While not reached in nature, the strong-interaction limit of exact (Hohenberg-Kohn-Levy-Lieb) DFT points the way towards the real physics in strongly correlated systems missed by semilocal or hybrid functionals, an example being that unlike the latter, it gets the dissociation of the H2 molecule right. The bottleneck so far towards integrating it into future XC functionals has been the lack of an efficient algorithm to compute this limit. After a brief introduction to this limit I will explain a new, simple, accurate, and extremely efficient method for computing the resulting limiting XC functional. Our method, genetic column generation (GenCol), relies on (i) the rigorous sparsity of optimal plans established by F. and Voegler [the support of optimizers is linear instead of exponential in the number N of electrons, SIAM J. Math. Anal. 2018], (ii) the method of column generation (CG) from discrete optimization which is novel in the present context, and (iii) basic ideas from machine leaning. In the GenCol method, the dual state within CG plays the role of an adversary, in loose similarity to Wasserstein GANs. On a sequence of benchmark problems with up to 120 gridpoints and up to 30 electrons, our method found the exact optimizers to machine precision. Moreover, empirically the number of computational steps needed to find them exhibits only slow polynomial growth when both N and L are simultaneously increased (while keeping their ratio fixed to mimic a thermodynamic limit of the particle system), appearing to break the curse of dimension.
Joint work with Andreas Schulz and Daniela Voegler, arXiv:2103.12624, to appear in SIAM JSC.