Track #21: Add partial information: Correcting bug stemming from the differences in QZ calculation under Octave and Matlab: modified PI_gensys to use C.Sims' Lapack zgges based qzcomplex() qz decomposition routine under Octave which gives nearly identical complex results to Matlab QZ (except for transpose of q matrix and few sign changes in the output matrices)
parent
057908baf6
commit
e8a193bbfb
|
@ -153,7 +153,16 @@ end
|
|||
|
||||
G0pi=eye(n+FL_RANK+NX);
|
||||
try
|
||||
[a b q z v]=qz(G0pi,G1pi);
|
||||
% In Matlab: [aa bb q z v w] = qz(a,b) s.t. qaz = aa, qbz = bb %
|
||||
% In Octave: [aa bb q z v w] = qz(a,b) s.t. q'az = aa, q'bz=bb %
|
||||
% and qzcomplex() extension based on lapack zgges produces same
|
||||
% qz output for Octave as Matlab qz() does for Matlab thus:
|
||||
if exist('OCTAVE_VERSION')
|
||||
[a b q z]=qzcomplex(G0pi,G1pi);
|
||||
q=q';
|
||||
else
|
||||
[a b q z]=qz(G0pi,G1pi);
|
||||
end
|
||||
catch
|
||||
try
|
||||
lerror=lasterror;
|
||||
|
|
Loading…
Reference in New Issue