Local fields, Gaussian measures, and Brownian motions

Abstract

These are lecture notes from a course given at the CRM in Montreal in 1992. They survey the author's attempts to find and understand canonical probabilistic entities in a local field (e.g. p-adic) setting. We propose answers to the related questions: ``What are the analogues for Gaussian measures?'' and ``What are the analogues for Brownian motion and its multiparameter relatives?'' There is a suitable concept of orthogonality in the local field setting, and one can mimic a classical abstract definition of Euclidean Gaussian measures. The resulting theory is, in some sense, an L^\infty rather than an L^2 one; and so although there are many similarities to the Euclidean case there are also some striking differences. Local field Brownian motion is the most natural local field Gaussian process that takes values in a local field vector space and is indexed by another local field vector space. Characterising the polar sets for such processes lead us to study the notion of additive functionals or homogeneous random measures and a related Riesz-like potential theory. We study local times, show they can be recovered by counterparts of classical intrinsic contstructions using dilation and Hausdorff measures, and establish an analogue of Trotter's theorem. We finish with a discussion of random series with local field Gaussian coefficients. Although we show that there are broad classes of random series which are stationary, we also find that there is no obvious counterpart of the representation of a general stationary Gaussian process on the circle as a random Fourier series.