Creating a parameterized Operator in Qiskit that cannot be decomposed into Qiskit supported gates

Question

I am trying to create a custom ansatz to use the built-in Qiskit VQE() function. My ansatz is composed of single qubit gates and a hamiltonian gate which cannot be decomposed into Qiskit supported gates. Here is my ansatz function:

def quantum_state_preparation(sites, reps):
qc = QuantumCircuit(sites)
num_params = reps*(2*sites+1)
params = ParameterVector('θ', num_params)  

H = 0*SparsePauliOp('I'*(sites))
for j in range(1, sites):
    H -= 1/2 * (SparsePauliOp('I'*(j-1) + 'XX' + 'I'*(sites-j-1)) + SparsePauliOp('I'*(j-1) + 'YY' + 'I'*(sites-j-1)))
ham_op = H.simplify()

for n in range(reps):
    ham_gate = HamiltonianGate(ham_op, params[n*(2*sites+1)]/2)
    qc.append(ham_gate, range(sites))

    for i in range(sites): #single qubit gates
        qc.p(params[n*(2*sites+1)+1+2*i] ,i)
        qc.rx(params[n*(2*sites+1)+2+2*i] ,i)
qc.parameter_bounds=[[0,2*np.pi]]*num_params 
return qc

When I plug quantum_state_preparation(sites, reps) into the VQE() function it fails. I am pretty sure it is because HamiltonianGate() is unable to handle unbound parameters but I don't know a way around this.

Answer 1

From HamiltonianGate docs ...which can be decomposed into basis gates if it is 2 qubits or less, or simulated directly in Aer for more qubits.

Try PauliEvolutionGate which is what the EvolvedOperatorAnsatz in the circuit library uses and is the base class for QAOAAnsatz there and UCC in Nature and they work with VQE.