Tensor network simulator backend about qiskit-aer HOT 16 CLOSED

@chriseclectic or @atilag , I don't think I have write permissions here, can you please assign @merav-aharoni for this Issue? She'll be responsible for checking the checkboxes above to make it aer-ready

from qiskit-aer.

Will add below some preliminary runtime results. All cases are random circuits with X,Y,Z,H,S,T,CX gates, CX percentage is 10%, all others are uniform. Each datapoint is an average of three runs. Time in all charts is in seconds. All datapoints are checked for correctness against the Aer Qasm simulator full statevector. Experiments were conducted mainly by @merav-aharoni

from qiskit-aer.

Full calculation of statevector at end of TN run:

TN vs. qasm benchmark_10.pdf

from qiskit-aer.

Without calculation of statevector at end of TN run (for use, e.g., for expectation value finding):

TN vs. qasm benchmark_10_no_statevector_expansion.pdf

from qiskit-aer.

Calculating expectation value of X on a single qubit:

TN vs. qasm benchmark_10_expectation_X.pdf

from qiskit-aer.

Here is a graph of running the tensor network simulator up to 128 qubits (without calculating the statevector). Qasm ran up to 26 qubits.

from qiskit-aer.

Here is a graph of a benchmark for calculating expectation value on subset of qubits.
The matrix M is a tensor product of 5 random paulis on a subset of 5 following qubits.
Each point in the graph is an average of results of 5 circuits.

TN vs. qasm expectaition value of 5 qubits benchmark 10% CNOTS.pdf

from qiskit-aer.

Continued chart from Elad - runtime for getting expectation value of random Paulis on five consecutive qubits, for random circuits of depth 120 and up to 150 qubits

from qiskit-aer.

Runtime does increase substantially with depth. Will show rigorous results of that soon

from qiskit-aer.

(and of course do the work behind the checkboxes :-) )

from qiskit-aer.

with @eladgoldman

from qiskit-aer.

Here two graphs of a benchmark for calculating expectation value on subset of qubits.
First graph, for 16 qubits, changing depth (20:10:620):

Second graph, for depth 120, changing number of qubits (10:5:700), Qasm is shown up to 23 qubits:

from qiskit-aer.

At some stage we will start benchmarking also on structured problems, in particular Aqua. For now we are mainly focused on infrastructure items and consistency, correctness, and extent of the algorithm. Hence the random benchmarks which in a sense provide the most simplest and straightforward results to analyze

from qiskit-aer.

Current status:
Tensor network simulator is implemented as a simulation method of qasm controller.
Supported operations: gates, snapshots (statevector and expectation value pauli).
Next steps:
Testing, benchmarking - @eladgoldman
Complete missing interface functions, additional code improvements - Merav
Review - @chriseclectic , @atilag

from qiskit-aer.

It's been a while since we update this issue. I guess it was intended to be used as a "meta" or "epic" issue to track MPS simulator progress.
@yehuda-naveh Can we close this and create issues for each of the remaining tasks?

from qiskit-aer.

@atilag , you can close this issue.

from qiskit-aer.