Thursday 26th November 2026, 10:00, room: ILWA-SR 28 (ILWA-U1-13), Jere Juhani Koskela (TU Graz):

"Bernoulli factories and well-posedness of Wright-Fisher stochastic heat equations with measurable drift". (Image: Jere Koskela)

 

Click for details.

“Bernoulli factories and well-posedness of Wright-Fisher stochastic heat equations with measurable drift”.

Abstract:
The 1D stochastic heat equation with Wright-Fisher noise is a canonical model of allele frequency evolution and front propagation in a spatially structured, randomly reproducing population. I will show existence and uniqueness in law of its solutions for a wide class of drift functions, which captures dynamics other than spatial migration and random reproduction, such as mutation or patterns of natural selection. The only requirements on the drift are measurability, and so-called polynomial boundedness which ensures solutions remain [0, 1]-valued. In particular, the resulting solution theory applies to drifts which are non-smooth or discontinuous. Key to the proof are Bernstein duality, used to determine moments of the SPDE, and the so-called Bernoulli factory, which answers the following, seemingly unrelated question: given a known function f mapping [0, 1] to itself, and an unbounded number of independent Ber(p)-distributed coins with unknown bias p, when is it possible to construct a Ber(f(p))-distributed coin? This is joint work with Oliver Tough at Durham University.


13.11.2026