`mps_log_data` seems to store data from previous runs?

[liu-zixiong]

Informations

• Qiskit Aer version: 0.17.1
• Python version: 3.12.9
• Operating system: MacOS Sequoia 15.6

What is the current behavior?

I'm not sure why, but it seems that the mps_log_data metadata stores the information from previous results, even though each time the circuit, simulator, result objects are freshly initialised.

Example 1: I create two separate circuits, with two separate sets of IDs for each circ, simulator, and res object involved in the computation. I simulate them separately, yet the MPS log information of circuit 1 appears in the log information of circuit 2, even though we're dealing with two separate sets of objects. Somehow circuit 2 contains the log of both cz gate and cx gate even though it was not defined with the cx.

## QISKIT BUG CIRCUIT? ##
import qiskit_aer
import qiskit

# Define two circuits
circ1 = qiskit.QuantumCircuit(3)
circ1.h(0)
circ1.cx(0, 2)
circ1.measure_all()
print(f"Circuit ID: {id(circ1)}")

simulator1 = qiskit_aer.AerSimulator(method='matrix_product_state', matrix_product_state_max_bond_dimension=2**10, mps_log_data=True, mps_lapack = False)
print(f"Simulator ID: {id(simulator1)}")
res1 = simulator1.run(circ1, shots=1).result()
print(f"Result ID: {id(res1)}")

print(f"Counts: {res1.get_counts()}")
print(f"MPS log data: {res1.results[0].metadata['MPS_log_data']}")
print(f"Length of log string: {len(res1.results[0].metadata['MPS_log_data'])}")
print()


circ2 = qiskit.QuantumCircuit(3)
circ2.h(0)
circ2.cz(0, 2)
circ2.measure_all()
print(f"Circuit ID: {id(circ2)}")

simulator2 = qiskit_aer.AerSimulator(method='matrix_product_state', matrix_product_state_max_bond_dimension=2**10, mps_log_data=True, mps_lapack = False)
print(f"Simulator ID: {id(simulator2)}")
res2 = simulator2.run(circ2, shots=1).result()
print(f"Result ID: {id(res2)}")

print(f"Counts: {res2.get_counts()}")
print(f"MPS log data: {res2.results[0].metadata['MPS_log_data']}")
print(f"Length of log string: {len(res2.results[0].metadata['MPS_log_data'])}")
print()
===========================================================================
Circuit ID: 4369141344
Simulator ID: 4576459328
Result ID: 4578347200
Counts: {'000': 1}
MPS log data: {internal_swap on qubits 2,1, BD=[1 1],  I0:cx on qubits 0,2, BD=[2 1],  }
Length of log string: 74

Circuit ID: 4578348688
Simulator ID: 4369906720
Result ID: 4578622016
Counts: {'001': 1}
MPS log data: {internal_swap on qubits 2,1, BD=[1 1],  I0:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I1:cz on qubits 0,2, BD=[1 1],  }
Length of log string: 146

Example2: The object IDs of circ, simulator, and res are different each time. And the circuit is simple enough that we know it should only involve two operations — an internal swap, followed by the 2q cx gate. However, it seems that each time the results are simulated, the MPS log data gets longer and longer.

## QISKIT BUG CIRCUIT? ##
import qiskit_aer
import qiskit

for _ in range(5):
    # Define arbitrary circuit
    circ = qiskit.QuantumCircuit(3)
    circ.h(0)
    circ.cx(0, 2)
    circ.measure_all()
    print(f"Circuit ID: {id(circ)}")

    simulator = qiskit_aer.AerSimulator(method='matrix_product_state', matrix_product_state_max_bond_dimension=2**10, mps_log_data=True, mps_lapack = False)
    print(f"Simulator ID: {id(simulator)}")
    res = simulator.run(circ, shots=1).result()
    print(f"Result ID: {id(res)}")
    
    print(f"Counts: {res.get_counts()}")
    print(f"MPS log data: {res.results[0].metadata['MPS_log_data']}")
    print(f"Length of log string: {len(res.results[0].metadata['MPS_log_data'])}")
    print()
===========================================================================
Circuit ID: 4697661344
Simulator ID: 4394192672
Result ID: 4533600912
Counts: {'000': 1}
MPS log data: {internal_swap on qubits 2,1, BD=[1 1],  I0:cx on qubits 0,2, BD=[2 1],  }
Length of log string: 74

Circuit ID: 4394192720
Simulator ID: 4697848816
Result ID: 4698142048
Counts: {'000': 1}
MPS log data: {internal_swap on qubits 2,1, BD=[1 1],  I0:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I1:cx on qubits 0,2, BD=[2 1],  }
Length of log string: 146

Circuit ID: 4394192672
Simulator ID: 4473157200
Result ID: 4698142288
Counts: {'101': 1}
MPS log data: {internal_swap on qubits 2,1, BD=[1 1],  I0:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I1:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I2:cx on qubits 0,2, BD=[2 1],  }
Length of log string: 218

Circuit ID: 4562328544
Simulator ID: 4697848816
Result ID: 4698142528
Counts: {'101': 1}
MPS log data: {internal_swap on qubits 2,1, BD=[1 1],  I0:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I1:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I2:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I3:cx on qubits 0,2, BD=[2 1],  }
Length of log string: 290

Circuit ID: 4697666672
Simulator ID: 4698141856
Result ID: 4698142768
Counts: {'101': 1}
MPS log data: {internal_swap on qubits 2,1, BD=[1 1],  I0:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I1:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I2:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I3:cx on qubits 0,2, BD=[2 1],  internal_swap on qubits 2,1, BD=[1 1],  I4:cx on qubits 0,2, BD=[2 1],  }
Length of log string: 362

Steps to reproduce the problem

See above.

What is the expected behavior?

I am hoping to be able to extract the MPS log information of just the single run. Would the team be able to advise on what I should be doing instead?

Suggested solutions

It seems that the MPS log information is somehow stored within the class and persists across separate instances. Either that, or I'm using qiskit-aer wrong.