bloch_redfield_solve fails with: can't convert complex to float

[Muir Kumph]

When using the Bloch Redfield solver (Qutip 4.2.0) sometimes the subroutine fails with "can't convert complex to float".  The work-around is to pass expectation operators e_ops=[] so that the routine just returns the state vector and then post-evaluate the expectation values with expect().  Maybe this will help someone else with the same problem.  If anyone knows why, I would be curious to hear why.

Software	Version
QuTiP	4.2.0
Numpy	1.13.1
SciPy	0.19.1
matplotlib	2.0.2
Cython	0.26
Number of CPUs	8
BLAS Info	INTEL MKL
IPython	6.1.0
Python	3.6.2 |Continuum Analytics, Inc.| (default, Jul 20 2017, 12:30:02) [MSC v.1900 64 bit (AMD64)]
OS	nt [win32]

[koon...@gmail.com]

Hi Muir,

I got the same problem today: I was trying to pass an imaginary term into the Hamiltonian (something like an abs(erf(1+1j*t)) etc.).

It first gives me the error with :Cannot assign type 'double complex' to 'double',

then I change all the 1j*t term to complex(0,t), and it gives me this error.

I tried your solution but it is not working though. Will you mind if I ask if you are trying to parse complex term into your Hamiltonian as well when you are calculating the Bloch-Redfield master equation? If so, do you have any tips/tricks for incorporating complex term into the Bloch-redfield master equation ?

Zak

[Paul Nation]

Erf only supports real number inputs. You can use np.erf to do complex

--
You received this message because you are subscribed to the Google Groups "QuTiP: Quantum Toolbox in Python" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qutip+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[koon...@gmail.com]

Hi Paul,

It works. Thank you. 

Zak

[ahmed ghareeb]

Script.special.erf

See 

https://github.com/numpy/numpy/issues/12515

[koon...@gmail.com]

Hi Ahmed and Paul,

It turns out the result for the simulation does not match up with the same simulation using mesolve if I use np.erf (no additional time dependent a_op terms of course).

How would you suggest to use scipy.special.erf into the time dependent Hamiltonian ?

Obviously, simply putting scipy.special.erf into the string input of the time dependent Hamiltonian does not work, hence is there any way of adding scipy or specifically scipy.special into cython code?

Zak 

[Paul Nation]

The erf function needs to be updated to support complex. Not really that hard. Let me see if i can get it done in the next day or so

[Paul Nation]

See PR #960 that introduces the zerf

[Alex Pitchford]

PR #960 just merged. So zerf now available in dev branch.

If the users here want to try and report back, that would be great.

Thanks

Alex

[koon...@gmail.com]

Hi guys,

The function zerf works properly now. I got the same behaviour as the result I get using the function based method. Thanks.

Zak