Failure when running QAOA on AQT-Simulator

[lasys]

Informations

• Qiskit AQT Provider version: 0.4.2
• Qiskit version: 0.28 and 0.27
• Python version: 3.8.0
• Operating system: Linux

What is the current behavior?

I wanted to execute QAOA on the AQT-Simulator, but unfortunately I have this error several times in a row:

FAILURE: Can not get job id, Resubmit the qobj to get job id.Error: 'list' object has no attribute 'qubits'

Steps to reproduce the problem

import networkx as nx
import numpy as np
from qiskit.algorithms.optimizers import SPSA
from qiskit.utils import QuantumInstance
from qiskit_optimization.algorithms import MinimumEigenOptimizer
from qiskit.algorithms import QAOA
from qiskit_aqt_provider import AQTProvider
from qiskit_optimization.applications import Maxcut

# create graph
n = 4
graph = nx.Graph()
graph.add_nodes_from(np.arange(0, n, 1))
elist = [(0, 1, 1), (1, 2, 1), (2, 3, 1), (3, 0, 1)]
graph.add_weighted_edges_from(elist)

# create Maxcut
max_cut = Maxcut(graph)
max_cut_qubo = max_cut.to_quadratic_program()

# connection to AQT
aqt = AQTProvider('<token>')
print(aqt.backends)
simulator_backend = aqt.backends.aqt_qasm_simulator
quantum_instance = QuantumInstance(backend=simulator_backend,shots=200)
qaoa = QAOA(optimizer=SPSA(maxiter=1), quantum_instance=quantum_instance)

# execute qaoa
MinimumEigenOptimizer(qaoa).solve(max_cut_qubo)

What is the expected behavior?

Running without failure.

Suggested solutions

[nonhermitian]

I can reproduce this, however I do not understand the error as the AQT jobs have a job_id that is returned by the API (explicitly verified this). I would open an issue in the qiskit-terra repo as the issue is likely in the quantum_instance job submission code.

[mtreinish]

I took a look at this and after adding a bunch of raises and prints to the QuantumInstance code the underlying issue is in the aqt provider: https://github.com/Qiskit-Partners/qiskit-aqt-provider/blob/master/qiskit_aqt_provider/aqt_job.py#L80 does not account for a list of circuits. backend.run() can take a QuantumCircuit or a list of QuantumCircuit objects and while the aqt provider only supports a single circuit we still should accept a list of 1 as input and test that path. It should be simple enough to fix (probably just converting a single entry list to a QuantumCircuit in the job).

[woodsp-ibm]

What could be tried, to check that out, is to set this environment variable that allows the circuits per job to be limited and see if that works - of course it may turn out to be a lot of jobs dependent on the algo in question. (I notice the name never got changed though!)

MAX_CIRCUITS_PER_JOB = os.environ.get("QISKIT_AQUA_MAX_CIRCUITS_PER_JOB", None)

[lasys]

I took a look at this and after adding a bunch of raises and prints to the QuantumInstance code the underlying issue is in the aqt provider: https://github.com/Qiskit-Partners/qiskit-aqt-provider/blob/master/qiskit_aqt_provider/aqt_job.py#L80 does not account for a list of circuits. backend.run() can take a QuantumCircuit or a list of QuantumCircuit objects and while the aqt provider only supports a single circuit we still should accept a list of 1 as input and test that path. It should be simple enough to fix (probably just converting a single entry list to a QuantumCircuit in the job).

Yes, you're right.
I set self.qobj to the given QuantumCircuit and it's working.

def _build_memory_mapping(self):
    qu2cl = {}
    if isinstance(self.qobj, QasmQobj):
        for instruction in self.qobj.experiments[0].instructions:
            if instruction.name == 'measure':
                qu2cl[instruction.qubits[0]] = instruction.memory[0]
        return qu2cl
    qubit_map = {}
    count = 0
    
    # NEW # 

    # check if self.qobj is list
    if isinstance(self.qobj, list) and len(self.qobj) != 0:
        # check if first element of self.qobj is QuantumCircuit
        if isinstance(self.qobj[0], QuantumCircuit):
            # set self.qobj to QuantumCircuit object
            self.qobj = self.qobj[0]
  
    for bit in self.qobj.qubits:
        qubit_map[bit] = count
        count += 1
    clbit_map = {}
    count = 0
    for bit in self.qobj.clbits:
        clbit_map[bit] = count
        count += 1
    for instruction in self.qobj.data:
        if instruction[0].name == 'measure':
            for index, qubit in enumerate(instruction[1]):
                qu2cl[qubit_map[qubit]] = clbit_map[instruction[2][index]]
    return qu2cl

[mtreinish]

@lasys please feel free to open a PR with your modification (I meant to last week but clearly forgot about it) to fix this issue. The one thing I'd say though is that the list will only have a single entry. The AQTBackend's configuration set the max_experiments attribute to 1: https://github.com/Qiskit-Partners/qiskit-aqt-provider/blob/master/qiskit_aqt_provider/aqt_backend.py#L49 so it should raise an exception if there is more than 1 circuit in the list: https://github.com/Qiskit-Partners/qiskit-aqt-provider/blob/056b203e97bb5eabcc77b08c359f711628c11efe/qiskit_aqt_provider/circuit_to_aqt.py#L77

[lasys]

@mtreinish all right. Then a simple fix like:

...
qubit_map = {}
count = 0

self.qobj = self.qobj[0]

for bit in self.qobj.qubits:
    qubit_map[bit] = count
    count += 1
clbit_map = {}
...

should be okay, shouldn't it?
I will open a PR.