Following UT course on Quantum Machine Learning, they have a notebook on QAOA. As part of their lecture, they perform QAOA using Qiskit. Unfortunately, it seems the new versions of Qiskit have changed a lot since then and the code is not working anymore.

Specifically, the instructions on the optimization / evaluation procedure say the following:

We now create a function evaluate_circuit that takes a single vector gamma_beta (the concatenation of gamma and beta) and returns $\langle H_c \rangle = \langle \psi | H_c | \psi \rangle$ where $\psi$ is defined by the circuit created with the function above.

In their code, evaluate_circuit looks like this:

def evaluate_circuit(beta_gamma):
    n = len(beta_gamma)//2
    circuit = create_circuit(beta_gamma[:n], beta_gamma[n:])
    return np.real(Hc.eval("matrix", circuit, get_aer_backend('statevector_simulator'))[0])

So basically they're calculating the expected value of $H_c$ (the hamiltonian we expect to evolve to at the end) with respect to $\psi$, which in this case is constructed using the function create_circuit (which simulates the evolution from a superposition state based on the applicacion of Hc and Hm operators with angles beta and gamma, respectively):

def create_circuit(beta, gamma):
    circuit_evolv = sum([evolve(Hc, beta[i], qr) + evolve(Hm, gamma[i], qr)
    for i in range(p)], evolve(identity, 0, qr))
        circuit = circuit_init + circuit_evolv
    return circuit

The problem here is that $H_c$ in this case is an object of WeightedPauliOperator, and it doesn't seem to have the function eval anymore.

What's the best way to calculate $\langle \psi | H_c | \psi \rangle$ using WeightedPauliOperator objects?


1 Answer 1


What you can do to compute $\langle \psi | H_c | \psi \rangle$ with Qiskit, considering what you have, is first convert your circuit from create_circuit into a state, using CircuitStateFn. Careful for the importation though, since Qiskit got a recent big release, you will need a different importation whether you are using Qiskit 0.25 or Qiskit 0.24 (and older versions), uncomment the one that is good for your Qiskit version :

#0.24 and before
#from qiskit.aqua.operators import CircuitStateFn 
#from qiskit.opflow import CircuitStateFn
psi = CircuitStateFn(circuit)

Next you can directly compute $\langle \psi | H_c | \psi \rangle$ :

val = psi.adjoint().compose(Hc).compose(psi).eval().real

A shorter and equivalent expression for val is this :

output = (~psi @ Hc @ psi )
val = output.eval().real

Also, just in case you need it later, you can also compute an approximation from a dict of counts from measurement of the circuit you are looking at, and a diagonal observable you study that can be described as a matrix (array), a list describing the diagonal of the matrix, or a dict describing the behaviour over basis states (for example, for $ZZ$, it would be $\{"00": 1, "11": 1, "01": -1, "10": -1\}$ with the eigenvalue for each of the eigenstates). So here you would do something like :

from qiskit.quantum_info.analysis import average_data
mean = average_data(observable, counts)

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.