Bulk-to-Brane Projection, Dynamical Nonclosure, and Observable Residues in Randall–Sundrum Gravity

The operational-equivalence obstruction

Jeremy RodgersSection 12 of 18Original PDF ↗Publication DOI ↗

The operational-equivalence obstruction

Sections 3–8 establish nonclosure of the instantaneous brane state. That statement must be sharply distinguished from the far stronger claim that no autonomous description using only brane observables can exist. The stronger claim is false when arbitrary history dependence is allowed.

Classical pushforward lemma

Let $\mathfrak E_T$ be the class of all experiments that prepare, intervene on, and read only Tier-1 (brane) variables. For each protocol $e\in\mathfrak E_T$ , a fully specified source theory defines a probability measure $\mu^S_e$ on its admissible source histories $\Gamma^e_S$ . Let $R_e:\Gamma^e_S\longrightarrow\Gamma^e_T\tag{94}$ be the measurable map returning the complete Tier-1 record.

Lemma 11.1 (Pushforward representation of Tier-1 operational laws).

Define $\boxed{\; \nu_e=(R_e)_*\mu^S_e, \qquad e\in\mathfrak E_T . \;}\tag{95}$ Then $M^\star_T=\{\nu_e:e\in\mathfrak E_T\}$ is a theory whose sample space contains only Tier-1 histories and whose prediction for every Tier-1 experiment coincides exactly with the source prediction: $d_{\mathrm{TV}}\bigl((R_e)_*\mu^S_e,\ \nu^\star_e\bigr)=0 \qquad\text{for every }e\in\mathfrak E_T ,\tag{96}$ where $\nu^\star_e:=\nu_e$ regarded as the surrogate theory’s law.

Proof. By the definition of pushforward, for every bounded measurable statistic $F:\Gamma^e_T\to\mathbb R$ , $\int_{\Gamma^e_T}F(\gamma)\,\nu_e(d\gamma) =\int_{\Gamma^e_S}F(R_e\omega)\,\mu^S_e(d\omega).\tag{97}$ Taking $F$ to be the indicator of an arbitrary measurable event proves equality of the complete record distributions, hence (96). Taking $F$ to encode an estimator, a randomized decision rule, a stopping rule, a correlation function, or a loss function proves equality of its distribution as well. Adaptive interventions are already covered, because each complete adaptive protocol is a member of $\mathfrak E_T$ and its record map is measurable. Marginalization, random mixtures of protocols, and causal consistency are inherited because pushforward commutes with measurable composition and with marginalization. ◻

Remark 11.2 (Status of Lemma 11.1).

The equality (96) is mathematically immediate from the definition of pushforward; nothing in the proof is deep, and no claim of mathematical originality is made for it. Its significance here is interpretive. It establishes the exact boundary on what brane-only observations can discriminate, and hence fixes the strongest conclusion that the positive results of Sections 3–7 can support. The substantive work lies in identifying which restrictions on the alternative class do permit discrimination (Section 11.3), and in exhibiting , through the four-dimensional CFT description of Section 10 , that the obstruction is realized concretely in RS2 and not merely as a formal possibility.

For standard Borel state spaces the finite-time laws disintegrate into causal history kernels, $\nu(dx_{0:n}\mid u_{0:n-1}) =K_0(dx_0)\prod_{j=0}^{n-1} K_{j+1}\bigl(dx_{j+1}\mid x_{0:j},u_{0:j}\bigr),\tag{98}$ an autonomous stochastic law on Tier-1 histories and controls. It need not be Markovian, finite-dimensional, local, simple, or state-independent. Those are explanatory restrictions, not consequences of autonomy in the unrestricted operational sense.

Quantum and field-theoretic forms

In quantum theory, equality of one-time reduced states, or of an ordinary in–out generating functional, is insufficient: interventions can expose temporal correlations. The exact Tier-1 object is the process tensor [14]. If $\mathcal A^{(j)}_{x_j}$ are Tier-1 instruments and $\mathcal U_{j+1:j}$ the joint source propagator, then $\Pr(\mathbf x\mid\mathbf A) =\operatorname{Tr}_{TE}\Bigl[ \mathcal A^{(n)}_{x_n}\,\mathcal U_{n:n-1}\cdots \mathcal A^{(0)}_{x_0}(\rho_{TE})\Bigr].\tag{99}$ Contracting every eliminated-sector index defines a positive, causally normalized multilinear functional $\Upsilon^T_{n:0}$ with Tier-1 legs only, such that $\Pr(\mathbf x\mid\mathbf A) =\Upsilon^T_{n:0}\bigl[ \mathcal A^{(n)}_{x_n},\mathcal A^{(n-1)}_{x_{n-1}},\ldots, \mathcal A^{(0)}_{x_0}\bigr].\tag{100}$

The field-theoretic counterpart is the Feynman–Vernon closed-time-path influence functional [12]. With $S[\phi,\chi]=S_T[\phi]+S_E[\chi]+S_{\mathrm{int}}[\phi,\chi]$ and a factorized initial environment state, $e^{iS_{\mathrm{IF}}[\phi_+,\phi_-]} =\int_{\chi_+(t_f)=\chi_-(t_f)} \mathcal D\chi_+\,\mathcal D\chi_-\; \rho_E(\chi^+_i,\chi^-_i)\, e^{iS_E[\chi_+]+iS_{\mathrm{int}}[\phi_+,\chi_+] -iS_E[\chi_-]-iS_{\mathrm{int}}[\phi_-,\chi_-]},\tag{101}$ $Z_T[J_+,J_-] =\int\mathcal D\phi_+\,\mathcal D\phi_-\; \rho_T(\phi^+_i,\phi^-_i)\, e^{iS_T[\phi_+]-iS_T[\phi_-] +iS_{\mathrm{IF}}[\phi_+,\phi_-] +iJ_+\phi_+-iJ_-\phi_-} .\tag{102}$ The retained action is counted exactly once. A correlated initial state replaces $\rho_E$ by the corresponding joint boundary functional; the process-tensor construction (100) covers that case with no factorization assumption. The resulting closed-time-path functional contains retarded response, fluctuations, and multi-time Tier-1 probabilities, and may be complex, nonlocal, and state-dependent while remaining a complete Tier-1 description.

The exact boundary of the conclusion

The witness of Section 3 proves $p\circ\Phi_t\ne U_t\circ p\tag{103}$ for every deterministic flow $U_t$ on the specified instantaneous brane state, and , for any preparation class that resolves the two witness states, so that the conditional future law differs within a readout fiber , it also defeats universal Markov lumpability. It does not defeat the pushed-forward law (95) on complete brane histories.

Accordingly, source projection can be empirically distinguished only relative to a restricted alternative class $\mathfrak M^R_T$ . If $\operatorname{Op}(M)$ denotes the complete family of protocol-conditioned Tier-1 laws of a model, the necessary and sufficient condition that the restricted class contain no exactly operationally equivalent model is $\boxed{\; \operatorname{Op}(M_S,p)\notin\bigl\{\operatorname{Op}(M):M\in\mathfrak M^R_T\bigr\}. \;}\tag{104}$

Remark 11.3 (Identifiability is not a finite-data testing theorem).

Condition (104) is exact identifiability. Uniform statistical discrimination from a composite alternative class additionally requires separation of the source law from the closure of that class in an operational metric appropriate to the allowed protocols , total variation, or a relative-entropy rate. If the source law lies outside the class but can be approximated arbitrarily closely by its members, exact equivalence is absent while finite-data discrimination may nonetheless be arbitrarily difficult.

Useful independent restrictions include: a fixed finite set of free or weakly coupled instantaneous fields at a stated perturbative order; a Markov or finite-memory condition; fixed finite-pole local quadratic dynamics; a finite EFT basis with coefficients fixed before the test; a prescribed preparation/boundary-state class. Locality alone, or finitely many interacting field species alone, is insufficient, because multiparticle states generate branch cuts. If arbitrary kernels, continua, initial correlations, and complete-history states are admitted, the pushforward theory belongs to the alternative class and discrimination is impossible.