ) (t,x) ( t , x ) plane and gives the decomposition
A k , L = A k , L c o l + δ A k , L t e r m . (329) \mathcal{A}_{k,L}=\mathcal{A}^{\mathrm{col}}_{k,L}+\delta\mathcal{A}^{\mathrm{term}}_{k,L}. \tag*{(329)} A k , L = A k , L col + δ A k , L term . (329)
The first-split bound controls the first spatial moment of the reference; mass conservation controls the terminal contribution. Together they give a sufficient smearing scale L s m e a r = L_{\mathrm{smear}}= L smear = Γ s p l i t − 1 / 2 ∼ ℓ s / g s \Gamma_{\mathrm{split}}^{-1/2} \sim\ell_s/g_s Γ split − 1/2 ∼ ℓ s / g s , without requiring finite integrated trace.
Setup and result
We work at fixed weak coupling in flat spacetime. The ensemble is invariant under boosts along the original string and under x ↦ − x x \mapsto-x x ↦ − x . The preterminal evolution is the collinear splitting cascade, with no joining, and only the two original folds carry negative energy. Each terminal conversion removes a positive-energy yo-yo from its tree and replaces it locally by free causal products with the same total four-momentum. Their momenta obey p j 0 > 0 p_j^0 > 0 p j 0 > 0 and p j 2 = − m j 2 ≤ 0 p_j^2 = -m_j^2 \le0 p j 2 = − m j 2 ≤ 0 ; their masses, multiplicity, and angular distribution are otherwise unrestricted. The conversion law is boost covariant. Massive free products are allowed here because the trace bound is applied to the collinear reference. Their actual trace need not vanish or have finite integrated response. The dilaton-response model of Appendix H instead assumes traceless products. Free propagation is used for the quantitative smearing estimate; the direct completion proof in Appendix F.3 needs only the conservation and positivity assumptions stated there.
Let W ∗ ≡ sup t > 0 ⟨ W t o t c o l ( t ) ⟩ W_* \equiv\sup_{t>0}\langle W_{\mathrm{tot}}^{\mathrm{col}}(t)\rangle W ∗ ≡ sup t > 0 ⟨ W tot col ( t )⟩ be the uniform bound on the reference trace weight, and let M t e r m \mathcal{M}_{\mathrm{term}} M term be the terminal-parent rest-mass intensity per unit momentum rapidity, defined in Eq. (F.31). They determine two lengths:
ℓ c o l = W ∗ 4 μ , ℓ a n g = M t e r m μ . (330) \ell_{\mathrm{col}} = \frac{W_*}{4\mu}, \qquad\ell_{\mathrm{ang}} = \frac{\mathcal{M}_{\mathrm{term}}}{\mu}. \tag*{(330)} ℓ col = 4 μ W ∗ , ℓ ang = μ M term . (330)
We will prove
sup b 0 ∣ A k , L ( s , b 0 ) ∣ ≤ C K F k ( s ) L d ℓ c o l + ℓ a n g L , L > 0 , (331) \sup_{\mathbf{b}_0} \left|A_{k,L}(s,\mathbf{b}_0)\right| \le C_K \frac{\mathcal{F}_k(s)}{L^d}\frac{\ell_{\mathrm{col}}+\ell_{\mathrm{ang}}}{L}, \qquad L>0, \tag*{(331)} b 0 sup ∣ A k , L ( s , b 0 ) ∣ ≤ C K L d F k ( s ) L ℓ col + ℓ ang , L > 0 , (331)
and bound both lengths by O ( Γ s p l i t − 1 / 2 ) O(\Gamma_{\mathrm{split}}^{-1/2}) O ( Γ split − 1/2 ) . Here A k , L A_{k,L} A k , L uses the ensemble-averaged stress tensor, d = D − 2 d=D-2 d = D − 2 , and C K C_K C K depends only on bounds on the kernel and its first two derivatives. It is independent of s s s , L L L , k k k and of terminal decay times and angular distributions. The residual can have either sign.
As in Sec. 7.2, k = ( 1 , n ^ ) k=(1,\hat{\mathbf{n}}) k = ( 1 , n ^ ) is normalized and s = − k ⋅ x > 0 s=-k\cdot x>0 s = − k ⋅ x > 0 labels the null planes. Write c = n ^ ⋅ x ^ c=\hat{\mathbf{n}}\cdot\hat{\mathbf{x}} c = n ^ ⋅ x ^ . A fold on x = σ t x=\sigma t x = σ t contributes − 2 μ s -2\mu s − 2 μ s whenever 1 − σ c > 0 1-\sigma c>0 1 − σ c > 0 . Thus the magnitude of the negative fold flux is
F k ( s ) = { 4 μ s , ∣ c ∣ < 1 , 2 μ s , ∣ c ∣ = 1. (332) \mathcal{F}_k(s)=
\begin{cases}
4\mu s, & |c|<1,\\
2\mu s, & |c|=1.
\end{cases} \tag*{(332)} F k ( s ) = { 4 μ s , 2 μ s , ∣ c ∣ < 1 , ∣ c ∣ = 1. (332)
At tangency the parallel fold carries no test flux; only the other fold contributes.
The collinear reference stays close to the folds
For a terminal parent with momentum P = ( E , P x , 0 ⊥ ) P=(E,P_x,0_\perp) P = ( E , P x , 0 ⊥ ) , the reference packets move along ± x \pm x ± x with nonnegative energies ( E + P x ) / 2 (E+P_x)/2 ( E + P x ) /2 and ( E − P x ) / 2 (E-P_x)/2 ( E − P x ) /2 . Their four-momenta sum to P P P . Keeping the original folds and all preterminal evolution defines a conserved reference stress with vanishing total energy and momentum. The positive reference packets preserve the branch-charge bound of Sec. 1: in each history, W t o t c o l ( t ) ≤ 8 μ t s W^{\mathrm{col}}_{\mathrm{tot}}(t)\leq8\mu t_s W tot col ( t ) ≤ 8 μ t s , where t s t_s t s is the first split time. Since the unbroken worldsheet up to t t t has null area 2 t 2 2t^2 2 t 2 , Pr ( t s > t ) = e − 2 Γ s p l i t t 2 \Pr(t_s>t)=e^{-2\Gamma_{\mathrm{split}}t^2} Pr ( t s > t ) = e − 2 Γ split t 2 , and hence
W ∗ ≤ 8 μ ⟨ t s ⟩ = 2 2 π μ Γ s p l i t . (333) W_* \leq8\mu\langle t_s\rangle= \frac{2\sqrt{2\pi}\,\mu}{\sqrt{\Gamma_{\mathrm{split}}}}. \tag*{(333)} W ∗ ≤ 8 μ ⟨ t s ⟩ = Γ split 2 2 π μ . (333)
No period averaging or assumption about subsequent decay is used.
Suppressing the transverse delta function, boost invariance gives the ensemble-averaged trace
T u v c o l = f c o l ( z ) , z = u v , u = t − x , v = t + x , (334) T^{\mathrm{col}}_{uv}=f_{\mathrm{col}}(z),\qquad z=uv,\qquad u=t-x,\qquad v=t+x, \tag*{(334)} T uv col = f col ( z ) , z = uv , u = t − x , v = t + x , (334)
where f c o l ≥ 0 f_{\mathrm{col}}\geq0 f col ≥ 0 and f c o l ( 0 ) = μ f_{\mathrm{col}}(0)=\mu f col ( 0 ) = μ . Define
I c o l ( z ) = ∫ 0 z f c o l ( w ) d w , I ∞ c o l = ∫ 0 ∞ f c o l ( w ) d w . (335) I_{\mathrm{col}}(z)=\int_0^z f_{\mathrm{col}}(w)\,dw,\qquad I^{\mathrm{col}}_{\infty}=\int_0^\infty f_{\mathrm{col}}(w)\,dw. \tag*{(335)} I col ( z ) = ∫ 0 z f col ( w ) d w , I ∞ col = ∫ 0 ∞ f col ( w ) d w . (335)
The latter may be infinite. Equation (13) applied to the reference gives
I c o l ( z ) z ≤ ⟨ W t o t c o l ( z ) ⟩ 4 z ≤ W ∗ 4 z . (336) \frac{I_{\mathrm{col}}(z)}{z}\leq\frac{\langle W^{\mathrm{col}}_{\mathrm{tot}}(\sqrt{z})\rangle}{4\sqrt{z}}\leq\frac{W_*}{4\sqrt{z}}. \tag*{(336)} z I col ( z ) ≤ 4 z ⟨ W tot col ( z )⟩ ≤ 4 z W ∗ . (336)
Conservation fixes the regular null components:
T u u c o l , r e g = I c o l ( z ) − z f c o l ( z ) u 2 , T v v c o l , r e g = I c o l ( z ) − z f c o l ( z ) v 2 . (337) T^{\mathrm{col,reg}}_{uu}=\frac{I_{\mathrm{col}}(z)-zf_{\mathrm{col}}(z)}{u^2},\qquad T^{\mathrm{col,reg}}_{vv}=\frac{I_{\mathrm{col}}(z)-zf_{\mathrm{col}}(z)}{v^2}. \tag*{(337)} T uu col , reg = u 2 I col ( z ) − z f col ( z ) , T v v col , reg = v 2 I col ( z ) − z f col ( z ) . (337)
Their positivity implies z f c o l ≤ I c o l zf_{\mathrm{col}}\leq I_{\mathrm{col}} z f col ≤ I col and ( I c o l / z ) ′ ≤ 0 (I_{\mathrm{col}}/z)'\leq0 ( I col / z ) ′ ≤ 0 . Together with I c o l ( z ) / z → μ I_{\mathrm{col}}(z)/z\to\mu I col ( z ) / z → μ as z → 0 z\to0 z → 0 , this yields
0 ≤ f c o l ( z ) ≤ I c o l ( z ) z ≤ μ ( z > 0 ) . (338) 0\leq f_{\mathrm{col}}(z)\leq\frac{I_{\mathrm{col}}(z)}{z}\leq\mu\qquad(z>0). \tag*{(338)} 0 ≤ f col ( z ) ≤ z I col ( z ) ≤ μ ( z > 0 ) . (338)
Away from the two negative folds, the positive energy density is
ρ + ( t , x ) = [ I c o l ( z ) − z f c o l ( z ) ] ( u − 2 + v − 2 ) + 2 f c o l ( z ) = ∂ x [ 2 x I c o l ( t 2 − x 2 ) t 2 − x 2 ] . (339) \begin{aligned}
\rho_+(t,x)=\left[I_{\mathrm{col}}(z)-zf_{\mathrm{col}}(z)\right]\left(u^{-2}+v^{-2}\right)+2f_{\mathrm{col}}(z) \\
&=\partial_x\left[\frac{2xI_{\mathrm{col}}(t^2-x^2)}{t^2-x^2}\right]. \tag*{(339)}
\end{aligned} ρ + ( t , x ) = [ I col ( z ) − z f col ( z ) ] ( u − 2 + v − 2 ) + 2 f col ( z ) = ∂ x [ t 2 − x 2 2 x I col ( t 2 − x 2 ) ] . (339)
Let r f = t − ∣ x ∣ r_f=t-|x| r f = t − ∣ x ∣ be the distance to the nearer fold. Integration by parts on each half of the string gives
∫ − t t d x r f ρ + = 2 ∫ 0 t 2 I c o l ( z ) z d z ≤ W ∗ t , \int_{-t}^{t}dx\,r_f\rho_+=2\int_0^{t^2}\frac{I_{\mathrm{col}}(z)}{z}\,dz\leq W_*t, ∫ − t t d x r f ρ + = 2 ∫ 0 t 2 z I col ( z ) d z ≤ W ∗ t ,
∫ − t t d x ρ + = 4 μ t . (340) \int_{-t}^{t}dx\,\rho_+=4\mu t. \tag*{(340)} ∫ − t t d x ρ + = 4 μ t . (340)
The boundary terms in the first identity vanish because I c o l ( z ) / z I_{\mathrm{col}}(z)/z I col ( z ) / z is bounded at z = 0 z=0 z = 0 . The inequality follows from Eq. (F.8). The second identity also follows from zero total energy and the two fold energies − 2 μ t -2\mu t − 2 μ t . By reflection symmetry, each half carries exactly the positive energy needed to cancel its nearest fold. In particular, the mean distance in this positive cancelling measure satisfies
⟨ r f ⟩ + ≡ ∫ − t t r f ρ + d x 4 μ t ≤ ℓ c o l . (341) \langle r_{\mathrm{f}}\rangle_{+} \equiv\frac{\int_{-t}^{t} r_{\mathrm{f}}\rho_{+}\,\mathrm{d}x}{4\mu t} \le\ell_{\mathrm{col}}. \tag*{(341)} ⟨ r f ⟩ + ≡ 4 μ t ∫ − t t r f ρ + d x ≤ ℓ col . (341)
For a test direction perpendicular to the string, the null plane meets the reference worldsheet at t = s t=s t = s . Pair each positive energy element with its nearest fold. Their kernel values differ by at most ∥ ∇ K ∥ ∞ r f / L d + 1 \lVert\nabla K\rVert_{\infty}r_{\mathrm{f}}/L^{d+1} ∥ ∇ K ∥ ∞ r f / L d + 1 , so
sup b 0 ∣ A k ⊥ , L c o l ∣ ≤ ∥ ∇ K ∥ ∞ L d + 1 ∫ − s s d x r f ρ + ≤ C K F k ⊥ ( s ) L d + 1 ℓ c o l . (342) \begin{aligned}
\sup_{b_{0}}\left|A^{\mathrm{col}}_{k_{\perp},L}\right| \le\frac{\lVert\nabla K\rVert_{\infty}}{L^{d+1}}\int_{-s}^{s}\mathrm{d}x\,r_{\mathrm{f}}\rho_{+} \\
&\le C_{K}\frac{\mathcal{F}_{k_{\perp}}(s)}{L^{d+1}}\ell_{\mathrm{col}}. \tag*{(342)}
\end{aligned} b 0 sup A k ⊥ , L col ≤ L d + 1 ∥ ∇ K ∥ ∞ ∫ − s s d x r f ρ + ≤ C K L d + 1 F k ⊥ ( s ) ℓ col . (342)
For ∣ c ∣ < 1 |c|<1 ∣ c ∣ < 1 , a boost of velocity c c c along the string takes k k k to 1 − c 2 k ⊥ \sqrt{1-c^{2}}\,k_{\perp} 1 − c 2 k ⊥ . It induces an isometry on the screen, and both the light-ray integral and the fold flux scale linearly under rescaling of k k k . Boost invariance and the time-uniform bound above therefore give the same estimate for every ∣ c ∣ < 1 |c|<1 ∣ c ∣ < 1 . At c = ± 1 c=\pm1 c = ± 1 the collinear reference cancels directly on the complete null ray. Including the fold,
∫ − ∞ U d u T u u c o l ( u , v ) = − I c o l ( U v ) U ⟶ 0 , U ⟶ ∞ , (343) \int_{-\infty}^{U}\mathrm{d}u\,T^{\mathrm{col}}_{uu}(u,v)=-\frac{I_{\mathrm{col}}(Uv)}{U}\longrightarrow0,\qquad U\longrightarrow\infty, \tag*{(343)} ∫ − ∞ U d u T uu col ( u , v ) = − U I col ( U v ) ⟶ 0 , U ⟶ ∞ , (343)
since its magnitude is at most ( W ∗ / 4 ) v / U (W_{*}/4)\sqrt{v/U} ( W ∗ /4 ) v / U by Eq. (F.8). The same holds after u ↔ v u\leftrightarrow v u ↔ v . No singular boost limit or finite I ∞ c o l I^{\mathrm{col}}_{\infty} I ∞ col is required at exact tangency.
Complete null-plane flux and the continuing source
We now establish completion directly for the actual stress tensor. The argument uses conservation, causal support inside the future light cone of nucleation, and the positivity properties of the sector other than the two original folds. Writing ρ = T 00 \rho=T^{00} ρ = T 00 and j i = T 0 i j^{i}=T^{0i} j i = T 0 i , this sector obeys ρ ≥ ∣ j ∣ \rho\geq|\mathbf{j}| ρ ≥ ∣ j ∣ , the null energy condition, and T 00 − T x x ≥ 0 T^{00}-T^{xx}\geq0 T 00 − T xx ≥ 0 . Positive string stress, positive folds, and causal particles have these properties. The original folds retain their prescribed null trajectories and energies. For longitudinal completion we also use boost invariance of the ensemble and two separately conserved descendant lineages whose positive sectors obey the same inequalities. Free propagation after terminal conversion is not needed for this argument; any further interactions must preserve these conservation and positivity properties.
Nonparallel null directions. Fix k = ( 1 , n ^ ) k=(1,\hat{\mathbf{n}}) k = ( 1 , n ^ ) and H s = { t − n ^ ⋅ x = s } H_s=\{t-\hat{\mathbf{n}}\cdot\mathbf{x}=s\} H s = { t − n ^ ⋅ x = s } , with s > 0 s>0 s > 0 . For ∣ c ∣ < 1 |c|<1 ∣ c ∣ < 1 , where c = n ^ ⋅ x ^ c=\hat{\mathbf{n}}\cdot\hat{\mathbf{x}} c = n ^ ⋅ x ^ , both negative folds cross H s H_s H s at finite times s / ( 1 ∓ c ) s/(1\mp c) s / ( 1 ∓ c ) . Choose T 0 > s / ( 1 − ∣ c ∣ ) T_0>s/(1-|c|) T 0 > s / ( 1 − ∣ c ∣ ) . The region ahead of the plane and its remaining null charge are
R t = { x : n ^ ⋅ x > t − s } , Q r e m ( t ) = ∫ R t d D − 1 x ( ρ − n ^ ⋅ j ) . (344) R_t=\{\mathbf{x}:\hat{\mathbf{n}}\cdot\mathbf{x}>t-s\},\qquad Q_{\mathrm{rem}}(t)=\int_{R_t}d^{D-1}x\,(\rho-\hat{\mathbf{n}}\cdot\mathbf{j}). \tag*{(344)} R t = { x : n ^ ⋅ x > t − s } , Q rem ( t ) = ∫ R t d D − 1 x ( ρ − n ^ ⋅ j ) . (344)
Let Φ k ( s ; T ) \Phi_k(s;T) Φ k ( s ; T ) denote the flux through H s H_s H s with Minkowski-time cutoff t ≤ T t\le T t ≤ T , integrated over the complete screen. Conservation and the absence of stress before nucleation give
d Q r e m d t = − ∫ H s ∩ { t } d D − 2 b T μ ν k μ k ν , Φ k ( s ; T ) = − Q r e m ( T ) . (345) \frac{dQ_{\mathrm{rem}}}{dt}=-\int_{H_s\cap\{t\}}d^{D-2}b\,T_{\mu\nu}k^\mu k^\nu,\qquad\Phi_k(s;T)=-Q_{\mathrm{rem}}(T). \tag*{(345)} d t d Q rem = − ∫ H s ∩ { t } d D − 2 b T μν k μ k ν , Φ k ( s ; T ) = − Q rem ( T ) . (345)
For t ≥ T 0 t\ge T_0 t ≥ T 0 the remaining region contains only positive stress, so Q r e m ≥ 0 Q_{\mathrm{rem}}\ge0 Q rem ≥ 0 and d Q r e m / d t ≤ 0 dQ_{\mathrm{rem}}/dt\le0 d Q rem / d t ≤ 0 . Introduce its positive energy moment
M r e m ( t ) = ∫ R t d D − 1 x ( s − t + n ^ ⋅ x ) ρ . (346) M_{\mathrm{rem}}(t)=\int_{R_t}d^{D-1}x\,(s-t+\hat{\mathbf{n}}\cdot\mathbf{x})\rho. \tag*{(346)} M rem ( t ) = ∫ R t d D − 1 x ( s − t + n ^ ⋅ x ) ρ . (346)
Energy conservation gives d M r e m / d t = − Q r e m dM_{\mathrm{rem}}/dt=-Q_{\mathrm{rem}} d M rem / d t = − Q rem , since the weight vanishes at the moving boundary. Integration and monotonicity of Q r e m Q_{\mathrm{rem}} Q rem yield
0 ≤ Q r e m ( T ) ≤ M r e m ( T 0 ) T − T 0 ≤ 4 μ s T 0 T − T 0 ⟶ 0. (347) 0\le Q_{\mathrm{rem}}(T)\le\frac{M_{\mathrm{rem}}(T_0)}{T-T_0}\le\frac{4\mu sT_0}{T-T_0}\longrightarrow0. \tag*{(347)} 0 ≤ Q rem ( T ) ≤ T − T 0 M rem ( T 0 ) ≤ T − T 0 4 μ s T 0 ⟶ 0. (347)
Indeed, causal support bounds the weight by s s s , and zero total energy fixes the positive energy to E + ( T 0 ) = 4 μ T 0 E_+(T_0)=4\mu T_0 E + ( T 0 ) = 4 μ T 0 . The estimate holds history by history and also controls the ensemble limit. It proves completion for each fixed nonparallel direction, but is not uniform as ∣ c ∣ → 1 |c|\to1 ∣ c ∣ → 1 .
Longitudinal null directions. For the parallel directions, integrate the actual stress over the coordinates y \mathbf{y} y transverse to the original string: T ˉ α β ( u , v ) ≡ ∫ d D − 2 y T α β ( u , v , y ) \bar{T}_{\alpha\beta}(u,v)\equiv\int d^{D-2}y\,T_{\alpha\beta}(u,v,\mathbf{y}) T ˉ α β ( u , v ) ≡ ∫ d D − 2 y T α β ( u , v , y ) . The transverse divergence in ∂ v T u u + ∂ u T u v = 1 2 ∂ a T a u \partial_vT_{uu}+\partial_uT_{uv}=\frac{1}{2}\partial_aT_{au} ∂ v T uu + ∂ u T uv = 2 1 ∂ a T a u integrates to zero, leaving the same conservation equation as in 1 + 1 1+1 1 + 1 dimensions. The relevant positive weight is the longitudinal projection
W ∥ ( t ) ≡ ∫ d D − 1 x ( T 00 − T x x ) = 4 ∫ d x T ˉ u v . (348) W_{\parallel}(t)\equiv\int d^{D-1}x\,(T^{00}-T^{xx})=4\int dx\,\bar{T}_{uv}. \tag*{(348)} W ∥ ( t ) ≡ ∫ d D − 1 x ( T 00 − T xx ) = 4 ∫ d x T ˉ uv . (348)
It differs from the full scalar trace weight by ∫ d D − 1 x ∑ a T a a \int d^{D-1}x\sum_aT^{aa} ∫ d D − 1 x ∑ a T aa . In particular, an oblique massless particle contributes p ⊥ 2 / E p_\perp^2/E p ⊥ 2 / E to W ∥ W_{\parallel} W ∥ despite having zero scalar trace. After the first split at ( u s , v s ) (u_s,v_s) ( u s , v s ) , label the two complete lineages by their original outer folds, rather than their spatial positions. Their conserved charges are Q R = ( E − P x ) R = 2 μ u s Q_R=(E-P_x)_R=2\mu u_s Q R = ( E − P x ) R = 2 μ u s and Q L = ( E + P x ) L = 2 μ v s Q_L=(E+P_x)_L=2\mu v_s Q L = ( E + P x ) L = 2 μ v s . Each negative outer fold contributes zero to its lineage charge and to W ∥ W_{\parallel} W ∥ . The positive sector obeys, for example, 2 Q R − W ∥ , R = ∫ R d D − 1 x ( T 00 + T x x − 2 T 0 x ) ≥ 0 2Q_R-W_{\parallel,R}=\int_Rd^{D-1}x\,(T^{00}+T^{xx}-2T^{0x})\ge0 2 Q R − W ∥ , R = ∫ R d D − 1 x ( T 00 + T xx − 2 T 0 x ) ≥ 0 ; the analogous left-lineage inequality has the opposite sign of T 0 x T^{0x} T 0 x . Thus
0 ≤ W ∥ ( t ) ≤ 2 ( Q R + Q L ) = 8 μ t s . (349) 0\le W_{\parallel}(t)\le2(Q_R+Q_L)=8\mu t_s. \tag*{(349)} 0 ≤ W ∥ ( t ) ≤ 2 ( Q R + Q L ) = 8 μ t s . (349)
Before the first split W ∥ ( t ) = 8 μ t ≤ 8 μ t s W_{\parallel}(t)=8\mu t\leq8\mu t_s W ∥ ( t ) = 8 μ t ≤ 8 μ t s , so the same bound holds at every time. Boost invariance gives ⟨ T u v ⟩ = f ∥ ( u v ) ≥ 0 \langle T_{uv}\rangle=f_{\parallel}(uv)\geq0 ⟨ T uv ⟩ = f ∥ ( uv ) ≥ 0 . With I ∥ ( z ) = ∫ 0 z f ∥ ( w ) d w I_{\parallel}(z)=\int_{0}^{z}f_{\parallel}(w)\,\mathrm{d}w I ∥ ( z ) = ∫ 0 z f ∥ ( w ) d w , the time-slice identity of § 1 yields
I ∥ ( z ) z ≤ ⟨ W ∥ ( z ) ⟩ 4 z ≤ 2 μ ⟨ t s ⟩ z . (350) \frac{I_{\parallel}(z)}{z}\leq\frac{\langle W_{\parallel}(\sqrt{z})\rangle}{4\sqrt{z}}\leq\frac{2\mu\langle t_s\rangle}{\sqrt{z}}. \tag*{(350)} z I ∥ ( z ) ≤ 4 z ⟨ W ∥ ( z )⟩ ≤ z 2 μ ⟨ t s ⟩ . (350)
Consequently, at fixed v > 0 v>0 v > 0 ,
∫ d D − 2 y ∫ − ∞ U d u ⟨ T u u ( u , v , y ) ⟩ = − I ∥ ( U v ) U ⟶ 0 , (351) \int\mathrm{d}^{D-2}y\int_{-\infty}^{U}\mathrm{d}u\,\langle T_{uu}(u,v,\mathbf{y})\rangle=-\frac{I_{\parallel}(Uv)}{U}\longrightarrow0, \tag*{(351)} ∫ d D − 2 y ∫ − ∞ U d u ⟨ T uu ( u , v , y )⟩ = − U I ∥ ( U v ) ⟶ 0 , (351)
with the analogous result after u ↔ v u\leftrightarrow v u ↔ v . The normalized k k k -flux is twice this expression and also vanishes.
Together the two arguments establish, for every fixed normalized null direction and s > 0 s>0 s > 0 ,
∫ d D − 2 b A k ( s , b ) = 0. (352) \int\mathrm{d}^{D-2}b\,A_k(s,\mathbf{b})=0. \tag*{(352)} ∫ d D − 2 b A k ( s , b ) = 0. (352)
After the visible negative folds have crossed, all additional flux is nonnegative. Its complete integral is therefore finite and equals F k ( s ) \mathcal{F}_k(s) F k ( s ) . No terminal lifetime, finite integrated trace, or collinear replacement is needed for this completion statement.
Longitudinal rays and positional averaging. For the collinear cascade, conservation gives, history by history, A ( v ; U ) = − ∫ 0 v T v w ( U , w ) d w ≤ 0 \mathcal{A}(v;U)=-\int_{0}^{v}T_{vw}(U,w)\,\mathrm{d}w\leq0 A ( v ; U ) = − ∫ 0 v T v w ( U , w ) d w ≤ 0 . After the original fold is crossed, the residual increases with U U U and is bounded below by − μ v -\mu v − μv . Equation (F.8) gives I ( z ) = o ( z ) I(z)=o(z) I ( z ) = o ( z ) , so its ensemble limit vanishes. Bounded convergence then implies saturation with probability one for each fixed v v v . Intersect these probability-one events for positive rational r r r . For any r > v r>v r > v , A ( r ; U ) ≤ A ( v ; U ) ≤ 0 \mathcal{A}(r;U)\leq\mathcal{A}(v;U)\leq0 A ( r ; U ) ≤ A ( v ; U ) ≤ 0 , which extends the result to all rays using the cumulative-integral representative. The argument applies both to the unabsorbed semiclassical cascade and to its traceless terminal completion; finite I ∞ I_{\infty} I ∞ is needed only for the stronger positional estimate below.
For N N N IFSs with origins ( u 0 , v 0 ) = ( a , − a ) (u_0,v_0)=(a,-a) ( u 0 , v 0 ) = ( a , − a ) averaged uniformly over − R ≤ a ≤ R -R\leq a\leq R − R ≤ a ≤ R , the residual on v = 0 v=0 v = 0 , cut off at u = U > 0 u=U>0 u = U > 0 , is
A ‾ N , R ( U ) = − N 2 R ∫ 0 min ( R , U ) I ( a ( U − a ) ) U − a d a . \overline{\mathcal{A}}_{N,R}(U)=-\frac{N}{2R}\int_{0}^{\min(R,U)}\frac{I\bigl(a(U-a)\bigr)}{U-a}\,\mathrm{d}a. A N , R ( U ) = − 2 R N ∫ 0 m i n ( R , U ) U − a I ( a ( U − a ) ) d a .
Now assume I ∞ < ∞ I_{\infty}<\infty I ∞ < ∞ . The trace bound (F.10) gives I ( z ) ≤ min ( μ z , I ∞ ) I(z)\leq\min(\mu z,I_{\infty}) I ( z ) ≤ min ( μ z , I ∞ ) . Splitting the integral at U − a = I ∞ / ( μ U ) U-a=I_{\infty}/(\mu U) U − a = I ∞ / ( μU ) when U ≤ 2 R U\leq2R U ≤ 2 R , and using U − a ≥ U − R U-a\geq U-R U − a ≥ U − R when U > 2 R U>2R U > 2 R , yields
sup U > 0 ∣ A ‾ N , R ( U ) ∣ ≤ N I ∞ 2 R [ 1 + ln ( 1 + 4 μ R 2 I ∞ ) ] ⟶ 0 ( R → ∞ ) . \sup_{U>0}\left|\overline{\mathcal{A}}_{N,R}(U)\right|\leq\frac{NI_{\infty}}{2R}\left[1+\ln\left(1+\frac{4\mu R^2}{I_{\infty}}\right)\right]\longrightarrow0\qquad(R\to\infty). U > 0 sup A N , R ( U ) ≤ 2 R N I ∞ [ 1 + ln ( 1 + I ∞ 4 μ R 2 ) ] ⟶ 0 ( R → ∞ ) .
This is a bound for the normalized positional average of the 1 + 1 1+1 1 + 1 -dimensional cascade at fixed finite N N N , uniform in the affine cutoff. This stronger uniformity does not follow from bounded W t o t W_{\mathrm{tot}} W tot alone and is not used in the transverse-smearing bound.
Terminal products of arbitrary mass
Consider one actual terminal packet and its collinear replacement. Let the parent have momentum P = m U P=mU P = m U and define its positive k k k -flux and Doppler factor by
F = − k ⋅ P = m ω , ω = − k ⋅ U > 0. (353) F=-k\cdot P=m\omega,\qquad\omega=-k\cdot U>0. \tag*{(353)} F = − k ⋅ P = mω , ω = − k ⋅ U > 0. (353)
For an outgoing product of momentum p j p_j p j with f j > 0 f_j>0 f j > 0 , let
f j = − k ⋅ p j , a j = Π k ( p j f j − P F ) . (354) f_j=-k\cdot p_j,\qquad\mathbf{a}_j=\Pi_k\left(\frac{p_j}{f_j}-\frac{P}{F}\right). \tag*{(354)} f j = − k ⋅ p j , a j = Π k ( f j p j − F P ) . (354)
Here Π k \Pi_k Π k projects the spatial part onto the screen perpendicular to n ^ \hat{\mathbf{n}} n ^ . Along a trajectory, p j / f j p_j/f_j p j / f j is its velocity with respect to the null clock s s s , so a j \mathbf{a}_j a j is the relative screen velocity. If conversion occurs at X e X_e X e , write s e = − k ⋅ X e s_e=-k\cdot X_e s e = − k ⋅ X e . For s > s e s>s_e s > s e a free product with f j > 0 f_j>0 f j > 0 crosses the plane once, at X e + ( s − s e ) p j / f j X_e+(s-s_e)p_j/f_j X e + ( s − s e ) p j / f j , and contributes flux f j f_j f j . A parallel product has f j = 0 f_j=0 f j = 0 and carries no test flux.
Four-momentum conservation gives the two identities which drive the argument:
∑ j f j = F , ∑ j f j a j = 0 , ∑ j f j ∣ a j ∣ 2 ≤ m 2 F = F ω 2 . (355) \sum_j f_j=F,\qquad\sum_j f_j\mathbf{a}_j=0,\qquad\sum_j f_j|\mathbf{a}_j|^2\leq\frac{m^2}{F}=\frac{F}{\omega^2}. \tag*{(355)} j ∑ f j = F , j ∑ f j a j = 0 , j ∑ f j ∣ a j ∣ 2 ≤ F m 2 = ω 2 F . (355)
The middle identity says that the packet is centered in positive k k k -flux. To derive the last bound for massive products, write h j = E j + n ^ ⋅ p j h_j=E_j+\hat{\mathbf{n}}\cdot p_j h j = E j + n ^ ⋅ p j . The mass shell gives ∣ p j ⊥ ∣ 2 = f j h j − m j 2 |p_{j\perp}|^2=f_jh_j-m_j^2 ∣ p j ⊥ ∣ 2 = f j h j − m j 2 . Expanding the centered square and using total four-momentum conservation therefore gives
∑ f j > 0 f j ∣ a j ∣ 2 = m 2 F − ∑ f j > 0 m j 2 f j − 2 E ∥ ≤ m 2 F . (356) \sum_{f_j>0}f_j|\mathbf{a}_j|^2=\frac{m^2}{F}-\sum_{f_j>0}\frac{m_j^2}{f_j}-2E_\parallel\leq\frac{m^2}{F}. \tag*{(356)} f j > 0 ∑ f j ∣ a j ∣ 2 = F m 2 − f j > 0 ∑ f j m j 2 − 2 E ∥ ≤ F m 2 . (356)
Here sums involving a j \mathbf{a}_j a j omit zero-flux products, and E ∥ E_\parallel E ∥ is the total energy of massless products exactly parallel to k k k ; these have zero screen momentum and zero test flux. Massive products have f j > 0 f_j>0 f j > 0 . At fixed parent momentum, the massless case therefore gives the largest allowed centered moment.
Suppose conversion occurs on slice s e s_e s e . On a later slice s s s , the relative displacement is ( s − s e ) a j (s-s_e)\mathbf{a}_j ( s − s e ) a j . Taylor expanding the smeared kernel around the parent continuation, the centered identity cancels the linear term and the last identity bounds the quadratic remainder:
∣ δ A k , L p a c k e t ∣ ≤ C K F L d ( s − s e ) 2 L 2 ω 2 . (357) \left|\delta\mathcal{A}^{\mathrm{packet}}_{k,L}\right|\leq C_K\frac{F}{L^d}\frac{(s-s_e)^2}{L^2\omega^2}. \tag*{(357)} δ A k , L packet ≤ C K L d F L 2 ω 2 ( s − s e ) 2 . (357)
There is also the trivial bound ∣ δ A k , L p a c k e t ∣ ≤ C K F / L d \left|\delta\mathcal{A}^{\mathrm{packet}}_{k,L}\right|\leq C_KF/L^d δ A k , L packet ≤ C K F / L d , because both packets are positive and have the same total flux. We apply the better of the two bounds only after summing the complete packet:
∣ δ A k , L p a c k e t ∣ ≤ C K m ω L d min { 1 , ( s − s e ) 2 L 2 ω 2 } . (358) \left|\delta\mathcal{A}^{\mathrm{packet}}_{k,L}\right|\leq C_K\frac{m\omega}{L^d}\min\left\{1,\frac{(s-s_e)^2}{L^2\omega^2}\right\}. \tag*{(358)} δ A k , L packet ≤ C K L d mω min { 1 , L 2 ω 2 ( s − s e ) 2 } . (358)
Both the actual and collinear packets obey this estimate relative to the same fictitious parent continuation. Their mutual difference therefore obeys Eq. (F.30), with a larger C K C_K C K . This capped form is important: a very old or highly transverse packet can contribute at most its total positive flux.
It remains to sum over all terminal events. Let m j m_j m j and χ j \chi_j χ j be the mass and longitudinal momentum rapidity of terminal parent j j j . Boost stationarity defines its mass intensity by
⟨ ∑ j ∈ t e r m m j h ( χ j ) ⟩ = M t e r m ∫ − ∞ ∞ d χ h ( χ ) (359) \left\langle\sum_{j\in\mathrm{term}} m_j h(\chi_j) \right\rangle= \mathcal{M}_{\mathrm{term}} \int_{-\infty}^{\infty} d\chi\,h(\chi) \tag*{(359)} ⟨ j ∈ term ∑ m j h ( χ j ) ⟩ = M term ∫ − ∞ ∞ d χ h ( χ ) (359)
for every nonnegative integrable test function h h h . The sum includes the complete conversion history. This measure retains correlations between mass, conversion time, and ancestry. The uniform source intensity in rapidity is preserved by the random shift from each root to its descendants.
For a parent of rapidity χ \chi χ ,
ω = cosh χ − c sinh χ . (360) \omega= \cosh\chi-c\sinh\chi. \tag*{(360)} ω = cosh χ − c sinh χ . (360)
All terminal events lie in the future cone, so a packet observed on slice s s s has 0 ≤ s e ≤ s 0\le s_e\le s 0 ≤ s e ≤ s . In the positive upper bound Eq. (F.30) we may replace s − s e s-s_e s − s e by s s s and then include the full event measure, even events with s e > s s_e>s s e > s . Hence, with a = s / L a=s/L a = s / L ,
sup b 0 ∣ δ A k , L t e r m ∣ ≤ C K M t e r m L d J c ( a ) , J c ( a ) ≡ ∫ − ∞ ∞ d χ ω min { 1 , a 2 ω 2 } . (361) \sup_{\mathbf{b}_0}\left|\delta A_{k,L}^{\mathrm{term}}\right|
\le C_K\frac{\mathcal{M}_{\mathrm{term}}}{L^d}J_c(a),
\qquad
J_c(a)\equiv\int_{-\infty}^{\infty}d\chi\,\omega\min\left\{1,\frac{a^2}{\omega^2}\right\}. \tag*{(361)} b 0 sup δ A k , L term ≤ C K L d M term J c ( a ) , J c ( a ) ≡ ∫ − ∞ ∞ d χ ω min { 1 , ω 2 a 2 } . (361)
This integral is finite uniformly in c c c . For ∣ c ∣ < 1 |c|<1 ∣ c ∣ < 1 , set ϵ = 1 − c 2 \epsilon=\sqrt{1-c^2} ϵ = 1 − c 2 and y = χ − arctanh c y=\chi-\operatorname{arctanh}c y = χ − arctanh c , so that ω = ϵ cosh y \omega=\epsilon\cosh y ω = ϵ cosh y . On either half-line put v = ϵ e ∣ y ∣ v=\epsilon e^{|y|} v = ϵ e ∣ y ∣ . Since v / 2 ≤ ω ≤ v v/2\le\omega\le v v /2 ≤ ω ≤ v and ∣ d χ ∣ = d v / v |d\chi|=dv/v ∣ d χ ∣ = d v / v ,
J c ( a ) ≤ 2 ∫ 0 ∞ d v min { 1 , 2 a 2 v 2 } = 4 2 a . (362) J_c(a)\le2\int_0^\infty dv\min\left\{1,\frac{2a^2}{v^2}\right\}=4\sqrt{2}\,a. \tag*{(362)} J c ( a ) ≤ 2 ∫ 0 ∞ d v min { 1 , v 2 2 a 2 } = 4 2 a . (362)
At exact tangency, ω = e ∓ χ \omega=e^{\mp\chi} ω = e ∓ χ and direct use of ω \omega ω as the integration variable gives J ± 1 ( a ) = 2 a J_{\pm1}(a)=2a J ± 1 ( a ) = 2 a . The grazing limit is therefore harmless: the large propagation distance of nearly parallel products is compensated by their small k k k -flux. Combining Eqs. (F.33) and (F.34), and using F k ( s ) ≥ 2 μ s \mathcal{F}_k(s)\ge2\mu s F k ( s ) ≥ 2 μ s , gives
sup b 0 ∣ δ A k , L t e r m ∣ ≤ C K F k ( s ) L d ℓ a n g L . (363) \sup_{\mathbf{b}_0}\left|\delta A_{k,L}^{\mathrm{term}}\right|
\le C_K\frac{\mathcal{F}_k(s)}{L^d}\frac{\ell_{\mathrm{ang}}}{L}. \tag*{(363)} b 0 sup δ A k , L term ≤ C K L d F k ( s ) L ℓ ang . (363)
Mass conservation and the smearing scale
It remains to bound the terminal mass intensity. At a splitting of a positive-energy yo-yo, the sum of daughter rest masses is at most the parent mass, by four-momentum conservation.
Terminal parents are disjoint leaves of the descendant tree. Therefore a first-generation yo-yo of mass m 1 m_1 m 1 obeys
∑ j ∈ term ( 1 ) m j ≤ m 1 . (364) \sum_{j\in\operatorname{term}(1)} m_j \le m_1. \tag*{(364)} j ∈ term ( 1 ) ∑ m j ≤ m 1 . (364)
The inequality holds for any finite set of terminal leaves, and thus for the full set by monotone convergence, even if some branches never terminate. It also follows directly from the exact staircase bound ∑ i z i ≤ 1 \sum_i z_i \le1 ∑ i z i ≤ 1 in Appendix § 12 .§ 2 .
To pass to intensities, write each terminal rapidity as χ j = χ 1 + δ χ j \chi_j=\chi_1+\delta\chi_j χ j = χ 1 + δ χ j . Boost covariance makes the law of the relative rapidities and masses independent of the root rapidity χ 1 \chi_1 χ 1 . Integrating a test function h ( χ 1 + δ χ j ) h(\chi_1+\delta\chi_j) h ( χ 1 + δ χ j ) over the stationary root rapidity therefore gives ∫ h ( χ ) d χ \int h(\chi)\,\mathrm{d}\chi ∫ h ( χ ) d χ for every descendant. Equation (F.36) then bounds M t e r m \mathcal{M}_{\mathrm{term}} M term by the first-generation mass intensity M 1 \mathcal{M}_1 M 1 . Using the raw source moment (B.21) at p = 1 p=1 p = 1 gives
M t e r m ≤ M 1 = 2 μ Γ s p l i t S 0 ( 1 ) = 2 π 3 μ Γ s p l i t . (365) \mathcal{M}_{\mathrm{term}} \le\mathcal{M}_1 = \frac{2\mu}{\sqrt{\Gamma_{\mathrm{split}}}} \mathcal{S}_0(1) = \frac{2\sqrt{\pi}}{3}\frac{\mu}{\sqrt{\Gamma_{\mathrm{split}}}}. \tag*{(365)} M term ≤ M 1 = Γ split 2 μ S 0 ( 1 ) = 3 2 π Γ split μ . (365)
This uses the first-generation source law, not a count or lifetime estimate for later generations. Equations (F.5) and (F.37) imply
ℓ c o l ≤ π 2 Γ s p l i t − 1 / 2 , ℓ a n g ≤ 2 π 3 Γ s p l i t − 1 / 2 . (366) \ell_{\mathrm{col}} \le\sqrt{\frac{\pi}{2}}\Gamma_{\mathrm{split}}^{-1/2}, \qquad\ell_{\mathrm{ang}} \le\frac{2\sqrt{\pi}}{3}\Gamma_{\mathrm{split}}^{-1/2}. \tag*{(366)} ℓ col ≤ 2 π Γ split − 1/2 , ℓ ang ≤ 3 2 π Γ split − 1/2 . (366)
Inserting these bounds in Eq. (F.3) and absorbing numerical constants into C K C_K C K yields
sup b 0 ∣ A k , L ( s , b 0 ) ∣ ≤ C K F k ( s ) L d 1 L Γ s p l i t , s > 0 , every normalized k . (367) \sup_{\mathbf{b}_0} \left|\mathcal{A}_{k,L}(s,\mathbf{b}_0)\right| \le C_K \frac{\mathcal{F}_k(s)}{L^d}\frac{1}{L\sqrt{\Gamma_{\mathrm{split}}}}, \qquad s>0,\quad\text{every normalized }k. \tag*{(367)} b 0 sup ∣ A k , L ( s , b 0 ) ∣ ≤ C K L d F k ( s ) L Γ split 1 , s > 0 , every normalized k . (367)
Thus a sufficient scale is
L s m e a r ≡ Γ s p l i t − 1 / 2 ∼ ℓ s g s − 1 . (368) L_{\mathrm{smear}} \equiv\Gamma_{\mathrm{split}}^{-1/2} \sim\ell_{s}g_s^{-1}. \tag*{(368)} L smear ≡ Γ split − 1/2 ∼ ℓ s g s − 1 . (368)
Accuracy ε \varepsilon ε requires L ≳ C K L s m e a r / ε L\gtrsim C_KL_{\mathrm{smear}}/\varepsilon L ≳ C K L smear / ε . This is a sufficient resolution scale, not an optimal-width claim. It applies to sharp or smooth terminal conversion within the stated assumptions, and also to the unabsorbed collinear cascade, where M t e r m = 0 \mathcal{M}_{\mathrm{term}}=0 M term = 0 .
Stronger terminal response and second-moment estimates
Finite integrated trace gives additional information, including the O ( U − 1 ) O(U^{-1}) O ( U − 1 ) longitudinal cutoff rate in § 6 . We record the sharp absorbing-model estimate separately. For m t e r m = O ( m s ) m_{\mathrm{term}}=O(m_s) m term = O ( m s ) , define
ξ ∗ ≡ 2 μ m t e r m Γ s p l i t ∼ g s − 1 . (369) \xi_* \equiv\frac{2\mu}{m_{\mathrm{term}}\sqrt{\Gamma_{\mathrm{split}}}} \sim g_s^{-1}. \tag*{(369)} ξ ∗ ≡ m term Γ split 2 μ ∼ g s − 1 . (369)
Under the entrance-weight assumptions of § 6 , the source-averaged internal-node and leaf counts satisfy ⟨ N i n t ⟩ , ⟨ N l e a f ⟩ = O ( ξ ∗ κ I F S ) \langle N_{\mathrm{int}}\rangle,\langle N_{\mathrm{leaf}}\rangle=O(\xi_*^{\kappa_{\mathrm{IFS}}}) ⟨ N int ⟩ , ⟨ N leaf ⟩ = O ( ξ ∗ κ IFS ) . Each internal node contributes expected integrated trace 2 μ / Γ s p l i t 2\mu/\Gamma_{\mathrm{split}} 2 μ / Γ split . Since d t d x = 1 2 d z d y dt\,dx = \frac{1}{2}dz\,dy d t d x = 2 1 d z d y and T 00 c o l − T x x c o l = 4 f c o l T^{\mathrm{col}}_{00}-T^{\mathrm{col}}_{xx}=4f_{\mathrm{col}} T 00 col − T xx col = 4 f col , the integrated trace per unit spacetime rapidity is 2 I ∞ c o l 2I^{\mathrm{col}}_{\infty} 2 I ∞ col . The source intensity therefore gives
I ∞ c o l μ ≲ 1 + ξ ∗ κ I F S Γ s p l i t . (370) \frac{I^{\mathrm{col}}_{\infty}}{\mu} \lesssim\frac{1+\xi_*^{\kappa_{\mathrm{IFS}}}}{\Gamma_{\mathrm{split}}}. \tag*{(370)} μ I ∞ col ≲ Γ split 1 + ξ ∗ κ IFS . (370)
The constant term includes the remnant and source-scale stages.
This controls a second spatial moment. On the right half of the string, r f = u r_f=u r f = u and r = u / v ∈ [ 0 , 1 ] r=u/v\in[0,1] r = u / v ∈ [ 0 , 1 ] , so r f 2 ρ + = I c o l ( 1 + r 2 ) − z f c o l ( 1 − r ) 2 ≤ 2 I ∞ c o l r_f^2\rho_+=I_{\mathrm{col}}(1+r^2)-zf_{\mathrm{col}}(1-r)^2\le2I^{\mathrm{col}}_{\infty} r f 2 ρ + = I col ( 1 + r 2 ) − z f col ( 1 − r ) 2 ≤ 2 I ∞ col . Reflection and integration give
⟨ r f 2 ⟩ + ≲ I ∞ c o l μ ≡ ℓ 2 2 , ℓ 2 ≲ ℓ s g − 1 − κ I F S / 2 = ℓ s g − 2 . (371) \langle r_f^2\rangle_+ \lesssim\frac{I^{\mathrm{col}}_{\infty}}{\mu}\equiv\ell_2^2,\qquad\ell_2\lesssim\ell_{\mathrm{sg}}^{-1-\kappa_{\mathrm{IFS}}/2}=\ell_{\mathrm{sg}}^{-\sqrt{2}}. \tag*{(371)} ⟨ r f 2 ⟩ + ≲ μ I ∞ col ≡ ℓ 2 2 , ℓ 2 ≲ ℓ sg − 1 − κ IFS /2 = ℓ sg − 2 . (371)
Under the same finite-response assumptions, the late pressure law proved in Appendix § 18 .§ 1 gives
ℓ 2 2 = 1 4 μ lim t → ∞ t ⟨ W t o t c o l ( t ) ⟩ = − 1 4 μ lim t → ∞ t ⟨ P t o t c o l ( t ) ⟩ . (372) \ell_2^2=\frac{1}{4\mu}\lim_{t\to\infty}t\langle W^{\mathrm{col}}_{\mathrm{tot}}(t)\rangle=-\frac{1}{4\mu}\lim_{t\to\infty}t\langle P^{\mathrm{col}}_{\mathrm{tot}}(t)\rangle. \tag*{(372)} ℓ 2 2 = 4 μ 1 t → ∞ lim t ⟨ W tot col ( t )⟩ = − 4 μ 1 t → ∞ lim t ⟨ P tot col ( t )⟩ . (372)
Thus the coefficient of the late negative-pressure tail also sets this upper bound on the RMS separation of positive energy from the folds.
The leaf count also refines the angular estimate in this model:
M t e r m ≲ m t e r m ξ ∗ κ I F S , ℓ a n g ≲ ℓ s g − κ I F S . (373) \mathcal{M}_{\mathrm{term}}\lesssim m_{\mathrm{term}}\xi_*^{\kappa_{\mathrm{IFS}}},\qquad\ell_{\mathrm{ang}}\lesssim\ell_{\mathrm{sg}}^{-\kappa_{\mathrm{IFS}}}. \tag*{(373)} M term ≲ m term ξ ∗ κ IFS , ℓ ang ≲ ℓ sg − κ IFS . (373)
For a smooth crossover, these additional estimates require the complete-tree response and terminal-mass matching of § 6 . They need not hold for a prolonged preterminal phase. The first-moment smearing bound (F.39) requires neither of them.
Gravitational emission from semiclassical yo-yos
A yo-yo loses energy gradually through gravitational radiation and redistributes it abruptly when it splits. We estimate the fraction radiated before the next split in weak, locally flat gravity. The long-wavelength calculation tests the neglect of radiation during the semiclassical cascade.
Radiation before the next split. In the rest frame set Δ u = Δ v = L \Delta u=\Delta v=L Δ u = Δ v = L , and write σ = 2 μ \sigma=2\mu σ = 2 μ for the tension of the doubled string. Its mass, period, and period-averaged splitting rate in our model are
m = σ L , T = L , λ s p l i t = Γ s p l i t L . (374) m=\sigma L,\qquad T=L,\qquad\lambda_{\mathrm{split}}=\Gamma_{\mathrm{split}}L. \tag*{(374)} m = σ L , T = L , λ split = Γ split L . (374)
For ℓ s ≪ L ≪ Γ s p l i t − 1 / 2 \ell_s\ll L\ll\Gamma_{\mathrm{split}}^{-1/2} ℓ s ≪ L ≪ Γ split − 1/2 , many oscillations precede a split. If P < Ω P_{<\Omega} P < Ω is the period-averaged power below frequency Ω \Omega Ω , the leading perturbative estimate of the fractional loss is
ε < Ω ≡ P < Ω m λ s p l i t = P < Ω σ Γ s p l i t L 2 . (375) \varepsilon_{<\Omega}\equiv\frac{P_{<\Omega}}{m\lambda_{\mathrm{split}}}=\frac{P_{<\Omega}}{\sigma\Gamma_{\mathrm{split}}L^2}. \tag*{(375)} ε < Ω ≡ m λ split P < Ω = σ Γ split L 2 P < Ω . (375)
This uses the unperturbed motion and splitting clock; small ϵ < Ω \epsilon_{<\Omega} ϵ < Ω makes that treatment self-consistent for the frequency band considered.
The classical spectrum. First consider four-dimensional gravity, with Newton constant G 4 G_4 G 4 . Let χ \chi χ be the cosine of the emission angle relative to the string, ω n = 2 π n / L \omega_n = 2\pi n/L ω n = 2 π n / L , and define the period-normalized Fourier tensor T n μ ν = L − 1 ∫ 0 L d t ∫ d 3 x e i ω n ( t − k ^ ⋅ x ) T μ ν T_n^{\mu\nu} = L^{-1}\int_0^L dt \int d^3x\,e^{i\omega_n(t-\hat{\mathbf{k}}\cdot\mathbf{x})}T^{\mu\nu} T n μν = L − 1 ∫ 0 L d t ∫ d 3 x e i ω n ( t − k ^ ⋅ x ) T μν . Only its 00, 01, 11 components are nonzero. Conservation gives T n 00 = χ 2 T n 11 T_n^{00} = \chi^2T_n^{11} T n 00 = χ 2 T n 11 and T n 01 = χ T n 11 T_n^{01} = \chi T_n^{11} T n 01 = χ T n 11 , so the graviton polarization sum reduces to T n μ ν ∗ T n μ ν − ∣ T n μ μ ∣ 2 / 2 = ∣ T n μ μ ∣ 2 / 2 T_n^{\mu\nu*}T_{n\mu\nu} - \lvert T_n^\mu{}_\mu\rvert^2/2 = \lvert T_n^\mu{}_\mu\rvert^2/2 T n μν ∗ T n μν − ∣ T n μ μ ∣ 2 /2 = ∣ T n μ μ ∣ 2 /2 . The null folds are essential for conservation, although their trace vanishes. The remaining bulk trace is − 2 σ -2\sigma − 2 σ inside the diamond 0 < u , v < L 0<u,v<L 0 < u , v < L ; its Fourier transform has magnitude
∣ T n μ μ ∣ = 4 σ L π 2 n 2 ( 1 − χ 2 ) sin 2 [ π n 2 ( 1 − χ ) ] . (376) \lvert T_n^\mu{}_\mu\rvert= \frac{4\sigma L}{\pi^2n^2(1-\chi^2)}\sin^2\left[\frac{\pi n}{2}(1-\chi)\right]. \tag*{(376)} ∣ T n μ μ ∣ = π 2 n 2 ( 1 − χ 2 ) 4 σ L sin 2 [ 2 π n ( 1 − χ ) ] . (376)
Using the weak-field radiation formula d P n / d Ω k ^ = G 4 ω n 2 ∣ T n μ μ ∣ 2 / ( 2 π ) dP_n/d\Omega_{\hat{\mathbf{k}}} = G_4\omega_n^2\lvert T_n^\mu{}_\mu\rvert^2/(2\pi) d P n / d Ω k ^ = G 4 ω n 2 ∣ T n μ μ ∣ 2 / ( 2 π ) (see, e.g., Ref. [46 ] ), the angular integral becomes
P n = 16 G 4 σ 2 π n ∫ 0 π n / 2 sin 4 y d y y 2 [ 1 − y / ( π n ) ] 2 ∼ 4 G 4 σ 2 n . (377) P_n = \frac{16G_4\sigma^2}{\pi n}\int_0^{\pi n/2}\frac{\sin^4 y\,dy}{y^2[1-y/(\pi n)]^2}\sim\frac{4G_4\sigma^2}{n}. \tag*{(377)} P n = π n 16 G 4 σ 2 ∫ 0 π n /2 y 2 [ 1 − y / ( π n ) ] 2 sin 4 y d y ∼ n 4 G 4 σ 2 . (377)
The asymptotic expression refers to the ideal classical source. Using [ 1 − y / ( π n ) ] − 2 ≤ 4 [1-y/(\pi n)]^{-2}\leq4 [ 1 − y / ( π n ) ] − 2 ≤ 4 and ∫ 0 ∞ sin 4 y d y / y 2 = π / 4 \int_0^\infty\sin^4 y\,dy/y^2=\pi/4 ∫ 0 ∞ sin 4 y d y / y 2 = π /4 gives the bound
P < Ω ≤ 16 G 4 σ 2 H N Ω , ϵ < Ω ≤ 16 G 4 σ Γ s p l i t L 2 H N Ω , N Ω = ⌊ Ω L 2 π ⌋ , (378) P_{<\Omega}\leq16G_4\sigma^2H_{N_\Omega},\qquad\epsilon_{<\Omega}\leq\frac{16G_4\sigma}{\Gamma_{\mathrm{split}}L^2}H_{N_\Omega},\qquad N_\Omega=\left\lfloor\frac{\Omega L}{2\pi}\right\rfloor, \tag*{(378)} P < Ω ≤ 16 G 4 σ 2 H N Ω , ϵ < Ω ≤ Γ split L 2 16 G 4 σ H N Ω , N Ω = ⌊ 2 π Ω L ⌋ , (378)
where H N = ∑ n = 1 N n − 1 ≤ 1 + ln N H_N=\sum_{n=1}^N n^{-1}\leq1+\ln N H N = ∑ n = 1 N n − 1 ≤ 1 + ln N for N ≥ 1 N\geq1 N ≥ 1 . Choose L − 1 ≪ Ω ≪ ℓ s − 1 L^{-1}\ll\Omega\ll\ell_s^{-1} L − 1 ≪ Ω ≪ ℓ s − 1 , within the four-dimensional effective regime: Ω \Omega Ω must also lie below the first relevant compactification excitation scale. At fixed compactification data, G 4 σ = O ( g s 2 ) G_4\sigma=O(g_s^2) G 4 σ = O ( g s 2 ) and Γ s p l i t = O ( g s 2 / ℓ s 2 ) \Gamma_{\mathrm{split}}=O(g_s^2/\ell_s^2) Γ split = O ( g s 2 / ℓ s 2 ) , so ϵ < Ω = O ( ( ℓ s / L ) 2 [ 1 + ln ( Ω L ) ] ) \epsilon_{<\Omega}=O((\ell_s/L)^2[1+\ln(\Omega L)]) ϵ < Ω = O (( ℓ s / L ) 2 [ 1 + ln ( Ω L )]) . Thus a macroscopic yo-yo splits before losing an appreciable fraction of its mass to these modes. The logarithm comes from radiation concentrated near the null-fold directions. For a fixed small Ω ℓ s \Omega\ell_s Ω ℓ s , its growth is ln ( L / ℓ s ) + O ( 1 ) \ln(L/\ell_s)+O(1) ln ( L / ℓ s ) + O ( 1 ) .
The ultraviolet contribution. The ideal four-dimensional spectrum has a logarithmically divergent harmonic sum. The total loss therefore depends on the quantum spectrum and any modes resolving the compact dimensions. Iengo’s maximal-angular-momentum folded-string example agrees with the classical spectrum at low emitted energy and falls off more rapidly at higher energy [47 ] ; extending this behavior to longitudinal yo-yos requires a calculation for that state.
The kinematic emission limit is ω = ( m 2 − m ′ 2 ) / ( 2 m ) ≤ m / 2 \omega=(m^2-m'^2)/(2m)\leq m/2 ω = ( m 2 − m ′2 ) / ( 2 m ) ≤ m /2 , potentially well above ℓ s − 1 \ell_s^{-1} ℓ s − 1 . If the quantum power is bounded by an envelope of order G 4 σ 2 / n G_4\sigma^2/n G 4 σ 2 / n , even extending it to n max = O ( m L ) = O ( ( L / ℓ s ) 2 ) n_{\max}=O(mL)=O((L/\ell_s)^2) n m a x = O ( m L ) = O (( L / ℓ s ) 2 ) gives only ln n max = 2 ln ( L / ℓ s ) + O ( 1 ) \ln n_{\max}=2\ln(L/\ell_s)+O(1) ln n m a x = 2 ln ( L / ℓ s ) + O ( 1 ) . The suppression of radiative losses then survives. Establishing that envelope, including the narrow fold-direction cones, is the remaining ultraviolet input. The controlled long-wavelength bound supports neglecting gravitational losses in its frequency range.
The splitting rate used in this comparison is the local IFS law (2.1), motivated by Ref. [7 ] . The relative importance of radiative and splitting channels depends on the string state; see Ref. [48 ] .
Terminal response and the crossover
Terminal decay cuts off the hierarchy of massive strings, producing the crossover from the semiclassical power to the late 1 / t 1/t 1/ t trace. The crossover shape is first calculated in the large mass-hierarchy limit. We then hold the microscopic hierarchy fixed and study how the continuing source builds its logarithmic accumulated response. The first limit determines the crossover kernel; the second determines the delay caused by a finite source onset.
Terminal kernels and the crossover
A local terminal law first determines the total response, which is then resolved in invariant radius. Throughout this subsection the background parameters are fixed, final conversion products are traceless, and χ = Γ s p l i t / ( 2 μ ) \chi=\Gamma_{\mathrm{split}}/(2\mu) χ = Γ split / ( 2 μ ) , κ = κ I F S \kappa=\kappa_{\mathrm{IFS}} κ = κ IFS .
Rest-frame conversion and the integrated response. Suppose conversion at proper-time rate ν d e c ( m ) \nu_{\mathrm{dec}}(m) ν dec ( m ) competes with the binary rate χ m \chi m χ m . Put η ( m ) = ν d e c ( m ) / ( χ m ) \eta(m)=\nu_{\mathrm{dec}}(m)/(\chi m) η ( m ) = ν dec ( m ) / ( χ m ) . Both rates acquire the same factor γ − 1 \gamma^{-1} γ − 1 in the laboratory frame, while ∫ W d t = m Δ τ \int W\,dt=m\Delta\tau ∫ W d t = m Δ τ . A node therefore contributes 1 / [ χ ( 1 + η ) ] 1/[\chi(1+\eta)] 1/ [ χ ( 1 + η )] in expectation and splits with probability 1 / ( 1 + η ) 1/(1+\eta) 1/ ( 1 + η ) . First-event conditioning gives the mass renewal equation
[ 1 + η ( m ) ] H ( m ) = 1 + 2 ∫ 0 1 d a ∫ 0 1 d b H ( m a b ) , H ( m ) = χ R t r e e ( m ) . (379) [1+\eta(m)]H(m)=1+2\int_0^1 da\int_0^1 db\,H\left(m\sqrt{ab}\right),\qquad H(m)=\chi R_{\mathrm{tree}}(m). \tag*{(379)} [ 1 + η ( m )] H ( m ) = 1 + 2 ∫ 0 1 d a ∫ 0 1 d b H ( m ab ) , H ( m ) = χ R tree ( m ) . (379)
The physical solution is the minimal nonnegative one. For example, a constant competing ratio η > 1 \eta>1 η > 1 below the terminal scale gives mean offspring number 2 / ( 1 + η ) < 1 2/(1+\eta)<1 2/ ( 1 + η ) < 1 and response H = 1 / ( η − 1 ) H=1/(\eta-1) H = 1/ ( η − 1 ) there.
A second example isolates the response accumulated after fragmentation stops. Stop binary splitting below m t e r m m_{\mathrm{term}} m term , but let each terminal yo-yo convert to traceless products at rest-frame rate ν t e r m ( m ) = η Γ s p l i t m / ( 2 μ ) \nu_{\mathrm{term}}(m)=\eta\Gamma_{\mathrm{split}}m/(2\mu) ν term ( m ) = η Γ split m / ( 2 μ ) , with fixed η > 0 \eta>0 η > 0 . Its expected integrated trace is m / ν t e r m = 2 μ / ( η Γ s p l i t ) m/\nu_{\mathrm{term}}=2\mu/(\eta\Gamma_{\mathrm{split}}) m / ν term = 2 μ / ( η Γ split ) , independent of boost: the time-dilation factor cancels between its trace and its lifetime. A stopped binary tree has one more terminal leaf than internal nodes, so this completion gives
R t r e e ( η ) ( ξ ) = 2 μ Γ s p l i t [ ( 1 + η − 1 ) N i n t ( ξ ) + η − 1 ] . (380) \mathcal{R}_{\mathrm{tree}}^{(\eta)}(\xi)=\frac{2\mu}{\Gamma_{\mathrm{split}}}\left[(1+\eta^{-1})\mathcal{N}_{\mathrm{int}}(\xi)+\eta^{-1}\right]. \tag*{(380)} R tree ( η ) ( ξ ) = Γ split 2 μ [ ( 1 + η − 1 ) N int ( ξ ) + η − 1 ] . (380)
It preserves the hierarchy exponent while quantifying the extra trace accumulated before conversion.
Sharp-cutoff radial kernel. Use the chronological coordinates and weighted path law of Eqs. (C.2)–(C.9). Under P κ \mathbb{P}_{\kappa} P κ , the fractions a , b a,b a , b are independent Beta ( h , 1 ) \operatorname{Beta}(h,1) Beta ( h , 1 ) variables with h = 2 h=\sqrt{2} h = 2 . The affine clock approaches its stationary law π \pi π , whose marginal coordinates U , V U,V U , V have Gamma ( h , 1 ) \operatorname{Gamma}(h,1) Gamma ( h , 1 ) laws. They remain correlated through their common exponential waits. At large boost the occupation mark is γ ( V + ϵ ) \gamma(V+\epsilon) γ ( V + ϵ ) . Trace weighting multiplies the Gamma ( h + 1 , 1 ) \operatorname{Gamma}(h+1,1) Gamma ( h + 1 , 1 ) density of V + ϵ V+\epsilon V + ϵ by ( V + ϵ ) − κ (V+\epsilon)^{-\kappa} ( V + ϵ ) − κ ; the resulting shape is h + 1 − κ = 3 − 2 h+1-\kappa=3-\sqrt{2} h + 1 − κ = 3 − 2 . Keeping masses above m t e r m m_{\mathrm{term}} m term requires V + ϵ ≥ T / γ V+\epsilon\ge T/\gamma V + ϵ ≥ T / γ , giving Eq. (135) . Let S n = ln ( m 0 / m n ) S_n=\ln(m_0/m_n) S n = ln ( m 0 / m n ) , L = ln ( m 0 / m t e r m ) L=\ln(m_0/m_{\mathrm{term}}) L = ln ( m 0 / m term ) , and X ˉ = E κ [ S n + 1 − S n ] = 1 / h \bar{X}=\mathbb{E}_{\kappa}[S_{n+1}-S_n]=1/h X ˉ = E κ [ S n + 1 − S n ] = 1/ h . For the expected number of above-cutoff splits inside radius ρ \rho ρ , the many-to-one identity gives
e − κ L E [ N r a d ( ρ ) ] = ∑ n ≥ 0 E κ [ e − κ ( L − S n ) 1 S n < L 1 R n e S n − L ≤ z ] , z = χ m t e r m ρ . (381) e^{-\kappa L}\mathbb{E}[N_{\mathrm{rad}}(\rho)]=\sum_{n\ge0}\mathbb{E}_{\kappa}\left[e^{-\kappa(L-S_n)}\mathbf{1}_{S_n<L}\mathbf{1}_{R_n e^{S_n-L}\le z}\right],\qquad z=\chi m_{\mathrm{term}}\rho. \tag*{(381)} e − κ L E [ N rad ( ρ )] = n ≥ 0 ∑ E κ [ e − κ ( L − S n ) 1 S n < L 1 R n e S n − L ≤ z ] , z = χ m term ρ . (381)
Here R n R_n R n is the split mark in Eq. (C.3). The marked renewal limit replaces the sum by X ˉ − 1 ∫ 0 ∞ e − κ u P π ( R ≤ z e u ) d u \bar{X}^{-1}\int_0^\infty e^{-\kappa u}\mathbb{P}_{\pi}(R\le ze^u)\,\mathrm{d}u X ˉ − 1 ∫ 0 ∞ e − κ u P π ( R ≤ z e u ) d u . One direct justification is to truncate the clock to its last K K K generations. The remaining ancestral mass sum is independent of that finite block; its renewal density is h ( 1 − e − 4 h s ) h(1-e^{-4hs}) h ( 1 − e − 4 h s ) , since each mass step is gamma distributed with shape two and rate 2 h 2h 2 h . First let L → ∞ L\to\infty L → ∞ , then K → ∞ K\to\infty K → ∞ . The affine contraction restores the stationary clock, and the integrable weight e − κ u e^{-\kappa u} e − κ u controls the large-u u u tail. Thus
F s h a r p ( z ) = E π [ min { 1 , ( z / R ) κ } ] , R = ( U + ϵ ) ( V + ϵ ) , F_{\mathrm{sharp}}(z)=\mathbb{E}_{\pi}\left[\min\{1,(z/R)^{\kappa}\}\right],\qquad R=\sqrt{(U+\epsilon)(V+\epsilon)}, F sharp ( z ) = E π [ min { 1 , ( z / R ) κ } ] , R = ( U + ϵ ) ( V + ϵ ) ,
e − κ L E [ N r a d ( ρ ) ] ⟶ F s h a r p ( z ) κ X ˉ . (382) e^{-\kappa L}\mathbb{E}[N_{\mathrm{rad}}(\rho)]\longrightarrow\frac{F_{\mathrm{sharp}}(z)}{\kappa\bar{X}}. \tag*{(382)} e − κ L E [ N rad ( ρ )] ⟶ κ X ˉ F sharp ( z ) . (382)
Equivalently, F s h a r p F_{\mathrm{sharp}} F sharp is the distribution function of Z r = R e − E κ Z_r=Re^{-E_{\kappa}} Z r = R e − E κ , where E κ ∼ Exp ( κ ) E_{\kappa}\sim\operatorname{Exp}(\kappa) E κ ∼ Exp ( κ ) is independent of the stationary clock and ϵ ∼ Exp ( 1 ) \epsilon\sim\operatorname{Exp}(1) ϵ ∼ Exp ( 1 ) .
Integrating the boost orbit of a radial event gives 2 arcosh ( t / ρ ) 1 ρ < t 2\operatorname{arcosh}(t/\rho)\mathbf{1}_{\rho<t} 2 arcosh ( t / ρ ) 1 ρ < t . Hence
Ψ s h a r p ( T ) = 2 E [ arcosh ( T / Z r ) 1 Z r < T ] = 2 ∫ 0 1 F s h a r p ( T u ) u 1 − u 2 d u . (383) \Psi_{\mathrm{sharp}}(T)=2\mathbb{E}\left[\operatorname{arcosh}(T/Z_r)\mathbf{1}_{Z_r<T}\right]=2\int_0^1\frac{F_{\mathrm{sharp}}(Tu)}{u\sqrt{1-u^2}}\,\mathrm{d}u. \tag*{(383)} Ψ sharp ( T ) = 2 E [ arcosh ( T / Z r ) 1 Z r < T ] = 2 ∫ 0 1 u 1 − u 2 F sharp ( T u ) d u . (383)
The unbounded logarithmic reward is controlled by a negative moment of any order 0 < r < κ 0<r<\kappa 0 < r < κ : R ≥ ϵ R\ge\epsilon R ≥ ϵ gives E [ R − r ] ≤ Γ ( 1 − r ) \mathbb{E}[R^{-r}]\le\Gamma(1-r) E [ R − r ] ≤ Γ ( 1 − r ) , and the renewal weight becomes e − ( κ − r ) u e^{-(\kappa-r)u} e − ( κ − r ) u . The same limit therefore applies to the accumulated response. For the fixed scaled entrance law of Eq. (C.5), source integration gives
A t e r m = 2 μ Γ s p l i t ξ ∗ κ S ( κ ) κ X ˉ , C e a r l y = B ( κ / 2 , 1 / 2 ) E [ R − κ ] , C s h a r p = 2 ln 2 − 2 E [ ln R ] + 2 κ . (384) \begin{aligned}
\mathcal{A}_{\mathrm{term}} &= \frac{2\mu}{\Gamma_{\mathrm{split}}}\frac{\xi_*^\kappa\mathcal{S}(\kappa)}{\kappa\bar{X}},\\
C_{\mathrm{early}} &= \mathrm{B}(\kappa/2,1/2)\mathbb{E}[R^{-\kappa}], \qquad C_{\mathrm{sharp}} = 2\ln2 - 2\mathbb{E}[\ln R] + \frac{2}{\kappa}. \tag*{(384)}
\end{aligned} A term C early = Γ split 2 μ κ X ˉ ξ ∗ κ S ( κ ) , = B ( κ /2 , 1/2 ) E [ R − κ ] , C sharp = 2 ln 2 − 2 E [ ln R ] + κ 2 . (384)
The stopped node-count bound provides domination by the entrance mass moment S ( κ ) \mathcal{S}(\kappa) S ( κ ) . The limit first sends ξ ∗ = 2 μ / ( m t e r m Γ s p l i t ) \xi_* = 2\mu/(m_{\mathrm{term}}\sqrt{\Gamma_{\mathrm{split}}}) ξ ∗ = 2 μ / ( m term Γ split ) to infinity at fixed T T T , with the scaled entrance law held fixed. The small- and large-T T T limits then describe this limiting kernel. At fixed string scale, 2 μ / Γ s p l i t ∼ g s − 2 2\mu/\Gamma_{\mathrm{split}} \sim g_s^{-2} 2 μ / Γ split ∼ g s − 2 and ξ ∗ ∼ g s − 1 \xi_* \sim g_s^{-1} ξ ∗ ∼ g s − 1 , so A t e r m ∼ g s − 2 − κ \mathcal{A}_{\mathrm{term}} \sim g_s^{-2-\kappa} A term ∼ g s − 2 − κ : the response per node and the number of nodes supply separate enhancements. Numerical evaluation of this limiting kernel is described in App. § 20 .§ 7 .
The late response for a general terminal law. For a positive finite radial trace measure, normalize its weight ∑ j m j d τ j \sum_j m_j\mathrm{d}\tau_j ∑ j m j d τ j to one and again denote the sampled dimensionless radius by Z r Z_r Z r . Equation (H.5) then holds with its corresponding distribution function, irrespective of the local conversion mechanism. If F ( z ) ∼ a z κ F(z) \sim az^\kappa F ( z ) ∼ a z κ at small z z z , the early power is a B ( κ / 2 , 1 / 2 ) T κ a\mathrm{B}(\kappa/2,1/2)T^\kappa a B ( κ /2 , 1/2 ) T κ . For the late limit, assume E [ ∣ ln Z r ∣ ] < ∞ \mathbb{E}[|\ln Z_r|] < \infty E [ ∣ ln Z r ∣ ] < ∞ , set C t e r m = 2 ln 2 − 2 E [ ln Z r ] C_{\mathrm{term}} = 2\ln2 - 2\mathbb{E}[\ln Z_r] C term = 2 ln 2 − 2 E [ ln Z r ] , and subtract the logarithm inside the expectation. The exact remainder is
Ψ ( T ) − 2 ln T − C t e r m = E [ D ( Z r / T ) ] , D ( x ) = { 2 ln [ ( 1 + 1 − x 2 ) / 2 ] , 0 < x < 1 , 2 ln ( x / 2 ) , x ≥ 1. (385) \begin{aligned}
\Psi(T) - 2\ln T - C_{\mathrm{term}} &= \mathbb{E}[D(Z_r/T)],\\
D(x) &=
\begin{cases}
2\ln[(1+\sqrt{1-x^2})/2], & 0<x<1,\\
2\ln(x/2), & x\ge1.
\end{cases} \tag*{(385)}
\end{aligned} Ψ ( T ) − 2 ln T − C term D ( x ) = E [ D ( Z r / T )] , = { 2 ln [( 1 + 1 − x 2 ) /2 ] , 2 ln ( x /2 ) , 0 < x < 1 , x ≥ 1. (385)
Since D ( x ) = O ( x 2 ) D(x) = O(x^2) D ( x ) = O ( x 2 ) at zero and grows logarithmically at infinity, any finite radial moment of order 0 < p ≤ 2 0 < p \le2 0 < p ≤ 2 gives
∣ Ψ ( T ) − 2 ln T − C t e r m ∣ ≤ K p E [ Z r p ] T p . (386) \left|\Psi(T) - 2\ln T - C_{\mathrm{term}}\right| \le\frac{K_p\mathbb{E}[Z_r^p]}{T^p}. \tag*{(386)} ∣ Ψ ( T ) − 2 ln T − C term ∣ ≤ T p K p E [ Z r p ] . (386)
For the sharp law, the gamma marginals of U + ϵ U+\epsilon U + ϵ , V + ϵ V+\epsilon V + ϵ give all positive moments of R R R , hence of Z r Z_r Z r . Using E [ U V ] = [ h / ( h + 1 ) ] 2 ( E [ U V ] + 2 h + 2 ) \mathbb{E}[UV] = [h/(h+1)]^2(\mathbb{E}[UV]+2h+2) E [ U V ] = [ h / ( h + 1 ) ] 2 ( E [ U V ] + 2 h + 2 ) and D ( x ) = − x 2 / 2 + O ( x 4 ) D(x) = -x^2/2 + O(x^4) D ( x ) = − x 2 /2 + O ( x 4 ) yields
E [ Z r 2 ] = κ κ + 2 E [ R 2 ] = 8 + 5 2 7 , Ψ s h a r p ( T ) = 2 ln T + C s h a r p − 8 + 5 2 14 T 2 + O ( T − 4 ) . (387) \begin{aligned}
\mathbb{E}[Z_r^2] &= \frac{\kappa}{\kappa+2}\mathbb{E}[R^2] = \frac{8+5\sqrt{2}}{7},\\
\Psi_{\mathrm{sharp}}(T) &= 2\ln T + C_{\mathrm{sharp}} - \frac{8+5\sqrt{2}}{14T^2} + O(T^{-4}). \tag*{(387)}
\end{aligned} E [ Z r 2 ] Ψ sharp ( T ) = κ + 2 κ E [ R 2 ] = 7 8 + 5 2 , = 2 ln T + C sharp − 14 T 2 8 + 5 2 + O ( T − 4 ) . (387)
The terminal law enters the offset and convergence through these radial moments, while the finite total response fixes the logarithmic slope.
Pressure on late time slices. The accumulated response also determines the instantaneous terminal pressure. For the collinear massless completion, Eq. (F.10) gives 0 ≤ z f ( z ) ≤ I ( z ) ≤ I ∞ 0 \le zf(z) \le I(z) \le I_{\infty} 0 ≤ z f ( z ) ≤ I ( z ) ≤ I ∞ . Writing S = t 2 S=t^2 S = t 2 in Eq. (115) ,
t W ( t ) 4 = ∫ 0 S f ( z ) 1 − z / S d z . \frac{tW(t)}{4}=\int_{0}^{S}\frac{f(z)}{\sqrt{1-z/S}}\,\mathrm{d}z. 4 t W ( t ) = ∫ 0 S 1 − z / S f ( z ) d z .
For fixed 0 < ε < 1 0<\varepsilon<1 0 < ε < 1 , dominated convergence on 0 ≤ z ≤ ( 1 − ε ) S 0\le z\le(1-\varepsilon)S 0 ≤ z ≤ ( 1 − ε ) S gives I ∞ I_{\infty} I ∞ as S → ∞ S\to\infty S → ∞ . The endpoint contribution is at most 2 I ∞ ε / ( 1 − ε ) 2I_{\infty}\sqrt{\varepsilon}/(1-\varepsilon) 2 I ∞ ε / ( 1 − ε ) , by f ( z ) ≤ I ∞ / z f(z)\le I_{\infty}/z f ( z ) ≤ I ∞ / z . Taking S → ∞ S\to\infty S → ∞ and then ε → 0 \varepsilon\to0 ε → 0 proves W ( t ) ∼ 4 I ∞ / t W(t)\sim4I_{\infty}/t W ( t ) ∼ 4 I ∞ / t , including the remnant.
Logarithmic emission and the accumulated response
The persistent remnant emits a constant mean number of trees per logarithmic interval of age. Their finite responses therefore add to a constant response per logarithmic interval at late times, even when individual trees have long and unequal lifetimes. We derive this statement at fixed microscopic parameters, including the effect of a finite source onset.
Let Q > 0 Q>0 Q > 0 be the optical age measured from the IFS birth, with Q ˙ > 0 \dot{Q}>0 Q ˙ > 0 , and start the stationary source at Q 0 > 0 Q_{0}>0 Q 0 > 0 : d ⟨ N 1 ⟩ = ν d Q e / Q e \mathrm{d}\langle N_{1}\rangle=\nu\,\mathrm{d}Q_{\mathrm{e}}/Q_{\mathrm{e}} d ⟨ N 1 ⟩ = ν d Q e / Q e , with ν = 2 \nu=2 ν = 2 for the two arms. Scale covariance gives the source-averaged tree response
d R t r e e = F ( y ) d y , y = Q / Q e − 1 ≥ 0 , R t r e e = ∫ 0 ∞ F ( y ) d y < ∞ , F ≥ 0. (388) \mathrm{d}\mathcal{R}_{\mathrm{tree}}=\mathcal{F}(y)\,\mathrm{d}y,\qquad y=Q/Q_{\mathrm{e}}-1\ge0,\qquad\mathcal{R}_{\mathrm{tree}}=\int_{0}^{\infty}\mathcal{F}(y)\,\mathrm{d}y<\infty,\qquad\mathcal{F}\ge0. \tag*{(388)} d R tree = F ( y ) d y , y = Q / Q e − 1 ≥ 0 , R tree = ∫ 0 ∞ F ( y ) d y < ∞ , F ≥ 0. (388)
In logarithmic age and delay, x = ln ( Q / Q 0 ) x=\ln(Q/Q_{0}) x = ln ( Q / Q 0 ) and z = ln ( Q / Q e ) z=\ln(Q/Q_{\mathrm{e}}) z = ln ( Q / Q e ) , define
g ( z ) = e z F ( e z − 1 ) , ∫ 0 ∞ g ( z ) d z = R t r e e . (389) g(z)=e^{z}\mathcal{F}(e^{z}-1),\qquad\int_{0}^{\infty}g(z)\,\mathrm{d}z=\mathcal{R}_{\mathrm{tree}}. \tag*{(389)} g ( z ) = e z F ( e z − 1 ) , ∫ 0 ∞ g ( z ) d z = R tree . (389)
Convolving the constant emission rate with this delay profile yields
J d e s c ( x ) = ν ∫ 0 x g ( z ) d z ⟶ ν R t r e e , \mathcal{J}_{\mathrm{desc}}(x)=\nu\int_{0}^{x}g(z)\,\mathrm{d}z\longrightarrow\nu\mathcal{R}_{\mathrm{tree}}, J desc ( x ) = ν ∫ 0 x g ( z ) d z ⟶ ν R tree ,
w d e s c ( t ) = Q ˙ Q J d e s c ( x ) ∼ ν R t r e e Q ˙ Q , (390) w_{\mathrm{desc}}(t)=\frac{\dot{Q}}{Q}\mathcal{J}_{\mathrm{desc}}(x)\sim\nu\mathcal{R}_{\mathrm{tree}}\frac{\dot{Q}}{Q}, \tag*{(390)} w desc ( t ) = Q Q ˙ J desc ( x ) ∼ ν R tree Q Q ˙ , (390)
R d e s c ( Q ) = ν ∫ 0 x ( x − z ) g ( z ) d z ∼ ν R t r e e ln ( Q / Q 0 ) . (391) \mathcal{R}_{\mathrm{desc}}(Q)=\nu\int_{0}^{x}(x-z)g(z)\,\mathrm{d}z\sim\nu\mathcal{R}_{\mathrm{tree}}\ln(Q/Q_{0}). \tag*{(391)} R desc ( Q ) = ν ∫ 0 x ( x − z ) g ( z ) d z ∼ ν R tree ln ( Q / Q 0 ) . (391)
The last limit follows by dominated convergence applied to ( 1 − z / x ) + g ( z ) (1-z/x)_{+}g(z) ( 1 − z / x ) + g ( z ) . These are large-optical-age limits, Q / Q 0 → ∞ Q/Q_{0}\to\infty Q / Q 0 → ∞ . The direct remnant adds its coefficient C 0 C_{0} C 0 to the logarithmic slope ν R t r e e \nu\mathcal{R}_{\mathrm{tree}} ν R tree .
The delay controls the offset from this limiting logarithm:
ν R t r e e x − R d e s c = ν [ ∫ 0 x z g ( z ) d z + x ∫ x ∞ g ( z ) d z ] . (392) \nu\mathcal{R}_{\mathrm{tree}}x-\mathcal{R}_{\mathrm{desc}}=\nu\left[\int_{0}^{x}zg(z)\,\mathrm{d}z+x\int_{x}^{\infty}g(z)\,\mathrm{d}z\right]. \tag*{(392)} ν R tree x − R desc = ν [ ∫ 0 x z g ( z ) d z + x ∫ x ∞ g ( z ) d z ] . (392)
If M 1 = ∫ 0 ∞ z g ( z ) d z < ∞ M_{1}=\int_{0}^{\infty}zg(z)\,\mathrm{d}z<\infty M 1 = ∫ 0 ∞ z g ( z ) d z < ∞ , the deficit tends to ν M 1 \nu M_{1} ν M 1 . A finite first moment in physical delay gives a rate of convergence, as the following estimate shows for sharp absorption.
Delay moments in the stopped binary cascade. In a stationary flat background, let a root of mass m 0 m_0 m 0 and energy E 0 E_0 E 0 be emitted at t e > 0 t_e > 0 t e > 0 . Stop each branch at m ≤ m t e r m m \le m_{\mathrm{term}} m ≤ m term , and put ξ = m 0 / m t e r m \xi= m_0/m_{\mathrm{term}} ξ = m 0 / m term and c = Γ s p l i t / ( 2 μ ) c = \Gamma_{\mathrm{split}}/(2\mu) c = Γ split / ( 2 μ ) . For p > κ I F S p > \kappa_{\mathrm{IFS}} p > κ IFS , M ( p ) = 2 / ( 1 + p / 2 ) 2 < 1 M(p) = 2/(1+p/2)^2 < 1 M ( p ) = 2/ ( 1 + p /2 ) 2 < 1 . If N n N_n N n counts above-cutoff nodes at generation n n n , comparison with the unrestricted mass tree gives
E [ N n ] ≤ ξ p M ( p ) n , E [ ∑ n ≥ 0 ( n + 1 ) N n ] ≤ ξ p [ 1 − M ( p ) ] 2 . (393) \mathbb{E}[N_n] \le\xi^p M(p)^n,\qquad
\mathbb{E}\left[\sum_{n\geq0}(n+1)N_n\right] \leq\frac{\xi^p}{[1-M(p)]^2}. \tag*{(393)} E [ N n ] ≤ ξ p M ( p ) n , E [ n ≥ 0 ∑ ( n + 1 ) N n ] ≤ [ 1 − M ( p ) ] 2 ξ p . (393)
Indeed, each retained node obeys 1 ≤ ( m v / m t e r m ) p 1 \leq(m_v/m_{\mathrm{term}})^p 1 ≤ ( m v / m term ) p . Energy conservation within the positive-energy tree also gives E v ≤ E 0 E_v \leq E_0 E v ≤ E 0 , so every retained node has
λ v = c m v 2 E v ≥ c m t e r m 2 E 0 , λ v − 1 ≤ τ 0 ≡ E 0 c m t e r m 2 . (394) \lambda_v = c\frac{m_v^2}{E_v} \geq\frac{c m_{\mathrm{term}}^2}{E_0},\qquad
\lambda_v^{-1} \leq\tau_0 \equiv\frac{E_0}{c m_{\mathrm{term}}^2}. \tag*{(394)} λ v = c E v m v 2 ≥ E 0 c m term 2 , λ v − 1 ≤ τ 0 ≡ c m term 2 E 0 . (394)
Conditional on the mass and momentum genealogy, the clocks are independent exponentials. For a node born at delay b v b_v b v and living for τ v \tau_v τ v , W v / λ v = 1 / c W_v/\lambda_v = 1/c W v / λ v = 1/ c gives
E [ W v ∫ 0 τ v ( b v + s ) d s ∣ genealogy ] = 1 c ( ∑ a ≺ v λ a − 1 + λ v − 1 ) . \mathbb{E}\left[W_v\int_0^{\tau_v}(b_v+s)\,\mathrm{d}s\mathrel{\bigg|}\text{genealogy}\right]
=\frac{1}{c}\left(\sum_{a\prec v}\lambda_a^{-1}+\lambda_v^{-1}\right). E [ W v ∫ 0 τ v ( b v + s ) d s genealogy ] = c 1 ( a ≺ v ∑ λ a − 1 + λ v − 1 ) .
Summing over nodes and using the depth bound yields
∫ 0 ∞ T ⟨ W t r e e ( T ) ⟩ d T ≤ E 0 ξ p c 2 m t e r m 2 [ 1 − M ( p ) ] 2 . (395) \int_0^\infty T\langle W_{\mathrm{tree}}(T)\rangle\,\mathrm{d}T
\leq\frac{E_0\xi^p}{c^2m_{\mathrm{term}}^2[1-M(p)]^2}. \tag*{(395)} ∫ 0 ∞ T ⟨ W tree ( T )⟩ d T ≤ c 2 m term 2 [ 1 − M ( p ) ] 2 E 0 ξ p . (395)
The source satisfies E 0 ≤ 2 μ t e E_0 \leq2\mu t_e E 0 ≤ 2 μ t e and has finite mass moments (B.21). Hence its averaged profile has a finite first delay moment,
K 1 ≡ ∫ 0 ∞ y F ( y ) d y = ⟨ 1 t e ∫ 0 ∞ T W t r e e ( T ) d T ⟩ s r c ≤ 2 μ ⟨ ξ p ⟩ s r c c 2 m t e r m 2 [ 1 − M ( p ) ] 2 < ∞ . (396) K_1 \equiv\int_0^\infty yF(y)\,\mathrm{d}y
= \left\langle\frac{1}{t_e}\int_0^\infty T W_{\mathrm{tree}}(T)\,\mathrm{d}T\right\rangle_{\mathrm{src}}
\leq\frac{2\mu\langle\xi^p\rangle_{\mathrm{src}}}{c^2m_{\mathrm{term}}^2[1-M(p)]^2}<\infty. \tag*{(396)} K 1 ≡ ∫ 0 ∞ y F ( y ) d y = ⟨ t e 1 ∫ 0 ∞ T W tree ( T ) d T ⟩ src ≤ c 2 m term 2 [ 1 − M ( p ) ] 2 2 μ ⟨ ξ p ⟩ src < ∞. (396)
This bound uses binary evolution from the root. An initial geometric stage adds its own delay moment and the birth delays of the entrance roots: D r o o t = D p r e + E [ ∑ j ( b j R j + D j ) ] D_{\mathrm{root}}=D_{\mathrm{pre}}+\mathbb{E}\left[\sum_j(b_jR_j+D_j)\right] D root = D pre + E [ ∑ j ( b j R j + D j ) ] , where R j , D j R_j,D_j R j , D j are their integrated response and first delay moment measured from birth. Finiteness of this sum, and of any added terminal delay moment, supplies the corresponding extension.
The delayed logarithmic onset. Since ln ( 1 + y ) ≤ y \ln(1+y)\leq y ln ( 1 + y ) ≤ y , M 1 ≤ K 1 M_1\leq K_1 M 1 ≤ K 1 . Moreover, K = ∫ 0 ∞ e z g ( z ) d z = R t r e e + K 1 < ∞ K=\int_0^\infty e^z g(z)\,\mathrm{d}z=R_{\mathrm{tree}}+K_1<\infty K = ∫ 0 ∞ e z g ( z ) d z = R tree + K 1 < ∞ , so the exact convolution gives
0 ≤ ν R t r e e − J d e s c ( x ) ≤ ν K e − x , 0\leq\nu R_{\mathrm{tree}}-J_{\mathrm{desc}}(x)\leq\nu K e^{-x}, 0 ≤ ν R tree − J desc ( x ) ≤ ν K e − x ,
0 ≤ R d e s c − ν R t r e e x + ν M 1 = ν ∫ x ∞ ( z − x ) g ( z ) d z ≤ ν K e e − x . (397) 0\leq R_{\mathrm{desc}}-\nu R_{\mathrm{tree}}x+\nu M_1
=\nu\int_x^\infty(z-x)g(z)\,\mathrm{d}z
\leq\frac{\nu K}{e}e^{-x}. \tag*{(397)} 0 ≤ R desc − ν R tree x + ν M 1 = ν ∫ x ∞ ( z − x ) g ( z ) d z ≤ e ν K e − x . (397)
The second inequality uses u e − u ≤ e − 1 u e^{-u} \le e^{-1} u e − u ≤ e − 1 . Equivalently,
R d e s c ( Q ) = ν R t r e e ln ( Q / Q 0 ) − ν M 1 + O ( Q 0 / Q ) , w d e s c = Q ˙ Q [ ν R t r e e + O ( Q 0 / Q ) ] . (398) \mathcal{R}_{\mathrm{desc}}(Q) = \nu R_{\mathrm{tree}} \ln(Q/Q_0) - \nu M_1 + O(Q_0/Q),
\qquad
w_{\mathrm{desc}} = \frac{\dot{Q}}{Q}\left[\nu R_{\mathrm{tree}} + O(Q_0/Q)\right]. \tag*{(398)} R desc ( Q ) = ν R tree ln ( Q / Q 0 ) − ν M 1 + O ( Q 0 / Q ) , w desc = Q Q ˙ [ ν R tree + O ( Q 0 / Q ) ] . (398)
Thus the finite mean delay shifts the onset scale to Q e f f = Q 0 exp ( M 1 / R t r e e ) Q_{\mathrm{eff}} = Q_0 \exp(M_1/R_{\mathrm{tree}}) Q eff = Q 0 exp ( M 1 / R tree ) , leaving the logarithmic slope unchanged. The constants depend on the microscopic hierarchy; these estimates are at fixed coupling. With only finite total response, Eq. (H.13) still gives the leading logarithm, while the offset and convergence rate require delay moments.
A bounded emission rate ν ( x ) → ν \nu(x) \to\nu ν ( x ) → ν gives the same limiting slope by dominated convergence in J d e s c ( x ) = ∫ 0 x ν ( x − z ) g ( z ) d z \mathcal{J}_{\mathrm{desc}}(x) = \int_0^x \nu(x-z)g(z)\,dz J desc ( x ) = ∫ 0 x ν ( x − z ) g ( z ) d z . An integrable source transient changes the cumulative response by a constant when M 1 < ∞ M_1 < \infty M 1 < ∞ ; a finite-response initial population contributes a bounded term. The instantaneous tail additionally uses the source-tail control of § 6.2 . Throughout this argument the tree profile is stationary in logarithmic delay. An evolving remnant source is retained explicitly in the cosmological transport of App. § 19 .
Cosmological transport and a sharp-cutoff bounce
The homogeneous response combines transport of the emitted strings with the source supplied by the remnant. The transport and the criterion for birth-time source matching are formulated first, followed by an application of the sharp-cutoff response to the ESP background. The local string and binary approximations are the same as in the stationary calculation; the long propagation interval is retained in the optical clock.
High-boost transport in a homogeneous background
§ 8.4 introduces the optical clock from the common background dependence of the trace and splitting rate. Here we derive its branching generator, retain the remnant’s emission history, and specify the conditions for source matching. The integrated trace measure w E d t = w s d t s = m d τ s w_E dt = w_s dt_s = m\,d\tau_s w E d t = w s d t s = m d τ s is preserved by the local frame conversion g μ ν ( s ) = Ω 2 g μ ν ( E ) g_{\mu\nu}^{(s)} = \Omega^2 g_{\mu\nu}^{(E)} g μν ( s ) = Ω 2 g μν ( E ) .
The small-string approximation gives the oscillation-averaged worldline action − m ∫ d τ s -m\int d\tau_s − m ∫ d τ s , with constant string-frame mass between splits. Spatial homogeneity conserves k = a s p s k = a_s p_s k = a s p s , even when the Einstein-frame mass Ω m \Omega m Ω m varies. Combining this motion with λ s = Γ t m / ( 2 μ γ ) \lambda_s = \Gamma_t m/(2\mu\gamma) λ s = Γ t m / ( 2 μ γ ) and the clock (8.16) gives
λ s d t s = m 2 2 μ k 2 + a s 2 m 2 d Q ˉ = m 2 2 μ k [ 1 + O ( γ − 2 ) ] d Q ˉ . (399) \lambda_s dt_s =
\frac{m^2}{2\mu\sqrt{k^2+a_s^2m^2}}\,d\bar{Q}
=
\frac{m^2}{2\mu k}\left[1+O(\gamma^{-2})\right]d\bar{Q}. \tag*{(399)} λ s d t s = 2 μ k 2 + a s 2 m 2 m 2 d Q ˉ = 2 μ k m 2 [ 1 + O ( γ − 2 ) ] d Q ˉ . (399)
The optical clock retains the history of a s Γ t a_s\Gamma_t a s Γ t ; finite boost changes the rate by the factor [ 1 + ( a s m / k ) 2 ] − 1 / 2 [1+(a_sm/k)^2]^{-1/2} [ 1 + ( a s m / k ) 2 ] − 1/2 . For the live descendants of an IFS born at s s s , the total splitting activity and integrated trace are
J s ( Q ) = ⟨ ∑ l i v e m 2 2 μ k ⟩ , w d e s c , s ( t ) ≃ 2 μ Γ t Q ˙ s ( t ) J s ( Q s ( t ) ) . (400) \mathcal{J}_s(Q)=\left\langle\sum_{\mathrm{live}}\frac{m^2}{2\mu k}\right\rangle,\qquad w_{\mathrm{desc},s}(t)\simeq\frac{2\mu}{\Gamma_t} \dot{Q}_s(t)\mathcal{J}_s(Q_s(t)). \tag*{(400)} J s ( Q ) = ⟨ live ∑ 2 μ k m 2 ⟩ , w desc , s ( t ) ≃ Γ t 2 μ Q ˙ s ( t ) J s ( Q s ( t )) . (400)
On an outward high-boost branch the two daughters have, to leading order,
( m 1 , k 1 ) = ( m a b , a k ) , ( m 2 , k 2 ) = ( m ( 1 − a ) ( 1 − b ) , ( 1 − a ) k ) , (m_1,k_1)=(m\sqrt{ab},ak),\qquad(m_2,k_2)=(m\sqrt{(1-a)(1-b)},(1-a)k), ( m 1 , k 1 ) = ( m ab , ak ) , ( m 2 , k 2 ) = ( m ( 1 − a ) ( 1 − b ) , ( 1 − a ) k ) ,
with independent uniform a , b a,b a , b . For an additive test observable, the branching generator per unit Q Q Q is consequently
( L F ) ( m , k ) = m 2 2 μ k { ∫ 0 1 d a ∫ 0 1 d b [ F ( m a b , a k ) + F ( m ( 1 − a ) ( 1 − b ) , ( 1 − a ) k ) ] − F ( m , k ) } . (401) (\mathcal{L}F)(m,k)=\frac{m^2}{2\mu k}\left\{\int_0^1 da\int_0^1 db\left[F(m\sqrt{ab},ak)+F(m\sqrt{(1-a)(1-b)},(1-a)k)\right]-F(m,k)\right\}. \tag*{(401)} ( L F ) ( m , k ) = 2 μ k m 2 { ∫ 0 1 d a ∫ 0 1 d b [ F ( m ab , ak ) + F ( m ( 1 − a ) ( 1 − b ) , ( 1 − a ) k ) ] − F ( m , k ) } . (401)
Away from absorption, F = m 2 / k F=m^2/k F = m 2 / k satisfies L F = 0 \mathcal{L}F=0 L F = 0 : high-boost splitting conserves this trace moment. Its leading loss is therefore set by terminal conversion, while cosmological redshift supplies the common observation weight. Set F = 0 F=0 F = 0 below a fixed absorbing mass to obtain L t e r m \mathcal{L}_{\mathrm{term}} L term . More generally, a competing proper-time conversion rate ν d e c ( s ) = Γ t m η ( m / m s ) / ( 2 μ ) \nu_{\mathrm{dec}}^{(s)}=\Gamma_t m\eta(m/m_s)/(2\mu) ν dec ( s ) = Γ t m η ( m / m s ) / ( 2 μ ) adds − m 2 η F / ( 2 μ k ) -m^2\eta F/(2\mu k) − m 2 η F / ( 2 μ k ) . The generator remains autonomous if η \eta η is independent of the background. A terminal rate with a different coupling dependence instead leaves a time-dependent η \eta η in the operator. Traceless conversion products contribute no further trace.
Let n s ( d m d k ; Q s ) n_s(dm\,dk;Q_s) n s ( d m d k ; Q s ) be the expected live descendant measure from an IFS born at s s s , and S s ( Q s ) S_s(Q_s) S s ( Q s ) its remnant’s emission measure per unit optical time. The leading homogeneous transport closes as
∂ Q s n s = L t e r m ∗ n s + S s ( Q s ) , n s ( Q ) = e Q L t e r m ∗ n s ( 0 ) + ∫ 0 Q e ( Q − q ) L t e r m ∗ S s ( q ) d q . (402) \partial_{Q_s}n_s=\mathcal{L}_{\mathrm{term}}^*n_s+S_s(Q_s),\qquad n_s(Q)=e^{Q\mathcal{L}_{\mathrm{term}}^*}n_s(0)+\int_0^Qe^{(Q-q)\mathcal{L}_{\mathrm{term}}^*}S_s(q)\,dq. \tag*{(402)} ∂ Q s n s = L term ∗ n s + S s ( Q s ) , n s ( Q ) = e Q L term ∗ n s ( 0 ) + ∫ 0 Q e ( Q − q ) L term ∗ S s ( q ) d q . (402)
The trace is obtained by integrating m 2 / ( 2 μ k ) m^2/(2\mu k) m 2 / ( 2 μ k ) against this measure and applying Eq. (I.2). The leading transport retains the time dependence of redshift and coupling through the optical clock and the remnant’s emission history. The remnant source is also explicit in the same near-null approximation.
Remnant records and emitted marks. Use string-frame conformal null coordinates, d s s 2 = − a s 2 d u d v ds_s^2=-a_s^2du\,dv d s s 2 = − a s 2 d u d v , on the IFS worldsheet. Near an outgoing arm set z = 2 u z=2u z = 2 u . To leading order across the thin strip, Γ t a s 2 d u d v = d z d Q \Gamma_{\mathrm{t}}a_{\mathrm{s}}^{2}du\,dv=dz\,dQ Γ t a s 2 d u d v = d z d Q . Thus its surviving width has record rate z z z per unit Q Q Q , and a record replaces z z z by Y z Yz Y z , with Y Y Y uniform on ( 0 , 1 ) (0,1) ( 0 , 1 ) . The density satisfies
∂ Q p ( z , Q ) = − z p ( z , Q ) + ∫ z ∞ p ( z ′ , Q ) d z ′ , p ( z , Q ) ∼ Q e − Q z . (403) \partial_{Q}p(z,Q)=-zp(z,Q)+\int_{z}^{\infty}p(z',Q)\,dz',\qquad p(z,Q)\sim Qe^{-Qz}. \tag*{(403)} ∂ Q p ( z , Q ) = − z p ( z , Q ) + ∫ z ∞ p ( z ′ , Q ) d z ′ , p ( z , Q ) ∼ Q e − Q z . (403)
Consequently ⟨ z ⟩ ∼ Q − 1 \langle z\rangle\sim Q^{-1} ⟨ z ⟩ ∼ Q − 1 , giving d ⟨ N 1 ⟩ / d Q ∼ 2 / Q d\langle N_{1}\rangle/dQ\sim2/Q d ⟨ N 1 ⟩ / d Q ∼ 2/ Q for two arms and w r e m , s ∼ 4 μ Ω a s / Q s w_{\mathrm{rem},s}\sim4\mu\Omega a_{\mathrm{s}}/Q_{\mathrm{s}} w rem , s ∼ 4 μ Ω a s / Q s . The source onset and strip-width corrections are separate from this late record law.
The second state variable is the comoving fold impulse K K K accumulated since the previous record. The tension supplies d K = 2 μ a s d t s = 2 μ d Q / Γ ( Q ) dK=2\mu a_{\mathrm{s}}dt_{\mathrm{s}}=2\mu\,dQ/\Gamma(Q) d K = 2 μ a s d t s = 2 μ d Q /Γ ( Q ) . Together, the deterministic evolution and record jump are
d z d Q = 0 , d K d Q = 2 μ Γ ( Q ) , jump rate = z , \frac{dz}{dQ}=0,\qquad\frac{dK}{dQ}=\frac{2\mu}{\Gamma(Q)},\qquad\text{jump rate}=z, d Q d z = 0 , d Q d K = Γ ( Q ) 2 μ , jump rate = z ,
( z , K ) ⟼ ( Y z , 0 ) , k e m = K , m e m 2 = 2 μ K ( 1 − Y ) z . (404) (z,K)\longmapsto(Yz,0),\qquad k_{\mathrm{em}}=K,\qquad m_{\mathrm{em}}^{2}=2\mu K(1-Y)z. \tag*{(404)} ( z , K ) ⟼ ( Y z , 0 ) , k em = K , m em 2 = 2 μ K ( 1 − Y ) z . (404)
The mass formula combines the long fold impulse with the locally short opposite null momentum. In particular, m e m 2 / ( 2 μ k e m ) = ( 1 − Y ) z m_{\mathrm{em}}^{2}/(2\mu k_{\mathrm{em}})=(1-Y)z m em 2 / ( 2 μ k em ) = ( 1 − Y ) z : the injected trace moment equals the lost width, independently of the long impulse history. The expected jump measure of the two arms is precisely S s S_{\mathrm{s}} S s in Eq. (4) . It is the first-generation source; any non-dilute entrance stage must still be propagated or matched as in Appendix § 12 .
Over a record interval [ Q i , Q i + 1 ] [Q_i,Q_{i+1}] [ Q i , Q i + 1 ] , put
1 Γ e f f = 1 Δ Q ∫ Q i Q i + 1 d q Γ ( q ) , X = z Δ Q , m e m 2 = 2 μ 2 X ( 1 − Y ) Γ e f f . (405) \frac{1}{\Gamma_{\mathrm{eff}}}=\frac{1}{\Delta Q}\int_{Q_i}^{Q_{i+1}}\frac{dq}{\Gamma(q)},\qquad X=z\Delta Q,\qquad m_{\mathrm{em}}^{2}=\frac{2\mu^{2}X(1-Y)}{\Gamma_{\mathrm{eff}}}. \tag*{(405)} Γ eff 1 = Δ Q 1 ∫ Q i Q i + 1 Γ ( q ) d q , X = z Δ Q , m em 2 = Γ eff 2 μ 2 X ( 1 − Y ) . (405)
At constant coupling the marks in Q Q Q are the flat-space marks even if the scale factor changes substantially. When the coupling varies, the only additional history in these leading marks is the interval average Γ e f f − 1 \Gamma_{\mathrm{eff}}^{-1} Γ eff − 1 . A fixed birth-time source coefficient is therefore appropriate when Γ e f f \Gamma_{\mathrm{eff}} Γ eff stays close to the birth value during formation. The later transport uses the full optical interval Q ( t ) − Q ( t e ) Q(t)-Q(t_{\mathrm{e}}) Q ( t ) − Q ( t e ) .
Terminal absorption selects the high-boost population, with γ ∼ Δ t s / t × ( s ) \gamma\sim\Delta t_{\mathrm{s}}/t_{\times}^{(\mathrm{s})} γ ∼ Δ t s / t × ( s ) in the stationary reference. Here Δ t s \Delta t_{\mathrm{s}} Δ t s is the string-frame age and t × ( s ) ≡ 2 μ / ( Γ t m t e r m ) = Ω t × t_{\times}^{(\mathrm{s})}\equiv2\mu/(\Gamma_{\mathrm{t}}m_{\mathrm{term}})=\Omega t_{\times} t × ( s ) ≡ 2 μ / ( Γ t m term ) = Ω t × is the corresponding string-frame crossover time. Local matching requires small ∣ H s ∣ t × ( s ) |H_{\mathrm{s}}|t_{\times}^{(\mathrm{s})} ∣ H s ∣ t × ( s ) , ∣ ∂ t s ln g s ∣ t × ( s ) |\partial_{t_{\mathrm{s}}}\ln g_{\mathrm{s}}|t_{\times}^{(\mathrm{s})} ∣ ∂ t s ln g s ∣ t × ( s ) , and ∣ ∂ t s H s ∣ t × ( s ) \sqrt{|\partial_{t_{\mathrm{s}}}H_{\mathrm{s}}|t_{\times}^{(\mathrm{s})}} ∣ ∂ t s H s ∣ t × ( s ) . The worldline action also requires negligible background variation across an oscillation in the string’s rest frame, including the boost of the local gradients. Within this regime, Q Q Q retains the subsequent background history. Corrections come from the trace outside the high-boost population and from accumulated finite-boost effects; a quantitative error bound for them remains to be established.
For the stationary optical source, the normalized trace moment has the late form J s ∼ A s / Q s J_{\mathrm{s}}\sim A_{\mathrm{s}}/Q_{\mathrm{s}} J s ∼ A s / Q s , with constant-background descendant coefficient C d e s c , s = 2 μ A s / Γ s C_{\mathrm{desc},\mathrm{s}}=2\mu A_{\mathrm{s}}/\Gamma_{\mathrm{s}} C desc , s = 2 μ A s / Γ s . The remnant strip has width proportional to Q s − 1 Q_s^{-1} Q s − 1 and the same observation weight Ω a s \Omega_{as} Ω a s . Together they give the late response (8.20), with C s = 4 μ / Γ s + 2 R t r e e , s C_s = 4\mu/\Gamma_s + 2R_{\mathrm{tree},s} C s = 4 μ / Γ s + 2 R tree , s , where R t r e e , s R_{\mathrm{tree},s} R tree , s is the source-averaged integrated response of a finite tree. Each semiclassical node contributes a response of order g s − 2 g_s^{-2} g s − 2 , and the hierarchy contains O ( g s − κ I F S ) O(g_s^{-\kappa_{\mathrm{IFS}}}) O ( g s − κ IFS ) nodes, giving C d e s c ∝ g s − 2 2 C_{\mathrm{desc}} \propto g_s^{-2\sqrt{2}} C desc ∝ g s − 2 2 .
Formation fixes this source normalization, while subsequent propagation is retained by Q s ( t ) − Q s ( t e ) Q_s(t)-Q_s(t_e) Q s ( t ) − Q s ( t e ) for emission time t e t_e t e . The birth-time coefficient applies when the splitting rate varies little over the formation intervals supplying most of the observed response. Later propagation may include substantial background evolution while the local transport conditions hold. Under this matching, the accumulated response in § 8.4 satisfies d F s ( Q s ) = ( Γ t / Γ s ) w e v , s d t d\mathcal{F}_s(Q_s) = (\Gamma_t/\Gamma_s)w_{\mathrm{ev},s}dt d F s ( Q s ) = ( Γ t / Γ s ) w ev , s d t . A varying source uses Eqs. (I.6) and (I.4). The following estimate identifies the formation intervals relevant to old IFS cohorts.
Emission times and local source matching
The terminal tail is carried mainly by descendants of first-generation yo-yos emitted early in the IFS’s history. Source matching concerns their formation interval t e − s t_e-s t e − s , whereas the optical clock transports their response over the subsequent interval t − t e t-t_e t − t e . We identify the emission window that supplies the terminal trace, formulate its matching condition, and derive the window’s scaling in the binary reference.
The response-weighted emission window. At fixed parameters, changing variables from emission time to y = Q / Q e − 1 y=Q/Q_e-1 y = Q / Q e − 1 in Eq. (H.10) gives
d R d e s c d Q = ν ∫ Q 0 Q d Q e Q e 2 F ( Q / Q e − 1 ) = ν Q ∫ 0 Q / Q 0 − 1 F ( y ) d y . (406) \frac{d\mathcal{R}_{\mathrm{desc}}}{dQ}
= \nu\int_{Q_0}^{Q} \frac{dQ_e}{Q_e^2}\mathcal{F}(Q/Q_e-1)
= \frac{\nu}{Q}\int_0^{Q/Q_0-1}\mathcal{F}(y)\,dy. \tag*{(406)} d Q d R desc = ν ∫ Q 0 Q Q e 2 d Q e F ( Q / Q e − 1 ) = Q ν ∫ 0 Q / Q 0 − 1 F ( y ) d y . (406)
Once the stationary tail is reached, the fraction supplied by any range of emission times is therefore measured by F ( y ) d y / R t r e e \mathcal{F}(y)dy/R_{\mathrm{tree}} F ( y ) d y / R tree . This weights emissions by their contribution to the observed trace.
Write D = Q / Q e D=Q/Q_e D = Q / Q e for the ratio of observation to emission age. For a fixed small ε > 0 \varepsilon>0 ε > 0 , choose its lower quantile in the response-weighted measure. In the sharp-cutoff binary reference,
Pr r e s p ( D < d ε ) = ε , d ε ( ξ ∗ ) = ξ ∗ 1 + o ( 1 ) , ξ ∗ → ∞ . (407) \Pr_{\mathrm{resp}}(D<d_\varepsilon)=\varepsilon,\qquad d_\varepsilon(\xi_*)=\xi_*^{1+o(1)},\qquad\xi_*\to\infty. \tag*{(407)} resp Pr ( D < d ε ) = ε , d ε ( ξ ∗ ) = ξ ∗ 1 + o ( 1 ) , ξ ∗ → ∞. (407)
Thus a fraction 1 − ε 1-\varepsilon 1 − ε of the terminal trace comes from trees emitted at optical ages Q e ≤ Q / d ε Q_e\le Q/d_\varepsilon Q e ≤ Q / d ε . The finite-coupling quantile specifies the accuracy; its leading power identifies the parametric separation. We formulate source matching in terms of this quantile and then derive its scaling.
Matching the source in an evolving background. For an IFS born at s s s , define t e , ϵ t_{e,\epsilon} t e , ϵ by
Q s ( t e , ϵ ) = Q s ( t ) d ϵ , ϵ s r c ( s , t ; ϵ ) = sup s ≤ u ≤ t e , ϵ ∣ ln Γ u Γ s ∣ . (408) Q_s(t_{e,\epsilon})=\frac{Q_s(t)}{d_\epsilon},\qquad\epsilon_{\rm src}(s,t;\epsilon)=\sup_{s\leq u\leq t_{e,\epsilon}}\left|\ln\frac{\Gamma_u}{\Gamma_s}\right|. \tag*{(408)} Q s ( t e , ϵ ) = d ϵ Q s ( t ) , ϵ src ( s , t ; ϵ ) = s ≤ u ≤ t e , ϵ sup ln Γ s Γ u . (408)
The stationary reference places a fraction 1 − ϵ 1-\epsilon 1 − ϵ of the terminal trace in this early emission window. For those emissions, ϵ s r c ≪ 1 \epsilon_{\rm src}\ll1 ϵ src ≪ 1 bounds every record’s Γ e f f / Γ s \Gamma_{\rm eff}/\Gamma_s Γ eff / Γ s between e − ϵ s r c e^{-\epsilon_{\rm src}} e − ϵ src and e ϵ s r c e^{\epsilon_{\rm src}} e ϵ src . At fixed record marks, m 2 m^2 m 2 and k k k change together, leaving the optical splitting rate r 0 = m 2 / ( 2 μ k ) r_0=m^2/(2\mu k) r 0 = m 2 / ( 2 μ k ) unchanged. The coupled binary histories thus have the same clock and differ only by a mass hierarchy factor within e ± ϵ s r c / 2 e^{\pm\epsilon_{\rm src}/2} e ± ϵ src /2 . Their terminal trace is bracketed by these two cutoff histories, whose large-hierarchy integrated normalizations differ by O ( κ I F S ϵ s r c ) O(\kappa_{\rm IFS}\epsilon_{\rm src}) O ( κ IFS ϵ src ) . The excluded source tail can be evaluated with Eq. (I.6); a uniform comparison bound M M M on its actual versus reference contribution gives total relative error O ( ϵ s r c + M ϵ ) O(\epsilon_{\rm src}+M\epsilon) O ( ϵ src + M ϵ ) . The source-onset truncation in Eq. (I.8) must also be small before using the stationary quantile.
Only the formation interval t e − s t_e-s t e − s enters this matching condition. The subsequent interval t − t e t-t_e t − t e can contain substantial evolution of g s g_s g s and H H H , retained by the high-boost generator through Q s ( t ) − Q s ( t e ) Q_s(t)-Q_s(t_e) Q s ( t ) − Q s ( t e ) under the local conditions of App. § 19 .§ 1 .
If ∣ ∂ t ln Γ ∣ ≲ t b g − 1 |\partial_t\ln\Gamma|\lesssim t_{\rm bg}^{-1} ∣ ∂ t ln Γ∣ ≲ t bg − 1 near birth, a sufficient source condition is t e , ϵ − s ≪ t b g t_{e,\epsilon}-s\ll t_{\rm bg} t e , ϵ − s ≪ t bg . When Q ˉ ˙ \dot{\bar Q} Q ˉ ˙ is approximately constant over that short formation window this reads
Q s ( t ) Q ˉ ˙ ( s ) ≪ d ϵ t b g . (409) \frac{Q_s(t)}{\dot{\bar Q}(s)}\ll d_\epsilon t_{\rm bg}. \tag*{(409)} Q ˉ ˙ ( s ) Q s ( t ) ≪ d ϵ t bg . (409)
If the full optical age is also comparable to Q ˉ ˙ ( s ) ( t − s ) \dot{\bar Q}(s)(t-s) Q ˉ ˙ ( s ) ( t − s ) , the familiar estimate is t − s ≪ g s ( s ) − 1 + o ( 1 ) t b g t-s\ll g_s(s)^{-1+o(1)}t_{\rm bg} t − s ≪ g s ( s ) − 1 + o ( 1 ) t bg . The optical condition is the appropriate criterion when the full background history is substantial. Expansion can suppress older cohorts through population dilution, whereas contraction supplies no such bound. The source condition should therefore be checked in optical age with the actual production history. For the ESP example below, the measured coupling variation is small across the whole production history, so source matching holds directly over that interval.
Emission times selected by the terminal tail. For a root of mass ratio ξ = m 0 / m t e r m > 1 \xi=m_0/m_{\rm term}>1 ξ = m 0 / m term > 1 and rate r 0 = m 0 2 / ( 2 μ k 0 ) r_0=m_0^2/(2\mu k_0) r 0 = m 0 2 / ( 2 μ k 0 ) , follow a trace-weighted daughter at each high-boost split. Its null fractions have density 2 b d a d b 2b\,da\,db 2 b d a d b , so A = − ln a A=-\ln a A = − ln a and C = − ln b C=-\ln b C = − ln b are independent exponentials of rates 1 and 2. Write
L = ln ξ , S n = 1 2 ∑ j = 1 n ( A j + C j ) , V n = ∑ j = 1 n C j , N L = inf { n : S n ≥ L } , L=\ln\xi,\qquad S_n=\frac{1}{2}\sum_{j=1}^{n}(A_j+C_j),\qquad V_n=\sum_{j=1}^{n}C_j,\qquad N_L=\inf\{n:S_n\geq L\}, L = ln ξ , S n = 2 1 j = 1 ∑ n ( A j + C j ) , V n = j = 1 ∑ n C j , N L = inf { n : S n ≥ L } ,
Θ L = ∑ n < N L ϵ n e V n , ϵ n ∼ Exp ( 1 ) . (410) \Theta_L=\sum_{n<N_L}\epsilon_n e^{V_n},\qquad\epsilon_n\sim\operatorname{Exp}(1). \tag*{(410)} Θ L = n < N L ∑ ϵ n e V n , ϵ n ∼ Exp ( 1 ) . (410)
The root’s expected normalized trace at delay Δ Q \Delta Q Δ Q is r 0 Pr ( Θ L > r 0 Δ Q ) r_0\Pr(\Theta_L > r_0\Delta Q) r 0 Pr ( Θ L > r 0 Δ Q ) . Thus a delay sampled from its integrated response has Z = r 0 Δ Q Z = r_0\Delta Q Z = r 0 Δ Q distributed as
p L ( z ) = Pr ( Θ L > z ) E [ Θ L ] , E r e s p [ Z r ] = E [ Θ L 1 + r ] ( 1 + r ) E [ Θ L ] , − 1 < r < 1. (411) p_L(z) = \frac{\Pr(\Theta_L > z)}{\mathbb{E}[\Theta_L]}, \qquad\mathbb{E}_{\mathrm{resp}}[Z^r] = \frac{\mathbb{E}[\Theta_L^{1+r}]}{(1+r)\mathbb{E}[\Theta_L]}, \qquad-1 < r < 1. \tag*{(411)} p L ( z ) = E [ Θ L ] Pr ( Θ L > z ) , E resp [ Z r ] = ( 1 + r ) E [ Θ L ] E [ Θ L 1 + r ] , − 1 < r < 1. (411)
Sampling a time from the integrated trace weights each tagged lifetime by its duration. A short moment argument suffices to determine its leading scale. The joint increment transform determines the renewal exponent:
E [ e q C − p ( A + C ) / 2 ] = 2 ( 1 + p / 2 ) ( 2 − q + p / 2 ) , ϑ ( q ) = q − 3 + ( q − 1 ) 2 + 8 . (412) \mathbb{E}\left[e^{qC-p(A+C)/2}\right] = \frac{2}{(1+p/2)(2-q+p/2)}, \qquad\vartheta(q) = q - 3 + \sqrt{(q-1)^2+8}. \tag*{(412)} E [ e q C − p ( A + C ) /2 ] = ( 1 + p /2 ) ( 2 − q + p /2 ) 2 , ϑ ( q ) = q − 3 + ( q − 1 ) 2 + 8 . (412)
For q q q near 1 and any η > 0 \eta> 0 η > 0 ,
E [ Θ L q ] ≤ C q , η e [ ϑ ( q ) + η ] L , E [ Θ L ] ≍ e κ I F S L . (413) \mathbb{E}[\Theta_L^q] \le C_{q,\eta}e^{[\vartheta(q)+\eta]L}, \qquad\mathbb{E}[\Theta_L] \asymp e^{\kappa_{\mathrm{IFS}}L}. \tag*{(413)} E [ Θ L q ] ≤ C q , η e [ ϑ ( q ) + η ] L , E [ Θ L ] ≍ e κ IFS L . (413)
Indeed, for p > ϑ ( q ) p > \vartheta(q) p > ϑ ( q ) , E [ e q V n ; S n < L ] ≤ e p L ( E [ e q C − p ( A + C ) / 2 ] ) n \mathbb{E}[e^{qV_n}; S_n < L] \le e^{pL}\left(\mathbb{E}[e^{qC-p(A+C)/2}]\right)^n E [ e q V n ; S n < L ] ≤ e p L ( E [ e q C − p ( A + C ) /2 ] ) n , a summable geometric bound. Subadditivity proves the first estimate for q ≤ 1 q \le1 q ≤ 1 . For q > 1 q > 1 q > 1 , Hölder’s inequality with normalized weights proportional to e − δ n e^{-\delta n} e − δ n inserts only e δ ( q − 1 ) n e^{\delta(q-1)n} e δ ( q − 1 ) n ; choosing δ \delta δ sufficiently small preserves summability. The mean is the expected above-cutoff node count, since every node contributes unit expected integrated normalized trace; its exponent is κ I F S = ϑ ( 1 ) \kappa_{\mathrm{IFS}} = \vartheta(1) κ IFS = ϑ ( 1 ) . Finally ϑ ′ ( 1 ) = 1 \vartheta'(1) = 1 ϑ ′ ( 1 ) = 1 . Applying Markov’s inequality to Eq. (I.13) with small positive and negative r r r proves, for each fixed δ > 0 \delta> 0 δ > 0 ,
Pr r e s p { Z ∉ [ ξ 1 − δ , ξ 1 + δ ] } ≤ C δ ξ − c δ , c δ > 0. (414) \Pr_{\mathrm{resp}}\{Z \notin[\xi^{1-\delta},\xi^{1+\delta}]\} \le C_\delta\xi^{-c_\delta}, \qquad c_\delta> 0. \tag*{(414)} resp Pr { Z ∈ / [ ξ 1 − δ , ξ 1 + δ ]} ≤ C δ ξ − c δ , c δ > 0. (414)
For example, the tail exponent is bounded by r 2 + 8 − 2 2 − ∣ r ∣ δ + η < 0 \sqrt{r^2+8}-2\sqrt{2}-|r|\delta+\eta< 0 r 2 + 8 − 2 2 − ∣ r ∣ δ + η < 0 after choosing ∣ r ∣ |r| ∣ r ∣ and then η \eta η small enough.
This leading power survives the first-generation source average. Relabel its two uniform fractions as U , V U,V U , V , with V V V the short null fraction. The late marks obey ξ = ξ ∗ B U V \xi= \xi_* \sqrt{BUV} ξ = ξ ∗ B U V and r 0 Q e = B V r_0Q_e = BV r 0 Q e = B V , where B B B has density B e − B Be^{-B} B e − B . Equation (B.21) and the corresponding mixed moments with ( B V ) − r (BV)^{-r} ( B V ) − r are finite near r = 0 r = 0 r = 0 . Averaging the numerator and denominator of Eq. (I.13) with these marks therefore gives the same concentration for D = Q / Q e = 1 + Z / ( B V ) D = Q/Q_e = 1 + Z/(BV) D = Q / Q e = 1 + Z / ( B V ) , with ξ \xi ξ replaced by ξ ∗ \xi_* ξ ∗ . This proves Eq. (I.9). A different entrance or terminal law uses its own response quantile; the same leading power follows when the additional delay and response weights obey these neighboring moment bounds.
A sharp-cutoff ESP bounce
We apply the crossover calculated in § 6.3 in the homogeneous ESP equations. The calculation uses the large-hierarchy binary reference from its first-generation source, with traceless terminal products and the local small-string transport of App. § 19 .§ 1 . The source fixes the amplitude. The reference model neglects non-dilute entrance and finite-hierarchy corrections.
The response entering the background equations. Let m s = 1 / α ′ m_s = 1/\sqrt{\alpha'} m s = 1/ α ′ , m t e r m = c m s m_{\mathrm{term}} = c m_s m term = c m s , and use
μ = 1 2 π α ′ , Γ s = g s ( s ) 2 α ′ , ξ s = 2 μ Γ s m t e r m = 1 π c g s ( s ) . (415) \mu= \frac{1}{2\pi\alpha'}, \qquad\Gamma_s = \frac{g_s(s)^2}{\alpha'}, \qquad\xi_s = \frac{2\mu}{\sqrt{\Gamma_s}m_{\mathrm{term}}} = \frac{1}{\pi c g_s(s)}. \tag*{(415)} μ = 2 π α ′ 1 , Γ s = α ′ g s ( s ) 2 , ξ s = Γ s m term 2 μ = π c g s ( s ) 1 . (415)
Both the tension and splitting rate carry the same inverse string area. For binary evolution from the first-generation source, Eq. (B.21) fixes S ( κ ) = Γ ( 2 + κ / 2 ) / ( 1 + κ / 2 ) 2 ≃ 0.62691 S(\kappa) = \Gamma(2+\kappa/2)/(1+\kappa/2)^2 \simeq0.62691 S ( κ ) = Γ ( 2 + κ /2 ) / ( 1 + κ /2 ) 2 ≃ 0.62691 , with κ = κ I F S \kappa= \kappa_{\mathrm{IFS}} κ = κ IFS and X ˉ = 1 / 2 \bar{X} = 1/\sqrt{2} X ˉ = 1/ 2 . Define
A s = 2 μ Γ s ξ s κ S ( κ ) κ X ˉ , Q × , s = 2 μ a ( s ) Ω ( s ) m t e r m , Q ∗ , s = a ( s ) Ω ( s ) Γ s . (416) \mathcal{A}_s = \frac{2\mu}{\Gamma_s}\frac{\xi_s^\kappa S(\kappa)}{\kappa\bar{X}}, \qquad Q_{\times,s} = \frac{2\mu a(s)\Omega(s)}{m_{\mathrm{term}}}, \qquad Q_{*,s} = a(s)\Omega(s)\sqrt{\Gamma_s}. \tag*{(416)} A s = Γ s 2 μ κ X ˉ ξ s κ S ( κ ) , Q × , s = m term 2 μ a ( s ) Ω ( s ) , Q ∗ , s = a ( s ) Ω ( s ) Γ s . (416)
The remnant function obeys Ξ ( 0 ) = 0 \Xi(0) = 0 Ξ ( 0 ) = 0 and Ξ ′ ( z ) = 2 D ( z ) \Xi'(z) = 2D(z) Ξ ′ ( z ) = 2 D ( z ) , where D ( z ) = e − z 2 ∫ 0 z e y 2 d y D(z) = e^{-z^2}\int_0^z e^{y^2}\,dy D ( z ) = e − z 2 ∫ 0 z e y 2 d y is Dawson’s integral.
With these normalizations, the accumulated response (163) gives the event trace through Eq. (162) . Its remnant term recovers the stationary result 8 μ D ( Γ t ) / Γ 8\mu D(\sqrt{\Gamma t})/\sqrt{\Gamma} 8 μ D ( Γ t ) / Γ , including both the early growth and the late 4 μ / ( Γ t ) 4\mu/(\Gamma t) 4 μ / ( Γ t ) tail. The convolution is evaluated using the cumulative kernel and its first moment. The continuation to the youngest cohorts and its numerical sensitivity are described in Appendix § 20 .§ 7 .
The full-history integral (159) can be initialized at a finite time by evolving any earlier event population separately and adding its trace. The energy boundary condition also retains any earlier traceless products. In the present benchmark both populations are absent at the initial ESP crossing.
Background and initial data. We evolve the scalar equation with the full optical history and use the shifted Hubble variable J = H + Π / ( 3 2 M P l ) J = H + \Pi/(3\sqrt{2}M_{\mathrm{Pl}}) J = H + Π/ ( 3 2 M Pl ) . Combining the Friedmann constraint, Raychaudhuri equation, and scalar equation eliminates the explicit trace force from its evolution:
J ˙ = − 2 H 2 + ρ χ + W χ , b a l 6 M P l 2 − H Π 2 M P l − Π 2 6 M P l 2 . (417) \dot{J} = -2H^2 + \frac{\rho_\chi+ W_{\chi,\mathrm{bal}}}{6M_{\mathrm{Pl}}^2} - \frac{H\Pi}{\sqrt{2}M_{\mathrm{Pl}}} - \frac{\Pi^2}{6M_{\mathrm{Pl}}^2}. \tag*{(417)} J ˙ = − 2 H 2 + 6 M Pl 2 ρ χ + W χ , bal − 2 M Pl H Π − 6 M Pl 2 Π 2 . (417)
This identity is exact within the negligible-potential approximation; the scalar equation retains the computed trace rather than imposing W = W χ , b a l W = W_{\chi,\mathrm{bal}} W = W χ , bal . The initial ESP crossing has a = 1 a = 1 a = 1 , H = − H 0 H = -H_0 H = − H 0 , Π = − 6 M P l H 0 \Pi= -\sqrt{6}M_{\mathrm{Pl}}H_0 Π = − 6 M Pl H 0 , and no earlier IFS or additional radiation. The ordinary potential is neglected and the ESP particles are stable on the interval. The fiducial parameters are
g s ( E S P ) = g 0 = 10 − 4 , H 0 = 10 − 22 M P l , H 0 ℓ s = 10 − 17 , λ χ = 10 − 3 , c = 1. g_s(\mathrm{ESP}) = g_0 = 10^{-4}, \qquad H_0 = 10^{-22}M_{\mathrm{Pl}}, \qquad H_0\ell_s = 10^{-17}, \qquad\lambda_\chi= 10^{-3}, \qquad c = 1. g s ( ESP ) = g 0 = 1 0 − 4 , H 0 = 1 0 − 22 M Pl , H 0 ℓ s = 1 0 − 17 , λ χ = 1 0 − 3 , c = 1.
Thus α ′ M P l 2 = 10 10 \alpha' M_{\mathrm{Pl}}^2 = 10^{10} α ′ M Pl 2 = 1 0 10 and μ / M P l 2 = 1.59155 × 10 − 11 \mu/M_{\mathrm{Pl}}^2 = 1.59155 \times10^{-11} μ / M Pl 2 = 1.59155 × 1 0 − 11 . The ESP production prescription fixes n χ , ∗ = ( λ χ ∣ Π E S P ∣ ) 3 / 2 / ( 2 π ) 3 = 4.88735 × 10 − 40 M P l 3 n_{\chi,*} = (\lambda_{\chi}|\Pi_{\mathrm{ESP}}|)^{3/2}/(2\pi)^3 = 4.88735 \times10^{-40}M_{\mathrm{Pl}}^3 n χ , ∗ = ( λ χ ∣ Π ESP ∣ ) 3/2 / ( 2 π ) 3 = 4.88735 × 1 0 − 40 M Pl 3 . The IFS production coefficient is β = 1 / ( 32 π 6 ) \beta= 1/(32\pi^6) β = 1/ ( 32 π 6 ) . For the numerical results below, define x ≡ H 0 ( t − t E S P ) x \equiv H_0(t-t_{\mathrm{ESP}}) x ≡ H 0 ( t − t ESP ) .
Appendix J.7 describes the response integration and convergence checks.
Small motion and the bounce. The first dilaton turn occurs at x m ≃ 0.050119 x_m \simeq0.050119 x m ≃ 0.050119 . The reference solution (155) is matched there using ( a m , J m , ϕ m ) (a_m,J_m,\phi_m) ( a m , J m , ϕ m ) , with no subsequent fit. To measure the settled motion, set ϕ c = ϕ ( x = 0.051 ) \phi_c = \phi(x=0.051) ϕ c = ϕ ( x = 0.051 ) and use I : 0.051 ≤ x ≤ 0.13 I: 0.051 \leq x \leq0.13 I : 0.051 ≤ x ≤ 0.13 :
ϵ K = sup I Π 2 2 ρ χ , ϵ ϕ = sup I ∣ ϕ − ϕ c ∣ ϕ E S P − ϕ c . (418) \epsilon_K = \sup_I \frac{\Pi^2}{2\rho_\chi}, \qquad\epsilon_\phi= \sup_I \frac{|\phi-\phi_c|}{\phi_{\mathrm{ESP}}-\phi_c}. \tag*{(418)} ϵ K = I sup 2 ρ χ Π 2 , ϵ ϕ = I sup ϕ ESP − ϕ c ∣ ϕ − ϕ c ∣ . (418)
The fiducial calculation gives ϵ K ≃ 3.4 × 10 − 7 \epsilon_K \simeq3.4 \times10^{-7} ϵ K ≃ 3.4 × 1 0 − 7 and ϵ ϕ ≃ 1.8 × 10 − 3 \epsilon_\phi\simeq1.8 \times10^{-3} ϵ ϕ ≃ 1.8 × 1 0 − 3 , with a bounce at x b ≃ 0.12653 x_b \simeq0.12653 x b ≃ 0.12653 . The frozen-dilaton reference bounces at x b ( 0 ) ≃ 0.12651 x_b^{(0)} \simeq0.12651 x b ( 0 ) ≃ 0.12651 ; on I I I , max ∣ H − H ( 0 ) ∣ / H 0 < 4 × 10 − 4 \max|H-H^{(0)}|/H_0 < 4 \times10^{-4} max ∣ H − H ( 0 ) ∣/ H 0 < 4 × 1 0 − 4 . The sampled range in ln Γ \ln\Gamma ln Γ from the turn through x = 0.13 x=0.13 x = 0.13 is 1.7 × 10 − 4 1.7 \times10^{-4} 1.7 × 1 0 − 4 , so source matching is locally appropriate over the whole production history. The maximum ∣ H ∣ t × |H|t_\times ∣ H ∣ t × is below 5 × 10 − 10 5 \times10^{-10} 5 × 1 0 − 10 . The direct remnant is retained, although descendants supply more than 99% of the trace at the bounce.
The variations in Table 2 preserve the bounce while changing the motion needed to supply its trace. Increasing the coupling weakens the cascade response and increases the motion; lowering the terminal threshold allows more descendants and decreases it.
Table 2. Sharp-cutoff response: fiducial case and parameter variations. All other physical inputs are held fixed; the frame normalization is g 0 = g s ( E S P ) g_0 = g_s(\mathrm{ESP}) g 0 = g s ( ESP ) in each run. Motion diagnostics use the interval specified in the text.
To test the uncertainty in the first-generation entrance normalization, we independently multiply only the descendant term of Eq. (163) by a constant f d e s c f_{\mathrm{desc}} f desc , leaving the direct remnant and all background and production parameters fixed. For f d e s c = 0.3 , 1 , 3 f_{\mathrm{desc}} = 0.3, 1, 3 f desc = 0.3 , 1 , 3 , the bounce times are respectively x b = 0.126551 , 0.126532 , 0.126523 x_b = 0.126551, 0.126532, 0.126523 x b = 0.126551 , 0.126532 , 0.126523 ; the settled kinetic fractions are ϵ K = ( 1.13 × 10 − 6 , 3.39 × 10 − 7 , 1.13 × 10 − 7 ) \epsilon_K = (1.13 \times10^{-6}, 3.39 \times10^{-7}, 1.13 \times10^{-7}) ϵ K = ( 1.13 × 1 0 − 6 , 3.39 × 1 0 − 7 , 1.13 × 1 0 − 7 ) , and the settled relative field drifts are ϵ ϕ = ( 3.34 , 1.83 , 1.06 ) × 10 − 3 \epsilon_\phi= (3.34, 1.83, 1.06) \times10^{-3} ϵ ϕ = ( 3.34 , 1.83 , 1.06 ) × 1 0 − 3 . In all three cases max I ∣ H − H ( 0 ) ∣ / H 0 < 7 × 10 − 4 \max_I |H-H^{(0)}|/H_0 < 7 \times10^{-4} max I ∣ H − H ( 0 ) ∣/ H 0 < 7 × 1 0 − 4 , with the reference still fixed at the first dilaton turn. Including the initial adjustment, the displacement from that turn through the bounce stays below 3.3 × 10 − 3 3.3 \times10^{-3} 3.3 × 1 0 − 3 of the initial distance to the ESP. Halving the integration step from 5 × 10 − 5 5 \times10^{-5} 5 × 1 0 − 5 to 2.5 × 10 − 5 2.5 \times10^{-5} 2.5 × 1 0 − 5 changes the two extreme bounce times by less than 5 × 10 − 9 5 \times10^{-9} 5 × 1 0 − 9 in x x x . The kinetic fraction is approximately inversely proportional to f d e s c f_{\mathrm{desc}} f desc , as expected from near force balance. The sweep isolates the overall response normalization; it does not calculate changes in the microscopic entrance dynamics or crossover shape.
Dual-fluid evolution and expanding limits
String-frame scaling and the dual-fluid transfer. The dilaton work in Eq. (150) is absorbed into ordinary continuity in the string frame. With Ω = g s / g 0 \Omega= g_s/g_0 Ω = g s / g 0 , the scale factor and time are a s = Ω a a_s = \Omega a a s = Ω a and d t s = Ω d t dt_s = \Omega dt d t s = Ω d t . The energy conversion E E = Ω E s E_E = \Omega E_s E E = Ω E s and the volume conversion V s = Ω 3 V E V_s = \Omega^3 V_E V s = Ω 3 V E give ρ ( s ) = Ω − 4 ρ I F S \rho^{(s)} = \Omega^{-4}\rho_{\mathrm{IFS}} ρ ( s ) = Ω − 4 ρ IFS and W ( s ) = Ω − 4 W W^{(s)} = \Omega^{-4}W W ( s ) = Ω − 4 W . Write q ϕ = Ω ˙ / Ω = Π / ( 2 M P l ) q_\phi= \dot{\Omega}/\Omega= \Pi/(\sqrt{2}M_{\mathrm{Pl}}) q ϕ = Ω ˙ /Ω = Π/ ( 2 M Pl ) and Θ s = d ln a s / d t = H + q ϕ = Ω H s \Theta_s = d\ln a_s/dt = H + q_\phi= \Omega H_s Θ s = d ln a s / d t = H + q ϕ = Ω H s . Then
d ρ ( s ) d t + 4 Θ s ρ ( s ) = Ω − 4 ( ρ ˙ I F S + 4 H ρ I F S ) = Θ s W ( s ) . \frac{d\rho^{(s)}}{dt} + 4\Theta_s\rho^{(s)} = \Omega^{-4}\left(\dot{\rho}_{\mathrm{IFS}} + 4H\rho_{\mathrm{IFS}}\right) = \Theta_s W^{(s)}. d t d ρ ( s ) + 4 Θ s ρ ( s ) = Ω − 4 ( ρ ˙ IFS + 4 H ρ IFS ) = Θ s W ( s ) .
or, on an interval with Θ s ≠ 0 \Theta_s \ne0 Θ s = 0 ,
d ρ ( s ) d ln a s + 4 ρ ( s ) = W ( s ) , ϵ ˉ W ≡ − 1 2 d ln W ( s ) d ln a s . (419) \frac{d\rho^{(s)}}{d\ln a_s} + 4\rho^{(s)} = W^{(s)}, \qquad\bar{\epsilon}_W \equiv-\frac{1}{2}\frac{d\ln W^{(s)}}{d\ln a_s}. \tag*{(419)} d ln a s d ρ ( s ) + 4 ρ ( s ) = W ( s ) , ϵ ˉ W ≡ − 2 1 d ln a s d ln W ( s ) . (419)
In Einstein-frame variables,
ϵ ˉ W = − 1 2 W ˙ / W − 4 q ϕ H + q ϕ , ϵ ˉ W ≃ H ϵ V ′ + 2 q ϕ H + q ϕ \bar{\epsilon}_W = -\frac{1}{2}\frac{\dot{W}/W - 4q_\phi}{H + q_\phi}, \qquad\bar{\epsilon}_W \simeq\frac{H\epsilon_{V'} + 2q_\phi}{H + q_\phi} ϵ ˉ W = − 2 1 H + q ϕ W ˙ / W − 4 q ϕ , ϵ ˉ W ≃ H + q ϕ H ϵ V ′ + 2 q ϕ
in potential force balance. {#eq:I.22}
This reduces to ϵ V ′ \epsilon_{V'} ϵ V ′ for slow coupling evolution. Constant Einstein-frame W W W does not by itself give ϵ ˉ W = 0 \bar{\epsilon}_W = 0 ϵ ˉ W = 0 .
For constant ϵ ˉ W ≠ 2 \bar{\epsilon}_W \ne2 ϵ ˉ W = 2 , integration gives ρ ( s ) = W ( s ) / ( 4 − 2 ϵ ˉ W ) + C s / a s 4 \rho^{(s)} = W^{(s)}/(4 - 2\bar{\epsilon}_W) + C_s/a_s^4 ρ ( s ) = W ( s ) / ( 4 − 2 ϵ ˉ W ) + C s / a s 4 . Transforming back yields the dual-fluid prescription including dilaton work,
ρ I F S = ρ I D E + ρ R , p I F S = p I D E + ρ R 3 , \rho_{\mathrm{IFS}} = \rho_{\mathrm{IDE}} + \rho_R, \qquad p_{\mathrm{IFS}} = p_{\mathrm{IDE}} + \frac{\rho_R}{3}, ρ IFS = ρ IDE + ρ R , p IFS = p IDE + 3 ρ R ,
ρ I D E = W 4 − 2 ϵ ˉ W , p I D E = ( − 1 + 2 ϵ ˉ W 3 ) ρ I D E . (420) \rho_{\mathrm{IDE}} = \frac{W}{4 - 2\bar{\epsilon}_W}, \qquad p_{\mathrm{IDE}} = \left(-1 + \frac{2\bar{\epsilon}_W}{3}\right)\rho_{\mathrm{IDE}}. \tag*{(420)} ρ IDE = 4 − 2 ϵ ˉ W W , p IDE = ( − 1 + 3 2 ϵ ˉ W ) ρ IDE . (420)
The homogeneous radiation mode scales as a − 4 a^{-4} a − 4 because Ω 4 a s − 4 = a − 4 \Omega^4 a_s^{-4} = a^{-4} Ω 4 a s − 4 = a − 4 . The trace/null split and this prescription coincide for ϵ ˉ W = 0 \bar{\epsilon}_W = 0 ϵ ˉ W = 0 ; in general, ρ N = ρ R + ϵ ˉ W ρ I D E / 2 \rho_N = \rho_R + \bar{\epsilon}_W\rho_{\mathrm{IDE}}/2 ρ N = ρ R + ϵ ˉ W ρ IDE /2 .
When ϵ ˉ W \bar{\epsilon}_W ϵ ˉ W varies, differentiating the same local definition of ρ I D E \rho_{\mathrm{IDE}} ρ IDE gives
ρ ˙ I D E + 4 H ρ I D E = ( H + q ϕ ) W + 2 W ϵ ˉ ˙ W ( 4 − 2 ϵ ˉ W ) 2 . (421) \dot{\rho}_{\mathrm{IDE}} + 4H\rho_{\mathrm{IDE}} = (H + q_\phi)W + \frac{2W\dot{\bar{\epsilon}}_W}{(4 - 2\bar{\epsilon}_W)^2}. \tag*{(421)} ρ ˙ IDE + 4 H ρ IDE = ( H + q ϕ ) W + ( 4 − 2 ϵ ˉ W ) 2 2 W ϵ ˉ ˙ W . (421)
Using 3 p I D E = ρ I D E − W 3p_{\mathrm{IDE}}=\rho_{\mathrm{IDE}}-W 3 p IDE = ρ IDE − W , the Bianchi equations for the two components are
ρ ˙ I D E + 3 H ( ρ I D E + p I D E ) = Π W 2 M P l − Q R , ρ ˙ R + 4 H ρ R = Q R , Q R = − ϵ ˉ ˙ W 2 − ϵ ˉ W ρ I D E = − 2 W ϵ ˉ ˙ W ( 4 − 2 ϵ ˉ W ) 2 . (422) \begin{aligned}
\dot{\rho}_{\mathrm{IDE}}+3H(\rho_{\mathrm{IDE}}+p_{\mathrm{IDE}})&=\frac{\Pi W}{\sqrt{2}M_{\mathrm{Pl}}}-Q_R,\\
\dot{\rho}_R+4H\rho_R&=Q_R,\qquad Q_R=-\frac{\dot{\bar{\epsilon}}_W}{2-\bar{\epsilon}_W}\rho_{\mathrm{IDE}}=-\frac{2W\dot{\bar{\epsilon}}_W}{(4-2\bar{\epsilon}_W)^2}. \tag*{(422)}
\end{aligned} ρ ˙ IDE + 3 H ( ρ IDE + p IDE ) ρ ˙ R + 4 H ρ R = 2 M Pl Π W − Q R , = Q R , Q R = − 2 − ϵ ˉ W ϵ ˉ ˙ W ρ IDE = − ( 4 − 2 ϵ ˉ W ) 2 2 W ϵ ˉ ˙ W . (422)
Their sum is the IFS Ward identity. For W > 0 W>0 W > 0 , a decreasing exponent sources positive ρ R \rho_R ρ R , and an increasing exponent sources a negative contribution. The accumulated radiation-like stress is
ρ R ( t ) = 1 a 4 ( t ) [ a ∗ 4 ρ R ( t ∗ ) + ∫ t ∗ t d u a 4 ( u ) Q R ( u ) ] . (423) \rho_R(t)=\frac{1}{a^4(t)}\left[a_*^4\rho_R(t_*)+\int_{t_*}^{t}du\,a^4(u)Q_R(u)\right]. \tag*{(423)} ρ R ( t ) = a 4 ( t ) 1 [ a ∗ 4 ρ R ( t ∗ ) + ∫ t ∗ t d u a 4 ( u ) Q R ( u ) ] . (423)
Here Q R Q_R Q R measures transfer between effective components of the same IFS stress. The microscopic radiation from terminal conversion is already included in that stress and is not counted separately by Q R Q_R Q R .
Regular evolution through the onset of production. For arbitrary W ( t ) W(t) W ( t ) , the total IFS density instead has the regular integral representation
ρ I F S ( t ) = 1 a 4 ( t ) [ a ∗ 4 ρ I F S , ∗ + ∫ t ∗ t d u a 4 ( u ) ( H ( u ) + Π ( u ) 2 M P l ) W ( u ) ] . (424) \rho_{\mathrm{IFS}}(t)=\frac{1}{a^4(t)}\left[a_*^4\rho_{\mathrm{IFS},*}+\int_{t_*}^{t}du\,a^4(u)\left(H(u)+\frac{\Pi(u)}{\sqrt{2}M_{\mathrm{Pl}}}\right)W(u)\right]. \tag*{(424)} ρ IFS ( t ) = a 4 ( t ) 1 [ a ∗ 4 ρ IFS , ∗ + ∫ t ∗ t d u a 4 ( u ) ( H ( u ) + 2 M Pl Π ( u ) ) W ( u ) ] . (424)
This expression and the trace/null split require neither division by W W W nor a monotone string-frame scale factor. By contrast, the local IDE parametrization uses
4 − 2 ϵ ˉ W = W ˙ + 4 H W Θ s W , ρ I D E = Θ s W 2 W ˙ + 4 H W . 4-2\bar{\epsilon}_W=\frac{\dot{W}+4HW}{\Theta_sW},\qquad\rho_{\mathrm{IDE}}=\frac{\Theta_sW^2}{\dot{W}+4HW}. 4 − 2 ϵ ˉ W = Θ s W W ˙ + 4 H W , ρ IDE = W ˙ + 4 H W Θ s W 2 .
The logarithmic exponent is undefined at W = 0 W=0 W = 0 or Θ s = 0 \Theta_s=0 Θ s = 0 , and ϵ ˉ W = 2 \bar{\epsilon}_W=2 ϵ ˉ W = 2 makes the local scaling particular solution singular. These are limitations of the parametrization, not necessarily of the total stress. For a constant exponent equal to 2, the string-frame solution contains a s − 4 [ C s + W ( s ) a ∗ 4 ln ( a s / a s , ∗ ) ] a_s^{-4}[C_s+W^{(s)}a_*^4\ln(a_s/a_{s,*})] a s − 4 [ C s + W ( s ) a ∗ 4 ln ( a s / a s , ∗ )] instead.
A local onset model illustrates this transfer without specifying the full production history. As an example, let W = A τ ν [ 1 + O ( τ ) ] W=A\tau^\nu[1+O(\tau)] W = A τ ν [ 1 + O ( τ )] , with τ = t − t 0 > 0 \tau=t-t_0>0 τ = t − t 0 > 0 , A > 0 A>0 A > 0 , ν > 0 \nu>0 ν > 0 , smooth background evolution, Θ s ( t 0 ) = Θ 0 ≠ 0 \Theta_s(t_0)=\Theta_0\ne0 Θ s ( t 0 ) = Θ 0 = 0 , and no inherited IFS stress. Then the exact energy integral and the local IDE definition give
ρ I F S = Θ 0 A ν + 1 τ ν + 1 + O ( τ ν + 2 ) , ρ I D E = Θ 0 A ν τ ν + 1 + O ( τ ν + 2 ) , ρ R = − Θ 0 A ν ( ν + 1 ) τ ν + 1 + O ( τ ν + 2 ) . (425) \begin{aligned}
\rho_{\mathrm{IFS}}&=\frac{\Theta_0A}{\nu+1}\tau^{\nu+1}+O(\tau^{\nu+2}),\\
\rho_{\mathrm{IDE}}&=\frac{\Theta_0A}{\nu}\tau^{\nu+1}+O(\tau^{\nu+2}),\qquad
\rho_R=-\frac{\Theta_0A}{\nu(\nu+1)}\tau^{\nu+1}+O(\tau^{\nu+2}). \tag*{(425)}
\end{aligned} ρ IFS ρ IDE = ν + 1 Θ 0 A τ ν + 1 + O ( τ ν + 2 ) , = ν Θ 0 A τ ν + 1 + O ( τ ν + 2 ) , ρ R = − ν ( ν + 1 ) Θ 0 A τ ν + 1 + O ( τ ν + 2 ) . (425)
Thus Q R = − Θ 0 A τ ν / ν + O ( τ ν + 1 ) Q_R=-\Theta_0A\tau^\nu/\nu+O(\tau^{\nu+1}) Q R = − Θ 0 A τ ν / ν + O ( τ ν + 1 ) builds a radiation-like correction even with zero initial density. Its sign depends on Θ 0 \Theta_0 Θ 0 , whereas ρ N = − W / 4 + O ( τ ν + 1 ) \rho_N=-W/4+O(\tau^{\nu+1}) ρ N = − W /4 + O ( τ ν + 1 ) is negative in either case.
These are local asymptotics for the assumed trace growth; the ESP calculation instead evolves its full retarded response. Through a later zero of 4 − 2 ϵ ˉ W 4-2\bar{\epsilon}_{W} 4 − 2 ϵ ˉ W , the total energy and trace/null equations provide the regular evolution variables.
Finally, in the absence of ESP particles, set ρ ϕ + I F S = Π 2 / 2 + V + ρ I F S \rho_{\phi+\mathrm{IFS}}=\Pi^2/2+V+\rho_{\mathrm{IFS}} ρ ϕ + IFS = Π 2 /2 + V + ρ IFS . The scalar and IFS equations give the exact conserved-sector identity
ρ ˙ ϕ + I F S + 4 H ρ ϕ + I F S = H ( 4 V + W − Π 2 ) . (426) \dot{\rho}_{\phi+\mathrm{IFS}}+4H\rho_{\phi+\mathrm{IFS}}=H(4V+W-\Pi^2). \tag*{(426)} ρ ˙ ϕ + IFS + 4 H ρ ϕ + IFS = H ( 4 V + W − Π 2 ) . (426)
In potential force balance, q ϕ W ≃ − V ˙ q_{\phi}W\simeq-\dot{V} q ϕ W ≃ − V ˙ : the work gained by the strings is lost by the dilaton. The dual-fluid transfer cancels internally and does not alter this accounting.
Dilution of an inherited cohort. Consider an older, finite comoving population after IFS production has ceased. At approximately constant coupling, the late one-event trace is proportional to Q ˙ s / Q s \dot{Q}_{s}/Q_{s} Q ˙ s / Q s and the density dilutes as a − 3 a^{-3} a − 3 . If radiation domination begins at t r t_{r} t r , the optical age is Q s ( t ) = Q s ( t r ) + C Q ∫ t r t a ( u ) d u Q_{s}(t)=Q_{s}(t_{r})+C_{Q}\int_{t_{r}}^{t}a(u)\,\mathrm{d}u Q s ( t ) = Q s ( t r ) + C Q ∫ t r t a ( u ) d u , with C Q = Ω 2 Γ C_{Q}=\Omega^{2}\Gamma C Q = Ω 2 Γ constant in this approximation. With a ( t ) ∝ t 1 / 2 a(t)\propto t^{1/2} a ( t ) ∝ t 1/2 (using the radiation-era time origin), its late behavior, once this increment dominates the inherited age, is
Q s ( t ) ∝ t 3 / 2 , Q ˙ s Q s ≃ 3 2 t . (427) Q_{s}(t)\propto t^{3/2},\qquad\frac{\dot{Q}_{s}}{Q_{s}}\simeq\frac{3}{2t}. \tag*{(427)} Q s ( t ) ∝ t 3/2 , Q s Q ˙ s ≃ 2 t 3 . (427)
The terminal trace consequently scales as W o l d ∝ a − 3 t − 1 ∝ a − 5 W_{\mathrm{old}}\propto a^{-3}t^{-1}\propto a^{-5} W old ∝ a − 3 t − 1 ∝ a − 5 , faster than ρ r a d ∝ a − 4 \rho_{\mathrm{rad}}\propto a^{-4} ρ rad ∝ a − 4 . This suppresses the inherited traceful dilaton force in a sufficiently developed radiation era, but it does not fix the amplitude of the signed a − 4 a^{-4} a − 4 mode in Eq. (150) . If the inherited optical age remains important or the coupling evolves, the general a − 3 ( Γ s / Γ t ) Q ˙ s / Q s a^{-3}(\Gamma_{s}/\Gamma_{t})\dot{Q}_{s}/Q_{s} a − 3 ( Γ s / Γ t ) Q ˙ s / Q s response must instead be retained.
Logarithmic response on the expanding force-balance branch. Suppose H > 0 H>0 H > 0 , the coupling and production rate vary slowly over a Hubble time, and H t × ≪ 1 Ht_{\times}\ll1 H t × ≪ 1 . Writing u = t − s u=t-s u = t − s and taking a ( t − u ) / a ( t ) ≃ e − H u a(t-u)/a(t)\simeq e^{-Hu} a ( t − u ) / a ( t ) ≃ e − H u , the optical clock for nearly constant coupling gives Q ˙ s / Q s ≃ H / ( 1 − e − H u ) \dot{Q}_{s}/Q_{s}\simeq H/(1-e^{-Hu}) Q ˙ s / Q s ≃ H / ( 1 − e − H u ) . Thus the terminal response and the population dilution give
A c a s ≃ β C ( g s ) [ ∫ t × ∞ d u H e − 3 H u 1 − e − H u + O ( 1 ) ] = β C ( g s ) [ ln 1 H t × + O ( 1 ) ] . (428) \begin{aligned}
A_{\mathrm{cas}}\simeq\beta C(g_{s})\left[\int_{t_{\times}}^{\infty}\mathrm{d}u\,\frac{He^{-3Hu}}{1-e^{-Hu}}+O(1)\right] \\
&=\beta C(g_{s})\left[\ln\frac{1}{Ht_{\times}}+O(1)\right]. \tag*{(428)}
\end{aligned} A cas ≃ β C ( g s ) [ ∫ t × ∞ d u 1 − e − H u H e − 3 H u + O ( 1 ) ] = β C ( g s ) [ ln H t × 1 + O ( 1 ) ] . (428)
The finite crossover contribution and the part at H u = O ( 1 ) Hu=O(1) H u = O ( 1 ) affect only the O ( 1 ) O(1) O ( 1 ) term; the logarithm comes from t × ≪ u ≪ H − 1 t_{\times}\ll u\ll H^{-1} t × ≪ u ≪ H − 1 . At fixed W V , b a l W_{\mathrm{V,bal}} W V , bal , the weak-coupling coefficient C ( g s ) ∝ g s − 2 2 C(g_{s})\propto g_{s}^{-2\sqrt{2}} C ( g s ) ∝ g s − 2 2 gives Π ∗ ∝ g s 2 ln [ 1 / ( H t × ) ] \Pi_{*}\propto g_{s}^{\sqrt{2}}\sqrt{\ln[1/(Ht_{\times})]} Π ∗ ∝ g s 2 ln [ 1/ ( H t × )] . This scaling requires H t × ≪ 1 Ht_{\times}\ll1 H t × ≪ 1 throughout. At fixed string-frame scales, t × ( s ) = Ω t × ∝ g s − 2 t_{\times}^{(s)}=\Omega t_{\times}\propto g_{s}^{-2} t × ( s ) = Ω t × ∝ g s − 2 , so it is not a uniform g s → 0 g_{s}\to0 g s → 0 limit at fixed H / Ω H/\Omega H /Ω .
Numerical methods and validation
This appendix collects the numerical tests of the cascade and the homogeneous bounce. It begins with the common treatment of ensemble averages and correlated errors, followed by the implementation and results for each observable. The analytical predictions are derived in the preceding sections and appendices.
Ensemble averages and statistical tests
Each of N N N independent histories supplies a complete observable vector X s X_s X s . We use N = 1,000 N = 1{,}000 N = 1 , 000 for the histogram, live-trace, invariant-mass, ANEC, and spatial-moment fits. Coordinates may be boost bins, time-interval averages, sampled snapshots, or paired observables; correlations within a history are retained. Time-domain fits use exact live-history integrals as specified in App. J.4. For the ANEC grid, integrate piecewise-linear interpolation in each logarithmic coordinate and divide by the interval width. This gives nonnegative weights H j i H_{ji} H j i , with ∑ i H j i = 1 \sum_i H_{ji} = 1 ∑ i H j i = 1 ; use their tensor product in ln U \ln U ln U and ln v \ln v ln v for both data and prediction. The histogram uses direct bin integrals. Estimate the mean and its covariance by
y = 1 N ∑ s X s , Σ ^ = 1 N ( N − 1 ) ∑ s ( X s − y ) ( X s − y ) T . (429) y = \frac{1}{N}\sum_s X_s,\qquad\widehat{\Sigma} = \frac{1}{N(N-1)}\sum_s (X_s-y)(X_s-y)^{\mathsf{T}}. \tag*{(429)} y = N 1 s ∑ X s , Σ = N ( N − 1 ) 1 s ∑ ( X s − y ) ( X s − y ) T . (429)
For the time-domain, ANEC, and spatial-moment fits, use this full covariance without shrinkage or eigenmode removal. For the histogram, write Σ ^ = D R D \widehat{\Sigma} = DRD Σ = D R D , where R R R is the empirical correlation matrix and D i i = Σ ^ i i D_{ii} = \sqrt{\widehat{\Sigma}_{ii}} D ii = Σ ii , and use C = D [ ( 1 − λ ) R + λ I ] D C = D[(1-\lambda)R+\lambda I]D C = D [( 1 − λ ) R + λ I ] D . This preserves the marginal variances of the mean. The histogram estimates λ \lambda λ by Ledoit–Wolf.
With the applicable covariance C C C and model vector m ( θ ) m(\theta) m ( θ ) , fit the free parameters by minimizing
χ 2 ( θ ) = [ y − m ( θ ) ] T C − 1 [ y − m ( θ ) ] . (430) \chi^2(\theta) = [y-m(\theta)]^{\mathsf{T}}C^{-1}[y-m(\theta)]. \tag*{(430)} χ 2 ( θ ) = [ y − m ( θ ) ] T C − 1 [ y − m ( θ )] . (430)
Quoted degrees of freedom count coordinates minus fitted parameters and are nominal: the probabilities below are calibrated by resampling.
Bootstrap calibration. For goodness of fit, draw N N N centered vectors X s − y X_s-y X s − y with replacement and add the fitted null mean m ( θ ^ ) m(\widehat{\theta}) m ( θ ) . Each bootstrap sample re-estimates the covariance with the same prescription, including any estimated shrinkage, and refits the model. We use B b o o t = 5,000 B_{\mathrm{boot}} = 5{,}000 B boot = 5 , 000 samples for the time-domain and spatial-moment tests, and 9,999 9{,}999 9 , 999 for the histogram, ANEC, and paired pressure–ANEC comparisons. We report p b o o t = ( 1 + n ≥ ) / ( B b o o t + 1 ) p_{\mathrm{boot}} = (1+n_{\geq})/(B_{\mathrm{boot}}+1) p boot = ( 1 + n ≥ ) / ( B boot + 1 ) , where n ≥ n_{\geq} n ≥ counts bootstrap statistics at least as large as the observed one. Nested tests generate samples under the constrained model, refit both models, and calibrate the improvement in the quadratic statistic.
Confidence intervals instead resample uncentered histories. Paired observables use the same seed indices in every resample, preserving their correlations in amplitude ratios and joint fits. Rebinning and window checks reuse the same histories and are correlated diagnostics.
Error bars show marginal standard errors. The interpretation of the coherent trace fluctuations and residuals is given in App. § 20 .§ 4 .
Independent chronological binary benchmark
To test the binary prediction separately from geometric matching, we simulate both daughters at every split, with independent uniform null fractions and exponential proper waits of rate c m cm c m . Lab-frame birth times are inherited along each explicitly simulated tree, independently of the stationary-clock evaluation. Roots have mass m 0 m_0 m 0 , are born at the origin, and have unit entrance intensity per signed rapidity. We integrate over ∣ y 0 ∣ ≤ 16.1181 |y_0| \le16.1181 ∣ y 0 ∣ ≤ 16.1181 , using 40,000 roots in 100 independent, rapidity-stratified batches. There is no generation cap; the 142.4 million processed nodes reach generation 33. These runs form a separate binary ensemble.
The omitted rapidity tail has a uniform analytic bound. Conservation of four-momentum at a binary split, together with convexity of p 2 / E p^2/E p 2 / E , implies ∑ i m i 2 / E i ≤ m p a r e n t 2 / E p a r e n t \sum_i m_i^2/E_i \le m_{\mathrm{parent}}^2/E_{\mathrm{parent}} ∑ i m i 2 / E i ≤ m parent 2 / E parent , so the total trace of a tree never exceeds its root trace m 0 / cosh y 0 m_0/\cosh y_0 m 0 / cosh y 0 . Here p p p denotes longitudinal momentum. For unit entrance intensity per signed rapidity, the ensemble trace W > Y ( t ) W_{>Y}(t) W > Y ( t ) omitted by the cutoff therefore obeys
0 ≤ W > Y ( t ) ≤ m 0 ∫ ∣ y 0 ∣ > Y d y 0 cosh y 0 = 4 m 0 arctan ( e − Y ) ≤ 4 m 0 e − Y . (431) 0 \le W_{>Y}(t) \le m_0 \int_{|y_0|>Y} \frac{dy_0}{\cosh y_0} = 4m_0\arctan(e^{-Y}) \le4m_0e^{-Y}. \tag*{(431)} 0 ≤ W > Y ( t ) ≤ m 0 ∫ ∣ y 0 ∣ > Y cosh y 0 d y 0 = 4 m 0 arctan ( e − Y ) ≤ 4 m 0 e − Y . (431)
At Y = 16.1181 Y = 16.1181 Y = 16.1181 , this is less than 4 × 10 − 7 m 0 4 \times10^{-7}m_0 4 × 1 0 − 7 m 0 at every snapshot. It is an absolute truncation bound, separate from the sampling uncertainties quoted below.
Figure 13 compares snapshots at c m 0 t = 10 3 cm_0t = 10^3 c m 0 t = 1 0 3 , 10 4 10^4 1 0 4 , 10 5 10^5 1 0 5 with the absolute prediction of App. § 13 .§ 4 , with normalization fixed by the specified root population. The reference reward is evaluated with eight independent samples of 65,536 stationary clocks. All 48 mass-bin residuals are within 1.6 combined standard errors. At the latest time, the maximum bin discrepancy is 0.46%, and the total trace differs by 0.18%. Uncertainties use independent batches, retaining sibling and ancestral correlations; no normalization or exponent is fitted to the profile.
The effective trace exponent increases from 0.17112 over the first decade to 0.17146 over the second. The latter has a paired-batch bootstrap 95% interval [0.17121, 0.17170], containing q I F S = 0.171573 q_{\mathrm{IFS}} = 0.171573 q IFS = 0.171573 . These checks test the binary chronological prediction at finite time.
Figure 13. Independent full-tree binary benchmark. Left: differences from absolute predicted mass-bin integrals in 16 logarithmic bins over 1 ≤ γ ≤ 100 1 \le\gamma\le100 1 ≤ γ ≤ 100 ; bars combine simulation and reference-integration standard errors. Right: the rescaled total trace and the predicted constant, with its reference-integration uncertainty shaded. No amplitude is fitted.
Mass-spectrum implementation
In the plotted units M ^ = Γ s p l i t M / ( 2 μ ) \widehat{M} = \sqrt{\Gamma_{\mathrm{split}}}M/(2\mu) M = Γ split M / ( 2 μ ) , the density template is
d M ^ f i t d ln γ = A f i t γ q I F S 1 − γ − 2 F κ o c c ( w γ ) F κ o c c ( 1 ) . (432) \frac{\mathrm{d}\widehat{M}_{\mathrm{fit}}}{\mathrm{d}\ln\gamma}
= A_{\mathrm{fit}} \frac{\gamma^{q_{\mathrm{IFS}}}}{\sqrt{1-\gamma^{-2}}}
\frac{\mathcal{F}_{\kappa}^{\mathrm{occ}}(w_{\gamma})}{\mathcal{F}_{\kappa}^{\mathrm{occ}}(1)}. \tag*{(432)} d ln γ d M fit = A fit 1 − γ − 2 γ q IFS F κ occ ( 1 ) F κ occ ( w γ ) . (432)
The low-boost singularity is integrable; the comparison uses its integral over each measured bin.
For Figure 7 , h r i h_{ri} h r i denotes the live invariant mass in bin i i i of history r r r , in the plotted M ^ \widehat{M} M units. Fit h ˉ i = N − 1 ∑ r h r i \bar{h}_i = N^{-1}\sum_r h_{ri} h ˉ i = N − 1 ∑ r h r i to a f i a f_i a f i , where
f i = ∫ γ i γ i + 1 γ R κ ( γ ) d ln γ . (433) f_i = \int_{\gamma_i}^{\gamma_{i+1}} \gamma\mathcal{R}_{\kappa}(\gamma)\,\mathrm{d}\ln\gamma. \tag*{(433)} f i = ∫ γ i γ i + 1 γ R κ ( γ ) d ln γ . (433)
The amplitude in Eq. (J.4) is A f i t = 2 a F κ o c c ( 1 ) / Z κ A_{\mathrm{fit}} = 2a\mathcal{F}_{\kappa}^{\mathrm{occ}}(1)/\mathcal{Z}_{\kappa} A fit = 2 a F κ occ ( 1 ) / Z κ . Only a a a is fitted, using Eq. (J.2) and the histogram covariance in App. J.1.
We evaluate the stationary recursion with eight independently scrambled Sobol samples of 2 18 2^{18} 2 18 points, truncated at 64 generations. Rapidity integration removes the endpoint singularity; changing the scramble shifts the fiducial 50-bin χ 2 \chi^2 χ 2 by less than 0.085. The geometric simulation uses μ = Γ s p l i t = 1 \mu= \Gamma_{\mathrm{split}} = 1 μ = Γ split = 1 with terminal absorption disabled, a generation cap of 50, and at most 5 × 10 6 5 \times10^6 5 × 1 0 6 explicit layers per history. A parent leaves the snapshot at its earliest geometric daughter-birth time; each daughter enters at its own birth time. These settings specify the finite-time comparison with the dilute binary limit of App. C.4. Across all 1,000 histories, the maximum realized generation was 38 and the maximum number of explicit layers was 534,349; neither computational cap was reached. Bootstrap intervals quantify sampling uncertainty in this finite-time geometric comparison.
Over 1 ≤ γ ≤ 10 5 1 \le\gamma\le10^{5} 1 ≤ γ ≤ 1 0 5 , the fiducial 50-bin fit gives χ 2 = 39.51 \chi^{2} = 39.51 χ 2 = 39.51 for 49 nominal degrees of freedom and p b o o t = 0.486 p_{\mathrm{boot}} = 0.486 p boot = 0.486 . A finer 63-bin comparison, with a half-width final bin, gives 53.81/62 and p b o o t = 0.404 p_{\mathrm{boot}} = 0.404 p boot = 0.404 . Using the full sample covariance for the 50-bin fit gives 51.63/49 and p b o o t = 0.473 p_{\mathrm{boot}} = 0.473 p boot = 0.473 . Each comparison fits one normalization to the same 1,000 histories.
Trace and invariant-mass fits
The ensemble contains paired live-trace and invariant-mass data for 1,000 seeds evolved without terminal absorption to t ^ = 10 6 \hat{t} = 10^{6} t ^ = 1 0 6 . The fiducial fit uses 20 logarithmic intervals from 10 3 10^{3} 1 0 3 to 10 6 10^{6} 1 0 6 . The same histories enter all time-domain fits. Exact interval integration makes the convention at the final time immaterial.
Time averages and covariance. The time-domain observables are exact integrals of each string’s live contribution against d ln t ^ d \ln\hat{t} d ln t ^ , obtained from its birth and death times. We pool complete fine intervals with their logarithmic widths as weights. This uses the whole live history, including changes between sampled snapshots. The model columns are therefore exact interval averages:
f j ( q ) = a j − q − b j − q q ln ( b j / a j ) , g j = a j − 1 − b j − 1 ln ( b j / a j ) (434) f_j(q) = \frac{a_j^{-q} - b_j^{-q}}{q \ln(b_j/a_j)}, \qquad g_j = \frac{a_j^{-1} - b_j^{-1}}{\ln(b_j/a_j)} \tag*{(434)} f j ( q ) = q ln ( b j / a j ) a j − q − b j − q , g j = ln ( b j / a j ) a j − 1 − b j − 1 (434)
where [ a j , b j ] [a_j,b_j] [ a j , b j ] is the j j j th time interval. The leading and corrected trace models are A f j A f_j A f j and A f j + C g j A f_j + C g_j A f j + C g j ; the mass model is M ∞ − B f j M_{\infty} - B f_j M ∞ − B f j . Concatenating the paired trace and mass intervals gives a 40-coordinate joint vector, with the full covariance of App. § 20 .§ 1 . The mass is fitted directly as M ∞ − B t ^ − q M_{\infty} - B\hat{t}^{-q} M ∞ − B t ^ − q ; the figure displays its deficit from the fitted plateau. The displayed deficit errors include the plateau uncertainty and its correlation with the measured mass. All quoted fit uncertainties are statistical.
Amplitudes and the mass–trace relation. The separate corrected-trace and mass fits give A = 4.30 A = 4.30 A = 4.30 [4.23, 4.37], C = − 3.48 C = -3.48 C = − 3.48 [−5.14, −1.88], B = 12.84 B = 12.84 B = 12.84 [11.94, 13.67], and M ∞ = 13.83 M_{\infty} = 13.83 M ∞ = 13.83 [13.61, 14.02] in units μ = Γ s p l i t = 1 \mu= \Gamma_{\mathrm{split}} = 1 μ = Γ split = 1 , with 95% bootstrap intervals in brackets. Thus M ^ ∞ = 6.91 \widehat{M}_{\infty} = 6.91 M ∞ = 6.91 in dimensionless mass units. The free plateau and deficit amplitude limit the mass fit’s sensitivity to changes in the exponent.
The joint fit with independent A A A , B B B , C C C , M ∞ M_{\infty} M ∞ gives χ 2 = 41.01 \chi^{2} = 41.01 χ 2 = 41.01 for 36 nominal degrees of freedom, p b o o t = 0.329 p_{\mathrm{boot}} = 0.329 p boot = 0.329 , and
B / A = 3.004 , B / A ∈ [ 2.774 , 3.236 ] ( 95 % interval ) . (435) B/A = 3.004, \qquad B/A \in[2.774, 3.236] \quad(95\% \text{ interval}). \tag*{(435)} B / A = 3.004 , B / A ∈ [ 2.774 , 3.236 ] ( 95% interval ) . (435)
Imposing the stationary prediction a κ = 3.0774 a_{\kappa}=3.0774 a κ = 3.0774 leaves A A A , C C C , M ∞ M_{\infty} M ∞ free and gives the joint row of Table 1. The increase Δ χ 2 = 0.440 \Delta\chi^{2}=0.440 Δ χ 2 = 0.440 has nested-bootstrap p Δ = 0.528 p_{\Delta}=0.528 p Δ = 0.528 .
Adding C / t ^ C/\hat{t} C / t ^ lowers χ 2 \chi^{2} χ 2 by 16.46, with nested-bootstrap p Δ = 0.0002 p_{\Delta}=0.0002 p Δ = 0.0002 . The fitted correction is negative and about 0.26% of the leading trace at t ^ = 10 3 \hat{t}=10^{3} t ^ = 1 0 3 , decreasing thereafter. It describes this finite window; the asymptotic subleading power remains undetermined.
Source fluctuations and correlated residuals. Early remnant fluctuations leave a persistent imprint on the late trace. For each history, averaging the remnant trace divided by its exact mean 8 D ( t ^ ) 8D(\hat{t}) 8 D ( t ^ ) against d ln t ^ d\ln\hat{t} d ln t ^ over 0.1 ≤ t ^ ≤ 10 0.1\leq\hat{t}\leq10 0.1 ≤ t ^ ≤ 10 gives an onset coefficient correlated with the last live-trace interval at r = 0.88 r=0.88 r = 0.88 (95% paired-bootstrap interval [0.86, 0.90]). A nearly constant amplitude mode accounts for 99.7% of the fractional trace covariance. Consequently the overall level is less precise than differences between times: marginal error bars can obscure well-resolved changes of shape, and a covariance-weighted amplitude fit can have residuals of one sign. All fiducial fits use the ensemble averages with these correlations retained.
Cutoff ANEC fits
The ANEC data consist of A r ( v b ; U a ) \mathcal{A}_{r}(v_b;U_a) A r ( v b ; U a ) on logarithmic grids of 1,000 cutoffs 1 ≤ U a ≤ 10 6 1\leq U_a\leq10^{6} 1 ≤ U a ≤ 1 0 6 and 301 ray coordinates 0.5 ≤ v b ≤ 100 0.5\leq v_b\leq100 0.5 ≤ v b ≤ 100 , for the same 1,000 histories as the pressure fit. All integrals in the analysis range are negative, and the full fitting window lies within the simulated cutoff range. Each cell uses the tensor product of the logarithmic averaging weights in App. § 20 .§ 1 , one for U U U and one for v v v . For α = − ( 1 + q ) / 2 \alpha=-(1+q)/2 α = − ( 1 + q ) /2 and β = ( 1 − q ) / 2 \beta=(1-q)/2 β = ( 1 − q ) /2 , the averaged data and prediction are
y r , j i = − ∑ a , b H i a ( U ) H j b ( v ) A r ( v b ; U a ) , (436) y_{r,ji}=-\sum_{a,b}H^{(U)}_{ia}H^{(v)}_{jb}\mathcal{A}_{r}(v_b;U_a), \tag*{(436)} y r , j i = − a , b ∑ H ia ( U ) H j b ( v ) A r ( v b ; U a ) , (436)
F j i ( q ) = ( ∑ a H i a ( U ) U a α ) ( ∑ b H j b ( v ) v b β ) . (437) F_{ji}(q)=\left(\sum_a H^{(U)}_{ia}U_a^{\alpha}\right)\left(\sum_b H^{(v)}_{jb}v_b^{\beta}\right). \tag*{(437)} F j i ( q ) = ( a ∑ H ia ( U ) U a α ) ( b ∑ H j b ( v ) v b β ) . (437)
We fit C A F ( q ) C_A F(q) C A F ( q ) by the common full-covariance procedure. We use a 5 × 6 5\times6 5 × 6 grid for the fiducial fit; a finer 6 × 8 6\times8 6 × 8 grid displays the same histories and checks sensitivity to the binning. At fixed q = 3 − 2 2 q=3-2\sqrt{2} q = 3 − 2 2 , the fiducial fit gives C A = 0.731 C_A=0.731 C A = 0.731 , χ 2 = 22.99 \chi^{2}=22.99 χ 2 = 22.99 for 29 nominal degrees of freedom and p b o o t = 0.797 p_{\mathrm{boot}}=0.797 p boot = 0.797 . The finer grid gives p b o o t = 0.870 p_{\mathrm{boot}}=0.870 p boot = 0.870 . Changing the starting cutoff between 10 2 10^{2} 1 0 2 and 10 4 10^{4} 1 0 4 and varying the grid leaves the normalization within 0.729–0.733, with compatible goodness of fit. These checks reuse the same histories.
Tests and confidence intervals use App. § 20 .§ 1 , with the fixed-q q q null for the nested comparison. The free-q q q fiducial fit has χ 2 = 22.18 \chi^{2}=22.18 χ 2 = 22.18 for 28 nominal degrees of freedom and p b o o t = 0.795 p_{\mathrm{boot}} = 0.795 p boot = 0.795 , with the boost relation between α \alpha α and β \beta β imposed. Starting at U = 10 2 U = 10^{2} U = 1 0 2 or 10 4 10^{4} 1 0 4 gives q = 0.17148 q = 0.17148 q = 0.17148 or 0.17163 0.17163 0.17163 , with respective 95% intervals [ 0.17097 , 0.17196 ] [0.17097, 0.17196] [ 0.17097 , 0.17196 ] and [ 0.17106 , 0.17218 ] [0.17106, 0.17218] [ 0.17106 , 0.17218 ] .
The pressure–ANEC ratio pairs the ANEC and 20-interval pressure fits over 10 3 ≤ t ^ ≤ 10 6 10^{3} \le\hat{t} \le10^{6} 1 0 3 ≤ t ^ ≤ 1 0 6 under this resampling, using the corrected trace model A W t ^ − q + C / t ^ A_{W}\hat{t}^{-q} + C/\hat{t} A W t ^ − q + C / t ^ to extract its leading amplitude. Its interval tests the predicted relation between the leading asymptotic amplitudes.
Spatial-moment fits
The hierarchy predicts leading powers and limiting ratios; a finite-window fit also tests whether subleading terms are negligible at the measured precision. We use complete signed moments from 1,000 unabsorbed histories with μ = Γ s p l i t = 1 \mu= \Gamma_{\mathrm{split}} = 1 μ = Γ split = 1 , selecting twelve times approximately uniform in ln t \ln t ln t over 10 3 ≤ t ≤ 5 × 10 5 10^{3} \le t \le5 \times10^{5} 1 0 3 ≤ t ≤ 5 × 1 0 5 . The 36-coordinate vector ( X 0 , X 2 ( W ) , X 2 ( E ) ) (X_{0}, X_{2}^{(W)}, X_{2}^{(E)}) ( X 0 , X 2 ( W ) , X 2 ( E ) ) uses the common full-covariance and whole-history bootstrap procedure of App. J.1. With independent leading amplitudes, the measured ratios are
C 2 ( W ) C 0 ( W ) = 0.54680 [ 0.54656 , 0.54703 ] , \frac{C_{2}^{(W)}}{C_{0}^{(W)}} = 0.54680 \quad[0.54656, 0.54703], C 0 ( W ) C 2 ( W ) = 0.54680 [ 0.54656 , 0.54703 ] ,
C 2 ( E ) C 0 ( W ) = − 1.32046 [ − 1.32053 , − 1.32039 ] , (438) \frac{C_{2}^{(E)}}{C_{0}^{(W)}} = -1.32046 \quad[-1.32053, -1.32039], \tag*{(438)} C 0 ( W ) C 2 ( E ) = − 1.32046 [ − 1.32053 , − 1.32039 ] , (438)
with 95% bootstrap intervals, compared with predictions 0.546918 and − 1.320377 -1.320377 − 1.320377 . The joint fit resolves small departures from these limits because common amplitude fluctuations cancel in differences between moments and times.
Finite-window corrections and sensitivity. A sensitivity test adds D / t D/t D / t to the zeroth trace moment. The exact moment identities then require, in the same units,
M 0 ( W ) = C P t − q + D / t , M 2 ( W ) = C P 2 − q t 2 − q + D t + B W , M 2 ( E ) = − 2 C P ( 1 − q ) ( 2 − q ) t 2 − q − 2 D t ln ( t / t r e f ) + B E t + C E . (439) \begin{aligned}
M_{0}^{(W)} &= C_{P}t^{-q} + D/t,\\
M_{2}^{(W)} &= \frac{C_{P}}{2-q}t^{2-q} + Dt + B_{W},\\
M_{2}^{(E)} &= -\frac{2C_{P}}{(1-q)(2-q)}t^{2-q} - 2Dt\ln(t/t_{\mathrm{ref}}) + B_{E}t + C_{E}. \tag*{(439)}
\end{aligned} M 0 ( W ) M 2 ( W ) M 2 ( E ) = C P t − q + D / t , = 2 − q C P t 2 − q + D t + B W , = − ( 1 − q ) ( 2 − q ) 2 C P t 2 − q − 2 D t ln ( t / t ref ) + B E t + C E . (439)
The integration constants allow for the history before the fitting window; changing t r e f = 10 4 t_{\mathrm{ref}} = 10^{4} t ref = 1 0 4 is absorbed in B E B_{E} B E . This five-parameter finite-time model respects the exact moment identities. Its comparison with the leading model gives
Table 1 (PDF p. 114).
The correction improves the fiducial fit (p Δ = 0.00040 p_{\Delta} = 0.00040 p Δ = 0.00040 ), but freeing the leading ratios improves even this model (p Δ = 0.0072 p_{\Delta} = 0.0072 p Δ = 0.0072 ). Window and time-grid checks likewise leave resolved departures. Thus the ratios are close to their predicted limits, while the tested correction does not fully describe their finite-time approach.
The covariance matrix of the fiducial fit has condition number 2.4 × 10 9 2.4 \times10^{9} 2.4 × 1 0 9 . Independent singular-value whitening reproduces its quadratic statistics within 2 × 10 − 7 2 \times10^{-7} 2 × 1 0 − 7 without removing modes. We exclude t = 10 6 t = 10^{6} t = 1 0 6 because the half-open interval convention does not include folds terminating exactly at that endpoint, and use t ≤ t max / 2 t \leq t_{\max}/2 t ≤ t m a x /2 without imposing conservation numerically. Zeroth- and first-energy residuals are below 5 × 10 − 10 5 \times10^{-10} 5 × 1 0 − 10 of 4 μ t 4\mu t 4 μ t and 4 μ t 2 4\mu t^{2} 4 μ t 2 , respectively. The quoted intervals measure sampling uncertainty; systematic errors in the signed second moments are not bounded by these zeroth- and first-moment checks.
Crossover sampling and cosmological integration
The sharp-cutoff kernel uses 600,000 independent stationary-clock samples after 70 contraction steps; the exponential variable E κ E_{\kappa} E κ is integrated out analytically. The same samples produce Figure 10 and the response used in the homogeneous evolution of App. § 19 .§ 3 .
Youngest cohorts. For numerical convolution we extend Ψ s h a r p \Psi_{\mathrm{sharp}} Ψ sharp to the origin. Its derivative behaves as T − q I F S T^{-q_{\mathrm{IFS}}} T − q IFS , with q I F S < 1 q_{\mathrm{IFS}} < 1 q IFS < 1 , so the youngest-event integral is finite. This extension is used only for the convolution; the microscopic early-time behavior is fixed instead by the exact onset. The remnant starts as 8 μ t 8\mu t 8 μ t , while detached descendants require two splits. Equation (B.13) gives N 1 ( t ) = 2 3 Γ 2 t 4 + O ( Γ 3 t 6 ) N_{1}(t) = \frac{2}{3}\Gamma^{2}t^{4} + O(\Gamma^{3}t^{6}) N 1 ( t ) = 3 2 Γ 2 t 4 + O ( Γ 3 t 6 ) , so the unabsorbed geometric descendant trace starts at order μ Γ 2 t 5 \mu\Gamma^{2}t^{5} μ Γ 2 t 5 , and its cumulative response at order t 6 t^{6} t 6 . Terminal absorption can suppress this onset further.
We test entrance sensitivity by replacing the cumulative descendant response below T ∗ , s = Q ∗ , s / Q × , s = ξ s − 1 T_{\ast,s} = Q_{\ast,s}/Q_{\times,s} = \xi_{s}^{-1} T ∗ , s = Q ∗ , s / Q × , s = ξ s − 1 with Ψ s h a r p ( T ∗ , s ) ( T / T ∗ , s ) p \Psi_{\mathrm{sharp}}(T_{\ast,s})(T/T_{\ast,s})^{p} Ψ sharp ( T ∗ , s ) ( T / T ∗ , s ) p , for p = 4 , 6 p = 4, 6 p = 4 , 6 , while preserving it above that scale. The sixth power has the unabsorbed descendant onset; both choices serve only as matching prescriptions. The primitive and its first moment are changed consistently in the convolution. The bounce and motion diagnostics agree to numerical precision for both replacements and after doubling the sixth-power matching scale. The fiducial calculation therefore uses the unmodified cumulative kernel together with the exact remnant response.
Numerical convergence. The source is integrated against the calculated cumulative kernel and its first moment, including the youngest birth interval. Variable-step BDF2 evolves the background. With x = H 0 ( t − t E S P ) x = H_0(t-t_{\mathrm{ESP}}) x = H 0 ( t − t ESP ) , maximum steps 10 − 4 10^{-4} 1 0 − 4 , 5 × 10 − 5 5 \times10^{-5} 5 × 1 0 − 5 , 2.5 × 10 − 5 2.5 \times10^{-5} 2.5 × 1 0 − 5 give convergent results. The last two runs differ in bounce time by less than 4 × 10 − 8 4 \times10^{-8} 4 × 1 0 − 8 relatively and in momentum by less than 5 × 10 − 6 5 \times10^{-6} 5 × 1 0 − 6 relatively on the common comparison interval. Reducing the stationary-clock sample changes the directly evaluated response quantities by less than 2 × 10 − 4 2 \times10^{-4} 2 × 1 0 − 4 on the response grid. These comparisons establish numerical convergence for the large-hierarchy binary response used here. The physical approximations are stated in App. § 19 .§ 3 .