Hermite Polynomial Expansion

• Hermite Polynomial Expansion is a method that expresses functions, stochastic processes, or densities as a linear combination of orthogonal Hermite polynomials with a Gaussian weight.
• It systematically reduces high-dimensional, Gaussian-structured problems into algebraic forms, facilitating analytic and numerical computation of moments and observables.
• The technique is widely applied in kinetic theory, quantum mechanics, and financial models, providing a unified framework for handling complex stochastic dynamics.

A Hermite polynomial expansion represents a function, stochastic process, operator, or probability density as a (possibly infinite) linear combination of Hermite polynomials, which form an orthogonal basis with respect to a Gaussian weight. This technique is foundational in the mathematical analysis of stochastic processes, kinetic theory, uncertainty quantification, quantum mechanics, spectral methods, and the statistical modeling of complex systems. The method leverages the completeness, recurrence, and generating function properties of Hermite polynomials to systematically reduce high-dimensional problems—often governed by Gaussian structures or symmetries—to algebraic manipulations of expansion coefficients, facilitating both analytic and computational treatment of observables such as moments, densities, and solutions to PDEs.

1. Fundamental Properties of Hermite Polynomials

Let $H_n(x)$ denote the "physicists'" Hermite polynomials, defined by the Rodrigues formula: $$H_n(x) = (-1)^n e^{x^2} \frac{d^n}{dx^n} e^{-x^2}, \quad n=0,1,2,\ldots$$ These polynomials satisfy orthogonality on $\mathbb{R}$ with respect to the Gaussian weight $e^{-x^2}$: $$\int_{-\infty}^{\infty} H_m(x) H_n(x) e^{-x^2} dx = \sqrt{\pi} 2^n n! \delta_{mn}$$ They also admit an explicit generating function: $$\exp(2xt - t^2) = \sum_{n=0}^{\infty} H_n(x) \frac{t^n}{n!}$$ and a three-term recurrence relation: $$H_{n+1}(x) = 2x H_n(x) - 2n H_{n-1}(x), \quad H_0(x)=1,\, H_1(x)=2x$$ Multivariate and tensor forms can be constructed to support higher-dimensional expansions, with several conventions adapted for probabilists' Hermite polynomials and various generalizations for use with arbitrary Gaussian measures or anisotropic covariance structures (Xu et al., 2019).

2. Hermite Expansion Methodology: Theory and Steps

Given a function or stochastic process—e.g., the time-dependent density $P(x,t)$ of a Lévy walk—expand it as

$$P(x,t) = \sum_{n=0}^{\infty} H_n(x) \tilde{T}_n(t) e^{-x^2}$$

where $\{ H_n(x) e^{-x^2} \}$ is an orthogonal basis in $L^2(\mathbb{R}, e^{-x^2} dx)$ and $\tilde{T}_n(t)$ are time-dependent coefficients. Projection onto the basis via orthogonality yields the coefficient formula: $$\tilde{T}_m(t) = \frac{1}{\sqrt{\pi} 2^{m} m!} \int_{-\infty}^{\infty} P(x,t) H_m(x)\, dx$$ The method relies on substituting the Hermite expansion ansatz into the governing integral equations (e.g., those coupling space and time in transport processes), resulting in infinite or truncated systems of Volterra-type equations for the coefficients. After Laplace or other integral transformations, these become algebraic recursion relations, often facilitating their computation when classical transform-inversion methods are inapplicable due to space-time coupling or nonlinearity (Xu et al., 2019).

3. Calculation and Interpretation of Moments and Observables

Given the Hermite expansion coefficients, moments of $x$ are computed directly. For symmetric Lévy walks, the characteristic function is

$$P(k,t) = \int_{-\infty}^{\infty} e^{-ikx} P(x,t) dx = \sqrt{\pi} e^{-k^2/4} \sum_{n=0}^\infty (-ik)^n \tilde{T}_n(t)$$

Thus, the $m$-th moment is

$$\langle x^m(t) \rangle = i^m \frac{d^m}{dk^m} P(k,t) \Big|_{k=0}$$

For the second moment in the symmetric case, only even $n$ contribute: $$\langle x^2(t) \rangle = 2\sqrt{\pi} \tilde{T}_2(t) + \frac12 \sqrt{\pi} \tilde{T}_0(t)$$ This renders the extraction of high-order moments transparent, requiring only differentiation of the finite-sum expansion, in contrast to computing contour integrals required by classical approaches. The same machinery extends systematically to computing higher correlators and first-passage statistics (Xu et al., 2019).

4. Advantages and Limitations of Hermite Polynomial Expansion

The Hermite expansion is particularly advantageous in:

• Resolving strong space-time coupling, velocity-dependent or position-dependent kernels where integral transform methods (e.g., Fourier-Laplace techniques) fail or the inverse is intractable.
• Enabling the systematic computation of moments of arbitrary order from only a finite subset of coefficients.
• Allowing for fully algorithmic refinement: higher-order corrections and new observables (e.g., first-passage densities) emerge via the same projection and recursion approach.

Limitations include:

• Practical computations necessitate truncation; convergence may be slow for densities with pronounced non-Gaussian structure or multi-modal behavior, requiring many Hermite modes.
• The method replaces a single (often complex) transform inversion with the need to solve an infinite system of coupled equations, albeit these become algebraic after transformation.
• For strongly non-Gaussian phenomena, the expansion may converge slowly and require careful numerical stabilization (Xu et al., 2019).

5. Applications and Comparative Context

Hermite polynomial expansions have been critical in the analysis of time-space coupled stochastic processes, such as Lévy walks, where classical integral-transform methods become ineffective. Detailed comparisons demonstrate agreement with standard solutions in cases where both methods are applicable. The expansion is complementary rather than universally superior and is employed strategically where the problem structure aligns with the Gaussian weight and polynomial basis. In cases of significant nonlinearity or non-separability, or when moments of high-order need to be accessed directly, the Hermite method proves especially powerful.

This approach is not limited to stochastic dynamics but is intrinsic to kinetic theory, quantum optics (e.g., via Wigner functions), mathematical finance (e.g., for expansions of probability laws and pricing functions), and signal processing, exploiting the rich algebraic and analytic structure of the Hermite basis (Xu et al., 2019).

6. Extensions and Future Directions

The methodology allows systematic generalization:

• Multi-variate, tensor, and anisotropic Hermite expansions accommodate complex geometries and correlated Gaussian structures.
• Extensions to non-Gaussian base weights and connections to other orthogonal polynomial systems (e.g., for different base processes).
• Algorithmic advances target more efficient computation and stabilization for high-dimensional or highly non-Gaussian scenarios.
• Application to new domains, including nonlinear partial differential equations, quantum lattice models, and data-driven model fitting for high-dimensional empirical distributions.

A plausible implication is that Hermite expansions provide an essential bridge between analytic tractability and direct numerical access to observables in systems where Gaussian structure governs the underlying dynamics or approximates the dominant behavior (Xu et al., 2019).