Bug in SymbolicHamiltonian.expectation_from_samples for Qibo 0.2.14

[chmwzc]

To Reproduce

from qibo import Circuit, gates
from qibo.hamiltonians import SymbolicHamiltonian
from qibo.symbols import Z

circuit = Circuit(3)
circuit.add(gates.M(_i) for _i in (0, 2))
result = circuit(nshots=10)
freq = result.frequencies() # Counter({'00': 10})

ham = SymbolicHamiltonian(Z(2))
qubit_map = [0, 2]

print(ham.expectation_from_samples(freq, qubit_map))

The above code should simply give 1.0, but fails in Qibo 0.2.14. The error message:

Traceback (most recent call last):
  File "/home/wongzc/testing/sample_expectation/qibo_0p2p14_bug/bug.py", line 19, in <module>
    print(ham.expectation_from_samples(freq, qubit_map))
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/wongzc/miniforge3/envs/qibochem/lib/python3.11/site-packages/qibo/hamiltonians/hamiltonians.py", line 662, in expectation_from_samples
    [
  File "/home/wongzc/miniforge3/envs/qibochem/lib/python3.11/site-packages/qibo/hamiltonians/hamiltonians.py", line 664, in <listcomp>
    * (-1) ** [state[qubit_map[q]] for q in qubits].count("1")
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/wongzc/miniforge3/envs/qibochem/lib/python3.11/site-packages/qibo/hamiltonians/hamiltonians.py", line 664, in <listcomp>
    * (-1) ** [state[qubit_map[q]] for q in qubits].count("1")
                     ~~~~~~~~~^^^
IndexError: list index out of range

I believe the problem is caused by the qubit_map argument in SymbolicHamiltonian.expectation_from_samples.

The documentation says that it should be a dictionary:

qibo/src/qibo/hamiltonians/hamiltonians.py

Line 638 in a518e99

qubit_map (dict): qubit map.

But the default value is still a list:

qibo/src/qibo/hamiltonians/hamiltonians.py

Lines 651 to 652 in a518e99

	if qubit_map is None:
	qubit_map = list(range(self.nqubits))

while the code that checks the binary strings from the measurement frequencies will work if qubit_map is a dictionary (like {0: 0, 2:1} for my example above)

qibo/src/qibo/hamiltonians/hamiltonians.py

Line 664 in a518e99

* (-1) ** [state[qubit_map[q]] for q in qubits].count("1")

Suggested fix

I would suggest keeping the qubit_map argument for SymbolicHamiltonian as a list (just edit the docs) to keep it consistent across all the different Hamiltonians available, then just change the line above to

* (-1) ** [state[qubit_map.index(q)] for q in qubits].count("1")

[MatteoRobbiati]

It seems to me more a mismatch between the dimension of the observable you are measuring and the number of measures added to the circuit. Here you try to measure a 3 particles observable over a state provided by two qubits only.
Adding a measure or reducing the dimension of ham make it work.

Let me know if I did't get your point :)

[chmwzc]

Thanks for the quick reply! But no, I believe the idea of the qubit_map argument is to allow the use of one set of suitable circuit measurements for obtaining the expectation values for multiple observables simultaneously.

Say for example I want to get the values of Z(0) and Z(2).
Instead of repeating the circuit twice to get the circuit measurements for each individual qubit (once after adding gates.M(0) and again after adding gates.M(2)), I can just run the circuit once after adding both gates.M(0) and gates.M(2), and use qubit_map to indicate that the frequencies of the binary strings measured (e.g. Counter({'00': 10})) are to be mapped to qubits 0 and 2, to get the expectation values of Z(0) and Z(2) respectively.

[BrunoLiegiBastonLiegi]

Thanks @chmwzc, as you already have an idea on how to fix this, would you mind opening a new PR and add me as a reviewer?