ptrace error with large kets

[ciri]

I'm hitting the limit of ints in python, which causes troubles with large ket vectors.

Minimal example (I can go up to size 16 before it crashes):

ket = tensor([basis(2,0) for i in range(0,20)])
ket.ptrace(1)

Returns:

OverflowError                             Traceback (most recent call last)
<ipython-input-22-8485eed3b16f> in <module>()
      1 ket = tensor([basis(2,0) for i in range(0,20)])
----> 2 ket.ptrace(1)

/usr/lib/python2.7/site-packages/qutip-4.3.0.dev0+0b068fc4-py2.7-linux-x86_64.egg/qutip/qobj.pyc in ptrace(self, sel)
   1282         """
   1283         q = Qobj()
-> 1284         q.data, q.dims, _ = _ptrace(self, sel)
   1285         return q.tidyup() if settings.auto_tidyup else q
   1286 

/usr/lib/python2.7/site-packages/qutip-4.3.0.dev0+0b068fc4-py2.7-linux-x86_64.egg/qutip/cy/ptrace.pyx in qutip.cy.ptrace._ptrace()
     89     perm = fast_csr_matrix((data,ind,ptr),shape=(M * M, N * N))
     90     # No need to sort here, will be sorted in reshape
---> 91     rhdata = zcsr_mult(perm, zcsr_reshape(rho.data, np.prod(rho.shape), 1), sorted=0)
     92     rho1_data = zcsr_reshape(rhdata, M, M)
     93     dims_kept0 = np.asarray(rho.dims[0], dtype=np.int32).take(sel)

/usr/lib/python2.7/site-packages/qutip-4.3.0.dev0+0b068fc4-py2.7-linux-x86_64.egg/qutip/cy/spconvert.pyx in qutip.cy.spconvert.zcsr_reshape()
    187 @cython.boundscheck(False)
    188 @cython.wraparound(False)
--> 189 def zcsr_reshape(object A not None, int new_rows, int new_cols):
    190     """
    191     Reshapes a complex CSR matrix.

OverflowError: value too large to convert to int

Similar code works fine in MS LiQui|> using the same workstation so my hardware should be able to support this state size. What is the current ket size limit that is officially supported in qutip? 16?

[nonhermitian]

There is no official size limit. The code probably just needs int64 support.

[ciri]

I spent some time looking into patching this last night, but it seems like it would require a complete overhaul of many parts of the code so not an easy fix unfortunately. Perhaps this is something to put on the todo/request list, because without this, most ML QIP research that I've been looking at would not be implementable (need around 10-20 qubits, just a 2 qubit QRAM already requires 11 working qubits).

[Ericgig]

From 4.4.0, ptrace works up to 31Qbit