You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the example page in the documentation it is reported that passing an MPO which is a sum of some MPOs as a vector to the dmrg function results in faster computation. For bigger system I found the speedup is non-negligible and I would like to use the same trick for excited states, but the implementation of dmrg for this particular case is missing.
Thanks for the report. We are currently working on a rewrite of the projected MPO code, it should be easy to add support for this once that is done (we have a plan for a more general and flexible design than the current one).
@emstoudenmire this is a good use case to keep in mind for the new projected MPO code.
In the example page in the documentation it is reported that passing an MPO which is a sum of some MPOs as a vector to the
dmrg
function results in faster computation. For bigger system I found the speedup is non-negligible and I would like to use the same trick for excited states, but the implementation ofdmrg
for this particular case is missing.Minimal runnable code
The first
dmrg
call works as expected, while the second one gives the error reported below.Output of minimal runnable code
Version information
versioninfo()
:using Pkg; Pkg.status("ITensors")
:The text was updated successfully, but these errors were encountered: