Quantum-inspired nonlinear Galerkin ansatz for high-dimensional HJB equations

Chuhao Sun,Asaf Cohen,James Stokes,Shravan Veerapaneni

2023-11-21

Abstract:Neural networks are increasingly recognized as a powerful numerical solution technique for partial differential equations (PDEs) arising in diverse scientific computing domains, including quantum many-body physics. In the context of time-dependent PDEs, the dominant paradigm involves casting the approximate solution in terms of stochastic minimization of an objective function given by the norm of the PDE residual, viewed as a function of the neural network parameters. Recently, advancements have been made in the direction of an alternative approach which shares aspects of nonlinearly parametrized Galerkin methods and variational quantum Monte Carlo, especially for high-dimensional, time-dependent PDEs that extend beyond the usual scope of quantum physics. This paper is inspired by the potential of solving Hamilton-Jacobi-Bellman (HJB) PDEs using Neural Galerkin methods and commences the exploration of nonlinearly parametrized trial functions for which the evolution equations are analytically tractable. As a precursor to the Neural Galerkin scheme, we present trial functions with evolution equations that admit closed-form solutions, focusing on time-dependent HJB equations relevant to finance.

Numerical Analysis,Computational Finance

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to use the Nonlinear Galerkin (NG) method to solve high - dimensional time - dependent Hamilton - Jacobi - Bellman (HJB) equations in the financial field. Specifically, the paper focuses on how to improve the selection of neural network architectures by designing test functions that are easy to be analytically controlled, especially for those high - dimensional PDE problems that are difficult to handle in traditional numerical methods. The paper proposes a new method, that is, using test functions of evolution equations with closed - form solutions. These test functions can be used as a starting point for neural network architectures based on normalizing flows, thus providing the possibility of designing customized neural network architectures for specific PDE problems.

The core contribution of the paper lies in proposing a framework, which not only provides an approximate closed - form solution method, but also helps to improve the interpretability of neural network models, which is crucial for understanding and optimizing the decision - making process in complex financial models. In particular, in incomplete markets, when perfect replication strategies are not feasible, the paper explores how to determine the utility indifference price of options by solving two utility maximization problems, that is, the price at which investors reach the same expected utility level whether they buy derivatives or not. This involves solving the derivative pricing problem related to the value of non - trading assets, allowing investors to strategically allocate initial capital according to whether they buy derivatives at the indifference price.