Superoperator noise for density matrix method #295

chriseclectic · 2019-07-26T14:24:39Z

Summary

Adds support for general superoperator noise for the density matrix simulation method.

Depends on #253

TODO

Add noise tests for density matrix method that check measure sampling is working (Blocked by enable sampling for readout error #222)
Revamp noisy qasm simulator method testing for other non-statevector methods (stabilizer, extended stabilizer, tensor network) [PR Add noise tests for different qasm simulator methods #312]

Details and comments

Adds Superoperator class (subclass of DensityMatrix/QubitVector).
Adds function to QuantumError class to convert to Superoperator via superoperator simulation of noise sub circuits
Adds option to noise model to sample noise as full error superoperators rather than default stochastic noise sampling.
Adds superoperator noise sampling to QasmSimulator for the density matrix method
Moves QubitVector and its subclasses into separate folder.
Move state methods to seperate src/state-methods folder, and only have controllers in src/simulators

Example

An example of where this is useful is running noisy small qubit circuits with Kraus errors (eg something like randomized benchmarking simulations).

For the following example timing results on my laptop are:

Statevector method: 25 seconds
Density matrix method: 0.025 seconds

import numpy as np
from qiskit import *
from qiskit.compiler import assemble
from qiskit.providers.aer import QasmSimulator
from qiskit.providers.aer.noise import NoiseModel
from qiskit.providers.aer.noise.errors import thermal_relaxation_error

sim = QasmSimulator()

# Error model
noise_model = NoiseModel()
error = thermal_relaxation_error(50, 100, 50)
noise_model.add_all_qubit_quantum_error(error, ['u1', 'u2', 'u3'], False)

# Test circuit with noisy random u3 gates
nq = 2
depth = 1000
shots = 2000
qr = QuantumRegister(nq)
cr = ClassicalRegister(nq)
circ = QuantumCircuit(qr, cr)
for j in range(depth):
    for j in range(nq):
        a, b, c = np.random.rand(3)
        circ.u3(a, b, c, qr[j])
    circ.barrier(qr)
circ.measure(qr, cr)
qobj = assemble(circ, sim, shots=shots)

# Execute on statevector simulator
opts_sv = {"method": "statevector"}
results_sv = sim.run(qobj,
                 noise_model=noise_model,
                 backend_options=opts_sv).result()
time_sv = np.round(results_sv.results[0].time_taken, 3)
print("Execution time: {}".format(time_sv))
# ~25 seconds on Macbook

# Execute on density matrix simulator
opts_dm = {"method": "density_matrix"}
results_dm = sim.run(qobj,
                 noise_model=noise_model,
                 backend_options=opts_dm).result()
time_dm = np.round(results_sv.results[0].time_taken, 3)
print("Execution time: {}".format(time_dm))
# ~0.025 seconds on Macbook

chriseclectic · 2019-07-30T23:59:14Z

@atilag I rebased to remove the src directory restructuring from this PR so its easier to review, I will add that as a final PR after the others are merged

atilag

Good job again!

src/noise/noise_model.hpp

atilag · 2019-08-01T14:04:57Z

src/simulators/superoperator/superoperator.hpp

+  //-----------------------------------------------------------------------
+
+  Superoperator() : Superoperator(0) {};
+  explicit Superoperator(size_t num_qubits);


What implicit conversion are we trying to avoid here? ... not sure if explicit makes sense.. but in any case, is not bothering.

we can assing num_qubits_ in the initializer list of the ctor:

explicit Superoperator(size_t num_qubits) : num_qubits_(set_num_qubits(num_qubits)){ }

this used to save some time at object construction time (at least the last time I saw it, It may not be the case now)

This doesn't compile for me

atilag · 2019-08-01T14:12:02Z

src/simulators/superoperator/superoperator.hpp

+// Constructors & Destructor
+//------------------------------------------------------------------------------
+
+template <typename data_t>


see comment above (in the ctor)

atilag · 2019-08-01T14:21:13Z

src/simulators/superoperator/superoperator_state.hpp

+  }
+
+  // Return the set of qobj snapshot types supported by the State
+  virtual stringset_t allowed_snapshots() const override {


In our TODO list, we have to decople Qobj as much as possible.. just in case we change the format in the future (Protobuffer... is coming...)

atilag · 2019-08-01T14:24:23Z

src/simulators/superoperator/superoperator_state.hpp

+                                 const std::vector<Operations::Op> &ops) {
+  // An n-qubit unitary as 2^4n complex doubles
+  // where each complex double is 16 bytes
+  (void)ops; // avoid unused variable compiler warning


We can remove this ungly c-casting style by setting a default value for the ops parameter...

..., onst std::vector<Operations::Op> &ops = {}) {

Did you have the chance to check if this generates another unsed variable warning?

Even if it has a default value, it will still be unused in the function body

atilag · 2019-08-01T14:25:47Z

src/simulators/superoperator/superoperator_state.hpp

+  // An n-qubit unitary as 2^4n complex doubles
+  // where each complex double is 16 bytes
+  (void)ops; // avoid unused variable compiler warning
+  size_t shift_mb = std::max<int_t>(0, num_qubits + 4 - 20);


Why - 20 ?
Can we add a comment or better, use a const with a name?

Shift by -20 converts bytes to MiB

* Was using the wrong qubit number to calculate qubitvector multiplication indexes.

atilag

Merging!

* Add superoperator simulator * Add superoperator sampling method to noise model * Add superoperator noise to density matrix qasm controller * Remove AbstractError and add superop to QuantumError * Fix measure sampling test for density matrix method * Fix bug in superoperator matrix multiplication - Was using the wrong qubit number to calculate qubitvector multiplication indexes. * Use QasmBackendConfiguration

chriseclectic requested a review from atilag as a code owner July 26, 2019 14:24

atilag added this to the 0.3 milestone Jul 30, 2019

chriseclectic force-pushed the feature/superoperator-noise branch 2 times, most recently from 63b7002 to 212506a Compare July 30, 2019 23:48

chriseclectic added 4 commits August 1, 2019 00:30

add superoperator simulator

2f88f31

remove AbstractError, add superop to QuantumError

2d32a6a

add superoperator sampling method to noise model

c4c87ca

add superoperator noise to density matrix qasm controller

b59e13b

chriseclectic force-pushed the feature/superoperator-noise branch from 212506a to b59e13b Compare August 1, 2019 04:30

atilag previously approved these changes Aug 1, 2019

View reviewed changes

chriseclectic added 4 commits August 1, 2019 11:44

Merge branch 'master' into feature/superoperator-noise

541bb71

juan style fixes

15480b5

fix measure sampling test for density matrix method

76def51

Fix bug in superoperator matrix multiplication

918fae4

* Was using the wrong qubit number to calculate qubitvector multiplication indexes.

chriseclectic dismissed atilag’s stale review via 918fae4 August 1, 2019 19:33

chriseclectic changed the title ~~[WIP] Superoperator noise for density matrix method~~ Superoperator noise for density matrix method Aug 1, 2019

chriseclectic added 3 commits August 1, 2019 15:45

Merge branch 'master' into feature/superoperator-noise

a351027

use QasmBackendConfiguration

6f33270

fixup

c183c44

atilag approved these changes Aug 2, 2019

View reviewed changes

atilag merged commit bb2ac97 into Qiskit:master Aug 2, 2019

chriseclectic mentioned this pull request Aug 2, 2019

Backend configuration should create type of QasmBackendConfiguration #189

Closed

chriseclectic deleted the feature/superoperator-noise branch October 4, 2019 15:31

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Superoperator noise for density matrix method #295

Superoperator noise for density matrix method #295

chriseclectic commented Jul 26, 2019 •

edited

Loading

chriseclectic commented Jul 30, 2019

atilag left a comment

atilag Aug 1, 2019

atilag Aug 1, 2019

chriseclectic Aug 1, 2019

atilag Aug 1, 2019

atilag Aug 1, 2019

atilag Aug 1, 2019

chriseclectic Aug 1, 2019

atilag Aug 1, 2019

chriseclectic Aug 1, 2019

atilag left a comment

Superoperator noise for density matrix method #295

Superoperator noise for density matrix method #295

Conversation

chriseclectic commented Jul 26, 2019 • edited Loading

Summary

TODO

Details and comments

Example

chriseclectic commented Jul 30, 2019

atilag left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

atilag left a comment

Choose a reason for hiding this comment

chriseclectic commented Jul 26, 2019 •

edited

Loading