No output when simulating Cirq circuits via Python interface without reading in files

[rmlarose]

I'm trying to simulate a Cirq circuit via the Python interface without reading in circuits/grids/orderings from files. I can run the QFlex Cirq example to get the expected output. I modified this example to the following which creates the grid and circuit without reading them in from files:

import sys
import os
sys.path.insert(
    1, os.path.realpath(os.path.dirname(os.path.realpath(__file__)) + '/../'))
import python.cirq_interface.qflex_simulator as qsim
import python.cirq_interface.qflex_virtual_device as qdevice
import python.cirq_interface.qflex_circuit as qcirc

import cirq

# Define a grid of qubits
n = 4
grid = "1 " * n

# Get a device from the grid of qubits
device = qdevice.QFlexVirtualDevice(qflex_grid=grid)

# Get the cirq.GridQubit's from the device
qreg = device.get_indexed_grid_qubits()

# Define the circuit to simulate in Cirq
circ = cirq.Circuit()
circ.append([cirq.ops.H.on(qbit) for qbit in qreg.values()])

# Get the QFlex Circuit from the Cirq circuit
qflex_circ = qcirc.QFlexCircuit(cirq_circuit=circ, device=device)

# Simulate the circuit using QFlex
sim = qsim.QFlexSimulator()
res = sim.compute_amplitudes(program=qflex_circ, bitstrings=["0000"])
print("Result:")
print(res)

The output below shows that no amplitudes are returned from sim.compute_amplitudes:

Ordering is being computed from the provided circuit ...
... Done!
Result:
[]
_Map_base::at

I think _Map_base::at indicates a memory leak? Any ideas what is going on here?

[95-martin-orion]

As far as I can tell, the only notable differences between this and the config_small example are the shape of the grid (the example is square, while this is a single line) and the depth of the circuit (the example has a handful of gates per qubit, while this only has one). I'll take a look at this tomorrow and see if I can narrow down possible reasons that might cause an error.

Out of curiosity, what is the ordering generated for this circuit? You should be able to check with these additions:

from python.ordering import order_circuit_simulation as auto_order
# ...define circuit...
auto_order.circuit_to_ordering(qflex_circ, qubit_names=sorted(qreg))

[rmlarose]

When I add

print(auto_order.circuit_to_ordering(qflex_circ, qubit_names=sorted(qreg)))

after creating the circuit, the output is ['# approximate fidelity: 1.0'].

I tried increasing the circuit depth and changing the grid to 2x2 instead of 1x4 but still have the same issue.

[95-martin-orion]

I see the same failure when running 2x2 circuits with the CNOT gates removed. Have any of your tests included CNOTs or any other two-qubit gate?

This appears to be a bug in core qFlex behavior, as the same issue appears when running via the C++ binary: if a qubit is not acted on by at least one two-qubit gate, the simulation will fail. I'll work on fixing this, but until then you can work around this by adding back-to-back CZ gates on pairs of qubits if one or both of them do not otherwise have a two-qubit gate. (For most circuit I don't think this will be an issue - lone disconnected qubits don't contribute to the complexity of a simulation.)

[95-martin-orion]

The root cause of this bug occurs during the flatten_grid_of_tensors step: if a qubit has no interaction with any other qubits, it will flatten down to a rank-0 tensor (i.e. a scalar). This wreaks havoc with certain tensor-cleanup methods.

PR #274 addresses this issue.