{"id":"c16585ad-f241-40f8-8378-e20f6b874ed9","arxiv_id":"1908.03820","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The auxiliary variational problems of the full-RSB free energy functional for spin glasses on random regular graphs are solved by a unique backward-stochastic-differential-equation solution, and step-shaped order parameters recover the discrete-RSB free energy.","lead":"This paper proves that the two inner optimization problems inside the full replica-symmetry-breaking free energy for spin glasses on random regular graphs have a unique solution, characterized by a backward stochastic differential equation. It also shows that the full theory collapses to the older discrete replica-symmetry-breaking theory when the order parameter is a step function.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The converse of Theorem 4.1 rests on a density claim that is false as stated: the DDE exponentials are independent of F_q, so their span cannot be dense in L^p_1(Ω); stationarity-to-BSDE needs a corrected fiberwise statement.","rationale":"The reader's conditional verdict is appropriate. The central existence/maximum story is independent of the density lemma: Section 5 constructs BSDE solutions by discrete-RSB recursion, and Theorem 4.4 proves maximality via Lemma 4.3. Thus a failure of the density lemma would not collapse all the theorems, contrary to the reader's phrasing; it would invalidate only the converse direction of Theorem 4.1. That converse is nonetheless a headline claim and is currently unsupported. The citation to Øksendal Lemma 4.3.2 is not enough because the literal density statement is false: the DDEs depend only on future increments and cannot approximate a variable depending on ω(0). A fiberwise version is likely true and would fix the proof, but it must be stated and proved. I also note that identity (65) is not derivable from the definition in (44) without a notation change; this needs correction in revision. Because the mathematical architecture is plausible and repairable, the reader's CONDITIONAL verdict should remain unchanged; the revision should supply the fiberwise density lemma, fix the definition of E(v;W^{xr*}|q), and explicitly verify the martingale/DDE property of the limiting process in Theorem 5.7.","tokens_in":24599,"tokens_out":25953,"duration_ms":288040,"concrete_test":"Verify the density claim by taking X=ω(q) (or ω(0) for q=0), which is F_q-measurable and in L^p_1(Ω). Every E(f;1|q) is independent of F_q and has conditional mean 1, so the L^p distance from X to any finite linear combination of these exponentials is bounded below by ||X−1||_{L^p(F_q)}>0; hence the literal claim that their span is dense in L^p_1(Ω) is false. Then repeat the proof of (66) using the fiberwise density of the shifted Brownian motion on [q,1]; if that fiberwise version holds, the converse can be repaired, and if not, Theorem 4.1's only-if direction fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing unresolved step is the converse of Theorem 4.1 (Section 4.2, after Eq. (66)). The proof requires that E[E(f;1|q)π|F_q] = ~E_{xr*}[π|F_q] for all deterministic f forces π to be F_q-measurable. The paper cites Lemma 4.3.2 of Øksendal for the density of span{E(f;1|q)} in L^p_1(Ω), but that density is false as stated: each E(f;1|q)=exp(∫_q^1 f·dω − 1/2∫_q^1|f|²) depends only on Brownian increments after q and is independent of F_q (in particular of the random initial point ω(0)), so its span cannot be dense in L^p_1(Ω), which contains F_q-dependent variables. The correct argument requires a fiberwise density statement for the shifted Brownian motion on [q,1]; that version is plausible but is neither stated nor proved. Additionally, the direction identity (65) uses an E(v;W^{xr*}|q) whose quadratic term is not the one defined in (44); read literally, (65) fails already for constant data such as x=1/2, r*=1, f=2. If either gap cannot be repaired, stationarity does not imply the BSDE. The existence, uniqueness and maximum theorems are proved directly from the BSDE and are not automatically invalidated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the auxiliary variational problem Γ(Ψ,x,r) that appears in the full replica symmetry breaking free energy for the Ising spin glass on random regular graphs, as formulated in the author's previous work [9]. For a bounded claim Ψ and an increasing Parisi order parameter x, the paper proposes a stationary condition for Γ, converts it into a backward stochastic differential equation (BSDE) for a pair (φ,r), proves uniqueness and a maximum principle for the BSDE, constructs existence first for piecewise constant x and then by continuity for general x, and argues that for piecewise constant x the construction reproduces the discrete-RSB free energy. The paper does not compute free energy values or compare predictions with physical data; it is a mathematical analysis of the variational formulation.","tokens_in":24767,"tokens_out":10215,"duration_ms":94652,"significance":"If the main equivalence is correct, the paper would supply a stochastic-calculus foundation for the full-RSB auxiliary problem, including a BSDE characterization, uniqueness in the D[0,1] quotient, a global maximum principle, and a recovery of discrete-RSB solutions. The piecewise-constant construction in Section 5.1 is explicit and checkable, and the continuity argument from χ◦ to χ is a reasonable strategy. These are genuine contributions. However, the central stationarity-to-BSDE implication rests on two unproved or misstated density/identity steps, and the existence theorem does not establish that the limiting control lies in the domain D[0,1] on which uniqueness and the maximum principle are formulated. The contribution is therefore conditional.","major_comments":[{"comment":"The stationary condition (52) is stated as vanishing of the directional derivative for every direction xδu in H^p_[0,1](Ω), immediately after the sentence 'We guess that the set of all the possible directions xδu ... is dense in H^p_[0,1](Ω).' The density of the set of directions actually generated by the path {r,u} is never proved, and this is not a harmless formatting choice: without it, Definition 4.2 may be stronger than the genuine variational stationarity of Γ, so the subsequent equivalence in Theorem 4.1 would characterize a different object. The paper should either prove this density or formulate the stationary condition with an explicit domain and show that the domain is rich enough for the BSDE argument.","section":"Section 4.2, Definition 4.2"},{"comment":"The statement that the linear span of {E(f;q′,ω|q), f ∈ L^p([0,1],R^n)} is dense in L^p_1(Ω), cited to Lemma 4.3.2 of [15], is false as written. Each DDE E(f;q′|q)=exp(∫_q^{q′} f(s)·dω(s) − 1/2∫_q^{q′}|f(s)|² ds) is independent of F_q and in particular of the random starting point ω(0), so its span cannot be dense in L^p_1(Ω), which contains F_q-dependent random variables. A fiberwise density statement for the shifted Brownian motion on [q,1], conditional on F_q, would be the natural repair, but it is neither stated nor proved. Since this density is the mechanism by which stationarity (52) implies F_q-measurability of the random RSB π and hence the BSDE (59), Theorem 4.1 is not established as written.","section":"Section 4.2, proof of Theorem 4.1, after Eq. (66)"},{"comment":"The identity ∫_q^1 xδu·dW^{xr*} = E(f;q′|q)/E(xr*;q′|q) − 1 is not consistent with the definition of the DDE E(xv;q′,W^{xr*}|q) in Eq. (44). With v = f − xr*, definition (44) gives a quadratic term (1/2)∫_q^{q′} x(s)||v(s)||² ds under W^{xr*}, while the ratio on the right equals E(v;W^{xr*}|q) with a unit quadratic term. For the constant choice x=1/2, r*=1, f=2, the exponent on the left is 3/2(W−W_q) − 9/16(q′−q), whereas the exponent on the right is 3/2(W−W_q) − 15/8(q′−q). Thus Eq. (65) fails as written, and the derivation of Eq. (66) needs either a corrected direction or a corrected identity.","section":"Section 4.2, Eq. (65)"},{"comment":"The limit pair (φ,r) is proved to satisfy the BSDE (59), with r ∈ H^p_[0,1](Ω), but the proof never shows r ∈ D[0,1](Ω), i.e., that the stochastic logarithm ζ(r,x;1,ω) is bounded and E(xr) is a true martingale. Proposition 5.3 establishes boundedness (113)–(114) only for piecewise constant x ∈ χ◦, and Theorem 5.8 extends Propositions 5.1, 5.4, 5.5, and 5.6 but not Proposition 5.3. Without this step, the existence result does not place the solution in the domain on which Theorem 4.2 (uniqueness) and Theorem 4.4 (maximum principle) are stated, so the central existence claim is incomplete.","section":"Section 5, Theorem 5.7"}],"minor_comments":[{"comment":"The index notation in the definition of piecewise constant x is inconsistent: x(q)=∑_{n=1}^{K+1} x_i 1_{(q_{i-1},q_i]}(q) mixes n and i; the sum should be over i with coefficients x_i on the intervals (q_{i-1},q_i].","section":"Section 5.1, Eq. (84)"},{"comment":"Several integrals are written as '∫ dq x(q)' without limits; they should be ∫_q^1 dq′ x(q′). The missing limits obscure the backward nature of the BSDE and make the computations hard to follow.","section":"Section 4.3, Eqs. (70)–(73)"},{"comment":"The notation for the direction is inconsistent: the directional derivative Π is defined with δu, but in Eqs. (53), (55), and (56) the integral is written with u instead of δu. This is confusing because u was not introduced as a standalone direction in those equations.","section":"Section 4.2, Eqs. (51)–(57)"},{"comment":"The displayed constant '(p K_p 2(p−1))^p' appears to be a typographical corruption of (p/(2(p−1)))^p K_p^p; it should be corrected to make the dependence on p transparent.","section":"Theorem 5.6, proof after Eq. (150)"},{"comment":"The manuscript contains numerous typographical errors, including 'Doéans-Dade', 'soluion', 'wright', and 'calim'; a thorough proofreading pass is needed.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The report is based on version v1. The central theorems are not yet reliable because of the three technical gaps in Section 4.2 and the missing extension of Proposition 5.3 to general x. If the author repairs these with a correct fiberwise density statement and matching DDE definitions, the paper could be publishable. I also note that the paper relies on the author's prior work [9] for the starting functional; the referee did not have [9]'s derivations available to verify the physical setup."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Francesco—quick take. The paper does something genuinely new: it writes the stationary condition of the auxiliary RSB variational problem as a BSDE, proves existence and uniqueness for that BSDE, and shows that piecewise constant order parameters reproduce the discrete-RSB equations of [9]. The discrete-case construction (Propositions 5.1–5.4) is explicit and looks correct: the martingale representation of J_n, the formula for r, and the boundedness estimates are standard and solid. The Lipschitz continuity in x (Theorems 5.6–5.7) is a plausible continuity argument, though it inherits the problems below.\n\nThe soft spots are real and load-bearing. The converse of Theorem 4.1 needs the span of {E(f; q'|q), f ∈ L^p} to be dense in L^p_1(Ω). As written, that is false: those exponentials only see Brownian increments after q, so they are independent of F_q and cannot be dense in the full F_1 space. What is needed is a fiberwise density statement for the post-q increments, equivalently a tensor-product density with F_q-random variables; that version is plausible and probably true, but it is neither stated nor proved. The citation to Øksendal's Lemma 4.3.2 does not cover this. On top of that, the DDE definition (44) puts an x(q) inside the quadratic variation of E(v; W^{xr}|q); with that definition the key identity (65) fails even for x=1/2, r*=1, f=2. If the x in (44) is meant to be absent, identity (65) works algebraically, so this is a fixable but consequential typo or nonstandard definition. Since the maximum and uniqueness results are proved directly from the BSDE, they can survive a correction; what breaks is the claim that every stationary point of the variational problem solves the BSDE, and hence the identification of the global maximizer with the BSDE solution.\n\nThere is also a smaller gap: in Theorem 5.7 the limiting r for continuous x is shown to be in H^p, but the DDE E(xr) is only known to be a true martingale in the piecewise-constant case (Proposition 5.3). Without that, the limit may not sit in the domain D[0,1](Ω), and the variational statement is incomplete.\n\nOverall: this deserves a serious referee but not acceptance as is. The core construction is original and likely repairable; the missing density lemma and the DDE definition fix should be mandatory. The comparison to [9] rather than to Mézard–Parisi is a minor concern, not a blocker.","headline":"A real, original attempt to solve the auxiliary variational problems of [9] via a BSDE, but the key stationarity-to-BSDE step relies on a density claim that is false as stated and a DDE definition that is internally inconsistent; it deserves a conditional acceptance, not a pass.","tokens_in":25429,"tokens_out":7003,"would_cite":false,"duration_ms":74607,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82B44","60H10"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that the auxiliary variational problems in the full replica symmetry breaking free energy of the Ising spin glass on a random regular graph are solved by the unique solution of a backward stochastic differential equation…","keywords":["Ising spin glass","random regular graph","full replica symmetry breaking","discrete replica symmetry breaking","backward stochastic differential equation","Doléans-Dade exponential","Girsanov transformation","Parisi order parameter"],"falsifier":"On one-dimensional Wiener space, take a nontrivial bounded terminal functional $\\Psi$ and try to construct a nonzero $\\xi\\in L^2_1(\\Omega)$ with $\\mathbb{E}[\\xi E(f;1,\\omega)]=0$ for every deterministic $f$; if such a $\\xi$ exists, the converse direction of Theorem 4.1 fails and the stationarity-to-BSDE step loses its support.","tokens_in":24191,"feed_emoji":"🧊","tokens_out":9163,"duration_ms":92614,"temperature":0.7,"pith_summary":"The paper studies the two auxiliary variational problems that sit inside the full replica symmetry breaking free energy of the Ising spin glass on a random regular graph. It claims that the maximizing control process $r$ is exactly the adapted solution of a backward stochastic differential equation whose terminal condition is the cavity functional $\\Psi$ and whose driving noise is a Brownian motion. For every bounded claim $\\Psi$ and every increasing Parisi order parameter $x$, the paper claims this equation has a unique solution, and that this solution attains the maximum of the auxiliary value functional. When $x$ is piecewise constant, the solution reduces to the discrete replica symmetry breaking recursion, so the full-RSB variational formulation is shown to contain the discrete-RSB theory as a special case. A sympathetic reader would care because the result turns a non-Markovian variational problem into a well-posed stochastic equation that could in principle be solved pathwise.","feed_headline":"A single backward equation solves the spin-glass free energy problem","feed_subtitle":"The auxiliary order parameter obeys a unique backward stochastic equation, and discrete RSB solutions are recovered.","key_machinery":"The load-bearing object is the Doléans-Dade exponential $E(xr;q,\\omega)=\\exp\\left(\\int_0^q x(s) r(s)\\cdot d\\omega(s) - \\frac12\\int_0^q x(s)^2\\|r(s)\\|^2\\,ds\\right)$, a positive martingale that defines a Girsanov change of measure with $E(xr)$ as its Radon–Nikodym derivative. This exponential is used to build convex combinations of control processes, to form the admissible variations $x\\delta u$, and to rewrite the directional derivative of $\\Gamma$ as an expectation under the tilted measure. The 'random RSB' $\\pi(\\Psi,x,r|q)$ emerging from that derivative is the object whose $\\mathcal{F}_q$-measurability is equivalent to the stationary condition, and the backward stochastic differential equation (59) is the expression of that measurability. The piecewise-constant analysis shows the same exponential reduces to the discrete-RSB backward iteration, which is why the full-RSB functional recovers discrete solutions.","core_discovery":"The central claim is that maximizing the auxiliary functional $\\Gamma(\\Psi,x,r)$ over control processes $r$ is equivalent to solving the backward stochastic differential equation $\\varphi(q) = \\Psi(1) - \\int_q^1 r(s)\\cdot d\\omega(s) + \\frac12\\int_q^1 x(s)\\|r(s)\\|^2\\,ds$ with terminal condition $\\varphi(1)=\\Psi(1)$. Theorem 4.1 establishes the key equivalence: $r^*$ satisfies the stationary condition exactly when the 'random RSB' quantity $\\pi(\\Psi,x,r^*|q)$ is measurable with respect to the information $\\mathcal{F}_q$ up to time $q$. A comparison identity then forces uniqueness of the solution pair and shows the solution is the global maximizer of the auxiliary problem. Existence is built in two steps: an explicit backward recursion when $x$ is piecewise constant, which coincides with the discrete-RSB equations, followed by a limiting argument that extends the solution to arbitrary increasing $x$ using uniform bounds on the controls. The paper therefore claims that the auxiliary variational problem is completely characterized by this BSDE, and that the full-RSB free energy functional encodes all discrete-RSB solutions when the order parameter is piecewise constant.","pith_inferences":["If the denseness of admissible directions is proved rather than assumed, the same stationarity-to-BSDE equivalence should carry over to other mean-field spin-glass models with a martingale representation of the free energy, not only random regular graphs.","The explicit backward recursion for piecewise constant order parameters points to a concrete numerical strategy: solve the BSDE on a time grid and refine the grid; Theorem 5.7 guarantees $L^p$ convergence, though the paper gives no rate, so a numerical study of the rate would be a natural next step.","The derivative formula (175) is a non-Markovian relative of the classical Parisi PDE derivative and may allow a variational proof of uniqueness of the physical order parameter in models where the full-RSB free energy applies."],"forward_implications":["For every bounded claim and every increasing Parisi order parameter, the auxiliary variational problem has a well-defined value given by the unique BSDE solution, so the RSB expectation $\\Sigma(\\Psi,x)$ is unambiguously defined.","Choosing a piecewise constant order parameter makes the full-RSB equations reduce to the discrete-RSB iteration, so the full-RSB functional is a genuine extension rather than a competing formulation.","The solution pair depends continuously on the order parameter in the uniform norm, uniformly in the order parameter, so nearby order parameters produce nearby free energy contributions.","Theorem 5.8 gives an explicit functional-derivative formula for the auxiliary value in terms of the control process, providing a direct handle on how the free energy changes under deformations of the Parisi order parameter."],"supporting_citations":[{"why":"Defines the full-RSB free energy functional and the auxiliary variational problems that this paper modifies and analyzes.","marker":"[9]"},{"why":"Supplies the cited Lemma 4.3.2 asserting the density of Doléans-Dade exponentials, which carries the converse direction of Theorem 4.1.","marker":"[15]"},{"why":"Provides the theory of backward stochastic differential equations used to cast the stationary equation (59) as a BSDE.","marker":"[19]"},{"why":"Supplies the martingale, Doléans-Dade exponential, Itô, and Burkholder-Davis-Gundy facts used throughout the derivations.","marker":"[14]"},{"why":"Provides the variational-representation template that motivates the two-level structure of the full-RSB functional.","marker":"[13]"},{"why":"Cameron-Martin theorem used together with Girsanov arguments to change measure via the DDE.","marker":"[17]"},{"why":"Girsanov theorem used to identify the shifted process as a Brownian motion under the tilted measure.","marker":"[18]"}],"fun_headline_variants":["Spin-glass free energy solved by a single backward equation","Unique backward equation underpins RSB free energy","Full RSB functional collapses to one BSDE","Backward SDE uniquely solves spin-glass order","Existence and uniqueness via backward equation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the admissible directions $x\\delta u$ are dense in the space of adapted controls, so that vanishing of the directional derivative along them forces the random RSB to be measurable with respect to the past; the paper labels this denseness a guess and relies on a cited lemma for a related exponential-density statement.","fun_headline_variants_meta":{"raw":{"variants":["Spin-glass free energy solved by a single backward equation","Unique backward equation underpins RSB free energy","Full RSB functional collapses to one BSDE","Backward SDE uniquely solves spin-glass order","Existence and uniqueness via backward equation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000185,"raw_usage":{"total_tokens":1291,"prompt_tokens":884,"completion_tokens":407,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":500,"completion_tokens_details":{"reasoning_tokens":337}},"tokens_in":500,"tokens_out":407,"duration_ms":4603,"temperature":1.0,"reasoning_tokens":337,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:03:03.114513+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"On one-dimensional Wiener space, take a nontrivial bounded terminal functional $\\Psi$ and try to construct a nonzero $\\xi\\in L^2_1(\\Omega)$ with $\\mathbb{E}[\\xi E(f;1,\\omega)]=0$ for every deterministic $f$; if such a $\\xi$ exists, the converse direction of Theorem 4.1 fails and the stationarity-to-BSDE step loses its support.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the full-RSB free energy functional and the auxiliary variational problems that this paper modifies and analyzes."},{"cited_title":"Spr inger, Berlin, Heidelberg (2003)","cited_arxiv_id":null,"evidence_quote":"Supplies the cited Lemma 4.3.2 asserting the density of Doléans-Dade exponentials, which carries the converse direction of Theorem 4.1."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the theory of backward stochastic differential equations used to cast the stationary equation (59) as a BSDE."},{"cited_title":"Springer, Berlin (1999)","cited_arxiv_id":null,"evidence_quote":"Supplies the martingale, Doléans-Dade exponential, Itô, and Burkholder-Davis-Gundy facts used throughout the derivations."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the variational-representation template that motivates the two-level structure of the full-RSB functional."},{"cited_title":"H., Martin, W .T .: Transformation of Wiener integrals under trans- lations","cited_arxiv_id":null,"evidence_quote":"Cameron-Martin theorem used together with Girsanov arguments to change measure via the DDE."},{"cited_title":"V .: On transforming a certain class of stoc hastic processes by abso- lutely continuous substitution of measures","cited_arxiv_id":null,"evidence_quote":"Girsanov theorem used to identify the shifted process as a Brownian motion under the tilted measure."}],"review_version":1}