Backend-agnostic matrix definitions

[Simone-Bordoni]

Store a backend agnostic matrix definition tht can be casted with a specific _cast function in the proper backend.
Necessary for PR #1202

Checklist:

• Reviewers confirm new code works as expected.
• Tests are passing.
• Coverage does not decrease.
• Documentation is updated.

[Simone-Bordoni]

@renatomello
I am getting an error for the CSXDG gate.
previously is was:

self.np.conj(self.CSX)

Now i changed to:

a = (1 - 1j) / 2
b = self.np.conj(a)
      return self._cast(
          [
              [1, 0, 0, 0],
              [0, 1, 0, 0],
              [0, 0, a, b],
              [0, 0, b, a],
          ],
          dtype=self.dtype,
      )

It should be the same thing as CSX is defined the same way but with a=(1+1j)/2

[renatomello]

@renatomello I am getting an error for the CSXDG gate. previously is was:

self.np.conj(self.CSX)

Now i changed to:

a = (1 - 1j) / 2
b = self.np.conj(a)
      return self._cast(
          [
              [1, 0, 0, 0],
              [0, 1, 0, 0],
              [0, 0, a, b],
              [0, 0, b, a],
          ],
          dtype=self.dtype,
      )

It should be the same thing as CSX is defined the same way but with a=(1+1j)/2

Which error though

[Simone-Bordoni]

@renatomello I am getting an error for the CSXDG gate. previously is was:

self.np.conj(self.CSX)

Now i changed to:

a = (1 - 1j) / 2
b = self.np.conj(a)
      return self._cast(
          [
              [1, 0, 0, 0],
              [0, 1, 0, 0],
              [0, 0, a, b],
              [0, 0, b, a],
          ],
          dtype=self.dtype,
      )

It should be the same thing as CSX is defined the same way but with a=(1+1j)/2

Which error though

Assertion error in test_csxdg and test_dagger (the tests that failed also on github)

[renatomello]

Which error though

Assertion error in test_csxdg and test_dagger (the tests that failed also on github)

It's only failing for numba. So, most likely a PR in qibojit is needed here adding

@cached_property
def CSXDG(self):
    return self.SXDG

I have not tested locally but I'd imagine this would fix it.

[stavros11]

It's only failing for numba. So, most likely a PR in qibojit is needed here adding

@cached_property
def CSXDG(self):
    return self.SXDG

I have not tested locally but I'd imagine this would fix it.

@renatomello do you have any idea why this is the case? I understand this PR is changing the calculations of some matrices a bit, but the returned values should remain the same. I don't see how this would break qibojit. Unless this was an existing issue that came up now for some reason.

[renatomello]

@renatomello do you have any idea why this is the case? I understand this PR is changing the calculations of some matrices a bit, but the returned values should remain the same. I don't see how this would break qibojit. Unless this was an existing issue that came up now for some reason.

I didn't dig much deeper, but I checked and numba is generating the same matrices as the other backends (via gates.Gate.matrix(backend)), but somehow when it applies the gate and its dagger to a state it does not recover the original state. At least numpy and tensorflow are passing this test for this gate. I didn't check cupy / cuquantum.
And yes, I agree this is strange and should've came up in other situations.

[Simone-Bordoni]

Which error though

Assertion error in test_csxdg and test_dagger (the tests that failed also on github)

It's only failing for numba. So, most likely a PR in qibojit is needed here adding

@cached_property
def CSXDG(self):
    return self.SXDG

I have not tested locally but I'd imagine this would fix it.

I have opened the PR in qibojit

[Simone-Bordoni]

@renatomello @stavros11 by adding the CSXDG gate in qibojit the tests fail only on linux...

[renatomello]

@renatomello @stavros11 by adding the CSXDG gate in qibojit the tests fail only on linux...

it's because only linux is testing tensorflow, and the issue is with other gates

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (5fa9c82) 100.00% compared to head (98cf3d3) 100.00%.

❗ Current head 98cf3d3 differs from pull request most recent head bfaf142. Consider uploading reports for the commit bfaf142 to get more accurate results

Additional details and impacted files

@@            Coverage Diff            @@
##            master     #1217   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           70        70           
  Lines        10379     10381    +2     
=========================================
+ Hits         10379     10381    +2     

Flag	Coverage Δ
unittests	100.00% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[renatomello]

This looks good to advance #1202, but it needs to be merged with qiboteam/qibojit#168.

It'd be also interesting to test at least locally if this is really enough for pytorch to work.

[Simone-Bordoni]

Now that the qibojit branch "add CSXDG" has been merged this PR is ready to be merged (after one missing review)