Add W_v output eigenvalues fix#256
Conversation
|
Question on this - I believe Wv should be Hermitian, but can accumulate numerical errors (like Wp). In vacuum, it is forced Hermitian with If this breaks, perhaps another option is to use |
|
I agree with @jhalpern30 |
|
So not sure if this fixes the lsymz = false case. (i tested it with lsymz false and got similar output, so that means it's fine? The simple fix might just be using a new variable ev_real ? The versions i tested reinstate ev as COMPLEX(r8). From the outputs i test, the ev_real addition seems to work best. vs Test Script: Test outputs: |
Potential Bug find and Fix.
Comparisons between W_t and reconstructed matrix from eigen values and vectors was not working.
Matrix reconstruction tests were done with this script. (both before and after changes are included in comments of script)
dcon_W_reconstruction_test.py
The changes to dcon/free.f and dcon/dcon_netcdf.f now reflect that 'ev' is defined as explicitly real eigenvalues.
This can be seen from the call to zheev line305, ev needs to be real zheev doc