Perelman's entropy functionals

To control a heat-like flow you usually want a quantity that always moves in one direction — a thermodynamic-style entropy that can only increase (or a energy that can only decrease), so that the system cannot oscillate forever or hide its bad behaviour. Ricci flow lacked such a quantity for two decades, which is why it was so hard to control. Perelman's breakthrough in 2002 was to discover the right monotone quantities — his F-energy and his W-entropy — and to recognize Ricci flow as their gradient flow. These functionals are the analytic heart of his proof; everything about ruling out bad singularities flows from their monotonicity.

There are two main functionals. The F-energy is F(g, f) = integral over M of ( S + |grad f|^2 ) e^{-f} dV, where S is scalar curvature and f is an auxiliary function; under Ricci flow coupled with a backward heat equation for f, F is nondecreasing, and Ricci flow is its gradient flow up to diffeomorphism. The more powerful W-entropy is W(g, f, tau) = integral over M of [ tau( S + |grad f|^2 ) + f - n ] (4 pi tau)^{-n/2} e^{-f} dV, a scale-invariant functional with a positive 'time' parameter tau. Perelman proved dW/dt >= 0 along the flow, with equality only on gradient shrinking solitons. The single most important consequence is the no-local-collapsing theorem: the monotonicity of W forbids the geometry from collapsing — volumes of balls of a given curvature scale stay bounded below — which rules out the worst hypothetical singularities (cigar-like collapses) and is exactly what was missing to make blow-up analysis and surgery work. A companion construction, the reduced volume, is also monotone and gives a more geometric proof of the same non-collapsing.

Why this is the crux: before Perelman, one could not exclude singularity models that collapse without curvature blowing up at a controlled rate, and without excluding them the canonical-neighborhood description fails and surgery cannot be justified. The entropies supply the missing rigidity. Two honest cautions. First, 'entropy' here is an analogy with statistical mechanics, not a literal thermodynamic entropy; the monotonicity is a theorem about a specific functional, not a law of physics. Second, the equality cases are everything — W is constant precisely on gradient shrinking Ricci solitons, and it is the rigidity in the equality case (not merely monotonicity) that classifies the blow-up limits. Treating these functionals as mere bookkeeping misses that their gradient-flow and soliton structure is what does the geometric work.