get_ground_state(sparse_operator) sometimes fails with SIGSEGV about openfermion

[jarrodmcc]

This particular example seems to converge relatively fast (and correctly) in both cases for me. To get to the root of this problem, can you tell me if eigsh work for you in other circumstances? For example on just a simple 4x4 matrix? If you watch your resources, are you running out of memory when you get the SIGSEGV error?

[babbush]

This example runs just fine for me as well. Which version of OpenFermion are you running? What about your scipy version?

[Konjkov]

Hello Jarrod, Ryan
I have scipy version 1.0.0 and eigsh never work fine for me even in _hydrogen_integration_test.py or _lih_integration_test.py where eigsh is used in get_ground_state() function
Openfermion is latest from master (2 march commit)

best vladimir

[jarrodmcc]

Hi Vladimir
It sounds like the test is failing at the level of ARPACK. If your python distribution was self-contained you may want to reinstall scipy and/or numpy. If it's linking to existing linear algebra packages or ARPACK, you may need to reinstall those and ensure they are working.

[Konjkov]

Hi Jarrod
I created a simpler test that reproduces an issue on my computer

#!/usr/bin/env python3
import numpy as np
from scipy.sparse.linalg import eigsh

m = np.matrix(
[[-2.77555756e-17+0.j,  0.00000000e+00+0.j,  0.00000000e+00+0.j,   0.00000000e+00+0.j],
 [ 0.00000000e+00+0.j, -4.66581850e-01+0.j,  0.00000000e+00+0.j,   0.00000000e+00+0.j],
 [ 0.00000000e+00+0.j,  0.00000000e+00+0.j, -1.58557755e-01+0.j,   0.00000000e+00+0.j],
 [ 0.00000000e+00+0.j,  0.00000000e+00+0.j,  0.00000000e+00+0.j,  -4.66581850e-01+0.j]]
)
energy, state = eigsh(m, 2, which='SA', maxiter=1e7)

I think I need to write to scipy tech support, sorry.