Prev: dissections 1+2=sqrt(5) of polyominoes
Next: help wanted on a trig equation #350; Correcting Math
From: csdhan on 26 Jan 2010 00:35
I have a simple-looking optimisation as follows:

min ||PAP' - B||^2_F
s.t. P'P = I

where " ' " means transpose, A, B, P are real n*n matrices, || . ||_F is the Frobenius norm, A,B are symmetric positive semi-definite matrices, and P is an unknown orthogonal matrix.

By introducing Langrange multipliers, differentiating and equating to zero, I obtain

2BPA = - P(L' +L),

where L is a matrix of Langrange multipliers. At this point I get stuck. I can rearrange to obtain 2BP + P(L'+L)A^-1 = 0 which looks a bit like the Sylvester equation except (L'+L) is unknown.

I would be very grateful for any help. Thanks in advance!
 | 
Pages: 1
Prev: dissections 1+2=sqrt(5) of polyominoes
Next: help wanted on a trig equation #350; Correcting Math