Add operation to trace oper-ket without reshaping #2126
Conversation
There was a problem hiding this comment.
Looks good.
I don't see a way to do this without adjusting the data layer somehow, since it doesn't have the information to decide whether a matrix is a stacked operator or not.
What happens in one wants to trace the adjoint of an oper_ket? I.e. an oper_bra? Or does QuTiP not really support that?
| @@ -225,12 +225,12 @@ def reset(self, hard=False): | |||
|
|
|||
| def _prob_func(self, state): | |||
| if self.issuper: | |||
| return _data.norm.trace(unstack_columns(state)) | |||
| return _data.trace_oper_ket(state).real | |||
There was a problem hiding this comment.
Should trace_oper_ket not be exposed under data.norm the way the other norm operations are?
There was a problem hiding this comment.
We have both data.trace and data.norm.trace.
data.norm.trace is defined as tr(sqrt(A @ A.adjoint()))...
For dm, they give the same result and the normal trace is faster, so I should have used that one from the start.
Not a fan of having 2 functions with the same name, I sometime get confused and use the wrong one.
But this one should not be in data.norm.
|
@hodgestar, is there anything else you want me to fix in this PR? ps. Right now, to trace oper-bra, you would need to get it's adjoint first, use this or unstack then trace. |
|
@Ericgig Apologies for the delay. I wanted to go look at |
Description
Add and operation to compute the trace of column stacked operators without reshaping them.
The reshape would otherwise do a temporary copy of the data.
Another addition to improve stochastic solvers timings.