Problem with BSE diagonalization and restart

[javad]

Dear Yambo developers

I tested the BSE yambo-4.0.2 and I have a problem with diagonalisation and plotting.

In yambo-3.4.1 I was habit to calculate the BSE, and then plot the results in different
directions or changing the number of points in the final plot without running again the calculations
but just reading the diagonalized matrix.

I notice that this is not possible any-more in yambo-4.0.2, the code always recalculates all the BSE matrix.

Why? Is this a bug? or there is a particular reason to do so?

regards
Javad

[Davide Sangalli]

Dear Javad,
I've just tried and yambo-4.0.2 correctly computes the the kernel only the first time.

If you change the number of points on the frequency axis only epsilon is recomputed
If you also change the field direction then the BSE matrix is also diagonalized again, as it should be.

In neither case it recomputes the BSE matrix, which is indeed stored on disk.

Can you post an example of the different behaviour you found out between 4.0.2 and any older version
and / or the report file of your run.

Best,
D.

[javad]

Dear Davide

you are right, I notice that this happens only when I change the number of processors.

Usually I run the BSE in parallel and then I plot the different spectra with small serial runs, and this was the problem

Javad

[Davide Sangalli]

You are right, in case the number of processor is different then the ndb with the BSE matrix is not recognised.
We are still working on the I/O of the BSE matrix.
Maybe in the future we will release the constrain of keeping the same number of CPUs.

Best,
D.