Re-examine the threshold deciding which algorithm to use for sample_measure in MPS

[merav-aharoni]

Informations

• Qiskit Aer version:
• Python version:
• Operating system:

What is the current behavior?

For a simple circuit creating an EPR state and measure 100 shots, a user measured time_taken is 42sec for 23 qubits, 88sec for 24qubits, 207sec for 25qubits, 510sec for 26qubits and 0.2sec for 27 qubits. Also 0.3sec for 28qubits .

Steps to reproduce the problem

import numpy as np
from qiskit import QuantumCircuit, QuantumRegister, ClassicalRegister
from qiskit import Aer, execute
from qiskit.providers.aer import QasmSimulator

simulator = Aer.get_backend('qasm_simulator')
backend_opts = {'method':"matrix_product_state"}
#backend_opts = {'method':"statevector"}
num_qubits = 25


qr = QuantumRegister(num_qubits)
cr = ClassicalRegister(num_qubits)
circ = QuantumCircuit(qr, cr)

circ.h(0)
for i in range(0, num_qubits-1):
    circ.cx(i, i+1)
circ.measure(qr, cr)
job_sim = execute([circ], QasmSimulator(), backend_options= backend_opts, shots=100 )
result = job_sim.result()
print("Time taken: {} sec".format(result.time_taken))
result.get_counts()

When we add the option:
'mps_sample_measure_qubits_opt':20, we get the expected performance.

What is the expected behavior?

The default value for this parameter should give the optimal performance.

Suggested solutions

Re-examine the default values for this parameter and also for mps_sample_measure_shots_opt.

[merav-aharoni]

Also add documentation for these thresholds in qasm_simulator.py.

[merav-aharoni]

I am running experiments to define better thresholds to decide between the two algorithms for sample_measure. The experiments include 3 parameters: number of qubits, number of shots, max bond dimension.
Here are the results for bond dimension <= 2.