Adds support for CCZ and CCX/TOF gates.

[MichaelBroughton]

This PR adds serialization and C++ implementation support for three qubit eigen gates. A lot of this code is copy pasted from the two qubit cases with tweaks to work for the three qubit cases. Currently tests will fail on our unitary op since the latest qsim (https://github.com/quantumlib/qsim/blob/278b217f94bca179ac5fafd85498ee28bbd3e236/lib/unitary_calculator_basic.h#L52) unitary calculator doesn't support three qubit gates. This is being worked on by @sergeisakov so we should be able to merge very soon once we can upgrade our qsim dependency. Fixes #480

[MichaelBroughton]

@jaeyoo Would you be able to take a look at this issue ? It seems like the ParameterShift differentiator is breaking with adding the new gates. I did some investigation and I think it may have actually been broken by one of the recent PRs we did (since it still breaks even without the new gates sometimes if I increase the depth of the circuits). Do you think you would be able to take a look and see if you can spot the fix ?

[jaeyoo]

@jaeyoo Would you be able to take a look at this issue ? It seems like the ParameterShift differentiator is breaking with adding the new gates. I did some investigation and I think it may have actually been broken by one of the recent PRs we did (since it still breaks even without the new gates sometimes if I increase the depth of the circuits). Do you think you would be able to take a look and see if you can spot the fix ?

Let me look at it during weekend, and let's talk about it at our next sync.

[jaeyoo]

Good news! I found out the fix.

[jaeyoo]

Good news! I fixed it Mike. The culprit was qsim::Cirq::GateKind::kCCX for Toffoli gate. It has its own enum value. :)

Please add the following line:

else if (circuit.gates[i].kind == qsim::Cirq::GateKind::kCCZPowGate ||
             circuit.gates[i].kind == qsim::Cirq::GateKind::kCCXPowGate ||
             circuit.gates[i].kind == qsim::Cirq::GateKind::kCCX) {

and then I saw the following test passed.

bazel test tensorflow_quantum/python/differentiators:gradient_test

[MichaelBroughton]

Unfortunately I don't think that is the fix :(. Like we talked about on our sync it is the parametershift differentiator that is having issues, not the adjoint differentiator. The tests fail somewhat randomly and unpredictably so If you can get the parametershift test to pass with --runs_per_test=50 and with many more circuits then I think that would be enough to say that the bug is fixed. When I run with this change I can still get the error if I change the number of circuits and ops.

[jaeyoo]

Oops, right, thank you for the catch. I am working on it, and found out the broken case in ParameterShift.

[jaeyoo]

Hi Mike, I want to ask you some questions.

I am trying to figure out the culprit circuit, but it seems not yet supported for multiple controlled qubit ops.
For example, at the test test_gradients_vs_cirq_finite_difference in gradients_test.py, I used

        # will this be too tight? time will tell.
        self.assertAllClose(cirq_grads, tfq_grads, rtol=2e-2, atol=2e-2,
                            msg=f"{util.from_tensor(circuit_batch)}")

and got this error:

Traceback (most recent call last):
  File "/usr/local/google/home/jaeyoo/tfq-venv/lib/python3.8/site-packages/absl/testing/parameterized.py", line 282, in bound_param_test
    return test_method(self, **testcase_params)
  File "/usr/local/google/home/jaeyoo/.cache/bazel/_bazel_jaeyoo/7da1e95e8bb125cc5976e7c817e098d4/execroot/__main__/bazel-out/k8-opt/bin/tensorflow_quantum/python/differentiators/gradie
nt_test.runfiles/__main__/tensorflow_quantum/python/differentiators/gradient_test.py", line 225, in test_gradients_vs_cirq_finite_difference
    msg=f"{util.from_tensor(circuit_batch)}")
  File "/usr/local/google/home/jaeyoo/.cache/bazel/_bazel_jaeyoo/7da1e95e8bb125cc5976e7c817e098d4/execroot/__main__/bazel-out/k8-opt/bin/tensorflow_quantum/python/differentiators/gradie
nt_test.runfiles/__main__/tensorflow_quantum/python/util.py", line 380, in from_tensor
    found_item = _parse_single(item)
  File "/usr/local/google/home/jaeyoo/.cache/bazel/_bazel_jaeyoo/7da1e95e8bb125cc5976e7c817e098d4/execroot/__main__/bazel-out/k8-opt/bin/tensorflow_quantum/python/differentiators/gradie
nt_test.runfiles/__main__/tensorflow_quantum/python/util.py", line 330, in _parse_single
    raise TypeError('Error decoding item: ' + str(item))
TypeError: Error decoding item: C1C1C0CZ**0.648900484520275*a((0, 2), (0, 4), (0, 1), (0, 3), (0, 0))

Any help please?

[MichaelBroughton]

Hmmm is circuit_batch something has had tfq.convert_to_tensor called on it ?

[yudeep-rajbhandari]

Is there any update on this?