The Wellposedness of Path-dependent Multidimensional Forward-backward SDE

Kaitong Hu

REVIEW 4 major objections 5 minor 22 references

The Wellposedness of Path-dependent Multidimensional Forward-backward SDE

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A bounded dominating ODE makes path-dependent FBSDEs globally well-posed.

desk verdict A credible extension of the decoupling-field program to path-dependent FBSDEs, with a solid local wellposedness proof and useful stability result, but the global patching and maximal-interval theorems have real gaps that need fixing. read the letter →

arxiv 1908.09208 v3 pith:YNCGOP37 submitted 2019-08-24 math.PR

Kaitong Hu This is my paper · ORCID

classification math.PR MSC 60H0760H3035R6034F05

keywords DecouplingRandomFieldForward-BackwardSDECharacteristicBSDEBackwardStochasticRiccatiEquationsStabilityTheorempath-dependentFBSDEdominatingODEwellposedness

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Path-dependent forward-backward SDEs — systems where the coefficients at time $t$ may depend on the whole history of the forward process — appear when solving stochastic control problems with path-dependent contracts. The paper sets out to prove that such equations are well posed on arbitrary time intervals, not just locally. Its route is a random decoupling field on path space: a function $u_t(\omega,x)$ with $Y_t = u_t(\omega,X_{\cdot\wedge t})$ along solutions. Provided a scalar dominating ODE for the associated characteristic BSDE stays bounded over $[0,T]$, local decoupling fields can be patched into one global field, giving a unique solution; a separate theorem quantifies how solutions move when the coefficients are perturbed. If true, this removes the Markovian restriction from a standard wellposedness toolkit and covers control problems whose contracts depend on the whole path.

What carries the argument

The load-bearing object is the decoupling field, a progressively measurable random function $u:[0,T]\times\Omega\times C([0,T],\mathbb{R}^d)\to\mathbb{R}^n$ with $Y_t = u_t(X)$ along any solution; 'regular' means Lipschitz in the $L^2$-type path norm $\|x-x'\|_{2,t}$. The argument is carried by two auxiliary objects built from differences of two solutions: the normalized ratio $H_t = |\Delta Y_t|^2/\|\Delta X\|_{2,t}^2$ and the characteristic BSDE $dH_t = -F_t(H_t)\,dt + N_t\,d\tilde W_t$, whose polynomial driver $F_t$ is bounded above by a deterministic dominating ODE. The comparison principle for quadratic BSDE transfers boundedness of the dominating ODE to boundedness of $H_t$, which is exactly a Lipschitz bound on the decoupling field; that bound lets the local existence theorem be reapplied step by step to cover all of $[0,T]$.

What would settle it

Find two forward solutions that agree on a nontrivial interval and differ later; if the ratio $|\Delta Y_t|^2/\|\Delta X\|_{2,t}^2$ has no bounded extension at the coincidence time for a system satisfying the hypotheses of Proposition 4.2, the comparison argument collapses and the global wellposedness criterion fails.

Watch

Extended reading notes

Core claim

The paper's central claim is Proposition 4.2: for a path-dependent multidimensional FBSDE satisfying the global Lipschitz Assumption 1, the existence of a continuously differentiable $G$ such that $\dot y_t = -G(t,y_t)$ is a dominating ODE for the characteristic BSDE, together with a bounded solution of that ODE on $[0,T]$, implies a unique regular decoupling field on all of $[0,T]$; by Theorem 3.1 this is exactly wellposedness. The supporting local result is Theorem 4.1: under the contraction condition $K_1|\nabla_z\sigma|_\infty<1$, a unique solution exists on a small time interval whose length depends only on the dimensions and Lipschitz constants. The paper also proves Theorem 5.4, a stability estimate showing that if $(b,\sigma,f,g)$ are replaced by nearby coefficients, the resulting solutions differ by at most a constant times the coefficient differences, measured by $\sup_t \mathbb{E}[\|\Delta X\|_{2,t}^2 + |\Delta Y_t|^2 + \int_t^T |\Delta Z_s|^2\,ds]$.

Load-bearing premise

The patching argument assumes the ratio $H_t=|\Delta Y_t|^2/\|\Delta X\|_{2,t}^2$ stays well defined and bounded even when the denominator vanishes because the two forward paths coincide.

Editorial extensions

If this is right

Every path-dependent FBSDE whose characteristic BSDE admits a bounded dominating ODE has a unique solution on $[0,T]$, regardless of how strongly the coefficients are coupled.
Decoupled path-dependent FBSDEs are well posed for every horizon $T$, because their characteristic BSDE admits a linear dominating ODE (Proposition 4.3).
For systems with $b=b_t(X,Y_t)$ and $\sigma=\sigma_t(X)$, the sign condition (21) on increments of the coefficients and terminal data guarantees a global unique solution.
Regularity of the decoupling field persists on the maximal interval, and if that interval is open at the left end, the product $|\nabla_x u|\cdot|\nabla_z\sigma|_\infty$ must approach 1 there (Proposition 4.7).
Small changes in the coefficients produce small changes in the solution triple, with the explicit bound of Theorem 5.4 in the norm $\sup_t \mathbb{E}[\|\Delta X\|_{2,t}^2 + |\Delta Y_t|^2 + \int_t^T |\Delta Z_s|^2\,ds]$.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The stability estimate suggests a natural numerical test: approximate path-dependent coefficients by piecewise-constant or Markovian ones and use Theorem 5.4 to turn coefficient error into an explicit solution error; the paper does not perform such an experiment.
The proof's normalized ratios divide by the path distance between two forward solutions, so at moments when the two paths coincide the argument is undefined; a limiting convention could likely repair this, but none is given.
Because the dominating ODE is scalar, its explosion time could be computed numerically for a concrete control problem, making $T_{\max}$ a quantitative horizon check before solving the FBSDE.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague — quick read of arXiv:1908.09208 (Hu). The headline: this is a credible attempt to extend the decoupling-field program to path-dependent FBSDEs, and the local wellposedness part is genuinely useful, but the global patching and maximal-interval results do not hold as written — there is a missing condition in Prop 4.2 and an imported theorem in Prop 4.6.

What's new: the path-space norm ||x−x'||_{2,t} = (∫|x−x'|² + |x(t)−x'(t)|²)^{1/2} and the variational/characteristic BSDE on path space. Theorem 4.1's contraction proof is actually written out, not waved away, and it correctly identifies K1|∇zσ|∞<1 as the smallness condition. The stability result is a nice addition. Citation habits are fine; the one self-citation is contextual.

Soft spots, in order: 1. Proposition 4.2: the patching argument needs to reapply Theorem 4.1 at each step, which requires the Lipschitz constant of the current decoupling field to satisfy L|∇zσ|∞<1. From the dominating ODE and characteristic BSDE, the proof only gets L ≤ Kmax = sqrt(max_t y_t). It never checks Kmax|∇zσ|∞<1. If the dominating ODE's bounded solution has max above 1/|∇zσ|∞, no ε̄ exists and the induction cannot start. That is a genuine gap, not a typo. Adding (max y_t)|∇zσ|∞<1 to the statement would fix it, but then the theorem is much weaker. 2. Proposition 4.6, which provides the maximal interval for the general case, is disposed of with "we refer to [5, Theorem 2]". That is a Markovian theorem; the path-dependent proof is exactly what this paper is supposed to supply. This is load-bearing. 3. Theorem 4.4's proof has an unverified inequality and notation slips; Theorem 5.4's displayed equation for dΔY_t is garbled. These are addressable, but they make verification harder.

The D_t=0 degeneracy in the characteristic BSDE is minor; a limiting convention should work.

Who this is for: stochastic analysts working on path-dependent control and contract theory. If the gaps are fixed, this becomes a solid contribution. Right now I would not cite it for the global theorems.

Recommendation: send it to a serious referee — the topic and the local result justify that. But the referee should be told to demand the missing patching condition and a real proof of Prop 4.6, likely major revision.

Referee Report

4 major / 5 minor

Summary. The paper studies wellposedness of multidimensional forward-backward SDEs whose coefficients may depend on the whole past path of the forward component. The authors introduce a decoupling random field on path space, prove local existence and uniqueness under the condition K1|∇zσ|∞<1 via a contraction argument, and then attempt to patch local decoupling fields using a characteristic BSDE and a dominating ODE. They state a global wellposedness criterion (Proposition 4.2), treat the decoupled and drift-controlled cases, introduce a maximal interval in the general case following Fromm-Imkeller, and prove a stability estimate for nearby path-dependent FBSDEs. The central claimed contribution is the extension of the decoupling-field method from Markovian to path-dependent coefficients.

Significance. If the technical gaps are closed, the paper would provide a useful extension of the decoupling-field framework to path-dependent FBSDEs, with natural applications to path-dependent stochastic control and principal-agent problems. The local contraction proof in Theorem 4.1 is written out in detail, and the stability theorem gives a concrete quantitative estimate of a type not present in the earlier Markovian literature. However, the global patching argument in Proposition 4.2 has a load-bearing gap, and several supporting results are either imported from the Markovian setting without adaptation or only sketched.

major comments (4)

[Section 4.2.1, proof of Proposition 4.2, inequalities (14)-(15)] The patching induction is incomplete. At each step, Theorem 4.1 can only be reapplied if the Lipschitz constant of the newly constructed terminal condition satisfies L_i |∇zσ|∞ < 1. From (14) the proof obtains L_i^2 ≤ y_{t_i} ≤ Kmax^2, hence only L_i ≤ Kmax; the strict inequality Kmax |∇zσ|∞ < 1 is never proved. If the bounded solution of the dominating ODE has sup y_t larger than |∇zσ|∞^{-2}, no uniform ε̄ exists and the iteration cannot start. This is a load-bearing gap in the paper's central global-wellposedness claim.
[Section 4.2.5, Proposition 4.6] The proof is delegated to [5, Theorem 2], but the object here is path-dependent: the decoupling field is defined on C([0,T],R^d) and the gradient ∇x u is a derivative on path space. The Markovian proof in [5] does not automatically cover this setting; an adaptation showing that the same maximal-interval argument works with the path norm ||·||_{2,t} is required. Since Proposition 4.7 and the general-case wellposedness rely on Proposition 4.6, this cannot remain a citation-only step.
[Section 4.2.1, Eq. (12) and preceding definitions] The quantities H_t=|Y_t|^2/D_t^2, α_t=Z_t/D_t and β_t=X_t/D_t are undefined whenever D_t=||ΔX||_{2,t}=0, for instance when the two forward paths coincide on an interval. No convention or limiting argument is given, and the Itô derivation of the characteristic BSDE requires a justification in this degenerate case. This affects every result that uses the characteristic BSDE, including Proposition 4.2.
[Section 4.2.3, proof of Theorem 4.4] The displayed Itô computation is not carried out, and the sign conclusion 'which is smaller than 0 by (21)' is not transparent; as printed, the identity has mismatched factors of H_t and D_t and does not clearly follow from condition (21). Since Theorem 4.4 supplies global wellposedness in the drift-controlled case b=b_t(x,y), σ=σ_t(x), this proof should be written out in detail.

minor comments (5)

[Section 6, proof of Theorem 3.1] The partition condition 'ti+1 − ti > δ' should be 'ti+1 − ti ≤ δ', since Definition 3.1 applies only to intervals of length at most δ.
[Section 5, proof of Theorem 5.4, Eqs. (32)-(33)] The display writes ∫_T^t ... dt and ∫_T^t ΔZ_s dW_s as if these were the terms of dΔY_t; the correct differential form is dΔY_t = −(... )dt + ΔZ_t dW_t. The current notation makes the proof difficult to follow.
[Section 4.2.4, text before Theorem 4.5] The statement 'ess sup H_t is a Lipschitz constant of u_t' should refer to ess sup sqrt(H_t), because inequality (14) bounds |u(t,X)−u(t,X′)|^2 by y_t times the path norm, so the Lipschitz constant is the square root of the bound.
[Section 4.2.3, condition (21)] The right-hand side of (21) uses x_T; since g is defined on the whole path space C([0,T],R^d), please clarify whether x_T denotes the terminal value x(T) or the whole path x, and similarly for the notation Δx_t used in the same condition.
[Section 4.2.1, proof of Proposition 4.2, after Eq. (13)] The comparison 'H_t ≤ y_t' is invoked without a citation or a proof; since this comparison principle for quadratic BSDEs is a key ingredient of the patching argument, a reference or a short justification should be provided.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the global decoupling-field construction rests on externally cited contraction and dominating-ODE results, not on the paper's own conclusions.

full rationale

The paper transfers the decoupling-field technique of Ma-Wu-Zhang-Zhang [12] and Fromm-Imkeller [5] to path-dependent FBSDEs. I checked each load-bearing step for circularity. Theorem 4.1 is a genuine contraction argument: the contraction ratio satisfies gamma(epsilon,T) converging to K1^2 |nabla_z sigma|_infty^2 + K0 epsilon as T -> 0, and the hypothesis K1|nabla_z sigma|_infty < 1 is used to make gamma < 1 for suitable epsilon and T; no fitted quantity is later renamed as a prediction. Proposition 4.2 takes a dominating ODE as an external input and attempts to patch local intervals; its validity depends on whether the Lipschitz bound Kmax satisfies Kmax|nabla_z sigma|_infty < 1, which is a correctness gap in the proof, not a circular reduction. Theorem 3.1 is close to the definition of a decoupling field, but it still supplies a partition-and-patching proof; it does not assume the global conclusion. The only self-citation is reference [6] (Hu-Ren-Touzi), cited in the introduction as motivation for path-dependent contracts; it does no logical work in the proofs. Proposition 4.6 is imported from [5], which is independent prior work by Fromm and Imkeller, not a self-citation. The D_t = 0 degeneracy in the characteristic BSDE is a technical gap concerning undefined normalized quantities, not a circular step. No parameter is fitted from the target result, no uniqueness theorem from the author's own prior work is invoked to forbid alternatives, and no known result is merely renamed. The derivation chain is therefore self-contained relative to its cited external inputs; correctness concerns are separate and do not constitute circularity.

Assumptions & free parameters 0 free parameters · 7 assumptions · 0 invented entities

No fitted constants and no new entities are introduced. The central proof depends on standard stochastic calculus, on external theorems from the cited literature, and on the domain assumptions in Assumption 1 and the condition K1|nabla_z sigma|_infty < 1.

assumptions (7)

standard math The filtered probability space carries an n-dimensional Brownian motion and F is its augmented natural filtration.
Used throughout Section 2 to define H^2 spaces, adaptedness, and the stochastic integrals in the FBSDE.
domain assumption Coefficients b, sigma, f and terminal g are progressively measurable and Lipschitz in the path norm ||x-x'||_{2,t} (Assumption 1).
This defines the class of path-dependent FBSDEs under study; every theorem in the paper assumes it.
domain assumption K1|nabla_z sigma|_infty < 1, where K1 is the Lipschitz constant of g and nabla_z sigma is the z-derivative of sigma.
This condition is necessary for the contraction in Theorem 4.1, and Example 4.1 shows the threshold is sharp.
standard math Standard existence, uniqueness, and a priori estimates for decoupled SDEs and BSDEs.
Used in Theorem 4.1 and Lemmas 5.1, 5.2, and 5.3; the proofs are cited to the literature.
standard math Kobylanski comparison principle for quadratic BSDEs and the dominating ODE comparison.
Used in Proposition 4.2 to bound the characteristic BSDE H_t by the solution of the dominating ODE.
standard math [5, Theorem 2] on existence, uniqueness, and regularity of decoupling fields on maximal intervals.
Proposition 4.6 cites this theorem directly; the paper does not re-prove the path-dependent version.
domain assumption Monotonicity condition (21) connecting b, f, sigma, and g in Theorem 4.4.
This sufficient condition is used to make the variational term non-positive and to obtain a linear dominating ODE.

how reviews work

0 comments

Cite this review

Pith. "Pith review of The Wellposedness of Path-dependent Multidimensional Forward-backward SDE." pith.science (2026). https://pith.science/paper/YNCGOP37

@misc{pith2026190809208,
  author       = {Pith},
  title        = {Pith review of: The Wellposedness of Path-dependent Multidimensional Forward-backward SDE},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YNCGOP37}},
  note         = {Machine review of arXiv:1908.09208}
}

read the original abstract

We study in this paper the wellposedness of path-dependent multidimensional forward-backward stochastic differential equations (FBSDE). By path-dependent we mean that the coefficients of the forward-backward SDE at time t can depend on the whole path of the forward process up to time t. These kinds of forward-backward SDE appear when solving path-dependent stochastic control problem by means of variational calculus. At the heart of our analysis is the construction of a decoupling random field on the path space. We first prove the existence and the uniqueness of decoupling field on small time interval. Then by introducing the characteristic BSDE, we show that a global decoupling field can be constructed by patching local solutions together as long as the solution of the characteristic BSDE remains bounded. Finally, we show that the solution of a path-dependent forward-backward SDE is stable.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

[5]

Fromm and P

A. Fromm and P. Imkeller. Existence, Uniqueness and Regularity o f Decoupling Fields to Multidi- mensional Fully Coupled FBSDEs. ArXiv e-prints , October 2013

work page 2013
[1]

Backward-forward stochastic diﬀerential equa tions

Fabio Antonelli. Backward-forward stochastic diﬀerential equa tions. Ann. Appl. Probab. , 3(3):777– 793, 08 1993

work page 1993
[2]

The steepest descent meth od for forward-backward sdes

Jaksa Cvitanic and Jianfeng Zhang. The steepest descent meth od for forward-backward sdes. Elec- tron. J. Probab. , 10:1468–1495, 2005

work page 2005
[3]

Contract Theory in Continuous-Time Models

Jaksa Cvitanic and Jianfeng Zhang. Contract Theory in Continuous-Time Models . Springer, 2012

work page 2012
[4]

F. Delarue. On the existence and uniqueness of solutions to fbsd es in a non-degenerate case. Stochas- tic Process, 99, 209-286, MR1918539, 2002

work page 2002
[6]

Continuous-time princip al-agent problem in degenerate systems, 2019

Kaitong Hu, Zhenjie Ren, and Nizar Touzi. Continuous-time princip al-agent problem in degenerate systems, 2019

work page 2019
[7]

Hu and S

Y. Hu and S. Peng. Solution of forward-backward stochastic diﬀ erential equations. Probability Theory and Related Fields , 103(2):273–283, Jun 1995

work page 1995
[8]

Backward stochastic diﬀerential equatio ns and partial diﬀerential equations with quadratic growth

Magdalena Kobylanski. Backward stochastic diﬀerential equatio ns and partial diﬀerential equations with quadratic growth. The Annals of Probability , Vol.28, No.2, 558-602, 2000

work page 2000

Show all 22 references

[9]

Ma and J

J. Ma and J. Yong. On linear, degenerate backward stochastic p artial diﬀerential equations. Prob- ability Theory and Related Fields , 113(2):135–170, Feb 1999

1999
[10]

Forward-Backward Stochastic Diﬀerential Equations and th eir Applica- tions

Ji Ma and Jiongmin Yong. Forward-Backward Stochastic Diﬀerential Equations and th eir Applica- tions. Springer, Berlin, Heidelberg, 2007

2007
[11]

Solving forward-backw ard stochastic diﬀerential equa- tions explicitly — a four step scheme

Jin Ma, Philip Protter, and Jiongmin Yong. Solving forward-backw ard stochastic diﬀerential equa- tions explicitly — a four step scheme. Probability Theory and Related Fields , 98(3):339–359, Sep 1994

1994
[12]

On well-pose dness of forward-backward SDEs–a uniﬁed approach

Jin Ma, Zhen Wu, Detao Zhang, and Jianfeng Zhang. On well-pose dness of forward-backward SDEs–a uniﬁed approach. Ann. Appl. Probab. , 25(4):2168–2214, 08 2015

2015
[13]

Adapted solution of a degenerate back ward spde, with applications

Jin Ma and Jiongmin Yong. Adapted solution of a degenerate back ward spde, with applications. Stochastic Processes and their Applications , 70(1):59 – 84, 1997

1997
[14]

Pardoux and S

E. Pardoux and S. Peng. Backward stochastic diﬀerential equ ations and quasilinear parabolic partial diﬀerential equations. pages 200–217, 1992

1992
[15]

Forward-backward stoc hastic diﬀerential equations and quasi- linear parabolic pdes

Etienne Pardoux and Shanjian Tang. Forward-backward stoc hastic diﬀerential equations and quasi- linear parabolic pdes. Probability Theory and Related Fields , 114:123–150, 01 1999

1999
[16]

Peng and Zhen

Shige. Peng and Zhen. Wu. Fully coupled forward-backward sto chastic diﬀerential equations and applications to optimal control. SIAM Journal on Control and Optimization , 37(3):825–843, 1999

1999
[17]

Probabilistic interpretation for a syste m of quasilinear parabolic partial diﬀerential equation combined with algebra equations

Zhen Wu and Zhiyong Yu. Probabilistic interpretation for a syste m of quasilinear parabolic partial diﬀerential equation combined with algebra equations. Stochastic Processes and their Applications , 124(12):3921 – 3947, 2014

2014
[18]

Finding adapted solutions of forward-backward stochastic diﬀerential equations: method of continuation

Jiongmin Yong. Finding adapted solutions of forward-backward stochastic diﬀerential equations: method of continuation. Probability Theory and Related Fields , 107:537–572, 1997

1997
[19]

Forward-backward stochastic diﬀerential equ ations with mixed initial-terminal con- ditions

Jiongmin Yong. Forward-backward stochastic diﬀerential equ ations with mixed initial-terminal con- ditions. Trans. Amer. Math. Soc. , 362, 2009

2009
[20]

Equivalent cost functionals and stochastic linear qu adratic optimal control problems

Zhiyong Yu. Equivalent cost functionals and stochastic linear qu adratic optimal control problems. ESAIM: Control, Optimisation and Calculus of Variations , 19(1):78–90, 2013

2013
[21]

The wellposedness of fbsdes

Jianfeng Zhang. The wellposedness of fbsdes. Discrete Contin. Dyn. Syst. Ser. B , 6:927–940, 2006

2006
[22]

The wellposedness of FBSDEs (ii)

Jianfeng Zhang. The wellposedness of FBSDEs (ii). arXiv:1708.05785, 2017. 16

2017 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [5]

Fromm and P

A. Fromm and P. Imkeller. Existence, Uniqueness and Regularity o f Decoupling Fields to Multidi- mensional Fully Coupled FBSDEs. ArXiv e-prints , October 2013

work page 2013

[2] [1]

Backward-forward stochastic diﬀerential equa tions

Fabio Antonelli. Backward-forward stochastic diﬀerential equa tions. Ann. Appl. Probab. , 3(3):777– 793, 08 1993

work page 1993

[3] [2]

The steepest descent meth od for forward-backward sdes

Jaksa Cvitanic and Jianfeng Zhang. The steepest descent meth od for forward-backward sdes. Elec- tron. J. Probab. , 10:1468–1495, 2005

work page 2005

[4] [3]

Contract Theory in Continuous-Time Models

Jaksa Cvitanic and Jianfeng Zhang. Contract Theory in Continuous-Time Models . Springer, 2012

work page 2012

[5] [4]

F. Delarue. On the existence and uniqueness of solutions to fbsd es in a non-degenerate case. Stochas- tic Process, 99, 209-286, MR1918539, 2002

work page 2002

[6] [6]

Continuous-time princip al-agent problem in degenerate systems, 2019

Kaitong Hu, Zhenjie Ren, and Nizar Touzi. Continuous-time princip al-agent problem in degenerate systems, 2019

work page 2019

[7] [7]

Hu and S

Y. Hu and S. Peng. Solution of forward-backward stochastic diﬀ erential equations. Probability Theory and Related Fields , 103(2):273–283, Jun 1995

work page 1995

[8] [8]

Backward stochastic diﬀerential equatio ns and partial diﬀerential equations with quadratic growth

Magdalena Kobylanski. Backward stochastic diﬀerential equatio ns and partial diﬀerential equations with quadratic growth. The Annals of Probability , Vol.28, No.2, 558-602, 2000

work page 2000