In my research, I want to use a meaningful and computationally tractable distance between two time-dependent probability distributions. For stationary distributions, several distance measures are used, such as KL divergence, JS divergence, Wasserstein distance, and so on.

Is anyone aware of a similar thing for two time-dependent distributions p1(x,t) and p2(x,t)?

In my problem, both distributions are the solutions of non-stationary Fokker-Planck equation,

dp1/dt = (-d/dx F1(x) + d2/dx2 D1(x) ) p1(x,t),  p1(0,t)=p10(x)
dp1/dt = (-d/dx F2(x) + d2/dx2 D2(x) ) p2(x,t),  p2(0,t)=p20(x)

Where the drift forces F1(x), F2(x), diffusion coefficients D1 and D2, and initial probability distributions p10(x) and p20(x) are computed from data. So, ideally, I want some kind of norm that can be expressed through the parameters of the Langevin equation.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.