A mean field approximation in data assimilation for non-linear dynamics Eyink, G. L. et al. • non-linear filtering/smoothing problems • Connections with other methods • Mean-field variational approach • Closure techniques • Toy example: a bistable double-well system 1
19
Embed
A mean field approximation in data assimilation for nonlinear dynamics
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
A mean field approximation in data assimilation for non-linear
dynamics
Eyink, G. L. et al.
• non-linear filtering/smoothing problems
• Connections with other methods
• Mean-field variational approach
• Closure techniques
• Toy example: a bistable double-well system
1
Non-linear estimation problem
dX(t) = f(X, t)dt + (2D)1/2(X, t)dW(t),
Y(t) = Z(X(t), t) + R1/2(t)η(t),
• X(t): state vectors at ti ≤ t ≤ tf
• f(X, t): a drift/dynamical vector
• W(t): a vector Wiener process
• D: a diffusion matrix;
• η(t): a white noise;
• Y(t): an observation process → data Y(t′) = {Y(t) : t ≤ t′}
• R(t): a covariance function;
• Z(·): a measurement function
2
−→ optimal estimation of the conditional probability
• P(x, t|Y(t)) with t < tf for filtering problem
• P(x, t|Y(tf )) with t < tf for smoothing problem
Discrete-Time Data:
yi = Y(ti) with ti ≤ t1 < t2 < ... < tM−1 < tM ≤ tf .
optimal filtering → P(x, t|Y(tf )) is the solver of the forward
Kolmogorov equation ( Kushner and Stratonovich )
∂tP(x, t) = L(t)P(x, t) with P(x, t = ti) = P0(x)
where
L(t) = −∑
k
∂[f(x, t)(·)]∂xk
+∑
ij
∂2[Dij(x, t)(·)]∂xi∂xj
.
3
At measurement times tm, P (x, t) satisfies the forward jump