Talagrand's convolution conjecture up to loglog via perturbed reverse heat

We prove that under the heat semigroup (P_τ) on the Boolean hypercube, any nonnegative function f: {-1,1}^n to R_+ exhibits a uniform tail bound that is better than that by Markov's inequality. Specifically, for any η> e^3 and τ> 0, align* P_{X \sim μ}\left( P_τf(X) > η\int f dμ\right) \leq c_τ \log \log η{η\log η}, align* where μ is the uniform measure on the Boolean hypercube {-1,1}^n and c_τ is a constant that only depends on τ. This resolves Talagrand's convolution conjecture up to a dimension-free loglog η factor. Its proof relies on properties of the reverse heat process on the Boolean hypercube and a coupling construction based on carefully engineered perturbations of this reverse heat process.