Qutip can't handle .jnp arrays

[nlwach]

I am trying to differentiate a quantum circuit which uses qutip. Whenever I try to create a qutip.Qobj using a jnp.array the created qutip.Qobj is empty. Somewhere the jnparrays differ too much from the usual np arrays. The creation of qutip.Qobj using the standard np.array's works perfectly fine.

Here the code for the jnp arrays:

state = jnp.array([1+0j, 1+0j])*1/jnp.sqrt(2)
Q_state = q.Qobj(state)
print(Q_state)

and the output

Quantum object: dims = [[1], [1]], shape = (1, 1), type = bra
Qobj data =
[[0.]]

And using NumPyarrays:

state = np.array([1+0j, 1+0j])*1/np.sqrt(2)
Q_state = q.Qobj(state)
print(Q_state)

and the output

Quantum object: dims = [[2], [1]], shape = (2, 1), type = ket
Qobj data =
[[0.70710678]
 [0.70710678]]

[jakevdp]

Hi - thanks for the question. It appears that Qutip does not support duck-typed ndarrays, as you can see in the source: https://github.com/qutip/qutip/blob/dc2abc4be6540a3a3c549ed21d8da83f3f0db5aa/qutip/qobj.py#L221-L343

JAX arrays do not meet any of the isinstance checks used there, so they appear to be incompatible with the package. I would suggest explicitly casting the input to numpy, i.e.

Q_state = q.Qobj(np.asarray(state))

If you would like qutip to support JAX and other numpy-like array inputs natively (perhaps by checking for an __array__ attribute on the input), I would suggest opening a feature request in the Qutip repository.