Abstract:In this paper, we first derive Milstein schemes for an interacting particle system associated with point delay McKean-Vlasov stochastic differential equations (McKean-Vlasov SDEs), possibly with a drift term exhibiting super-linear growth in the state component. We prove strong convergence of order one and moment stability, making use of techniques from variational calculus on the space of probability measures with finite second order moments. Then, we introduce an antithetic multilevel Milstein scheme, which leads to optimal complexity estimators for expected functionals of solutions to delay McKean-Vlasov equations without the need to simulate Lévy areas.
What problem does this paper attempt to address?
The problems that this paper attempts to solve mainly focus on two aspects:
1. **Development of numerical methods**: First, the paper derives the Milstein scheme for McKean - Vlasov stochastic differential equations (SDEs) with point delays and the related interacting particle systems. These equations may contain drift terms with super - linear growth in the state variables. The author proves the first - order strong convergence and moment stability of this scheme, using the variational calculus techniques on the space of probability measures.
2. **Design of efficient estimators**: Based on the above - mentioned Milstein scheme, the paper introduces an anti - symmetric multilevel Monte Carlo (MLMC) scheme for estimating the expected functionals of the solutions to the delayed McKean - Vlasov equations. This scheme can obtain the optimal complexity estimator without simulating the Lévy area.
### Specific problem description
- **Delayed McKean - Vlasov SDEs**:
- The form of these equations is:
\[
dX(t) = b(X(t), X(t-\tau), \mathcal{L}_{X(t)}, \mathcal{L}_{X(t-\tau)}) dt + \sigma(X(t), X(t-\tau), \mathcal{L}_{X(t)}, \mathcal{L}_{X(t-\tau)}) dW(t),
\]
where \( W(t) \) is a standard Brownian motion, \( \mathcal{L}_{X(t)} \) represents the marginal distribution of the process \( X(t) \) at time \( t \), and \( \tau \) is a given deterministic delay parameter.
- **Challenges of numerical methods**:
- When dealing with SDEs with super - linear growth drift terms, the moments of the discretization process of the standard Euler - Maruyama method will explode as the grid size decreases. Therefore, it is necessary to develop higher - order and stable numerical methods.
- The simulation of delayed SDEs requires the calculation of iterative stochastic integrals, which increases the computational complexity.
### Main contributions
1. **Derivation of the Milstein scheme**:
- The author derives the Milstein scheme applicable to McKean - Vlasov SDEs with point delays and proves the stability and convergence of this scheme under first - order strong convergence.
- The drift term is allowed to have super - linear growth in the current state variable but does not depend on the delay value.
2. **Anti - symmetric multilevel Monte Carlo scheme**:
- An anti - symmetric multilevel Milstein scheme is introduced, which avoids the need to directly simulate iterative stochastic integrals while maintaining high - order convergence.
- Through this method, efficient estimation of the expected functionals of the solutions to the delayed McKean - Vlasov equations can be achieved without increasing the computational complexity.
### Application background
- **Medicine, biology and finance**:
- Delayed McKean - Vlasov SDEs have wide applications in these fields. For example, the incubation period of infectious diseases, the signal transmission delay in neural networks, and the hysteresis effect in financial systems.
By solving these problems, the paper provides an efficient and stable tool for the numerical simulation of delayed McKean - Vlasov SDEs, which is helpful for more accurate modeling and analysis of complex systems in practical applications.