Pauli-X measurement instruction

[yaelbh]

Summary

This PR replaces the old PR #5311, while addressing the comments there.

Details and comments

1. This is only measure_x, when it's finalized (after review) I'll write measure_y.
2. Afterwards we can consider @itoko's suggestion Pauli measurement #5311 (comment).

Ready for review if the CI passes.
@1ucian0 @nkanazawa1989 @itoko

[coveralls]

Pull Request Test Coverage Report for Build 1915762000

• 44 of 56 (78.57%) changed or added relevant lines in 8 files are covered.
• 18 unchanged lines in 2 files lost coverage.
• Overall coverage increased (+0.01%) to 83.451%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
qiskit/visualization/text.py	14	15	93.33%
qiskit/circuit/measure_x.py	20	24	83.33%
qiskit/visualization/matplotlib.py	2	9	22.22%

Files with Coverage Reduction	New Missed Lines	%
qiskit/transpiler/passes/routing/stochastic_swap.py	5	96.18%
qiskit/init.py	13	67.19%

Totals
Change from base Build 1904273099:	0.01%
Covered Lines:	52444
Relevant Lines:	62844

---

💛 - Coveralls

[nkanazawa1989]

LGTM. I like this implementation in calibration perspective. Since we can attach calibration per every instruction, a general Pauli measure instruction is difficult to manage because schedule may change depends on the basis.

(but I will hold my approval since I'm not expert of circuit instruction)

[ikkoham]

I have a question. Should QC.remove_final_measurements remove this instruction or not?

[yaelbh]

@ikkoham This is a good question. See also here: #5311 (comment).

[1ucian0]

I have a question. Should QC.remove_final_measurements remove this instruction or not?

I think RemoveFinalMeasurements should remove this measurement. The original motivation was to create a statevector out of the circuit #2903

[ikkoham]

Thank you, it would be useful to have an option to leave the Hadamard gate when removing the X measurement.

[ajavadia]

Since we can attach calibration per every instruction, a general Pauli measure instruction is difficult to manage because schedule may change depends on the basis.

What do you mean by this @nkanazawa1989 ?

Also is the experiments usecase here that there is a separate measure_x calibration that is different from just a Hadamard+measure?

[yaelbh]

The purpose of this PR is to speed-up Aer (statevector method) for VQE circuits. The idea is that VQE circuits will express their expectation value using the new instructions. Consequently, Aer will be able to efficiently simulate multi-qubit Pauli measurements.

I'm not sure if the last couple of sentences make sense; at least they made sense some time ago, when I started this work. Possibly, since then, the interface between VQE and Aer has changed, especially in light of the new Estimator primitive.

I'm closing this PR, because currently I'm not focused on classical simulation for VQE. I find the addition of instructions less simple than it looks, for reasons detailed in previous comments. Back then, we (@chriseclectic and I) considered two alternatives. The one that we decided to pursue is to add instructions to Terra, and let Aer benefit from them. The other alternative is to build a mechanism in Aer that detects Pauli measurement instructions, in circuits that don't contain them explicitly.