Abstract:Fractional Brownian motions(fBMs) are not semimartingales so the classical theory of Itô integral can't apply to fBMs. Wick integration as one of the applications of Malliavin calculus to stochastic analysis is a fine definition for fBMs. We consider the fractional forward backward stochastic differential equations(fFBSDEs) driven by a fBM that have the Hurst parameter in (1/2,1) where $\int_{0}^{t} f_s \, dB_s^H$ is in the sense of a Wick integral, and relate our fFBSDEs to the system of partial differential equations by using an analogue of the Itô formula for Wick integrals. And we develop a deep learning algorithm referred to as the RNN-BSDE method based on recurrent neural networks which is exactly designed for solving high-dimensional fractional BSDEs and their corresponding partial differential equations.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: how to use deep - learning methods to solve high - dimensional fractional - order backward stochastic differential equations (fBSDEs), especially those equations driven by fractional Brownian motion with Hurst parameter \(H\in(\frac{1}{2}, 1)\). Since fractional Brownian motion is not a semimartingale, the classical Ito integral theory is not applicable to fractional Brownian motion, so Wick integral needs to be adopted to define these equations.
Specifically, the goals of the paper are:
1. **Develop a deep - learning algorithm based on recurrent neural network (RNN)**, called the RNN - BSDE method, for solving high - dimensional fractional - order backward stochastic differential equations and their corresponding partial differential equations (PDEs).
2. **Establish the connection between fractional - order forward - backward stochastic differential equations (fFBSDEs) and partial differential equations**, and use methods similar to the Ito formula to deal with Wick integrals.
3. **Verify the effectiveness of the proposed RNN - BSDE method through numerical experiments**, and compare it with the existing methods for solving standard BSDEs and PDEs.
### Main problems
- **Special properties of fractional Brownian motion**: Fractional Brownian motion is not a semimartingale, so the traditional Ito integral cannot be directly applied, and Wick integral needs to be introduced.
- **Complexity of high - dimensional problems**: As the dimension increases, the computational complexity of traditional numerical methods rises sharply, while deep - learning methods can effectively deal with high - dimensional problems.
- **Long - memory characteristics**: When the Hurst parameter \(H\in(\frac{1}{2}, 1)\), fractional Brownian motion has long - memory characteristics, which makes it difficult for traditional recurrent neural networks to handle, so more powerful structures such as LSTM need to be used.
### Solutions
The paper proposes a deep - learning method based on recurrent neural network (RNN) - the RNN - BSDE method, which is especially suitable for dealing with fractional - order backward stochastic differential equations. This method is implemented through the following steps:
1. **Time discretization**: Discretize the BSDE into a series of time steps for the convenience of numerical solution.
2. **Wick product processing**: Use Wick integral to deal with the non - semimartingale characteristics in fractional Brownian motion.
3. **Recurrent neural network structure**: Adopt a multi - layer LSTM or stacked RNN structure to make full use of historical information and improve the expressive ability of the model.
4. **Loss function optimization**: Train the model parameters by minimizing the objective function (such as Euclidean distance).
Through the above methods, the paper aims to provide an efficient and accurate tool for solving high - dimensional fractional - order backward stochastic differential equations and verify its application effect in practical problems.