Abstract:In this paper we introduce a multilevel Picard approximation algorithm for semilinear parabolic partial integro-differential equations (PIDEs). We prove that the numerical approximation scheme converges to the unique viscosity solution of the PIDE under consideration. To that end, we derive a Feynman-Kac representation for the unique viscosity solution of the semilinear PIDE, extending the classical Feynman-Kac representation for linear PIDEs. Furthermore, we show that the algorithm does not suffer from the curse of dimensionality, i.e. the computational complexity of the algorithm is bounded polynomially in the dimension $d$ and the reciprocal of the prescribed accuracy $\varepsilon$. We also provide a numerical example in up to 10'000 dimensions to demonstrate its applicability.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to develop a Multilevel Picard Approximation (MLP) algorithm for solving semilinear parabolic partial integro - differential equations (PIDEs). Specifically, the main objectives of the paper include:
1. **Algorithm Development**: Propose a multilevel Picard approximation algorithm that can effectively solve high - dimensional semilinear PIDEs.
2. **Convergence Analysis**: Prove that the proposed numerical approximation scheme can converge to the unique viscosity solution of the PIDE.
3. **Complexity Analysis**: Prove that the algorithm is not affected by the curse of dimensionality, that is, the computational complexity of the algorithm is polynomially bounded between the dimension \(d\) and the reciprocal of the required precision \(\epsilon\).
4. **Numerical Verification**: Demonstrate the applicability of the algorithm in cases up to 10,000 dimensions through a numerical example.
### Main Contributions
- **Feynman - Kac Representation**: Derive the Feynman - Kac representation of the unique viscosity solution of the semilinear PIDE, which is an extension of the classical Feynman - Kac representation to nonlinear PIDEs.
- **Algorithm Design**: Based on the Feynman - Kac representation, define a multilevel Picard iteration sequence and propose the corresponding MLP algorithm.
- **Theoretical Guarantees**: Provide convergence and complexity analyses of the algorithm, proving the effectiveness and feasibility of the algorithm in high - dimensional cases.
- **Practical Applications**: Through an example of financial derivative pricing, demonstrate the application of the algorithm in practical problems.
### Mathematical Background
The mathematical tools and techniques involved in the paper include:
- **Partial Integro - Differential Equations (PIDEs)**: Describe the dynamics of stochastic processes with jump terms.
- **Feynman - Kac Formula**: Relate the solution of the PIDE to the expected value of a stochastic process.
- **Multilevel Picard Approximation**: Combine the Monte Carlo method and time discretization to gradually approximate the solution of the PIDE.
- **Viscosity Solution Theory**: Used to handle the existence and uniqueness of solutions to nonlinear PIDEs.
### Formula Examples
Some of the key formulas used in the paper include:
1. **The form of PIDE**:
\[
\frac{\partial u_d(t, x)}{\partial t}+\langle\nabla u_d(t, x), \mu_d(t, x)\rangle+\frac{1}{2}\text{Trace}(\sigma_d(t, x)[\sigma_d(t, x)]^T\text{Hess}_x(u_d)(t, x))+f_d(t, x, u_d(t, x))+\int_{\mathbb{R}^d}(u_d(t, x + \eta_d(t, x, z))-u_d(t, x)-\langle\nabla u_d(t, x), \eta_d(t, x, z)\rangle)\nu_d(dz) = 0
\]
2. **Feynman - Kac Representation**:
\[
u_d(t, x)=\mathbb{E}[g_d(X_{d,t,x}^T)]+\int_t^T\mathbb{E}[f_d(s, X_{d,t,x}^s, u_d(s, X_{d,t,x}^s))]ds
\]
3. **Multilevel Picard Approximation**:
\[
u_d^{k}(t, x)=(\Phi_d\circ u_d^{k - 1})(t, x)
\]
where,
\[
(\Phi_d\circ v_d)(t, x)=\m