{"id":"arxiv/2603.16186","url":"https://arxiv.org/abs/2603.16186","tags":["math.NA"],"score":0,"title":"Weak Adversarial Neural Pushforward Method for the McKean-Vlasov / Mean-Field Fokker-Planck Equation","domain":"math.NA","source":"arxiv","authors":["Andrew Qing He","Wei Cai"],"reviews":0,"subject":"math.NA","summary":"We extend the Weak Adversarial Neural Pushforward Method (WANPM) to the McKean--Vlasov mean-field Fokker--Planck equation, covering both the stationary and time-dependent cases. The key observation is that the mean-field nonlinearity -- an expectation under the solution distribution -- is naturally estimated by Monte Carlo sampling from the pushforward network, requiring no change to the architecture and only minor modifications to the training loop. For the quadratic (granular media) interaction kernel, the interaction term reduces to the batch sample mean, eliminating secondary sampling…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-03-17","reactions":{}}