Library: | kalman |
See also: | rICfil calibrIC ICerzsep |
Quantlet: | iteras | |
Description: |
Auxiliary routine for rICfil: - if possible - solves for Lambda1~N(0,S1),Lambda2~N(0,S2) indep. A^{-1} =E [ Lambda1 Lambda1' w_b ] + E [ Lambda2 Lambda2' ] (1) w_b=min(1,b/|A Lambda1|) using a fixed-point-algorithm |
Usage: | {A,V,ctrl}=iteras(A0,S1,S2,b,N,eps,itmax,expl) | |
Input: | ||
A0 | p x p; starting value for A; if 0 is entered I^{-1} is taken if p>1 and dim(A0)==1 I^{-1}*A0 is taken | |
S1 | p x p; Covariance of the first (clipped) component | |
S2 | p x p; Covariance of the second (unclipped) component | |
b | numeric; clipping height | |
N | integer; MC-sample size / integration grid-points | |
eps | numeric; exactitude | |
itmax | integer; maximal number of iterastions | |
expl | numeric; threshold for the changes in abs. value of A: beyond this value convergence is uncertain ~ 4 | |
Output: | ||
A | p x p; Lagrange-Multiplyer solving (1) | |
V | p x p; corresponding Covariance | |
ctrl | integer; tells if convergence "happened" |
Is called by ICerzsep.
Calles [within kalman.lib] numint2m stointpm.
to be looked up in rICfil
Library: | kalman |
See also: | rICfil calibrIC ICerzsep |