Abstract:We present a deep learning algorithm for the numerical solution of parametric families of high-dimensional linear Kolmogorov partial differential equations (PDEs). Our method is based on reformulating the numerical approximation of a whole family of Kolmogorov PDEs as a single statistical learning problem using the Feynman-Kac formula. Successful numerical experiments are presented, which empirically confirm the functionality and efficiency of our proposed algorithm in the case of heat equations and Black-Scholes option pricing models parametrized by affine-linear coefficient functions. We show that a single deep neural network trained on simulated data is capable of learning the solution functions of an entire family of PDEs on a full space-time region. Most notably, our numerical observations and theoretical results also demonstrate that the proposed method does not suffer from the curse of dimensionality, distinguishing it from almost all standard numerical methods for PDEs.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the efficient computation of numerical solutions for high - dimensional linear Kolmogorov partial differential equations (PDEs). Specifically, the author proposes an algorithm based on deep learning for numerically solving parameterized high - dimensional Kolmogorov PDEs. Such equations frequently appear in multiple scientific fields such as physics and financial engineering, but traditional numerical methods are usually affected by the curse of dimensionality, that is, as the dimension of the problem increases, the computational cost grows exponentially.
### Main contributions:
1. **Algorithm innovation**: A new deep - learning algorithm is proposed, which can simultaneously solve the entire parameter family of Kolmogorov PDEs in a complete space - time region. This is the first method to achieve this goal.
2. **Theoretical guarantee**: By non - trivially applying the Feynman - Kac formula, the problem of solving parameterized PDEs is transformed into a supervised learning problem, and it is proved that this method is not affected by the curse of dimensionality.
3. **Practical application**: The effectiveness and efficiency of this method in the heat equation and the Black - Scholes option pricing model are demonstrated, especially in high - dimensional cases.
### Specific problem description:
Consider the linear parabolic partial differential equations (Kolmogorov PDEs) in the following form:
\[
\frac{\partial u_\gamma}{\partial t}=\frac{1}{2}\text{Trace}(\sigma_\gamma [\sigma_\gamma]^* \nabla_x^2 u_\gamma)+\langle \mu_\gamma, \nabla_x u_\gamma \rangle, \quad u_\gamma(x, 0)=\phi_\gamma(x)
\]
where $\phi_\gamma: \mathbb{R}^d \to \mathbb{R}$ is the initial condition, $\sigma_\gamma: \mathbb{R}^d \to \mathbb{R}^{d\times d}$ and $\mu_\gamma: \mathbb{R}^d \to \mathbb{R}^d$ are coefficient mappings, and these functions are implicitly determined by the real - parameter vector $\gamma \in D$, where $D$ is a compact set in Euclidean space.
### Method overview:
1. **Problem transformation**: Through the Feynman - Kac formula, the numerical approximation problem of Kolmogorov PDEs is transformed into a supervised learning problem. Specifically, a predictor variable $\Lambda = (\Gamma, X, T)$ and a target variable $Y = \phi_\Gamma(S_\Lambda)$ are constructed, where $S_\Lambda$ is the solution of the stochastic differential equation (SDE).
2. **Deep - learning model**: Train a deep neural network $\Phi: D\times [v, w]^d\times [0, T]\to \mathbb{R}$ to approximate the parameterized solution mapping $\bar{u}(\gamma, x, t)=u_\gamma(x, t)$.
3. **Sample generation**: Use the Euler - Maruyama scheme to generate training data, which can effectively simulate the solution of SDE in high - dimensional cases without being affected by the curse of dimensionality.
### Experimental results:
- **Black - Scholes model**: It is shown that the neural network can learn the parameterized Black - Scholes option pricing model and achieves an average L1 error of 0.0011 after 24000 gradient updates.
- **Basket Put Option**: In the more complex Basket Put Option pricing problem, good results are also obtained, which verifies the effectiveness of this method in high - dimensional cases.
- **Heat equation**: Experiments are carried out in the 10 - dimensional and 150 - dimensional heat equations with parabolic and Gaussian initial conditions respectively.