Support loading Parameter Expressions

[albi3ro]

Currently, the load function to convert a Qiskit circuit cannot convert Qiskit ParameterExpression's with more than a simple, unaltered Parameter. See error raised at converter.py#L205

PennyLane now, as of v0.14.0, supports classical processing of arguments within a circuit. With this change, PennyLane-Qiskit should update their support of ParameterExpression's.

We should be able to do the following:

from qiskit import QuantumCircuit
from qiskit.circuit import Parameter
from pennylane_qiskit import load

a = Parameter("a")
b = Parameter("b")

qc = QuantumCircuit(1,1)
qc.rz(a*b, [0])

circ = load(qc)(params={a: 0.1, b: 0.1})

Potential Solution

ParameterExpression's are based on SymPy, and thus we can use SymPy functions with them.

from sympy import lambdify

expr = a*b
f = lambdify([a,b], expr, 'numpy')
f(0.1, 0.1)

The resulting function is differentiable in pennylane:

import pennylane as qml

qml.grad(f)(1.0, 1.0)

[antalszava]

@albi3ro has this been resolved?

[lillian542]

This is supported now - for example, we can do:

from qiskit import QuantumCircuit
from qiskit.circuit import Parameter
from pennylane_qiskit import load

dev = qml.device("default.qubit")

a = Parameter("a")
b = Parameter("b")

qc = QuantumCircuit(1,1)
qc.rx(a*b, [0])

@qml.qnode(dev)
def circuit(x, y):
    load(qc)(params={a: x, b: y})
    return qml.expval(qml.PauliZ(0))

and run

>>> circuit(0, np.pi)
tensor(1., requires_grad=True)

>>> circuit(1, np.pi)
tensor(-1., requires_grad=True)

>>> circuit(0.5, np.pi)
tensor(2.22044605e-16, requires_grad=True)

The circuit is differentiable.