What it is
The classical Harnack inequality bounds a positive solution of the heat equation at one point by its value at another, later point. Li and Yau (1986) proved a differential version on manifolds with non-negative Ricci curvature: ∣∇u∣2/u2−ut/u≤n/2t. Integrating it along paths gives the classical inequality, with the correct Gaussian dependence.
Hamilton (1993) proved a matrix version for Ricci flow itself. On a complete solution with bounded, non-negative curvature operator, a quadratic expression in curvature and its derivatives is non-negative. Its trace, displayed above for any vector field V, already implies that tR is non-decreasing along the flow. Equality holds exactly on expanding gradient solitons, which is how the Harnack inequality detects self-similarity. Cao proved the Kähler version under non-negative bisectional curvature. Brendle extended Hamilton's inequality to a weaker curvature condition, positive isotropic curvature on M×R2.
Why it matters
Harnack inequalities control ancient solutions. If an ancient solution has non-negative curvature operator, letting the initial time go to −∞ shows that scalar curvature is pointwise non-decreasing in time. This is used throughout the analysis of κ-solutions and singularity models. Hamilton used the equality case to show that certain eternal solutions are steady solitons. Perelman's reduced distance began as a Li–Yau–Hamilton-type quantity, and his Harnack estimate for the conjugate heat equation underlies pseudolocality.
Where it stands
Harnack inequalities now exist for many geometric flows (mean curvature, Gauss curvature, Kähler–Ricci, Yamabe), each under its own positivity assumption. Open directions include Harnack estimates without curvature sign assumptions, and on singular or weak flows, where heat kernel methods are taking over.