arxiv: 2603.09757 · v2 · submitted 2026-03-10 · 🧮 math.PR · cs.NA· math.NA

Recognition: 2 theorem links

· Lean Theorem

Two-grid Penalty Approximation Scheme for Doubly Reflected BSDEs

Wonjae Lee , Hyungbin Park

Authors on Pith no claims yet

Pith reviewed 2026-05-15 13:15 UTC · model grok-4.3

classification 🧮 math.PR cs.NAmath.NA

keywords doubly reflected BSDEspenalty approximationtwo-grid schemetime discretizationerror boundsbackward SDEsfinancial barriers

0 comments

The pith

A two-grid penalty scheme approximates doubly reflected BSDEs at the optimal O(Δt^{1/2}) rate for Z-independent drivers.

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

The paper develops a numerical method for doubly reflected backward stochastic differential equations with two time-dependent obstacles. Penalization with parameter λ is combined with implicit Euler time discretization, but the forward process approximation introduces an error that gets amplified by λ. To control this, the forward SDE is solved on a finer grid and projected onto the backward grid. Under structural assumptions on the obstacles motivated by financial barriers, explicit error bounds in the three parameters are obtained, along with tuning rules that recover the target convergence rate. Numerical tests on a one-dimensional game put confirm the predicted behavior.

Core claim

For Z-independent drivers, the choice λ ≃ Δt^{-1/2} together with a finer forward step Δ̃t = O(Δt / λ²) produces an overall approximation error of order O(Δt^{1/2}). An explicit error bound in (Δt, Δ̃t, λ) is derived, and a uniform O(λ^{-1}) bound holds for the value process when the obstacles satisfy the structural conditions that prevent any single shift from eliminating both barriers simultaneously.

What carries the argument

The two-grid projection: the forward SDE is simulated on a finer time mesh Δ̃t and its values are projected onto the coarser backward grid Δt, which removes the λ-amplification of the obstacle evaluation error that arises when both lower and upper barriers are present.

If this is right

An explicit error bound is obtained that depends on the three discretization parameters Δt, Δ̃t and λ.
A uniform O(λ^{-1}) bound holds for the value process under the structural obstacle assumptions.
Nonsmooth barriers are handled via a multivariate Itô-Tanaka formula and local-time-on-surfaces arguments.
The scheme produces observed errors consistent with O(n^{-1/2}) in grid-refinement tests for the Black-Scholes game put.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The two-grid idea may carry over to other reflected BSDE approximations where multiple constraints amplify discretization error.
The observed pre-asymptotic regime with respect to λ indicates that practical tuning may require additional analysis beyond the asymptotic rates.
The local-time argument used for nonsmooth barriers suggests the method could extend to higher-dimensional obstacles if the surface local-time construction generalizes.

Load-bearing premise

The structural assumptions on the obstacles that allow sharpening the penalization rates and removing λ-amplification via the two-grid projection, because no single shift eliminates both obstacles at once.

What would settle it

Numerical runs with the stated tuning λ ~ Δt^{-1/2} and Δ̃t = O(Δt/λ²) that fail to exhibit O(n^{-1/2}) grid-refinement error, or that violate the uniform O(λ^{-1}) bound on the value process for payoffs outside the assumed barrier class.

Figures

Figures reproduced from arXiv: 2603.09757 by Hyungbin Park, Wonjae Lee.

**Figure 2.** Figure 2: Relative error as a function of the penalty parameter [PITH_FULL_IMAGE:figures/full_fig_p031_2.png] view at source ↗

read the original abstract

We study penalization coupled with time discretization for decoupled Markovian doubly reflected BSDEs with obstacles \(p_b(t,X_t)\le Y_t\le p_w(t,X_t)\). The DRBSDE is approximated by a penalized BSDE with parameter \(\lambda\) and discretized by an implicit Euler scheme with step \(\Delta t\). A key difficulty is that the forward approximation used to evaluate the obstacles generates an error term that is amplified by \(\lambda\). In the single-obstacle case this amplification can be removed by the shift \(Y-p_b(t,X)\), but no analogous transformation eliminates both obstacles simultaneously; this motivates simulating the forward SDE on a finer grid \(\tilde{\Delta t}\) and projecting onto the backward grid (two-grid scheme). Under structural assumptions motivated by financial barriers we sharpen penalization rates and obtain a uniform \(O(\lambda^{-1})\) bound for the value process. We derive an explicit error bound in \((\Delta t,\tilde{\Delta t},\lambda)\) and tuning rules; for \(Z\)-independent drivers, \(\lambda\asymp \Delta t^{-1/2}\) with \(\tilde{\Delta t}=O(\Delta t/\lambda^2)\) yields the target \(O(\Delta t^{1/2})\) rate. Nonsmooth barriers/payoffs are handled via a multivariate It\^o--Tanaka and local-time-on-surfaces argument. We also provide numerical experiments for a one-dimensional game put under the Black--Scholes model. The observed grid-refinement errors are consistent with the predicted \(O(n^{-1/2})\) behavior, while the penalty sweep indicates that the tested regime remains pre-asymptotic with respect to the penalty parameter.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The two-grid penalty scheme gives a concrete way to handle both obstacles in reflected BSDEs with explicit error bounds, but the rates rest on structural assumptions about the barriers.

read the letter

The paper's main advance is the two-grid projection that lets you run the forward SDE on a finer step and project onto the backward grid, which keeps the lambda-amplified error from the obstacle approximation under control for both upper and lower barriers at once. A single shift works for one obstacle but not both, so this projection is the practical fix they propose. Under the financial-barrier assumptions they state, they sharpen the penalization to get a uniform O(1/lambda) bound on the value process and an explicit error estimate in (Delta t, Delta tilde t, lambda). For Z-independent drivers the tuning lambda ~ Delta t^{-1/2} with Delta tilde t = O(Delta t / lambda^2) recovers the target O(Delta t^{1/2}) rate. The multivariate Ito-Tanaka plus local-time argument for nonsmooth barriers is also new in this setting and useful for payoffs that are not smooth. The one-dimensional game-put experiments line up with the predicted rate, which is the right kind of check to include. The soft spot is exactly the one the stress-test flags: everything depends on those structural assumptions on p_b and p_w. If the barriers do not satisfy them, the sharpened rates and the removal of lambda amplification do not go through, and you lose the uniform bound. The paper itself notes that the tested regime is still pre-asymptotic in lambda, so the full asymptotic picture is not yet visible in the numerics. Without the full proofs it is hard to judge how much slack is in the constants, but the derivations appear to be direct rather than circular. This is for people who already work on numerical schemes for reflected BSDEs in mathematical finance, especially game options. It has enough new technique, explicit bounds, and validation to deserve a serious referee, even if the assumptions limit the scope and some readers will want more general cases or tighter numerics.

Referee Report

3 major / 2 minor

Summary. The manuscript develops a two-grid penalty approximation for decoupled Markovian doubly reflected BSDEs with obstacles p_b(t,X_t) ≤ Y_t ≤ p_w(t,X_t). Penalization with parameter λ is combined with implicit Euler discretization of step Δt on the backward equation; the forward SDE is simulated on a finer grid of step Δ̃t and projected onto the coarse grid to control λ-amplification of approximation errors. Under structural assumptions on the obstacles motivated by financial barriers, the authors derive an explicit error bound in (Δt, Δ̃t, λ), obtain a uniform O(λ^{-1}) bound on the value process, and give tuning rules (λ ≃ Δt^{-1/2}, Δ̃t = O(Δt/λ²)) that recover the target O(Δt^{1/2}) rate for Z-independent drivers. Nonsmooth barriers are treated via a multivariate Itô-Tanaka formula and local-time-on-surfaces arguments. Numerical experiments on a one-dimensional game put under Black-Scholes are reported to be consistent with the predicted O(n^{-1/2}) behavior.

Significance. If the explicit error bounds and tuning rules hold under the stated structural assumptions, the work provides a concrete, implementable scheme with verifiable rates for DRBSDEs, which arise in game options and other stochastic control problems. The two-grid construction directly addresses the λ-amplification issue that cannot be removed by a single shift, and the derivation of parameter-free tuning rules together with the local-time treatment of nonsmooth payoffs constitute genuine technical advances. The numerical consistency with theory is a positive indicator of practical utility, though the pre-asymptotic regime noted in the penalty sweep limits immediate confirmation of the uniform bound.

major comments (3)

[Derivation of the error bound and tuning rules] The structural assumptions on the obstacles p_b and p_w are load-bearing for both the uniform O(λ^{-1}) bound on the value process and the removal of λ-amplification in the two-grid projection. These assumptions must be stated as a numbered assumption block with precise conditions (e.g., on the signs of the derivatives or barrier regularity), and the proof of the error bound should explicitly flag each place where they are invoked to control the forward-approximation term.
[Nonsmooth barriers section] The multivariate Itô-Tanaka and local-time-on-surfaces argument for nonsmooth barriers is invoked to justify the penalization rates, yet it is unclear how the local-time terms interact with the projection step from the fine forward grid to the coarse backward grid. An expanded derivation (perhaps in an appendix) showing that no additional λ-dependent remainder appears after projection is required to support the claimed O(λ^{-1}) bound.
[Numerical experiments] The numerical experiments report consistency with O(n^{-1/2}) but note that the tested regime remains pre-asymptotic with respect to λ. To corroborate the uniform O(λ^{-1}) bound and the tuning rule λ ≃ Δt^{-1/2}, additional tables or figures with successively larger λ (while keeping Δt fixed) and the corresponding measured errors are needed; the current penalty sweep alone does not yet confirm the asymptotic regime.

minor comments (2)

[Notation and scheme definition] The notation Δ̃t for the fine time step should appear explicitly in every equation that involves the two-grid projection; currently the dependence is sometimes implicit.
[Introduction] A short remark clarifying whether the Markovian assumption is essential for the projection argument or whether the scheme extends verbatim to non-Markovian drivers would improve readability.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thorough review and constructive suggestions. We address each major comment below and will revise the manuscript accordingly to incorporate the requested clarifications and additional material.

read point-by-point responses

Referee: Derivation of the error bound and tuning rules: The structural assumptions on the obstacles p_b and p_w must be stated as a numbered assumption block with precise conditions (e.g., on the signs of the derivatives or barrier regularity), and the proof of the error bound should explicitly flag each place where they are invoked to control the forward-approximation term.

Authors: We agree that the structural assumptions are central to the uniform bound and error analysis. In the revised manuscript we will introduce a dedicated numbered Assumption block that states the precise conditions on the derivatives of p_b and p_w and on barrier regularity. We will also insert explicit cross-references in the proof of the error bound (Section 4) that flag each invocation of these assumptions when controlling the forward-approximation term after projection. revision: yes
Referee: Nonsmooth barriers section: The multivariate Itô-Tanaka and local-time-on-surfaces argument is invoked to justify the penalization rates, yet it is unclear how the local-time terms interact with the projection step from the fine forward grid to the coarse backward grid. An expanded derivation (perhaps in an appendix) showing that no additional λ-dependent remainder appears after projection is required.

Authors: We acknowledge that the interaction between the local-time terms arising from the multivariate Itô-Tanaka formula and the two-grid projection step needs explicit verification. We will add an appendix that derives the projected local-time contribution in detail, confirming that the projection operator introduces no extra λ-dependent remainder and thereby preserves the claimed O(λ^{-1}) bound on the value process. revision: yes
Referee: Numerical experiments: The numerical experiments report consistency with O(n^{-1/2}) but note that the tested regime remains pre-asymptotic with respect to λ. Additional tables or figures with successively larger λ (while keeping Δt fixed) and the corresponding measured errors are needed to corroborate the uniform O(λ^{-1}) bound and the tuning rule.

Authors: We agree that further numerical evidence is required to reach the asymptotic regime for the penalty parameter. In the revised version we will include new tables and figures that fix Δt and increase λ successively, reporting the measured errors and demonstrating convergence to the predicted O(λ^{-1}) behavior, thereby confirming the tuning rule λ ≃ Δt^{-1/2}. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivations are self-contained under stated assumptions

full rationale

The paper derives explicit error bounds in (Δt, Δ̃t, λ) and tuning rules λ ≃ Δt^{-1/2}, Δ̃t = O(Δt/λ²) for the two-grid scheme directly from the penalized BSDE discretization and projection step. The structural assumptions on p_b and p_w are introduced as inputs (motivated by financial barriers) to control λ-amplification and obtain the uniform O(λ^{-1}) bound; they are not derived from the target rates. No equation reduces by construction to a fitted parameter or prior self-citation, and the Itô-Tanaka/local-time argument is applied as an independent tool for nonsmooth barriers. The central claims rest on new analysis rather than renaming or self-referential fitting.

Axiom & Free-Parameter Ledger

3 free parameters · 1 axioms · 0 invented entities

The scheme depends on discretization parameters chosen by tuning rules and on domain assumptions about the obstacles; no new physical entities are introduced.

free parameters (3)

λ (penalty parameter)
Tuned as λ ≃ Δt^{-1/2} to balance penalization and discretization errors.
Δt (coarse time step)
Primary discretization parameter whose square-root rate is targeted.
Δ̃t (fine-grid time step)
Set to O(Δt/λ²) to control forward-process approximation error.

axioms (1)

domain assumption Structural assumptions on the obstacles p_b(t,X_t) and p_w(t,X_t) motivated by financial barriers
Invoked to obtain the uniform O(λ^{-1}) bound and to sharpen penalization rates.

pith-pipeline@v0.9.0 · 5615 in / 1538 out tokens · 55835 ms · 2026-05-15T13:15:17.401249+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Under structural assumptions motivated by financial barriers we sharpen penalization rates and obtain a uniform O(λ^{-1}) bound for the value process... λ≍Δt^{-1/2} with Δ̃t=O(Δt/λ²) yields the target O(Δt^{1/2}) rate.
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Nonsmooth barriers/payoffs are handled via a multivariate Itô–Tanaka and local-time-on-surfaces argument.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages · 1 internal anchor

[1]

International Journal of Stochastic Analysis2009, 695798 (2009)

Bielecki, T.R., Crépey, S., Jeanblanc, M., Rutkowski, M.: Defaultable game options in a hazard process model. International Journal of Stochastic Analysis2009, 695798 (2009). DOI 10.1155/2009/695798

work page doi:10.1155/2009/695798 2009
[2]

Stochastic Processes and their Applications118(12), 2269–2293 (2008)

Bouchard, B., Chassagneux, J.F.: Discrete-time approximation for continuously and dis- cretely reflected BSDEs. Stochastic Processes and their Applications118(12), 2269–2293 (2008). DOI 10.1016/j.spa.2007.12.007

work page doi:10.1016/j.spa.2007.12.007 2008
[3]

Stochastic Processes and their Applications118(1), 53–75 (2008)

Bouchard, B., Elie, R.: Discrete-time approximation of decoupled forward-backward SDEs with jumps. Stochastic Processes and their Applications118(1), 53–75 (2008). DOI 10.1016/j.spa.2007.03.010

work page doi:10.1016/j.spa.2007.03.010 2008
[4]

Stochastic Processes and their Applications 111(2), 175–206 (2004)

Bouchard, B., Touzi, N.: Discrete-time approximation and Monte-Carlo simulation of backward stochastic differential equations. Stochastic Processes and their Applications 111(2), 175–206 (2004). DOI 10.1016/j.spa.2004.01.001

work page doi:10.1016/j.spa.2004.01.001 2004
[5]

Advances in Applied Probability41(1), 101–130 (2009)

Chassagneux, J.F.: A discrete-time approximation for doubly reflected BSDEs. Advances in Applied Probability41(1), 101–130 (2009). DOI 10.1239/aap/1240319578 32

work page doi:10.1239/aap/1240319578 2009
[6]

Journal of Financial Economics7(3), 229–263 (1979)

Cox, J.C., Ross, S.A., Rubinstein, M.: Option pricing: A simplified approach. Journal of Financial Economics7(3), 229–263 (1979). DOI 10.1016/0304-405X(79)90015-1

work page doi:10.1016/0304-405x(79)90015-1 1979
[7]

The Annals of Probability24(4), 2024–2056 (1996)

Cvitanić, J., Karatzas, I.: Backward stochastic differential equations with reflection and Dynkin games. The Annals of Probability24(4), 2024–2056 (1996). DOI 10.1214/aop/ 1041903216

work page doi:10.1214/aop/ 2024
[8]

Electronic Journal of Probability21(64), 1–32 (2016)

Dumitrescu, R., Quenez, M.C., Sulem, A.: Generalized Dynkin games and doubly reflected BSDEs with jumps. Electronic Journal of Probability21(64), 1–32 (2016). DOI 10.1214/ 16-EJP4568

work page 2016
[9]

Journal of Mathematical Analysis and Applications 47(2), 324–353 (1974)

Ekeland, I.: On the variational principle. Journal of Mathematical Analysis and Applications 47(2), 324–353 (1974). DOI 10.1016/0022-247X(74)90025-0

work page doi:10.1016/0022-247x(74)90025-0 1974
[10]

Elie, R.: Contrôle stochastique et méthodes numériques en finance mathématique. Ph.D. thesis, Université Paris-Dauphine (2006)

work page 2006
[11]

Proba- bility Theory and Related Fields132(2), 237–264 (2005)

Hamadène, S., Hassani, M.: BSDEs with two reflecting barriers: The general result. Proba- bility Theory and Related Fields132(2), 237–264 (2005). DOI 10.1007/s00440-004-0395-2

work page doi:10.1007/s00440-004-0395-2 2005
[12]

Kloeden, P.E., Platen, E.: Numerical Solution of Stochastic Differential Equations,Applica- tions of Mathematics, vol. 23. Springer, Berlin (1992)

work page 1992
[13]

Numerische Mathematik138(1), 219–239 (2018)

Leobacher, G., Szölgyenyi, M.: Convergence of the euler–maruyama method for multidi- mensional sdes with discontinuous drift and degenerate diffusion coefficient. Numerische Mathematik138(1), 219–239 (2018). DOI 10.1007/s00211-017-0903-9

work page doi:10.1007/s00211-017-0903-9 2018
[14]

The Review of Financial Studies14(1), 113–147 (2001)

Longstaff, F.A., Schwartz, E.S.: Valuing american options by simulation: A simple least- squares approach. The Review of Financial Studies14(1), 113–147 (2001). DOI 10.1093/ rfs/14.1.113

work page 2001
[15]

Springer (1999)

Ma, J., Yong, J.: Forward-Backward Stochastic Differential Equations and Their Applica- tions. Springer (1999)

work page 1999
[16]

The Annals of Applied Probability12(4), 1390–1418 (2002)

Ma, J., Zhang, J.: Representation theorems for backward stochastic differential equations. The Annals of Applied Probability12(4), 1390–1418 (2002). DOI 10.1214/aoap/1037125868

work page doi:10.1214/aoap/1037125868 2002
[17]

Universitext

Øksendal, B.: Stochastic Differential Equations: An Introduction with Applications, 6 edn. Universitext. Springer, Berlin (2003)

work page 2003
[18]

Deep Penalty Methods: A Class of Deep Learning Algorithms for Solving High Dimensional Optimal Stopping Problems

Peng, Y., Wei, P., Wei, W.: Deep penalty methods: A class of deep learning algorithms for solving high dimensional optimal stopping problems (2024). DOI 10.48550/arXiv.2405.11392

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2405.11392 2024
[19]

In: Séminaire de Probabilités XL,Lecture Notes in Mathematics, vol

Peskir, G.: A change-of-variable formula with local time on surfaces. In: Séminaire de Probabilités XL,Lecture Notes in Mathematics, vol. 1899, pp. 70–96. Springer, Berlin, Heidelberg (2007). DOI 10.1007/978-3-540-71189-6_2

work page doi:10.1007/978-3-540-71189-6_2 2007
[20]

Journal of Computational and Applied Mathematics 236(5), 1137–1154 (2011)

Xu, M.: Numerical algorithms and simulations for reflected backward stochastic differential equations with two continuous barriers. Journal of Computational and Applied Mathematics 236(5), 1137–1154 (2011). DOI 10.1016/j.cam.2011.07.035 33

work page doi:10.1016/j.cam.2011.07.035 2011