Potential Hessian Ascent: The Sherrington-Kirkpatrick Model

We present the first iterative spectral algorithm to find near-optimal solutions for a random quadratic objective over the discrete hypercube, resolving a conjecture of Subag [Sub21]. The algorithm is a randomized Hessian ascent in the solid cube, with the objective modified by subtracting an instance-independent potential function [CPS18, Sub24]. Using tools from free probability theory, we construct an approximate projector into the top eigenspaces of the Hessian, which serves as the covariance matrix for the random increments. With high probability, the iterates’ empirical distribution approximates the solution to the primal version of the Auffinger-Chen SDE [AC15b]. The per-iterate change in the modified objective is bounded via a Taylor expansion, where the derivatives are controlled through Gaussian concentration bounds and smoothness properties of a semiconcave regularization of the Fenchel-Legendre dual to the Parisi PDE [AC15a]. These results lay the groundwork for (possibly) demonstrating low-degree sum-of-squares certificates over high-entropy step distributions for a relaxed version of the Parisi formula [SS24, Open Question 1.8].