Code that never executes in MPS::sample_measure_single_qubit

[aromanro]

Informations

• Qiskit Aer version: Latest development version
• Python version: Doesn't matter, it's in the C++ source
• Operating system: All

What is the current behavior?

There is some code in src/simulators/matrix_product_state/matrix_product_state_internal.cpp' in MPS::sample_measure_single_qubit` that never executes, more precisely, here:

qiskit-aer/src/simulators/matrix_product_state/matrix_product_state_internal.cpp

Line 1686 in 1e86d65

if (qubit != 0) // multiply mat by left lambda

Steps to reproduce the problem

I was looking over the sources to learn more about the qiskit aer internals when I've spotted this. Might be harmless, but it looks suspicious, so I labelled it as a bug.

What is the expected behavior?

I need to look more carefully over the code and perhaps over the referred paper (and also over 'Andrew J. Ferris and Guifre Vidal, “Perfect sampling with unitary tensor networks,” Phys. Rev. B 85, 165146 (2012)') to be sure...

Suggested solutions

Either the

    if (qubit != 0) // multiply mat by left lambda
      for (uint_t col = 0; col < mat.GetColumns(); col++)
        for (uint_t row = 0; row < mat.GetRows(); row++)
          mat(row, col) *= lambda_reg_[qubit - 1][row];

code should not be there or it should not be there but the

      for (uint_t col = 0; col < mat.GetColumns(); col++)
        for (uint_t row = 0; row < mat.GetRows(); row++)
          mat(row, col) *= lambda_reg_[qubit - 1][row];

should be in the else branch.

[aromanro]

Without yet looking over the referred article(s), it looks like the multiplication by left lambda should not be done, all lambdas are multiplied along into `mat' by the code that follows and multiplies the 'right lambda'.

This does not seem to be a bug, but simply a small enhancement by removing the useless code. Related with this, the shots parameter for the sample_measure function is useless and it should be removed.
Also I think that the documentation states that the complexity is O(1) for a shot, which is not true as the matrices are multiplied together for each shot.

Could somebody change the label to 'enhancement', it seems that I cannot edit that.

[gadial]

This code was added in #1377, specifically in da1ff73. Originally the first check was whether we have is_first_qubit==True , where is_first_qubit was a parameter to the method, so the dead code might have been activated. In the current state I think it indeed can be removed.

[aromanro]

Related with the same code, I think there still is an enhancement possible, avoiding the same matrix multiplication twice if zero is measured on a qubit.

What I mean is exemplified - because it's in a single function - in my implementation of a MPS simulator:
https://github.com/aromanro/QCSim/blob/798e33092d11db1053376e29207d3a3c914858a1/QCSim/MPSSimulatorImpl.h#L136
more precisely, this: https://github.com/aromanro/QCSim/blob/798e33092d11db1053376e29207d3a3c914858a1/QCSim/MPSSimulatorImpl.h#L168