[Paper Review] The differentiation of hypoelliptic diffusion semigroups

Marc Arnaudon, Anton Thalmaier|SPIRE - Sciences Po Institutional REpository|Apr 13, 2010

Advanced Mathematical Modeling in Engineering5 references4 citations

TL;DR

This paper develops stochastic integration by parts formulas for hypoelliptic diffusion semigroups using Malliavin calculus and local martingale techniques, enabling derivative formulas for semigroups and harmonic functions without requiring derivatives of the test functions. The key contribution is a representation of the derivative of the semigroup in terms of the derivative flow, Malliavin covariance, and its inverse, valid under hypoellipticity conditions via the Hörmander condition.

ABSTRACT

Basic derivative formulas are presented for hypoelliptic heat semigroups and harmonic functions extending earlier work in the elliptic case. Emphasis is placed on developing integration by parts formulas at the level of local martingales. Combined with the optional sampling theorem, this turns out to be an efficient way of dealing with boundary conditions, as well as with finite lifetime of the underlying diffusion. Our formulas require hypoellipticity of the diffusion in the sense of Malliavin calculus (integrability of the inverse Malliavin covariance) and are formulated in terms of the derivative flow, the Malliavin covariance and its inverse. Finally some extensions to the nonlinear setting of harmonic mappings are discussed.

Motivation & Objective

• To develop derivative formulas for hypoelliptic diffusion semigroups that do not require derivatives of the test functions.
• To extend integration by parts techniques from the elliptic to the hypoelliptic setting using local martingales and Malliavin calculus.
• To handle boundary conditions and finite lifetime diffusions via optional sampling and stochastic control methods.
• To generalize results to harmonic mappings with non-Euclidean targets, particularly in the hypoelliptic regime.
• To provide probabilistic representations of derivatives under the Hörmander condition, ensuring smooth transition densities.

Proposed method

• Uses integration by parts at the level of local martingales to derive derivative formulas for semigroups and harmonic functions.
• Applies the optional sampling theorem to manage boundary conditions and finite lifetime diffusions.
• Employs the derivative flow $X_{tullet}$, Malliavin covariance, and its inverse to express derivatives of semigroups.
• Introduces a local martingale $n_s$ constructed via the derivative flow and stochastic integral with respect to Brownian motion.
• Utilizes the process $a_s = (X_{sullet}^{-1}A)_x^* \mathbf{1}_{\{s \leq \tau\}}$ to generate local martingales for perturbation analysis.
• Derives a representation involving $\mathbb{E}\left[\left(\int_0^\sigma \Theta_{0,s}^{-1} \nabla \Theta_{0,s} \, dY_s\right) C^{-1}_\sigma(x) v\right]$ as an additional term in the nonlinear case.

Experimental results

Research questions

• RQ1 How can integration by parts formulas be adapted to hypoelliptic diffusions where the generator fails to be elliptic?
• RQ2 Can derivative formulas for semigroups be expressed without involving derivatives of the test function?
• RQ3 What role does the Malliavin covariance and its inverse play in representing derivatives under hypoellipticity?
• RQ4 How can boundary conditions and finite lifetime diffusions be treated within a stochastic integration by parts framework?
• RQ5 Is it possible to extend derivative formulas to harmonic mappings with non-Euclidean targets under hypoelliptic conditions?

Key findings

• The paper establishes a derivative formula for the semigroup $P_t f$ that depends only on the derivative flow, Malliavin covariance, and its inverse, not on derivatives of $f$.
• Under the Hörmander condition, the semigroup is strongly Feller and admits a smooth transition density, ensuring regularity for the derived formulas.
• The integration by parts formula at the level of local martingales enables handling of boundary conditions and finite lifetime via optional sampling.
• For harmonic functions, the derivative $du$ is represented without derivatives of the boundary data, using stochastic integrals and local martingales.
• In the nonlinear case with non-Euclidean targets, an additional term involving $\int_0^\sigma \Theta_{0,s}^{-1} \nabla \Theta_{0,s} \, dY_s$ appears, which cannot yet be eliminated.
• The method relies on perturbations of the Brownian motion and change of measure, with alternative variations (e.g., time change, rotation) noted as potentially useful but not yet fully exploitable.