On Approximate Representation of Fractional Brownian Motion

Basic Information

• Paper ID: 2503.04575
• Title: On Approximate Representation of Fractional Brownian Motion
• Author: Konstantin A. Rybakov
• Classification: math.PR (Probability Theory)
• Publication Date: March 2025 (arXiv preprint, version v2 on October 14, 2025)
• Paper Link: https://arxiv.org/abs/2503.04575

Abstract

This paper investigates the orthogonal expansion of fractional Brownian motion with respect to Legendre polynomials. This expansion possesses both theoretical significance and practical value, as it can be utilized for approximation and simulation of fractional Brownian motion in continuous time. The paper provides relationships for mean-square approximation errors and compares them with previously obtained results.

Research Background and Motivation

1. Core Problem: How to effectively approximate and simulate fractional Brownian motion (FBM) in continuous time
2. Importance: Fractional Brownian motion has widespread applications in financial modeling, physics, engineering, and other fields. Its simulation is crucial for validating theoretical results and solving practical problems
3. Limitations of Existing Methods:
   • Traditional discrete-time grid methods only guarantee accuracy at grid nodes
   • Existing general Gaussian process simulation methods have high computational complexity
   • Lack of continuous-time methods capable of precisely calculating approximation errors
4. Research Motivation: Building upon the author's previous work 2, propose simpler methods for computing expansion coefficients, reducing computational complexity and improving accuracy

Core Contributions

1. Theoretical Contribution: Derives explicit relationships for expansion coefficients of kernel functions related to fractional Brownian motion's integral representation (with respect to Legendre polynomials)
2. Method Improvement: Provides simpler expansion coefficient computation methods compared to previous work 2, reducing computational complexity
3. Error Analysis: Establishes equations for precisely calculating mean-square errors in polynomial approximations of fractional Brownian motion
4. Continuous-Time Approximation: Achieves approximate representation of fractional Brownian motion in continuous time, rather than only at discrete time points

Detailed Methodology

Problem Definition

Given fractional Brownian motion $B_H(t)$ (Hurst exponent $H \in (0,1)$), find its expansion coefficients with respect to the orthogonal basis of Legendre polynomials to achieve continuous-time approximation:

$B_H(t) = \sum_{i=0}^{\infty} B_H^i \hat{P}(i,t)$

where $\hat{P}(i,t)$ are normalized Legendre polynomials.

Core Mathematical Framework

Fractional Brownian Motion Definition:

• Covariance function: $R_H(t,\tau) = \frac{t^{2H} + \tau^{2H} - |t-\tau|^{2H}}{2}$
• Integral representation: $B_H(t) = \int_0^t k_H(t,\tau) dB(\tau)$

Kernel Function: $k_H(t,\tau) = a_H(t-\tau)^{H-1/2} \,_2F_1\left(\frac{1}{2}-H, H-\frac{1}{2}, H+\frac{1}{2}, 1-\frac{t}{\tau}\right)\mathbf{1}(t-\tau)$

where $a_H = \sqrt{\frac{\pi H(1-2H)}{\Gamma(2-2H)\cos\pi H}}$ (when $H \neq 1/2$).

Main Theoretical Results

Theorem 1 (Explicit Expression for Expansion Coefficients): For the $j$-th column of matrix $K^H$: $K^H_{*j} = a_H\Gamma(3/2-H)\sqrt{\frac{2j+1}{T}} \sum_{k=0}^j \frac{l_{jk}(3/2-H)_k}{T^k(H+1/2+k)k!} F^{H+1/2+k}$

For $H \neq 1/2$, the expansion coefficients satisfy: $K^H_{ij} = a_H\Gamma(1/2-H)\sqrt{\frac{2j+1}{T}} F^{H+1/2}_i \sum_{k=0}^j (-1)^{j-k} \frac{1/2-H+k}{H+1/2+k} \Pi^{(ij)}_k$

Algorithm Implementation

1. Coefficient Calculation: Use recurrence relations (12)-(13) to compute expansion coefficients of power functions
2. Matrix Construction: Construct expansion coefficient matrix $K^H$ through Theorem 1
3. Random Variable Generation: Generate independent standard normal random variables $V_i$
4. Approximation Construction: $\tilde{B}_H(t) = \sum_{i=0}^{L-1} \tilde{B}^H_i \hat{P}(i,t)$

Experimental Setup

Parameter Configuration

• Time Interval: $T = [0,1]$
• Hurst Exponent: $H = 0.1, 0.2, \ldots, 0.9$
• Truncation Length: $L = 4, 8, 16, 32, 64, 128$

Evaluation Metrics

Mean-Square Approximation Error: $\varepsilon = E\int_T (B_H(t) - \tilde{B}_H(t))^2 dt$

Theoretical Formula: $\varepsilon = \|k_H(\cdot)\|^2_{L^2(T^2)} - \|\bar{K}^H\|^2$

where $\|k_H(\cdot)\|^2_{L^2(T^2)} = \frac{T^{2H+1}}{2H+1}$

Experimental Results

Main Results

Table 1: Mean-Square Approximation Error $\varepsilon$

Key Findings

1. Convergence Analysis:
   • For $H = 0.5$, convergence is fastest
   • When $H$ approaches boundary values (0 or 1), convergence rate decreases significantly
   • Relative error is $\varepsilon/(2H+1)^{-1}$
2. Method Comparison:
   • New method yields qualitatively similar results to reference 2
   • Errors between the two methods are very close even for small $L$ values
   • Indicates that matrix product representation has minimal impact on accuracy but provides computational stability
3. Convergence Rate Analysis:
   • Convergence rate for power function $t^\alpha$ expansion is $O(L^{-(2\alpha+1)})$
   • When $\alpha \to -1/2$, convergence rate approaches zero, explaining low convergence at boundary Hurst exponents

Related Work

1. Classical Methods Review:
   • Mandelbrot-Van Ness representation 25
   • Decreusefond-Üstünel integral representation 15
   • Survey of various simulation methods 4-7
2. Relationship to Related Work:
   • Based on author's previous spectral representation method 2
   • Draws from spectral form mathematical descriptions in control systems 16,17
   • Extends Legendre polynomial representation of Wiener processes 29,30
3. Application Extensions:
   • Applicable to fractional Ornstein-Uhlenbeck processes
   • Suitable for fractional Brownian bridge simulation
   • Extensible to iterated random integral representations

Conclusions and Discussion

Main Conclusions

1. Successfully derives explicit formulas for expansion coefficients of fractional Brownian motion kernel functions
2. Establishes precise methods for calculating mean-square approximation errors
3. Achieves approximate representation of fractional Brownian motion in continuous time
4. Validates method effectiveness and computational advantages

Limitations

1. Computational Precision Requirements: New method requires high-precision floating-point or symbolic computation
2. Boundary Effects: Slower convergence when Hurst exponent approaches 0 or 1
3. Computational Complexity: While coefficient calculation is simplified, matrix operations remain necessary

Future Directions

1. Extension to multidimensional fractional Brownian motion
2. Application to more complex stochastic processes (e.g., fractional stochastic differential equations)
3. Development of more efficient numerical algorithms

In-Depth Evaluation

Strengths

1. Theoretical Rigor: Provides complete mathematical derivations and rigorous error analysis
2. Method Innovation: Simplifies expansion coefficient calculation, enhancing method practicality
3. Comprehensive Experiments: Validates method effectiveness across multiple Hurst exponents and truncation lengths
4. Practical Value: Provides new tools for continuous-time simulation of fractional Brownian motion

Weaknesses

1. Computational Requirements: High numerical precision requirements may limit practical applications
2. Boundary Performance: Convergence performance at extreme Hurst exponents requires improvement
3. Comparative Analysis: Insufficient detailed performance comparison with other existing methods

Impact

1. Academic Contribution: Makes important contributions to fractional Brownian motion theory and computational methods
2. Practical Value: Has application potential in financial mathematics, physical modeling, and other fields
3. Methodological Significance: Provides reference framework for orthogonal expansions of other stochastic processes

Applicable Scenarios

1. Applications requiring continuous-time fractional Brownian motion simulation
2. Scenarios requiring precise control of approximation errors
3. Numerical solution of fractional stochastic differential equations
4. Financial derivative pricing and risk management

References

This paper cites 37 related references, primarily including:

• Foundational theory of fractional Brownian motion 22-25
• Stochastic process simulation methods 1,3,4-7
• Spectral methods and orthogonal expansions 16,17,26
• Fractional calculus theory 32