{"id":"e320cc5f-2928-40d3-b6f4-3d19d040d321","arxiv_id":"2506.10213","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A coupling variance estimate for fully coupled FBSDEs on the Wiener space yields time regularity and D^{1,2} Malliavin differentiability of solutions.","lead":"This paper extends the coupling method for stochastic differential equations to forward-backward SDEs with random coefficients, proving estimates for the coupling variance and deriving time regularity and Malliavin differentiability for the solution. The main results give a unified way to transfer regularity of coefficients and terminal data to the solution, with the arbitrary-horizon results requiring a regular decoupling field.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The arbitrary-horizon coupling-variance estimate is conditional on an unproven regular decoupling field with quantitative smallness; the abstract and applications state the result more broadly than the hypotheses justify.","rationale":"My reading agrees with the reader's weakest-assumption analysis. The local estimate Theorem 5.5 is derived in detail and is internally plausible, but the extension to arbitrary intervals in Theorem 5.7 depends entirely on the regular decoupling field: the recursive bounds (5.22)-(5.23) use the decoupling property (4.7) on every subinterval, and Lemma 5.6 is the only mechanism transferring that structure to the coupled side. The manuscript itself signals that Lemma 4.7 is stated without proof and that Lemma 5.6's approximation step is only sketched, so the decoupling-field assumption is doubly load-bearing: it is both an unverified external input and the point at which the proof is most abbreviated. I do not see an internal contradiction in the main estimate once the decoupling-field hypothesis and the intended corrections to Lemma 5.6's indexing are granted, but I also do not see evidence that the hypothesis is satisfiable for the paper's advertised class of fully coupled random-coefficient FBSDEs. This is a genuine limitation of scope rather than a local computational error, so the reader's CONDITIONAL verdict is appropriate and should not be changed.","tokens_in":29088,"tokens_out":26652,"duration_ms":293533,"concrete_test":"Verify whether [14, Theorems 7.3 and 7.4] (and [30, Theorem 8.3.5]) actually supply a regular decoupling field under the paper's assumptions, i.e. with random coefficients satisfying Assumption 4.1, H(p), and 5.1, non-Markovian drivers, and the quantitative bound L_w L_{μ,3}<1. If the cited theorems require deterministic or Markovian structure, the paper should add the existence of w as an explicit standing assumption and exhibit at least one fully coupled random-coefficient FBSDE meeting Theorem 5.7's hypotheses; otherwise the arbitrary-horizon coupling-variance estimate is conditional only.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The bridge from the small-interval estimate to the advertised arbitrary-horizon results is Theorem 5.7, and that theorem requires, beyond the Lipschitz/random-coefficient assumptions, the existence of a regular decoupling field w with the quantitative condition (4.3) (with L_g replaced by L_w; in the p=2 case, L_w L_{μ,3}<1). The gluing proof (5.22)-(5.23) cannot start without Lemma 5.6, which transfers the decoupling field to the coupled FBSDE and identifies w^{φ,i}(X_{t_i}) with the local coupled solution's Y_{t_i}; the lemma itself is the most compressed step, and its statement/proof contain an index inconsistency (w^{φ,i} appears as the terminal condition on [t_i,t_{i+1}] although property (1) couples w^{φ,i} to w(t_i,·)). The paper does not prove existence of such a w for its random-coefficient setting; it only cites [14]. The cited results are developed mainly for structural/deterministic cases, and the manuscript does not verify that they deliver L_w L_{μ,3}<1 under Assumptions 4.1/H(p)/5.1. Since strongly coupled FBSDEs can lack global solutions (Antonelli's counterexample), this is a genuine structural restriction, not a technical convenience. The abstract, Theorem 6.1(2), and Theorem 6.3 do not flag this condition, so the claimed scope for random coefficients is broader than what is proved.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a coupling method for fully coupled forward-backward stochastic differential equations (FBSDEs) with random Lipschitz coefficients. It defines a p-coupling variance CV_p([t,T]) and proves two main estimates: a small-time estimate in Theorem 5.5 and an arbitrary-horizon estimate in Theorem 5.7 under an additional regular decoupling field assumption. These estimates are then applied to derive L^p time regularity for the solution (Theorem 6.1) and Malliavin Sobolev D^{1,2} differentiability (Theorem 6.3), using the coupling characterization of D^{1,2} from [8]. The abstract and introduction present the results as applying to FBSDEs with random Lipschitz coefficients, without flagging the decoupling-field condition needed for the arbitrary-horizon result.","tokens_in":29387,"tokens_out":14832,"duration_ms":168590,"significance":"If the technical gaps are repaired, the paper would provide a meaningful extension of the Geiss--Ylinen coupling method from BSDEs and SDEs to fully coupled FBSDEs, yielding explicit bounds on the coupling variance and, as a consequence, regularity results for all three solution components X, Y, and Z. The small-time estimate in Theorem 5.5 is proved in detail and appears to be a genuine contribution. The idea of using a decoupling field to glue small-time estimates is also appropriate. However, the arbitrary-horizon theorem depends on a structural existence assumption that is not verified, and the key gluing lemma contains a time-index inconsistency. These issues must be resolved before the advertised scope can be accepted.","major_comments":[{"comment":"There is a time-index inconsistency in Lemma 5.6. Property (1) states that w^{φ,i} is the coupling of w(t_i,·), but Eq. (5.15) uses w^{φ,i}(X_{t_{i+1}}) as the terminal condition on [t_i,t_{i+1}], and property (4) claims Y_{t_i}=w^{φ,i}(ξ). The transference argument in the proof transfers the FBSDE (5.18), whose terminal condition is w(t_{i+1},X_{t_{i+1}}); the coupled equation should therefore have terminal condition w^{φ,i+1}(X_{t_{i+1}}), where w^{φ,i+1}∈C^{φ,0}(w(t_{i+1},·)). With the indexing as written, Eq. (5.15) does not follow from the transference, and property (5.16) is unsupported. This is load-bearing because Theorem 5.7's gluing estimate (5.23) relies exactly on the identity w^{φ}(t_i,X_{t_i})=w^{φ,i}(X_{t_i})=Y_{t_i}.","section":"Section 5.2, Lemma 5.6 and Eq. (5.15)"},{"comment":"The arbitrary-horizon result is conditional on an existence assumption that is not verified. Theorem 5.7 conditions (2)--(3) require an existing regular decoupling field w satisfying the quantitative condition (4.3) with L_g replaced by L_w. The paper does not prove that such a field exists under the random-coefficient Assumptions 4.1, H(p), and 5.1; Remark 4.8 only cites [14, Theorems 7.3 and 7.4] without checking their hypotheses or verifying the inequality L_w L_{μ,3}<1. Consequently, the abstract and the introductory statement in Section 1, which advertise regularity for FBSDEs with 'random and Lipschitz coefficients', overstate the proven scope. Theorems 6.1(2) and 6.3 inherit this condition. Since strongly coupled FBSDEs can fail to be solvable on long intervals, this is a substantive restriction and should be stated explicitly in the abstract and in the main application theorems.","section":"Section 5.3, Theorem 5.7 and Remark 4.8"},{"comment":"The proof of Lemma 4.7 is omitted, with the text saying only that it follows by a similar approach as in the L^2 case and citing [14, Theorem 7.3]. Since this lemma is the paper's route to L^p well-posedness on arbitrary intervals, the omission is a missing support for a claimed result. The authors should either provide the proof or give a precise reference that covers exactly the present random-coefficient setting, including the role of Assumption H(p) and the quantitative Lipschitz condition involving L_w.","section":"Section 4.2, Lemma 4.7"}],"minor_comments":[{"comment":"Inequality (5.14) contains a typo: both sides use w^{φ,i}(ω,x); the second argument should be y, so the inequality should read |w^{φ,i}(ω,x)-w^{φ,i}(ω,y)|≤L_w|x-y|.","section":"Lemma 5.6(2)"},{"comment":"Inequality (6.1) is stated as a triangle inequality followed by ≤3∥Y_r−Y_s∥_p, which is tautological and does not by itself justify the later decomposition. The proof actually uses the two-term decomposition (6.7); that decomposition should be stated and proved directly rather than introduced through (6.1).","section":"Section 6.1, Eq. (6.1)"},{"comment":"The reduction 'W.l.o.g., T−t=Nδ' is not automatic because δ is fixed by Lemma 5.6 and Theorem 5.5. The argument can be repaired by choosing a step size h=(T−t)/N≤δ and working with a uniform partition of step h, but the constants in the recursion should then be stated uniformly for h≤δ.","section":"Theorem 5.7, proof"},{"comment":"In the proof, the notation Y^{x0,φ,[t1,t2]}_{t_i} uses the undefined interval [t1,t2]; this should be [t_i,t_{i+1}]. There is also a typo in the line following (5.19), where the interval [t1,t2] appears again.","section":"Lemma 5.6, proof"},{"comment":"The remark states that the constants c_1, c_2 and c_3 depend on the data, but only c_1 and c_2 appear in the displayed estimate; either add c_3 or correct the sentence.","section":"Remark 6.2"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely salvageable: the small-time theorem appears sound, and the arbitrary-horizon result could be repaired by correcting the time index in Lemma 5.6 and by making the decoupling-field assumption explicit throughout. The main risk is whether a regular decoupling field with the required quantitative smallness actually exists for the random-coefficient setting claimed in the abstract; the authors should verify the cited results from [14] or restrict the statements. I would not reject on the current evidence, but the manuscript needs a substantive revision before the advertised scope is justified."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing here is the transfer of the Geiss–Ylinen coupling program from BSDEs and decoupled FBSDEs to fully coupled FBSDEs with random Lipschitz coefficients. The small-time coupling-variance estimate, Theorem 5.5, is the backbone, and it is largely proved in detail: the decomposition Σ in Lemma 5.3, the rewriting of the coupled FBSDE under the coupling operator, and the handling of the Z-terms are all real work. The transference of the decoupling field (Lemma 5.6) and the gluing argument (Theorem 5.7) are also genuinely new steps, and the two applications, time regularity and D^{1,2} differentiability, are natural and mostly follow from the main estimate. I would credit the paper for engaging seriously with the existing literature and for being honest about its reliance on decoupling fields, at least in the body.\n\nThe soft spots are real but not fatal if you read the paper as the authors wrote it. The biggest one is that Theorem 5.7, the arbitrary-horizon result advertised in the abstract, is conditional on the existence of a regular decoupling field w with the quantitative condition L_w L_{μ,3} < 1. That is not a technical convenience: strongly coupled FBSDEs can fail to have global solutions (Antonelli's counterexample), and the paper does not prove existence of such a w for its random-coefficient setting. It cites [14], but the cited results are mostly structural/deterministic, and the manuscript does not verify that they deliver the required smallness under Assumptions 4.1, H(p), and 5.1. The abstract and the statements of Theorems 6.1(2) and 6.3 do not flag this condition, which is a scope problem.\n\nThere are also several compressed steps. Lemma 4.7 is stated without proof; the gluing argument in Theorem 5.7 assumes w.l.o.g. that T−t = Nδ, hides the recursion constants behind a generic c, and never tracks the dependence of the final constant on L_w. Lemma 5.6 has an index inconsistency: w^{φ,i} is used as the terminal condition on [t_i, t_{i+1}], but property (1) couples w^{φ,i} to w(t_i,·). None of this destroys the small-time result, and the reader's skepticism about the decoupling field is legitimate, but on reading the paper I think the central mechanism is sound as far as it goes.\n\nThis paper is for researchers in stochastic analysis working on FBSDE regularity or on the coupling method; it extends a known program to a new class of equations. It deserves a serious referee, but the referee should push for a proof or explicit assumption of the decoupling field in the random-coefficient setting, and for the abstract to state the condition. I would not desk-reject it.","headline":"Credible extension of the coupling method to fully coupled FBSDEs, but the arbitrary-horizon theorem is conditional on an unproven regular decoupling field with quantitative smallness, and the abstract oversells the scope.","tokens_in":29978,"tokens_out":1448,"would_cite":true,"duration_ms":18940,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H10","60H07"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves a coupling-variance estimate that turns coefficient data into time and Malliavin regularity for fully coupled forward-backward stochastic differential equations with random coefficients.","keywords":["forward-backward stochastic differential equations","coupling method","Wiener space","Malliavin Sobolev space","decoupling field","coupling variance","path regularity","random coefficients"],"falsifier":"Compute the coupling-variance bound in the one-dimensional linear example with $b=f=0$, $\\mu(u,x,y,z)=z$, and $g(x)=x$, taking $\\varphi=\\mathbf{1}_{(s,r]}$. The solution satisfies $Y_s=X_s$, so a regular decoupling field is $w(s,x)=x$ with $L_w=1$ and $L_{\\mu,3}=1$, placing the strict inequality $L_wL_{\\mu,3}<1$ exactly at its boundary. If inequality (5.20) holds for this equation, the strict condition is unnecessary; if it fails, that condition is the dividing line between coupling-variance control and its absence.","tokens_in":28852,"feed_emoji":"🔀","tokens_out":17022,"duration_ms":150443,"temperature":0.7,"pith_summary":"Forward-backward stochastic differential equations (FBSDEs) couple a forward process driven by Brownian motion to a backward process fixed by a terminal condition, and their solutions are hard to study directly because the auxiliary process $Z$ is part of the unknown. This paper adapts the coupling method to fully coupled FBSDEs with random, Lipschitz coefficients: on a product space carrying two Brownian motions, one replaces the driving noise by a correlated copy and compares the original solution to its twin. The main object is the $p$-coupling variance $CV_p$, which measures the $L^p$ size of the difference between the two solutions plus a weighted $Z$-term. The paper proves that $CV_p$ is controlled by the coupling distance of the initial and terminal data together with coefficient potential terms, on short intervals and, with a decoupling field, on arbitrary finite intervals. From that estimate it derives $L^p$ regularity in time for $X$, $Y$, and the integrated $Z$, and Malliavin-Sobolev $D_{1,2}$ differentiability of $X_s$ and $Y_s$.","feed_headline":"Coupling estimate gives time and Malliavin regularity for FBSDEs","feed_subtitle":"Comparing a solution with its twin under a second Brownian motion controls both path regularity and differentiability.","key_machinery":"The load-bearing mechanism is the coupling operator $C^\\varphi$ on the Wiener space: given a measurable $\\varphi:[0,T]\\to[0,1]$, it replaces $W$ with $W^\\varphi=\\int\\sqrt{1-\\varphi^2}\\,dW+\\int\\varphi\\,dW'$ and transfers random variables and processes while preserving their joint law and their Lipschitz constants. Applied to the FBSDE, it produces a twin equation with solution $(X^\\varphi,Y^\\varphi,Z^\\varphi)$, and the $p$-coupling variance $CV_p([t_1,t_2])$ collects the $L^p$ size of the difference between original and twin plus a term in which the factor $1-\\sqrt{1-\\varphi^2}$ multiplies $|Z_u|^2$. Lemma 5.3 rewrites the twin equation's diffusion coefficient as a two-component map $\\Sigma(s,\\alpha,x,y,(z_1,z_2))$ with the same Lipschitz constants, so the local stability estimate of Lemma 4.4 applies directly. For long intervals, Lemma 5.6 transfers a decoupling field $w$ through $C^\\varphi$ and preserves the field property, so the local bounds glue across a partition; the final constant grows like a power of the number of subintervals.","core_discovery":"Under Assumption 4.1, Assumption H(p), and Assumption 5.1—so the diffusion coefficient splits as $\\sigma(x,y)+A(z)$ with $A$ linear—and under the small-interval condition (4.3), Theorem 5.5 gives $CV_p([t,T]) \\le c\\{E[|\\xi-\\xi^\\varphi|^p + |g(X_T)-g^\\varphi(X_T)|^p] + E[U_p([t,T])] + S_p([t,T])\\}$ whenever $T-t$ is small. For arbitrary finite horizons, Theorem 5.7 proves the same bound provided the equation admits a regular decoupling field $w$, a random function with $Y_s=w(s,X_s)$ on small subintervals, whose Lipschitz constant $L_w$ satisfies $L_w L_{\\mu,3}<1$. The right-hand side involves only the data: the transferred initial value, the transferred terminal function, and the potential terms $U_p$ and $S_p$ built from the coefficients and the coupling function $\\varphi$. Specializing $\\varphi=\\mathbf{1}_{(s,r]}$ gives explicit $L^p$ time-regularity estimates for $X$ and $Y$, and specializing $\\varphi_r\\equiv r$ recovers the coupling characterization of the Malliavin-Sobolev space $D_{1,2}$, yielding $X_s,Y_s\\in D_{1,2}$ for all $s\\in[t,T]$.","pith_inferences":["An extension the paper leaves implicit: because $CV_p$ contains $\\int(1-\\sqrt{1-\\varphi^2})|Z_u|^2\\,du$, the same machinery likely yields a fractional Sobolev or Besov regularity scale for $Z$ itself, although the paper only states $L^p$ information for the integrated $Z$.","An extension the paper does not develop: any numerical scheme that respects the coupling operator and the local stability estimate could inherit the same gluing argument, giving discretization-error bounds measured in the $CV_p$ metric.","The strict inequality $L_w L_{\\mu,3}<1$ enters through the small-interval $L^p$ theory; testing boundary cases such as the linear example in the falsifier would show whether the inequality is necessary for the coupling-variance bound or only for the proof."],"forward_implications":["On short time intervals, the coupling-variance estimate needs no decoupling field: Theorem 5.5 holds whenever the coefficients are Lipschitz and integrable and condition (4.3) holds, so it covers strongly coupled FBSDEs locally.","With a regular decoupling field satisfying $L_w L_{\\mu,3}<1$, the same estimate holds on an interval of arbitrary finite length, so the regularity conclusions are not confined to small-time solvability.","Taking $\\varphi=\\mathbf{1}_{(s,r]}$ turns the bound into explicit $L^p$ time regularity: $E|X_s-X_r|^p$ and $E|Y_s-Y_r|^p$ are bounded by powers of $r-s$ times coefficient potentials, and $E[(\\int_s^r |Z_u|^2\\,du)^{p/2}]$ is bounded by $CV_p([t,T])$.","Taking $\\varphi_r\\equiv r$ and dividing by $r^2$ recovers the known coupling characterization of $D_{1,2}$, so under condition (6.8) both $X_s$ and $Y_s$ lie in the Malliavin-Sobolev space for every $s\\in[t,T]$.","For deterministic coefficients, condition (6.8) holds automatically, and the estimates reduce to the previously known BSDE and SDE coupling results as special cases."],"supporting_citations":[{"why":"Supplies the coupling operator, its transference properties, and the basic estimate linking $\\xi-\\xi^\\varphi$ to conditional expectations; the whole method builds on this.","marker":"[7]"},{"why":"Provides the coupling characterization of $D_{1,2}$ and the SDE transference theorem used in the regularity applications.","marker":"[8]"},{"why":"Gives the small-interval $L^p$ existence and uniqueness of FBSDEs and the Lipschitz condition (4.3) on which Theorem 5.5 rests.","marker":"[28]"},{"why":"Supplies the $L^p$ a priori estimate for the backward component used in the proof of Lemma 4.4.","marker":"[3]"},{"why":"Justifies the regular decoupling field assumption and the extension from local to global solvability used in Theorem 5.7.","marker":"[14]"},{"why":"Provides the decoupling-field framework and one-dimensional existence results that motivate the $L^p$ extension in Section 4.2.","marker":"[30]"}],"fun_headline_variants":["Coupling twin Brownian motions proves FBSDE regularity","Malliavin differentiability and time regularity for FBSDEs via coupling","Coupling method yields time and Malliavin regularity for FBSDEs","New coupling estimates give FBSDE regularity in time and Malliavin sense","Twin coupling proves FBSDE differentiability and time regularity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that, for arbitrary time intervals, the FBSDE admits a regular decoupling field $w$ with Lipschitz constant $L_w$ such that $L_w L_{\\mu,3}<1$ (Theorem 5.7, conditions (2)-(3)); the paper cites earlier work for existence of such fields rather than proving them, and the gluing argument for $CV_p$ relies completely on this condition.","fun_headline_variants_meta":{"raw":{"variants":["Coupling twin Brownian motions proves FBSDE regularity","Malliavin differentiability and time regularity for FBSDEs via coupling","Coupling method yields time and Malliavin regularity for FBSDEs","New coupling estimates give FBSDE regularity in time and Malliavin sense","Twin coupling proves FBSDE differentiability and time regularity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001178,"raw_usage":{"total_tokens":4864,"prompt_tokens":935,"completion_tokens":3929,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":551,"completion_tokens_details":{"reasoning_tokens":3834}},"tokens_in":551,"tokens_out":3929,"duration_ms":29886,"temperature":1.0,"reasoning_tokens":3834,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:33:35.053282+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the coupling-variance bound in the one-dimensional linear example with $b=f=0$, $\\mu(u,x,y,z)=z$, and $g(x)=x$, taking $\\varphi=\\mathbf{1}_{(s,r]}$. The solution satisfies $Y_s=X_s$, so a regular decoupling field is $w(s,x)=x$ with $L_w=1$ and $L_{\\mu,3}=1$, placing the strict inequality $L_wL_{\\mu,3}<1$ exactly at its boundary. If inequality (5.20) holds for this equation, the strict condition is unnecessary; if it fails, that condition is the dividing line between coupling-variance control and its absence.","supporting_citations":[{"cited_title":"Geiss and J","cited_arxiv_id":null,"evidence_quote":"Supplies the coupling operator, its transference properties, and the basic estimate linking $\\xi-\\xi^\\varphi$ to conditional expectations; the whole method builds on this."},{"cited_title":"Regularity of stochastic differential equations on the Wiener space by coupling","cited_arxiv_id":"2412.10836","evidence_quote":"Provides the coupling characterization of $D_{1,2}$ and the SDE transference theorem used in the regularity applications."},{"cited_title":"Yong,L p theory of the forward-backward stochastic differential equations","cited_arxiv_id":null,"evidence_quote":"Gives the small-interval $L^p$ existence and uniqueness of FBSDEs and the Lipschitz condition (4.3) on which Theorem 5.5 rests."},{"cited_title":"Briand, B","cited_arxiv_id":null,"evidence_quote":"Supplies the $L^p$ a priori estimate for the backward component used in the proof of Lemma 4.4."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Justifies the regular decoupling field assumption and the extension from local to global solvability used in Theorem 5.7."},{"cited_title":"Zhang, Backward Stochastic Differential Equations: From Linear to Fully Nonlinear The- ory","cited_arxiv_id":null,"evidence_quote":"Provides the decoupling-field framework and one-dimensional existence results that motivate the $L^p$ extension in Section 4.2."}],"review_version":1}