amilsted / evomps Goto Github PK

Dan reported this on slack.

It's broken.

Expokit and also the eps_maps_c module need BLAS. Get known working library paths and names from numpy/scipy configuration to avoid user config requirements.

At the moment there is too much duplication.

Possibilities:

• Return them from the methods that set them and remove the object attributes
• Set them to NaN when they're invalid (difficult to catch all cases, but could do the major ones)
• Provide getters that detect when they are valid (same as above)

For large local dimensions, we can make much better use of multiple cores, compared to the use of threaded BLAS, by parallelising the loop over the physical dimension in eps_l and eps_r.

This would be a straightforward modification of the C versions of the functions, within which the GIL is already released.

Avoiding holding the GIL throughout functions like eps_l and eps_r (especially for the multi-site versions) should increase the benefits of using python's threading.

Hi,
Thanks for making this code available!

I tried running the heisenberg_af_sandwich_scattering.py file from the example folder.
I find that there are strong unphysical effects at the boundaries of the nonuniform region, as can be seen in the plot.
These begin to occur right from the beginning of time evolution, before any disturbance had enough time to propagate to the edge, so I think this definitely should not happen.
I also tried the impurity hamiltonian from heisenberg_af_sandwich_impurity.py by changing the time evolution to a complex time step, and I see similar unphysical boundary effects.

Thanks

heisn_af_sandwich_scat.pdf