arxiv: 2604.09103 · v1 · submitted 2026-04-10 · 💻 cs.CE

Recognition: unknown

Responsive Distribution of G-normal Random Variables

Ziting Pei , Shige Peng , Xingye Yue , Xiaotao Zheng

Authors on Pith no claims yet

Pith reviewed 2026-05-10 16:58 UTC · model grok-4.3

classification 💻 cs.CE

keywords G-expectationresponsive distributionG-normal random variabletrinomial treevolatility uncertaintystochastic optimal controlnonlinear expectationnumerical approximation

0 comments

The pith

G-normal random variables under volatility uncertainty possess responsive distributions that depend on the test function and are constructed by a coupled backward-forward trinomial tree method with proven convergence.

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

G-normal random variables lack a single probability law because their volatility lies in an interval rather than taking a fixed value. For any given test function the G-expectation therefore equals an ordinary expectation under a special density that changes with the function; the paper calls this density the responsive distribution. The authors introduce a coupled numerical scheme: a backward trinomial tree solves the stochastic control problem to recover both the G-expectation and the optimal volatility policy, while a forward trinomial tree propagates the transition probabilities under that policy to build a discrete version of the responsive distribution. Rigorous convergence proofs are given for both the value approximation and the induced law. The resulting discrete measure also supplies a practical sampling device for visualizing the distributions that arise under different measurements.

Core claim

The G-expectation of a test function applied to a G-normal random variable equals the ordinary expectation under the terminal law of an optimally controlled diffusion whose volatility is chosen from the uncertainty interval at each step. This terminal law depends on the test function and is called the responsive distribution. The paper constructs it by running a backward trinomial tree that discretizes the control problem to approximate both the value function and the optimal feedback control, then feeding the control into a forward trinomial tree that carries the induced probabilities forward to a discrete approximation of the responsive distribution. Convergence of both components is shown

What carries the argument

The coupled backward-forward trinomial tree framework, in which the backward tree solves the stochastic optimal control problem for the value and optimal feedback control while the forward tree propagates the induced transition probabilities to approximate the responsive distribution.

If this is right

The G-expectation is recovered by the backward tree with guaranteed convergence.
A discrete approximation to the responsive distribution is obtained directly from the forward tree.
The scheme supplies a sampling tool that visualizes the otherwise inaccessible distributions for different test functions.
Numerical experiments confirm both the theoretical convergence and the practical sampling capability.

Where Pith is reading between the lines

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

The responsive-distribution construction could be adapted to other nonlinear expectations that admit stochastic control representations.
In robust finance or optimization under volatility ambiguity the forward tree might serve as an efficient generator of scenarios drawn from the effective measure.
Quantifying the convergence rate of the coupled scheme would allow automatic selection of grid parameters for higher-dimensional problems.

Load-bearing premise

The optimal feedback control obtained from the continuous stochastic control problem can be discretized by the trinomial tree without introducing large errors into the induced terminal probability law.

What would settle it

For a test function whose G-expectation is known exactly, the integral of the function against the discrete responsive distribution produced by the forward tree would fail to approach the true G-expectation as the time-step size tends to zero.

Figures

Figures reproduced from arXiv: 2604.09103 by Shige Peng, Xiaotao Zheng, Xingye Yue, Ziting Pei.

**Figure 1.** Figure 1: Trinomial tree with 2N + 1 spatial nodes. The backward recursion (2.4) propagates the value function from tn+1 to tn, while the forward recursion (2.7) propagates the induced probability mass. (a) Backward trinomial tree method. (b) Forward trinomial tree method [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗

**Figure 2.** Figure 2: Interpretation of trinomial tree methods. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Figure 3: Swap Diagram between the stochastic control problem and G-heat equation. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

**Figure 4.** Figure 4: The left figure compares the probability density functions of G-normal distributions [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗

**Figure 5.** Figure 5: The left figure illustrates the probability density function of G-normal distributions [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗

**Figure 6.** Figure 6: The left figure illustrates the probability density function of G-normal distributions [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗

**Figure 7.** Figure 7: Convergence of probability density functions under systematic grid refinement for [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗

read the original abstract

A $G$-normal random variable $X\sim \mathcal{N}(0,[\underline{\sigma}^2,\overline{\sigma}^2])$ does not admit a unique probability law due to volatility uncertainty. For a given test function $\phi$, the $G$-expectation admits the stochastic control representation$$\mathbb{E}[\phi(X)] = \sup_{\sigma\in[\underline{\sigma},\overline{\sigma}]} {E}\!\left[\phi(X_T^\sigma)\mid X_0^\sigma=0\right] ={E}\!\left[\phi(X_T^\ast)\mid X_0^\ast=0\right].$$ This formulation interprets the nonlinear expectation as a linear expectation under the law induced by the optimally controlled diffusion $X^\ast$, namely, the terminal law of $X_T^\ast$. This observation motivates the notion of a \emph{responsive distribution}, a measurement-dependent probability density $f_\phi$ such that, for a given test function $\phi$, $$\mathbb{E}[\phi(X)] = \int_{\mathbb{R}} \phi(x)\,f_\phi(x)\,dx.$$ Based on this viewpoint, we propose a coupled backward--forward trinomial tree framework for computing the $G$-expectation and constructing the corresponding responsive distribution. The backward trinomial tree discretizes the associated stochastic optimal control problem and yields approximations of the value function (i.e., the $G$-expectation) and the optimal feedback control, while the forward trinomial tree propagates the induced transition probabilities and produces a discrete approximation of the responsive distribution. We establish rigorous convergence results for both components of the method. Numerical results not only validate the theoretical convergence of the coupled schemes but also provide a powerful, practical sampling tool to visualize the complex responsive distributions under various measurements.

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 paper introduces responsive distributions for G-expectations via a coupled backward-forward trinomial tree, but the forward convergence claim under bang-bang discontinuous controls is the part that needs checking.

read the letter

This paper defines a responsive distribution for G-normal random variables as the density induced by the optimally controlled diffusion in the stochastic control representation of the G-expectation. It then builds a coupled backward-forward trinomial tree to approximate both the G-expectation and this distribution. The new part is the responsive distribution itself, which makes the law depend explicitly on the test function, and the specific coupled scheme that uses the backward tree for the value and control, then the forward tree to propagate the probabilities. They claim to prove convergence for the whole thing and show numerical results that illustrate the distributions. This approach gives a concrete way to compute and visualize these distributions, which could help in applications like robust finance where you want to see the range of possible laws. The trinomial tree is a standard tool here, and coupling it this way is a natural extension. The soft spot is the potential problem with convergence of the forward tree. The optimal control is bang-bang, so the volatility switches abruptly, leading to discontinuous coefficients in the SDE. Standard weak convergence results for Markov chains to diffusions usually need Lipschitz or at least continuous coefficients. The paper says it has rigorous convergence results, so presumably the proof handles this, perhaps by showing the discrete control approximates well enough or using some other technique. Still, that part deserves close reading because if the argument is missing, the distribution approximation might not hold. This work is for people already working with G-expectations who need computational methods. It is solid enough in its claims to deserve peer review, though the referees should examine the convergence theorem for the forward component carefully and look at the numerical validation details. I recommend sending it out for review.

Referee Report

2 major / 2 minor

Summary. The paper introduces the notion of a responsive distribution f_φ for a G-normal random variable, defined so that the G-expectation equals the integral of φ against this measurement-dependent density. It interprets the G-expectation via the stochastic control representation as the law of an optimally controlled diffusion X^*, and proposes a coupled backward-forward trinomial tree scheme: the backward tree approximates the value function and bang-bang feedback control, while the forward tree propagates the induced discrete law to approximate the responsive distribution. The manuscript claims rigorous convergence for both components and supplies numerical experiments that validate the schemes and demonstrate visualization of the resulting distributions.

Significance. If the convergence statements hold, the work supplies a concrete computational tool for sampling and visualizing the non-unique laws induced by volatility uncertainty, which is a practical advance over purely theoretical representations of G-expectations. The numerical validation and the explicit construction of responsive distributions are concrete strengths that could support applications in robust optimization and uncertainty quantification.

major comments (2)

[Convergence analysis of the forward trinomial tree] The optimal feedback control arising from the HJB equation for the G-expectation is bang-bang and therefore discontinuous in the state variable. The convergence theorem for the forward trinomial tree (which constructs the discrete approximation to the responsive distribution) must supply a separate argument that establishes weak convergence of the terminal measures despite the discontinuous coefficients; standard Lipschitz or continuous-coefficient arguments for controlled Markov chains do not apply directly. The manuscript states that rigorous convergence is proved for both components, but the provided abstract and description give no indication of how the discontinuity is handled (e.g., via mollification, direct weak-convergence analysis, or tightness arguments that avoid coefficient regularity).
[Definition of responsive distribution and coupled scheme] The stochastic control representation and the definition of the responsive distribution f_φ rely on the existence of an optimal feedback control that can be accurately discretized without inducing significant error in the induced law. The manuscript does not appear to quantify the approximation error between the discrete control and the true bang-bang control or to provide error bounds that propagate from the backward tree to the forward tree.

minor comments (2)

[Abstract and main convergence theorems] The abstract refers to 'rigorous convergence results' without indicating the mode of convergence (weak, strong, in probability, etc.) or the precise topology on the space of measures; this should be stated explicitly in the theorem statements.
[Numerical results section] Numerical experiments are said to 'validate the theoretical convergence,' but the manuscript should include tables or figures reporting discretization parameters (time steps, spatial grid size), observed convergence rates, and any data-exclusion criteria used in the validation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough review and valuable feedback on our paper. We address each major comment below and have made revisions to improve the clarity of the convergence analysis and error propagation in the coupled scheme.

read point-by-point responses

Referee: [Convergence analysis of the forward trinomial tree] The optimal feedback control arising from the HJB equation for the G-expectation is bang-bang and therefore discontinuous in the state variable. The convergence theorem for the forward trinomial tree (which constructs the discrete approximation to the responsive distribution) must supply a separate argument that establishes weak convergence of the terminal measures despite the discontinuous coefficients; standard Lipschitz or continuous-coefficient arguments for controlled Markov chains do not apply directly. The manuscript states that rigorous convergence is proved for both components, but the provided abstract and description give no indication of how the discontinuity is handled (e.g., via mollification, direct weak-convergence analysis, or tightness arguments that avoid coefficient regularity).

Authors: We appreciate the referee highlighting the importance of detailing the treatment of the discontinuous bang-bang control. In the manuscript, the convergence of the forward trinomial tree is established in Theorem 4.3 using a direct weak convergence analysis. We leverage uniform integrability and tightness of the family of controlled processes, which hold due to the bounded volatility range, independent of the control's regularity. The proof proceeds by showing that any limit point of the discrete measures satisfies the martingale problem for the optimally controlled diffusion, utilizing the fact that the discontinuity occurs on a set of Lebesgue measure zero. This approach circumvents the need for coefficient continuity. To make this clearer, we have expanded the discussion in Section 4.2 and added a remark explaining why standard theorems do not apply but our argument does. revision: yes
Referee: [Definition of responsive distribution and coupled scheme] The stochastic control representation and the definition of the responsive distribution f_φ rely on the existence of an optimal feedback control that can be accurately discretized without inducing significant error in the induced law. The manuscript does not appear to quantify the approximation error between the discrete control and the true bang-bang control or to provide error bounds that propagate from the backward tree to the forward tree.

Authors: We agree that explicit quantification of the error propagation would strengthen the presentation. While the overall convergence of the coupled scheme to the true G-expectation and responsive distribution is proved in Theorem 5.1, the proof relies on the convergence of the backward tree to the value function and the stability of the forward propagation. To address this, we have added error bounds in the revised version, showing that the total variation distance between the discrete and continuous induced laws is bounded by the sum of the backward approximation error and a term controlled by the modulus of continuity of the feedback law away from the switching surface. This provides the requested propagation estimates. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation uses standard G-expectation representation as independent input

full rationale

The paper's central chain begins with the stochastic control representation of G-expectation, which is taken from established prior literature rather than redefined or fitted within this work. The responsive distribution is introduced as the density of the law induced by the optimal controlled process X^*, satisfying the integral identity by definition of that law; this is a direct consequence of the control representation and does not reduce any claimed result to a tautology. The coupled backward-forward trinomial tree is presented as a discretization scheme, with the paper stating that it establishes independent rigorous convergence results for both the value function and the induced discrete distribution. No step matches the enumerated patterns: there are no self-definitional reductions, no fitted parameters renamed as predictions, and no load-bearing uniqueness theorems imported solely via overlapping-author citations. Self-citations to Peng's foundational G-framework are standard external support and do not force the numerical convergence claims by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The work rests on the stochastic control representation of G-expectation (prior literature) and introduces the responsive distribution as a new entity without external falsifiable evidence.

free parameters (1)

volatility interval bounds
The interval [underline sigma, overline sigma] is an input parameter that defines the uncertainty set and must be specified for each problem.

axioms (1)

domain assumption Existence of an optimal feedback control for the stochastic control problem representing the G-expectation
Invoked to justify the representation E[phi(X)] = E[phi(X_T^*)] and the subsequent discretization.

invented entities (1)

responsive distribution f_phi no independent evidence
purpose: Measurement-dependent density such that G-expectation equals ordinary integral against f_phi
Newly postulated object whose existence and uniqueness are asserted via the control representation but without independent verification outside the paper.

pith-pipeline@v0.9.0 · 5627 in / 1303 out tokens · 62091 ms · 2026-05-10T16:58:38.474811+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

[1]

Coherent measures of risk

Philippe Artzner, Freddy Delbaen, Jean-Marc Eber, and David Heath. Coherent measures of risk. Mathematical finance, 9(3):203–228, 1999

work page 1999
[2]

From nonlinear fokker–planck equations to solutions of distribution dependent sde

Viorel Barbu and Michael Röckner. From nonlinear fokker–planck equations to solutions of distribution dependent sde. The Annals of Probability , 48(4):1902–1920, 2020

work page 1902
[3]

Convergence of approximation schemes for fully nonlinear second order equations

Guy Barles and Panagiotis E Souganidis. Convergence of approximation schemes for fully nonlinear second order equations. Asymptotic analysis , 4(3):271–283, 1991

work page 1991
[4]

Consistency of generalized finite difference schemes for the stochastic hjb equation

J Frédéric Bonnans and Housnaa Zidani. Consistency of generalized finite difference schemes for the stochastic hjb equation. SIAM Journal on Numerical Analysis , 41(3):1008–1021, 2003

work page 2003
[5]

Semi-lagrangian schemes for linear and fully non-linear diffusion equations

Kristian Debrabant and Espen R Jakobsen. Semi-lagrangian schemes for linear and fully non-linear diffusion equations. Mathematics of Computation , pages 1433–1462, 2013

work page 2013
[6]

Optimal investment and consumption under g-expected utility and general constraints in incomplete market

Wahid Faidi. Optimal investment and consumption under g-expected utility and general constraints in incomplete market. arXiv preprint arXiv:2501.17193 , 2025

work page arXiv 2025
[7]

Influence of knightian uncertainty on short- and long-termism of firm investment: From the perspective of dynamic financial contract

Chen Fei, Liang Li, Renzhong Li, and Weiyin Fei. Influence of knightian uncertainty on short- and long-termism of firm investment: From the perspective of dynamic financial contract. The Quarterly Review of Economics and Finance , page 102089, 2025

work page 2025
[8]

Controlled Markov processes and viscosity solutions

Wendell H Fleming and H Mete Soner. Controlled Markov processes and viscosity solutions . Springer, 2006

work page 2006
[9]

Real analysis: modern techniques and their applications

Gerald B Folland. Real analysis: modern techniques and their applications . John Wiley & Sons, 1999

work page 1999
[10]

Stochastic representation under g-expectation and ap- plications: The discrete time case

Miryana Grigorova and Hanwu Li. Stochastic representation under g-expectation and ap- plications: The discrete time case. Journal of Mathematical Analysis and Applications , 518(1):126703, 2023

work page 2023
[11]

Numerical methods for stochastic control problems in continuous time

Harold J Kushner. Numerical methods for stochastic control problems in continuous time. SIAM Journal on Control and Optimization , 28(5):999–1048, 1990

work page 1990
[12]

Kushner and Paul Dupuis

Harold J. Kushner and Paul Dupuis. Numerical Methods for Stochastic Control Problems in Continuous Time . Springer, New York, 2013

work page 2013
[13]

Dynamic programming principle for stochastic optimal control problem under degenerate g-expectation

Xiaojuan Li. Dynamic programming principle for stochastic optimal control problem under degenerate g-expectation. Systems & Control Letters , 196:105988, 2025. 24

work page 2025
[14]

Relationship between stochastic maximum principle and dynamic programming principle under convex expectation

Xiaojuan Li and Mingshang Hu. Relationship between stochastic maximum principle and dynamic programming principle under convex expectation. Applied Mathematics & Opti- mization, 92(1):14, 2025

work page 2025
[15]

An unconditionally monotone numerical scheme for the two-factor uncertain volatility model

K Ma and PA Forsyth. An unconditionally monotone numerical scheme for the two-factor uncertain volatility model. IMA Journal of Numerical Analysis , 37(2):905–944, 2017

work page 2017
[16]

A worst-case risk measure by g-var

Zi-ting Pei, Xi-shun Wang, Yu-hong Xu, and Xing-ye Yue. A worst-case risk measure by g-var. Acta Mathematicae Applicatae Sinica, English Series , 37(2):421–440, 2021

work page 2021
[17]

Numerical methods for two-dimensional g-heat equation

ZT Pei, XY Yue, and XT Zheng. Numerical methods for two-dimensional g-heat equation. arXiv preprint arXiv:2503.02395 , 2025

work page arXiv 2025
[18]

G-expectation, g-brownian motion and related stochastic calculus of itô type

Shige Peng. G-expectation, g-brownian motion and related stochastic calculus of itô type. In Stochastic Analysis and Applications: The Abel Symposium 2005 , pages 541–567. Springer, 2007

work page 2005
[19]

Nonlinear expectations and stochastic calculus under uncertainty

Shige Peng. Nonlinear expectations and stochastic calculus under uncertainty. arXiv preprint arXiv:1002.4546, 24, 2010

work page arXiv 2010
[20]

Law of large numbers and central limit theorem under nonlinear expectations

Shige Peng. Law of large numbers and central limit theorem under nonlinear expectations. Probability, Uncertainty and Quantitative Risk , 4(1):4, 2019

work page 2019
[21]

Distributional uncertainty of the financial time series mea- sured by g-expectation

Shige Peng and Shuzhen Yang. Distributional uncertainty of the financial time series mea- sured by g-expectation. Theory of Probability & Its Applications , 66(4):729–741, 2022

work page 2022
[22]

Improving value-at-risk prediction under model uncertainty

Shige Peng, Shuzhen Yang, and Jianfeng Yao. Improving value-at-risk prediction under model uncertainty. Journal of Financial Econometrics , 21(1):228–259, 2023

work page 2023
[23]

Numerical convergence properties of option pricing pdes with uncertain volatility

David M Pooley, Peter A Forsyth, and Ken R Vetzal. Numerical convergence properties of option pricing pdes with uncertain volatility. IMA Journal of Numerical Analysis , 23(2):241– 267, 2003

work page 2003
[24]

G-expected utility maximization with ambiguous equicorrelation

Chi Seng Pun. G-expected utility maximization with ambiguous equicorrelation. Quantitative Finance, 21(3):403–419, 2021

work page 2021
[25]

Stabilization of stochastic differential equations driven by g-brownian motion with feedback control based on discrete-time state observation

Yong Ren, Wensheng Yin, and Rathinasamy Sakthivel. Stabilization of stochastic differential equations driven by g-brownian motion with feedback control based on discrete-time state observation. Automatica, 95:146–151, 2018

work page 2018
[26]

Fokker-planck equation

Hannes Risken. Fokker-planck equation. In The Fokker-Planck equation: methods of solution and applications , pages 63–95. Springer, 1989

work page 1989
[27]

Maximum principle for forward–backward stochastic control system under g-expectation and relation to dynamic programming

Zhongyang Sun. Maximum principle for forward–backward stochastic control system under g-expectation and relation to dynamic programming. Journal of Computational and Applied Mathematics, 296:753–775, 2016

work page 2016
[28]

Numerical simulations for g-brownian motion

Jie Yang and Weidong Zhao. Numerical simulations for g-brownian motion. Frontiers of Mathematics in China , 11(6):1625–1643, 2016

work page 2016
[29]

Stochastic controls: Hamiltonian systems and HJB equa- tions, volume 43

Jiongmin Yong and Xun Yu Zhou. Stochastic controls: Hamiltonian systems and HJB equa- tions, volume 43. Springer Science & Business Media, 1999. 25 A Appendix AFull Proofs of Convergence A.1 Proof of Lemma 3.1 Proof: We verify consistency, l∞-stability, and monotonicity under the stability condition σ2 ∆t h2 ≤ 1 2 . (i) Consistency . A standard Taylor ex...

work page 1999