QCross: Quantum
Cross-Platform Testing

Supervisors:

Dr. Shaukat Ali and Dr. Christoph Laaber

Simula Research Laboratory

Applying differential and metamorphic testing techniques to test IBM Qiskit, Rigetti PyQuil, and Google Cirq

About me

Arfat Salman

Master in Informatics: Programming and System Architecture

Arrange

Assert

Act

Arrange

Assert

Act

Quantum Computer

Uses quantum physics concepts like superposition and entanglement to achieve computation

IBM Quantum scientist Dr. Maika Takita in lab

Proposed in the 1980s by Richard Feynman and Yuri Manin

 A new kind of computing machine 

*Not a replacement for classical computer

Why?

Due to their unique computational paradigm, quantum computers can solve certain problems exponentially faster and more efficiently than their classical counterparts.

Examples

  • Shor's Algorithm: Factors an integer N in polylogarithmic time.
  • Deutsch–Jozsa algorithm: determines whether a binary function is constant or balanced (returning equal numbers of 0s and 1s) one single operation

Why?

Deutsch–Jozsa algorithm

def balanced_fn(n_bit_arg):
    if some_condition(n_bit_arg):
        return 0
    return 1
    

balanced_fn('001') # => 1
balanced_fn('011') # => 0
balanced_fn('010') # => 0

balanced: Half of inputs return 1, other half 0

constant: 0 or 1 on all inputs

Classical Computer:

2^{n-1} + 1

Quantum Computer:

1

We shouldn’t be asking ‘where do quantum speedups come from?’ we should say ‘all computers are quantum, [...]’ and ask ’where do classical slowdowns come from?’

How?

https://quantum.country/

Qubit

\lvert 0 \rangle
\lvert 1 \rangle

Quantum Bit

Qubit

\lvert 0 \rangle
\lvert 1 \rangle

Quantum Bit

dead

alive

Qubit

\lvert 0 \rangle
\lvert 1 \rangle
0.6\lvert 0 \rangle + 0.8\lvert 1 \rangle

Quantum Bit

dead

alive

Qubit

\lvert 0 \rangle
\lvert 1 \rangle
0.6\lvert 0 \rangle + 0.8\lvert 1 \rangle

Quantum Bit

dead

alive

Ket

Qubit

\lvert 0 \rangle
\lvert 1 \rangle
0.6\lvert 0 \rangle + 0.8\lvert 1 \rangle

Quantum Bit

Superposition

dead

alive

Qubit

\lvert 0 \rangle
\lvert 1 \rangle
0.6\lvert 0 \rangle + 0.8\lvert 1 \rangle

Quantum Bit

Superposition

\lvert \psi \rangle = 0.6 \begin{bmatrix} 1 \\ 0 \end{bmatrix} + 0.8 \begin{bmatrix} 0 \\ 1 \end{bmatrix} = \begin{bmatrix} 0.6 \\ 0.8 \end{bmatrix}

dead

alive

Qubit

\lvert 0 \rangle
\lvert 1 \rangle
0.6\lvert 0 \rangle + 0.8\lvert 1 \rangle

Quantum Bit

Superposition

\lvert 0.6 \rvert ^ 2 + \lvert 0.8 \rvert ^ 2 = 1

dead

alive

Qubit

\lvert 0 \rangle
\lvert 1 \rangle
0.6\lvert 0 \rangle + 0.8\lvert 1 \rangle

Quantum Bit

Superposition

\lvert 0.6 \rvert ^ 2 + \lvert 0.8 \rvert ^ 2 = 1

dead

alive

Qubit

\lvert 0 \rangle
\lvert 1 \rangle
0.6\lvert 0 \rangle + 0.8\lvert 1 \rangle

Quantum Bit

Superposition

\lvert 0.6 \rvert ^ 2 + \lvert 0.8 \rvert ^ 2 = 1

probability of observing a 0

probability of observing a 1

Schrödinger's cat

dead

alive

Quantum Gate

A quantum gate is a unitary matrix that acts on a quantum state and changes it.

NOT( \alpha \lvert 0 \rangle + \beta \lvert 1 \rangle ) = \alpha \lvert 1 \rangle + \beta \lvert 0 \rangle

Qubit

Flipped Probailities

NOT = \begin{bmatrix} 0 & 1 \\ 1 & 0 \end{bmatrix}

Other quantum gates

Hadamard Gate (H):

Single qubit gate that puts a quantum state in superposition

Controlled-NOT gate:

Two-qubit quantum gate 

\begin{array}{c} CNOT \lvert00\rangle = \lvert00\rangle \\ CNOT\lvert01\rangle = \lvert01\rangle \\ CNOT\lvert10\rangle = \lvert11\rangle \\ CNOT\lvert11\rangle = \lvert10\rangle \\ \end{array}

Other quantum gates

Hadamard Gate (H):

Single qubit gate that puts a quantum state in superposition

Controlled-NOT gate:

Two-qubit quantum gate 

\begin{array}{c} CNOT \lvert00\rangle = \lvert00\rangle \\ CNOT\lvert01\rangle = \lvert01\rangle \\ CNOT\lvert10\rangle = \lvert11\rangle \\ CNOT\lvert11\rangle = \lvert10\rangle \\ \end{array}

Other quantum gates

Hadamard Gate (H):

Single qubit gate that puts a quantum state in superposition

Controlled-NOT gate:

Two-qubit quantum gate 

\begin{array}{c} CNOT \lvert00\rangle = \lvert00\rangle \\ CNOT\lvert01\rangle = \lvert01\rangle \\ CNOT\lvert10\rangle = \lvert11\rangle \\ CNOT\lvert11\rangle = \lvert10\rangle \\ \end{array}

Quantum Computation

Quantum computation is a change in the quantum state (i.e., the state of a qubit, the quantum bit).

  • A pure quantum computation is reversible. (applying the reverse of the gate will inverse the operation)
     
  • A quantum state cannot be copied due to No-cloning theorem (most assignment operations become impossible)
     
  • Intermediate quantum state cannot be read as it leads to collapse of quantum state (no debugging or print statements)

Quantum Circuit

QSP

Quantum Software Platforms

https://thequantuminsider.com/2022/09/05/quantum-computing-companies-ultimate-list-for2022/

Standalone or embedded quantum programming language or API

Quantum simulator that emulates instructions on a classical device

Optimizing compiler that translates high-level language into quantum gate instructions

Software controller that sends analog signals to quantum hardware

pyQuil

Cirq

Bugs in QSPs

Microsoft Q#

Google Cirq

Rigetti PyQuil

IBM Qiskit

% of bug-labelled GitHub issues

Bugs in QSPs

Microsoft Q#

Google Cirq

Rigetti PyQuil

IBM Qiskit

% of bug-labelled GitHub issues

30

15

24

40

😵

Challenges in Testing QSPs

C1: The need for a significant quantity of quantum test programs.

C2: Cross-platform testing is hard due to varying support and APIs, needing manual porting.

C3: Presence of stochasticity.

Lack of generalised automated testing techniques within quantum settings

Testing Overview

Quantum Software Testing (and testing at large) is facing two fundamental problems:

Oracle Problem

The Reliable Test-Set problem

Testing Overview

Quantum Software Testing (and testing at large) is facing two fundamental problems:

Oracle Problem

The Reliable Test-Set problem

situations where it is extremely difficult, or impossible, to verify the test result of a given test case

the challenge of creating a set of tests that can adequately and accurately assess the functionality and performance of a software system.

Differential Testing

This technique involves supplying identical input to comparable applications or distinct implementations of the same application and observing discrepancies in their performance and behaviour.

Solves the Oracle Problem

Python

Source

CPython

PyPy

Same Result

Differential Testing

This technique involves supplying identical input to comparable applications or distinct implementations of the same application and observing discrepancies in their performance and behaviour.

Solves the Oracle Problem

Quantum

Algorithm

Qiskit

Cirq

Same Result

after necessary translations

Metamorphic Testing

property-based software testing technique 

Metamorphic Testing

property-based software testing technique 

Algorithm

passing

test case

Metamorphic Testing

property-based software testing technique 

Algorithm

new test case

using algorithm property

passing

test case

Metamorphic Testing

property-based software testing technique 

Algorithm

new test case

using algorithm property

Same Result

passing

test case

Metamorphic Testing

property-based software testing technique 

min

min(4,5) 

min(5,4)

swapping arguments order does not change answer

Same Result

Algorithm

passing

test case

new test case

using algorithm property

Same Result

Metamorphic Testing

property-based software testing technique 

min

min(4,5) 

min(5,4)

swapping arguments order does not change answer

Same Result

Algorithm

passing

test case

new test case

using algorithm property

Same Result

Source

Follow-up

Prior Work

QDiff

MorphQ

  • 6 hand-written quantum programs
  • Evaluated on 3 platforms( Qiskit, Cirq, PyQuil) and quantum hardware
  • Uses mutation testing to create increased program divergence
  • Generates random Qiskit quantum programs
  • Establishes 10 metamorphic relations
  • Uses metamorphic testing to test
    • Only Qiskit

Contribution of the thesis

  • Used MorphQ to generate random Qiskit Programs
  • Provide basis for translating metamorphic relations in Cirq and PyQuil.
  • Built a MR-aware quantum program translator that translates MorphQ Qiskit programs to  Cirq and PyQuil programs. 
  • As part of translation, created a cross-platform quantum gate library (bloqs)

Research Questions

  • RQ1: How many syntactically different but correct programs can be translated by QCross’s converter?
     
  • RQ2: What has QCross found via cross-platform testing of the widely-used QSSes? i.e., how many warnings and errors do QCross produce?
     
  • RQ3: How does QCross compare to prior work on testing quantum computing platforms?
     
  • RQ4: How useful is Bloqs?

QCross's Program Translation Process

Each quantum program can be logically divided into six parts:

  1. import statements
  2. Declaration of circuits / qubits / classical bits
  3. Application of gates and addition of terminal measurements
  4. Preparation of the circuit to be executed on simulators / optimiziations
  5. Execution of the program
  6. Collection of the results
qr = QuantumRegister(7, name='qr') 
cr = ClassicalRegister(7, name='cr') 
qc = QuantumCircuit(qr, cr, name='qc')
qc = cirq.Circuit() 

qr = [cirq.NamedQubit('q' + str(i)) for i in range(7)]
qc = Program() 

cr = qc.declare("ro", "BIT", 7)

Declarations

Cirq

Qiskit

PyQuil

qr = QuantumRegister(7, name='qr') 
cr = ClassicalRegister(7, name='cr') 
qc = QuantumCircuit(qr, cr, name='qc')
qc = cirq.Circuit() 

qr = [cirq.NamedQubit('q' + str(i)) for i in range(7)]
qc = Program() 

cr = qc.declare("ro", "BIT", 7)

Declarations (Circuit)

Cirq

Qiskit

PyQuil

qr = QuantumRegister(7, name='qr') 
cr = ClassicalRegister(7, name='cr') 
qc = QuantumCircuit(qr, cr, name='qc')
qc = cirq.Circuit() 

qr = [cirq.NamedQubit('q' + str(i)) for i in range(7)]
qc = Program() 

cr = qc.declare("ro", "BIT", 7)

Declarations (Quantum Register)

Cirq

Qiskit

PyQuil

qr = QuantumRegister(7, name='qr') 
cr = ClassicalRegister(7, name='cr') 
qc = QuantumCircuit(qr, cr, name='qc')
qc = cirq.Circuit() 

qr = [cirq.NamedQubit('q' + str(i)) for i in range(7)]
qc = Program() 

cr = qc.declare("ro", "BIT", 7)

Declarations (Classical Register)

Cirq

Qiskit

PyQuil

from qiskit import QuantumCircuit, ClassicalRegister, QuantumRegister

qr = QuantumRegister(7, name='qr') 
cr = ClassicalRegister(7, name='cr') 
qc = QuantumCircuit(qr, cr, name='qc')
import cirq

qc = cirq.Circuit() 

qr = [cirq.NamedQubit('q' + str(i)) for i in range(7)]
from pyquil import Program

qc = Program() 

cr = qc.declare("ro", "BIT", 7)

Import statements

Cirq

Qiskit

PyQuil

Application of gates

from qiskit.circuit.library.standard_gates import *

qc.append(ZGate(), qargs=[qr[2]], cargs=[]) 
qc.append(CHGate(), qargs=[qr[2], qr[0]], cargs=[]) 
qc.append(RYYGate(5.398622178940033), qargs=[qr[0], qr[2]], cargs=[])
import cirq 

qc.append(cirq.Z( qr[2] ))
from pyquil.gates import *

qc.inst(Z(2))

Cirq

Qiskit

PyQuil

Application of gates

from qiskit.circuit.library.standard_gates import *

qc.append(ZGate(), qargs=[qr[2]], cargs=[]) 
qc.append(CHGate(), qargs=[qr[2], qr[0]], cargs=[]) 
qc.append(RYYGate(5.398622178940033), qargs=[qr[0], qr[2]], cargs=[])
import cirq 

qc.append(cirq.Z( qr[2] ))
from pyquil.gates import *

qc.inst(Z(2))

Cirq

Qiskit

PyQuil

Application of gates

from qiskit.circuit.library.standard_gates import *

qc.append(ZGate(), qargs=[qr[2]], cargs=[]) 
qc.append(CHGate(), qargs=[qr[2], qr[0]], cargs=[]) 
qc.append(RYYGate(5.398622178940033), qargs=[qr[0], qr[2]], cargs=[])
import cirq 

qc.append(cirq.Z( qr[2] ))
qc.append(cirq.H.controlled()( qr[2], qr[0] ))
from pyquil.gates import *

qc.inst(Z(2))
# control qubit 2 and target qubit 0 
qc.inst(H(0).controlled(2))

Cirq

Qiskit

PyQuil

Application of gates

from qiskit.circuit.library.standard_gates import *

qc.append(ZGate(), qargs=[qr[2]], cargs=[]) 
qc.append(CHGate(), qargs=[qr[2], qr[0]], cargs=[]) 
qc.append(RYYGate(5.398622178940033), qargs=[qr[0], qr[2]], cargs=[])
import cirq 

qc.append(cirq.Z( qr[2] ))
qc.append(cirq.H.controlled()( qr[2], qr[0] ))
# ??
from pyquil.gates import *

qc.inst(Z(2))
# control qubit 2 and target qubit 0 
qc.inst(H(0).controlled(2))
# ??

Cirq

Qiskit

PyQuil

Application of gates (using bloqs)

from qiskit.circuit.library.standard_gates import *

qc.append(ZGate(), qargs=[qr[2]], cargs=[]) 
qc.append(CHGate(), qargs=[qr[2], qr[0]], cargs=[]) 
qc.append(RYYGate(5.398622178940033), qargs=[qr[0], qr[2]], cargs=[])
import cirq 
from bloqs.ext.cirq import Gates

qc.append(cirq.Z( qr[2] ))
qc.append(cirq.H.controlled()( qr[2], qr[0] ))
qc.append(Gates.RYYGate(5.398622178940033)( qr[0], qr[2] ))
from pyquil.gates import *
from bloqs.ext.PyQuil import Gates

qc.inst(Z(2))
# control qubit 2 and target qubit 0 
qc.inst(H(0).controlled(2))
qc.inst(Gates.RYYGate(5.398622178940033)( 0, 2 ))

Cirq

Qiskit

PyQuil

Application of gates (using bloqs)

from qiskit.circuit.library.standard_gates import *

qc.append(ZGate(), qargs=[qr[2]], cargs=[]) 
qc.append(CHGate(), qargs=[qr[2], qr[0]], cargs=[]) 
qc.append(RYYGate(5.398622178940033), qargs=[qr[0], qr[2]], cargs=[])
import cirq 
from bloqs.ext.cirq import Gates

qc.append(cirq.Z( qr[2] ))
qc.append(cirq.H.controlled()( qr[2], qr[0] ))
qc.append(Gates.RYYGate(5.398622178940033)( qr[0], qr[2] ))
from pyquil.gates import *
from bloqs.ext.PyQuil import Gates

qc.inst(Z(2))
# control qubit 2 and target qubit 0 
qc.inst(H(0).controlled(2))
qc.inst(Gates.RYYGate(5.398622178940033)( 0, 2 ))

Cirq

Qiskit

PyQuil

Application of gates (using bloqs)

from qiskit.circuit.library.standard_gates import *

qc.append(ZGate(), qargs=[qr[2]], cargs=[]) 
qc.append(CHGate(), qargs=[qr[2], qr[0]], cargs=[]) 
qc.append(RYYGate(5.398622178940033), qargs=[qr[0], qr[2]], cargs=[])
import cirq 
from bloqs.ext.cirq import Gates

qc.append(cirq.Z( qr[2] ))
qc.append(cirq.H.controlled()( qr[2], qr[0] ))
qc.append(Gates.RYYGate(5.398622178940033)( qr[0], qr[2] ))
from pyquil.gates import *
from bloqs.ext.PyQuil import Gates, get_custom_get_definitions
qc = Program()
ryy_defn = get_custom_get_definitions("RYYGate") 
qc += ryy_defn

qc.inst(Z(2))
qc.inst(H(0).controlled(2))
qc.inst(Gates.RYYGate(5.398622178940033)( 0, 2 ))

Cirq

Qiskit

PyQuil

Application of gates (measurement gates)

qc.append(ZGate(), qargs=[qr[2]], cargs=[]) 
qc.append(CHGate(), qargs=[qr[2], qr[0]], cargs=[]) 
qc.append(RYYGate(5.398622178940033), qargs=[qr[0], qr[2]], cargs=[])

qc.measure(qr, cr)
qc.append(cirq.Z( qr[2] ))
qc.append(cirq.H.controlled()( qr[2], qr[0] ))
qc.append(Gates.RYYGate(5.398622178940033)( qr[0], qr[2] ))

qc.append(cirq.measure(qr[0], key='q0'))
qc.append(cirq.measure(qr[1], key='q1'))
qc.inst(Z(2))
qc.inst(H(0).controlled(2))
qc.inst(Gates.RYYGate(5.398622178940033)( 0, 2 ))

qc += MEASURE(0, qr[0]) 
qc += MEASURE(1, qr[1])

Cirq

Qiskit

PyQuil

Preparation of the circuit

from qiskit import transpile

qc = transpile(qc, optimization_level=2)
# other options are coupling_map, basis_gates etc.
import cirq 

qc = cirq.drop_empty_moments(qc)
qc = cirq.merge_k_qubit_unitaries(qc)
qc = cirq.eject_z(qc)
qc = cirq.drop_negligible_operations(qc)
executable = qvm.compile(qc, protoquil=True, optimize=True)

# some optimization available via quilc and qvm programs

Cirq

Qiskit

PyQuil

Execution of the circuit

from qiskit import Aer, execute

qasm_sim_backend = Aer.get_backend('qasm_simulator') 
result = execute(qc, backend=qasm_sim_backend, shots=2048)
import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)
from pyquil import get_qc

qc.wrap_in_numshots_loop(2048)

qc = get_qc("11q-qvm") 
result = qc.run(executable)

Cirq

Qiskit

PyQuil

Execution of the circuit

from qiskit import Aer, execute

qasm_sim_backend = Aer.get_backend('qasm_simulator') 
result = execute(qc, backend=qasm_sim_backend, shots=2048)
import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)
from pyquil import get_qc

qc.wrap_in_numshots_loop(2048)

qc = get_qc("11q-qvm") 
result = qc.run(executable)

Cirq

Qiskit

PyQuil

[
    AerSimulator("aer_simulator"),
    AerSimulator("aer_simulator_statevector"),
    AerSimulator("aer_simulator_density_matrix"),
    AerSimulator("aer_simulator_stabilizer"),
    AerSimulator("aer_simulator_matrix_product_state"),
    AerSimulator("aer_simulator_extended_stabilizer"),
    AerSimulator("aer_simulator_unitary"),
    AerSimulator("aer_simulator_superop"),
    QasmSimulator("qasm_simulator"),
    StatevectorSimulator("statevector_simulator"),
    UnitarySimulator("unitary_simulator"),
    PulseSimulator("pulse_simulator"),
]

Execution of the circuit

from qiskit import Aer, execute

qasm_sim_backend = Aer.get_backend('qasm_simulator') 
result = execute(qc, backend=qasm_sim_backend, shots=2048)
import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)
from pyquil import get_qc

qc.wrap_in_numshots_loop(2048)

qc = get_qc("11q-qvm") 
result = qc.run(executable)

Cirq

Qiskit

PyQuil

import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)

Execution of the circuit

from qiskit import Aer, execute

qasm_sim_backend = Aer.get_backend('qasm_simulator') 
result = execute(qc, backend=qasm_sim_backend, shots=2048)
import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)
from pyquil import get_qc

qc.wrap_in_numshots_loop(2048)

qc = get_qc("11q-qvm") 
result = qc.run(executable)

Cirq

Qiskit

PyQuil

import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)

Execution of the circuit

from qiskit import Aer

qasm_sim_backend = Aer.get_backend('qasm_simulator') 
result = execute(qc, backend=qasm_sim_backend, shots=2048)
import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)
from pyquil import get_qc

qc.wrap_in_numshots_loop(2048)

qc = get_qc("11q-qvm") 
result = qc.run(executable)

Cirq

Qiskit

PyQuil

import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)

Collection of results

result = execute(qc, backend=qasm_sim_backend, shots=2048)

counts = result.get_counts(qc)

# {'010': 66, # '000': 198, # '110': 63 ... }
import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)
result = qc.run(executable)
counts = result.readout_data.get('ro')

# [ [0, 1, 1], [1, 0, 1], [1, 0, 0], [1, 1, 0], ... ]

Cirq

Qiskit

PyQuil

result = simulator.run(circuit, repetitions=2048)
print(result)
# q0=11011110100010101111 
# q1=01100101011001110001

Collection of results

result = execute(qc, backend=qasm_sim_backend, shots=2048)

counts = result.get_counts(qc)

# {'010': 66, # '000': 198, # '110': 63 ... }
import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)
result = qc.run(executable)
counts = result.readout_data.get('ro')

# [ [0, 1, 1], [1, 0, 1], [1, 0, 0], [1, 1, 0], ... ]

Cirq

Qiskit

PyQuil

result = simulator.run(circuit, repetitions=2048)
print(result)
# q0=11011110100010101111 
# q1=01100101011001110001

Little- and Big-endian

qr = QuantumRegister(3, name='qr') 
cr = ClassicalRegister(3, name='cr') 
qc = QuantumCircuit(qr, cr, name='qc')

qc.append(HGate(), qargs=[qr[2]], cargs=[])

qc.measure(qr, cr)

# ... backend selection ...

counts = execute(qc, backend=b, shots=1024) 
	.result()
	.get_counts(qc)
    
# {'100': 505, '000': 519}

Collection of results (using bloqs)

result = execute(qc, backend=qasm_sim_backend, shots=2048)

counts = result.get_counts(qc)

# {'010': 66, # '000': 198, # '110': 63 ... }
import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)
from bloqs.ext.PyQuil.utils import get_qiskit_like_output

result = qc.run(executable)
counts = result.readout_data.get('ro')
output = get_qiskit_like_output(data)

# { ... }

Cirq

Qiskit

PyQuil

from bloqs.ext.cirq.utils import get_qiskit_like_output

result = simulator.run(circuit, repetitions=2048)

counts = get_qiskit_like_output(result, keys=['cr0', 'cr1', 'cr2'])

# {'010': 65, # '000': 199, # '110': 66 ... }

Collection of results

result = execute(qc, backend=qasm_sim_backend, shots=2048)

counts = result.get_counts(qc)

# {'010': 66, # '000': 198, # '110': 63 ... }
import cirq 

simulator = cirq.Simulator() 
result = simulator.run(circuit, repetitions=2048)
from bloqs.ext.PyQuil.utils import get_qiskit_like_output

result = qc.run(executable)
counts = result.readout_data.get('ro')
output = get_qiskit_like_output(data)

# { ... }

Cirq

Qiskit

PyQuil

from bloqs.ext.cirq.utils import get_qiskit_like_output

result = simulator.run(circuit, repetitions=2048)

counts = get_qiskit_like_output(result, keys=['cr0', 'cr1', 'cr2'])

# {'010': 65, # '000': 199, # '110': 66 ... }

MorphQ Paper

Source Program

as passing test case

Metamorphic Testing

MorphQ Paper

Source Program

as passing test case

based on the previous MRs

Metamorphic Testing

Follow-up Porgram

MorphQ Paper

Source Program

as passing test case

Follow-up Porgram

based on the previous MRs

Execute

both and

compare

Metamorphic Testing

Metamorphic Relations (MR)

MR Qiskit Cirq PyQuil
Change Qubit Order Y Y Y
Inject null-effect operation Y Y Y
Add quantum register Y N -
Inject parameters Y N Y
Partitioned execution Y Y Y
Intermediary language roundtrip Y Y Y
Roundtrip conversion via QASM3 Y - -
Serialization roundtrip Y Y -
Change of coupling map Y Y N
Change of gate set Y N N
Change of optimization level Y Y Y
Change of backend Y Y Y

Y = QSP supports given MR

N = No valid API

 -  = Not a valid MR in the given platform

Change Qubit Order

requires post-processing of the result

maps the qubit indices of source program to new positions and then creates a follow up program by adapting the sequence of gates to the newly mapped qubit indices.

Change Qubit Order

requires post-processing of the result

Change Qubit Order

requires post-processing of the result

Change Qubit Order

requires post-processing of the result

Change Qubit Order

requires post-processing of the result

Change Qubit Order

requires post-processing of the result

Change Qubit Order

requires post-processing of the result

Inject null-effect operation

Inserting into the main circuit a sub-circuit that performs a sequence of gate operations followed by its inverse

Inject null-effect operation

Inserting into the main circuit a sub-circuit that performs a sequence of gate operations followed by its inverse

Inject null-effect operation

qc.rx(math.pi, 2) 
qc.x(0)

subcircuit = QuantumCircuit(qr, cr, name='subcircuit') 
subcircuit.append(HGate(), qargs=[qr[0]], cargs=[])

qc.append(subcircuit, qargs=qr, cargs=cr) 
qc.append(subcircuit.inverse(), qargs=qr, cargs=cr)

qc.h(1) qc.cx(0,1)

Inserting into the main circuit a sub-circuit that performs a sequence of gate operations followed by its inverse

Cirq

Qiskit

PyQuil

qc.append(subcircuit, qargs=qr, cargs=cr) 
qc.append(subcircuit.inverse(), qargs=qr, cargs=cr)
subcircuit = Program() 

# add gates to sub-circuit

qc.inst(subcircuit) 
qc.inst(subcircuit.dagger())
qc = cirq.Circuit() # add gates to qc

subcircuit = cirq.Circuit() # add gates to sub-circuit

qc.append(subcircuit) 
qc.append(cirq.inverse(subcircuit))

Inject null-effect operation

Add quantum register

Enlarging the set of available qubits by adding a new and unused quantum register should not affect the computation on the existing qubits.

# Qiskit
unused_register = QuantumRegister(5, name='extra_registers') 
qc.add_register(unused_register)

A comparable API does not exist in Cirq and PyQuil.

Inject parameters

Parameterized quantum circuits are quantum circuits that contain one or more parameters (for example, the angle provided to RXGate) that can be adjusted without changing the overall structure of the circuit.

There are some advantages to using parameters:

  • Efficiency: When using parameters, the circuit structure remains fixed, and only the parameter values change.
  • Circuit Compilation: Parametrized circuits can be transpiled and optimized before the parameters are bound to specific values.

Qiskit

from qiskit.circuit import Parameter

theta = Parameter('theta') 
gamma = Parameter('gamma') 

qc.append(RZGate(theta), qargs=[qr[0]], cargs=[]) 
qc.append(U2Gate(theta, 2.12), qargs=[qr[2]], cargs=[]) 
qc.append(CRZGate(gamma), qargs=[qr[1], qr[0]], cargs=[]) 

qc.measure(qr, cr)

# before execution 
qc = qc.bind_parameters({
  theta: 4.2641612072511235,
  gamma: 2.5163050709890156,
})

Inject parameters

Qiskit

from qiskit.circuit import Parameter

theta = Parameter('theta') 
gamma = Parameter('gamma') 

qc.append(RZGate(theta), qargs=[qr[0]], cargs=[]) 
qc.append(U2Gate(theta, 2.12), qargs=[qr[2]], cargs=[]) 
qc.append(CRZGate(gamma), qargs=[qr[1], qr[0]], cargs=[]) 

qc.measure(qr, cr)

# before execution 
qc = qc.bind_parameters({
  theta: 4.2641612072511235,
  gamma: 2.5163050709890156,
})

Inject parameters

Qiskit

from qiskit.circuit import Parameter

theta = Parameter('theta') 
gamma = Parameter('gamma') 

qc.append(RZGate(theta), qargs=[qr[0]], cargs=[]) 
qc.append(U2Gate(theta, 2.12), qargs=[qr[2]], cargs=[]) 
qc.append(CRZGate(gamma), qargs=[qr[1], qr[0]], cargs=[]) 

qc.measure(qr, cr)

# before execution 
qc = qc.bind_parameters({
  theta: 4.2641612072511235,
  gamma: 2.5163050709890156,
})

Inject parameters

Qiskit

from qiskit.circuit import Parameter

theta = Parameter('theta') 
gamma = Parameter('gamma') 

qc.append(RZGate(theta), qargs=[qr[0]], cargs=[]) 
qc.append(U2Gate(theta, 2.12), qargs=[qr[2]], cargs=[]) 
qc.append(CRZGate(gamma), qargs=[qr[1], qr[0]], cargs=[]) 

qc.measure(qr, cr)

# before execution 
qc = qc.bind_parameters({
  theta: 4.2641612072511235,
  gamma: 2.5163050709890156,
})

Inject parameters

Cirq

Qiskit

from qiskit.circuit import Parameter

theta = Parameter('theta') 
gamma = Parameter('gamma') 

qc.append(RZGate(theta), qargs=[qr[0]], cargs=[]) 
qc.append(U2Gate(theta, 2.12), qargs=[qr[2]], cargs=[]) 
qc.append(CRZGate(gamma), qargs=[qr[1], qr[0]], cargs=[]) 

qc.measure(qr, cr)

# before execution 
qc = qc.bind_parameters({
  theta: 4.2641612072511235,
  gamma: 2.5163050709890156,
})
import cirq from sympy import Symbol

theta = Symbol('theta') 
gamma = Symbol('gamma')

qc.append(cirq.rz(theta)(qr[0])) 
qc.append(Gates.CRZGate(gamma)( qr[1], qr[2] ))

qc = cirq.resolve_parameters(qc, {
  "theta": 4.2641612072511235,
  "gamma": 2.5163050709890156,
  "_lambda": 2.586208953975239, 
})

Inject parameters

Cirq

Qiskit

PyQuil

from qiskit.circuit import Parameter

theta = Parameter('theta') 
gamma = Parameter('gamma') 

qc.append(RZGate(theta), qargs=[qr[0]], cargs=[]) 
qc.append(U2Gate(theta, 2.12), qargs=[qr[2]], cargs=[]) 
qc.append(CRZGate(gamma), qargs=[qr[1], qr[0]], cargs=[]) 

qc.measure(qr, cr)

# before execution 
qc = qc.bind_parameters({
  theta: 4.2641612072511235,
  gamma: 2.5163050709890156,
})
theta = qc.declare('theta', 'REAL') 
gamma = qc.declare('gamma', 'REAL')

qc.inst(RZ(theta, 0)) 
qc.inst(Gates.CRZGate(gamma, 1, 2 )) 

params = {
  "theta": 4.2641612072511235, 
  "gamma": 2.5163050709890156,
}
for param, value in params.items():
  qc.write_memory(region_name=param, value=value)
import cirq from sympy import Symbol

theta = Symbol('theta') 
gamma = Symbol('gamma')

qc.append(cirq.rz(theta)(qr[0])) 
qc.append(Gates.CRZGate(gamma)( qr[1], qr[2] ))

qc = cirq.resolve_parameters(qc, {
  "theta": 4.2641612072511235,
  "gamma": 2.5163050709890156,
  "_lambda": 2.586208953975239, 
})

Inject parameters

Partitioned execution

Source programs might have two subsets of qubits that never interact with each other.

Partitioned execution

Source programs might have two subsets of qubits that never interact with each other.

  • Post-Processing of results required
  • No extra API required as QCross calls the circuit generation function twice.

Intermediary language roundtrip

  • Qiskit has OpenQASM2.

  • Cirq has parital support for OpenQASM2

  • PyQuil has QUIL

Cirq

Qiskit

PyQuil

qc = qc.from_qasm_str(
  qc.qasm()
)
from PyQuil.parser import parse

quil_str = qc.out() 
qc = parse(quil_str)
import cirq 
from cirq.contrib.qasm_import import circuit_from_qasm


qasm_out = cirq.qasm(qc)
qc = circuit_from_qasm(qasm_out)

Round-trip conversion via QASM3

QASM 3 format is anticipated to supersede QASM 2 in the near future

Qiskit

from qiskit.qasm3 import loads, dumps 

qasm3_out = dumps(qc) 
qc = loads(qasm3_out)

Serialization roundtrip

Serialization is defined as the process of converting the state of an object into a form that can be persisted or transported.

Cirq

Qiskit

from qiskit.circuit import QuantumCircuit from qiskit import qpy

qc = QuantumCircuit(2, name='Bell') 
qc.h(0) 
qc.cx(0, 1) 
qc.measure_all()

with open('bell.qpy', 'wb') as fd:
  qpy.dump(qc, fd)

with open('bell.qpy', 'rb') as fd:
  new_qc = qpy.load(fd)[0]
qc = cirq.read_json(
  cirq.to_json(qc)
)

Change of coupling map

  • A coupling map is defined as a representation of the connectivity between the qubits in a quantum computing device.
  • It describes which qubits are physically connected and can directly interact with each other through two-qubit gates, such as the controlled NOT (CNOT) gate.

q0

q1

q2

Change of coupling map

  • A coupling map is defined as a representation of the connectivity between the qubits in a quantum computing device.
  • It describes which qubits are physically connected and can directly interact with each other through two-qubit gates, such as the controlled NOT (CNOT) gate.

q0

q1

q2

Change of coupling map

  • A coupling map is defined as a representation of the connectivity between the qubits in a quantum computing device.
  • It describes which qubits are physically connected and can directly interact with each other through two-qubit gates, such as the controlled NOT (CNOT) gate.

q0

q1

q2

Change of coupling map

Cirq

Qiskit

# Define a custom coupling map 
custom_coupling_map = [[0, 1], [1, 2], [2, 3]]

# Transpile the circuit using the custom coupling map 
transpiled_qc = transpile(qc, coupling_map=custom_coupling_map)
import networkx as nx

def edge_list_to_cirq_graph(edge_list, nodes=None):
    if nodes is None:
        num_qubits = len(
            set(item for sublist in edge_list for item in sublist)
        )
        nodes = [
            cirq.NamedQubit("q" + str(i)) for i in range(num_qubits)
        ]

    graph = nx.Graph()
    for n in nodes:
        graph.add_node(n)
    for e in edge_list:
        graph.add_edge(nodes[e[0]], nodes[e[1]])

    return graph

graph = edge_list_to_cirq_graph(qc)
router = cirq.RouteCQC(graph)
routed_qc = router(qc)

Change of gate set

This transformation exercise this translation step by replacing the circuit gates in the program with a universal gate set, such as ["rx", "ry", "rz", "p", "cx"]

Qiskit

from qiskit import QuantumCircuit, transpile 

qc = QuantumCircuit(2) 

qc.h(0)
qc.cx(0, 1)

custom_basis_gates = ['rx', 'ry', 'rz', 'p', 'cx']

transpiled_qc = transpile(qc, basis_gates=custom_basis_gates)

Change of optimization level

Similar to modifying the optimization level of a traditional compiler, we can change the optimization level of the quantum transpilation process.

Qiskit

from qiskit import QuantumCircuit, transpile 

qc = QuantumCircuit(2) 

qc.h(0) 
qc.cx(0, 1)

transpiled_qc = transpile(qc, optimization_level=1)

Cirq

qc = cirq.eject_phased_paulis(qc)
  • cirq.align_left / cirq.align_right
  • cirq.drop_empty_moments / cirq.drop_negligible_operations
  • cirq.eject_phased_paulis
  • cirq.eject_z
  • cirq.expand_composite
  • cirq.merge_k_qubit_unitaries
  • cirq.stratified_circuit

Change of backend

Different simulators typically have completely different implementations, such as one based on state vectors or density matrices.

  • Qiskit
    • many
  • Cirq
    • Simulator
    • DensityMatrixSimulator
  • PyQuil
    • 9q-square-qvm
    • Xq-qvm

Differential Testing

  • Unitary Matrix Comparison
  • Cirq Qiskit Rountrip
  • Output comparison
    • All Source Output
    • All Follow-up Outputs

Unitary Matrix Comparison

  • A quantum gate can be represented by a unitary matrix.
  • The application of multiple quantum gates preserves unitarity.
  • When two quantum circuits possess the same gates and an equal number of qubits, their respective unitary matrices will be equivalent, modulo a global phase and a same starting qubit states.

Qiskit

Cirq

UNITARY = cirq.unitary(qc)
backend = Aer.get_backend('unitary_simulator') 
result = execute(qc.reverse_bits(), backend=backend).result() 
UNITARY = result.get_unitary(qc).data
from pyquil.simulation.tools import program_unitary
# 6 is the number of qubits
UNITARY = program_unitary(circuit, 6)

PyQuil

numpy.allcose or cirq.equal_up_to_global_phase

Cirq Qiskit Rountrip

  1. Cirq’s follow-up program is converted to QASM using the cirq.qasm function.
  2. It is ingested by QuantumCircuit.from_qasm_str to create a Qiskit circuit.
  3. The Qiskit circuit is then exported to QASM using the qasm method.
  4. Finally, Qiskit’s QASM is fed into Cirq to generate a Cirq circuit using the circuit_from_qasm.

Cirq has limited but growing support for QASM 2 interoperability. Therefore, this rountrip tests Cirq QASM’s generation capabilities, and Qiksit’s foreign QASM ingestion capabilities.

import cirq
from cirq.testing import *

expanded_circuit = cirq.expand_composite(circuit)
qasm_from_qiskit = qiskit.QuantumCircuit.from_qasm_str(
  cirq.qasm(expanded_circuit)
).qasm()
cirq_circuit_from_qiskit_qasm = circuit_from_qasm(
  qasm_from_qiskit
)

# since, importing in cirq uses different qubit names, 
# we need to change the qubit names
circuit_transformed = cirq_circuit_from_qiskit_qasm.transform_qubits(
  lambda q: cirq.NamedQubit(q.name.replace("q_", "q"))
)

assert_circuits_with_terminal_measurements_are_equivalent(
  circuit_transformed, 
  circuit
)

Outputs Equivalence Check

\binom{3}{2} = 3

Given that each set contains three programs (3 sources and 3 followfups), we carry out                  pairwise assessments per set, resulting in a total of six comparisons.

Comparing Execution Behavior

  • We use the Kolmogorov-Smirnov test  to assess the statistical significance of the difference between the two distributions, as done in previous work, using a significance level of α = 5%. We call any pair of programs with a p-value below α a statistically significant distribution difference.
  • Source or follow-up crash is termed as crash difference.

Arrange

Assert

Act

We expanded MorphQ’s Qiskit gate-set by incorporating following new available gates: CCZGate, CSGate, CSdgGate, RGate, and RVGate.

Tools and Test-bed

All experiments were run on an Apple M1 Pro 14-inch (2021 model) machine. It has ten cores (eight high-performance and two energy efficient), and 16 GB RAM. The OS at the time of evaluation was MacOS Ventura 13.3.1 (22E261)

Arrange

Assert

Act

Research Questions

  • RQ1: How many syntactically different but correct programs can be translated by QCross’s converter?
     
  • RQ2: What has QCross found via cross-platform testing of the widely-used QSSes? i.e., how many warnings and errors do QCross produce?
     
  • RQ3: How does QCross compare to prior work on testing quantum computing platforms?
     
  • RQ4: How useful is Bloqs?

RQ1

We did not record any crashes during the translation process.

How many syntactically different but correct programs can be translated by QCross’s converter?

The program converter successfully translates only valid quantum programs, ensures that any possible metamorphic relation is maintained, and QCross is effective in producing numerous warnings and crashes in follow-up programs of different platforms when executed.

RQ2

QCross identified 14 bugs and 2 potential issues across quantum programming platforms, with Qiskit (8) having the most bugs, followed by PyQuil (4) and Cirq (2)

What has QCross found via cross-platform testing of the widely-used QSSes? i.e., how many warnings and errors do QCross produce?

QCross identified 14 bugs and 2 potential issues across quantum programming platforms, with Qiskit (8) having the most bugs, followed by PyQuil (4) and Cirq (2).

Metamorphic testing proved more effective than differential testing, and ten of the reported bugs have been confirmed as novel by developers.

RQ3

We regard QCross as an "evolutionary successor" to MorphQ and QDiff. It has uncovered new bugs that were not detected in the previous works, demonstrating its complementary value to the existing research.

How does QCross compare to prior work on testing quantum computing platforms?
 

  • More platforms
  • New bugs
  • New MRs and extra gates
  • bloqs as a utility library

RQ4

The bloqs library was developed to overcome gate set limitations in Cirq and PyQuil, enabling successful translation of Qiskit programs. The library proved essential, as it was required in approximately 94% of translations.

How useful is Bloqs?

Future Works

  • Ensure the generated programs are more "real-world" programs.
  • Extend the number of tested platforms to include non-python platforms as well.
  • Execute the programs on quantum hardware to establish a source of truth or to find bugs in the hardware.
  • Devise a method to better test the divergence of program outputs such that false positives are minimized.

  • Analyse the existing divergent programs to find out the source of divergence.

Thank you!

Made with Slides.com