[Paper Review] Gaussian Estimates: A Brief History

D. G. Aronson|arXiv (Cornell University)|Jul 14, 2017

Gaussian Processes and Bayesian Inference3 citations

TL;DR

This paper traces the historical development of two-sided Gaussian estimates for fundamental solutions of second-order linear parabolic equations with bounded, measurable coefficients. It details how Aronson established the first truly Gaussian bounds under minimal regularity, while Fabes and Stroock later derived these estimates directly from Nash's ideas, proving Hölder continuity and the Harnack inequality as consequences, thus unifying and simplifying the foundational theory of parabolic PDEs.

ABSTRACT

Two-sided Gaussian estimates for the fundamental solution of a second order linear parabolic differential equation are upper and lower bounds in terms of the fundamental solution of the classical heat conduction equation. In his seminal 1958 paper Nash stated, without proof, two-sided non-Gaussian bounds for the fundamental solution of a uniformly parabolic divergence structure equation assuming only boundedness of the coefficients. In his 1967-1968 papers Aronson derived truly Gaussian estimates for the fundamental solutions of a large class of linear parabolic equations (including the divergence structure equation) under minimal non-regularity assumptions on the coefficients. Subsequently in 1986 Fabes & Stroock derived Gaussian estimates for the divergence structure equation directly from the ideas of Nash and went on to prove Nash's continuity theorem and the Harnack inequality as a consequence of their estimate. In this note I describe these results together with various extensions.

Motivation & Objective

• To trace the evolution of two-sided Gaussian estimates for fundamental solutions of linear parabolic equations.
• To clarify the foundational role of Nash's unpublished bounds and their subsequent rigorous derivation by Aronson and Fabes-Stroock.
• To demonstrate how Gaussian estimates imply key regularity results such as Hölder continuity and the Harnack inequality.
• To unify the theoretical framework for divergence-form parabolic equations under minimal coefficient assumptions.
• To extend the scope of Gaussian estimates to general parabolic equations with lower-order terms under structural hypotheses (H).

Proposed method

• Deriving two-sided Gaussian bounds using energy estimates and the Harnack inequality, as in Aronson's 1967 work.
• Reconstructing Nash's sketch using a refined version of his moment estimate and the semigroup property of the fundamental solution.
• Applying Davis' Riemannian distance technique to sharpen the upper bound in the Gaussian estimate for time-independent coefficients.
• Using a technical lemma on solutions supported outside a ball to control the decay of the fundamental solution.
• Establishing the lower bound via Nash's logarithmic integral inequality and the semigroup property.
• Deriving Hölder continuity and the Harnack inequality as consequences of the Gaussian estimate, rather than as prerequisites.

Experimental results

Research questions

• RQ1 How can two-sided Gaussian estimates for the fundamental solution of a parabolic PDE be derived without assuming smoothness of the coefficients?
• RQ2 What is the precise relationship between Nash's non-Gaussian bounds and the subsequent Gaussian estimates of Aronson and Fabes-Stroock?
• RQ3 Can the Harnack inequality and Hölder continuity be derived directly from the Gaussian estimate, rather than as intermediate steps?
• RQ4 How do structural assumptions on the coefficients (H) affect the constants in the Gaussian bounds?
• RQ5 To what extent can the Gaussian estimate be extended to parabolic equations with lower-order terms and time-dependent coefficients?

Key findings

• Two-sided Gaussian estimates of the form $ C^{-1}g_1(x- au,t- au) \leq \Gamma(x,t;\xi,\tau) \leq C g_2(x-\xi,t-\tau) $ hold for the fundamental solution of equation (1) under the structural hypothesis (H), with constants depending only on $ N, T $, and the structure of the equation.
• The upper bound $ \Gamma(x,t;\xi,\tau) \leq C(t-\tau)^{-N/2} $ is derived from a moment estimate and the semigroup property, independent of the Harnack inequality.
• The lower bound is established using Nash's logarithmic integral inequality and the semigroup property, yielding a Gaussian decay structure.
• Fabes and Stroock showed that the Gaussian estimate implies both Hölder continuity of solutions and the Harnack inequality, reversing the traditional logical order.
• For time-independent coefficients with $ N \geq 3 $, integrating the Gaussian estimate yields $ K^{-1}|x-\xi|^{2-N} \leq G(x,\xi) \leq K|x-\xi|^{2-N} $, matching classical potential theory results.
• The constants $ \alpha_1, \alpha_2, C $ in the estimate depend only on $ \nu, M, M_0, N, T $, and the integrability structure of the coefficients, not on pointwise smoothness.