Редактиране

Споделяне чрез

How to debug and test your quantum code

  • Статия

As with classical programming, it is essential to be able to check that quantum programs act as intended, and to be able to diagnose incorrect behavior. This article discusses the tools offered by the Azure Quantum Development Kit for testing and debugging quantum programs.

Debug your Q# program

The Azure Quantum Development Kit (QDK) Visual Studio Code extension includes a debugger for Q# programs. You can set breakpoints, step through your code and into each function or operation, and track not only the local variables, but the quantum state of the qubits as well.

Note

The VS Code debugger only works with Q# (.qs) files and doesn't work with Q# cells in a Jupyter Notebook. For testing Jupyter Notebook cells, see Test your code.

The following example demonstrates the basic features of the debugger. For complete information about using VS Code debuggers, see Debugging.

In VS Code, create and save a new .qs file with the following code:

import Microsoft.Quantum.Arrays.*;
import Microsoft.Quantum.Convert.*;

operation Main() : Result {
    use qubit = Qubit();
    H(qubit);
    let result = M(qubit);
    Reset(qubit);
    return result;
}
  1. Set a breakpoint on the line H(qubit) by clicking to the left of the line number.
  2. Select the debugger icon to open the debugger pane and select Run and Debug. The debugger controls are displayed at the top of the screen.
  3. Select F5 to start debugging and continue to the breakpoint. In the debugger Variables pane, expand the Quantum State category. You can see that the qubit has been initialized in the |0> state.
  4. Step into (F11) the H operation and the source code for the H operation displays. As you step through the operation, note the quantum value changes as the H operation puts the qubit into superposition.
  5. As you step over (F10) the M operation, the quantum value is resolved to either |0> or |1> as a result of the measurement, and the value of the classical variable result is displayed.
  6. As you step over the Reset operation, the qubit is reset to |0>.

Test your code

Although the VS Code Q# debugger is not available for Q# cells in a Jupyter Notebook, the Azure QDK provides some expressions and functions that can help troubleshoot your code.

Fail expression

The fail expression ends the computation entirely, corresponding to a fatal error that stops the program.

Consider this simple example that validates a parameter value:

# import qsharp package to access the %%qsharp magic command
import qsharp 

// use the %%qsharp magic command to change the cell type from Python to Q#
%%qsharp 
function PositivityFact(value : Int) : Unit {
    if value <= 0 {
        fail $"{value} isn't a positive number.";
    }   
}
PositivityFact(0);

Error: program failed: 0 isn't a positive number.
Call stack:
    at PositivityFact in line_2
Qsc.Eval.UserFail

  × runtime error
  ╰─▶ program failed: 0 isn't a positive number.
   ╭─[line_2:5:1]
 5 │ 
 6 │             fail $"{value} isn't a positive number.";
   ·             ────────────────────┬───────────────────
   ·                                 ╰── explicit fail
 7 │     }   
   ╰────

Here, the fail expression prevents the program from continuing to run with invalid data.

Fact() function

You can implement the same behavior as the previous example using the Fact() function from the Microsoft.Quantum.Diagnostics namespace. The Fact() function evaluates a given classical condition and throws an exception if it is false.

import qsharp 

%%qsharp
function PositivityFact(value : Int) : Unit {
    Fact(value > 0, "Expected a positive number."); 
}
PositivityFact(4);

Error: program failed: Expected a positive number.
Call stack:
    at Microsoft.Quantum.Diagnostics.Fact in diagnostics.qs
    at PositivityFact in line_4
Qsc.Eval.UserFail

  × runtime error
  ╰─▶ program failed: Expected a positive number.
    ╭─[diagnostics.qs:29:1]
 29 │         if (not actual) {
 30 │             fail message;
    ·             ──────┬─────
    ·                   ╰── explicit fail
 31 │         }
    ╰────

DumpMachine() function

DumpMachine() is a Q# function that allows you to dump information about the current state of the target machine to the console and continue to run your program.

Note

With the release of the Azure Quantum Development Kit, the DumpMachine() function now uses big-endian ordering for its output.

import qsharp

%%qsharp
import Microsoft.Quantum.Diagnostics.*;
operation MultiQubitDumpMachineDemo() : Unit {
    use qubits = Qubit[2];
    X(qubits[1]);
    H(qubits[1]);
    DumpMachine();

    R1Frac(1, 2, qubits[0]);
    R1Frac(1, 3, qubits[1]);
    DumpMachine();
    
    ResetAll(qubits);
}
MultiQubitDumpMachineDemo();

Basis State
(|𝜓₁…𝜓ₙ⟩)	Amplitude	Measurement Probability	Phase
|00⟩	0.7071+0.0000𝑖	 50.0000%	↑	0.0000
|01⟩	−0.7071+0.0000𝑖	 50.0000%	↓	-3.1416

Basis State
(|𝜓₁…𝜓ₙ⟩)	Amplitude	Measurement Probability	Phase
|00⟩	0.7071+0.0000𝑖	 50.0000%	↑	0.0000
|01⟩	−0.6533−0.2706𝑖	 50.0000%	↙	-2.7489

dump_machine() function

dump_machine is a Python function that returns the current allocated qubit count and a Python dictionary of sparse state amplitudes that you can parse. Using either of these functions in a Jupyter Notebook allows you to step through your operations much like a debugger. Using the previous example program:

import qsharp 

%%qsharp
use qubits = Qubit[2];
X(qubits[0]);
H(qubits[1]);

dump = qsharp.dump_machine()
dump


Basis State
(|𝜓₁…𝜓ₙ⟩)	Amplitude	Measurement Probability	Phase
|10⟩	0.7071+0.0000𝑖	 50.0000%	↑	0.0000
|11⟩	0.7071+0.0000𝑖	 50.0000%	↑	0.0000

%%qsharp
R1Frac(1, 2, qubits[0]);
R1Frac(1, 3, qubits[1]);

dump = qsharp.dump_machine()
dump

Basis State
(|𝜓₁…𝜓ₙ⟩)	Amplitude	Measurement Probability	Phase
|10⟩	0.5000+0.5000𝑖	 50.0000%	↗	0.7854
|11⟩	0.2706+0.6533𝑖	 50.0000%	↗	1.1781    

# you can print an abbreviated version of the values
print(dump)

STATE:
|10⟩: 0.5000+0.5000𝑖
|11⟩: 0.2706+0.6533𝑖

# you can access the current qubit count
dump.qubit_count

2

# you can access individual states by their index
dump[2]

(0.5+0.5000000000000001j)

dump[3]

(0.27059805007309845+0.6532814824381883j)

CheckZero() and CheckAllZero() operations

CheckZero() and CheckAllZero() are Q# operations that can check whether the current state of a qubit or qubit array is $\ket{0}$. CheckZero() returns true if the qubit is in the $\ket{0}$ state, and false if it is in any other state. CheckAllZero() returns true if all qubits in the array are in the $\ket{0}$ state, and false if the qubits are in any other state.

import Microsoft.Quantum.Diagnostics.*;

operation Main() : Unit {
    use qs = Qubit[2];
    X(qs[0]); 
    if CheckZero(qs[0]) {
        Message("X operation failed");
    }
    else {
        Message("X operation succeeded");
    }
    ResetAll(qs);
    if CheckAllZero(qs) {
        Message("Reset operation succeeded");
    }
    else {
        Message("Reset operation failed");
    }
}

dump_operation() function

dump_operation is a Python function that takes an operation, or operation definition, and a number of qubits to use, and returns a square matrix of complex numbers representing the output of the operation.

You import dump_operation from qsharp.utils.

import qsharp
from qsharp.utils import dump_operation

This example prints the matrix of a single-qubit identity gate and the Hadamard gate.

res = dump_operation("qs => ()", 1)
print(res)
res = dump_operation("qs => H(qs[0])", 1)
print(res)

[[(1+0j), 0j], [0j, (1+0j)]]
[[(0.707107+0j), (0.707107+0j)], [(0.707107+0j), (-0.707107-0j)]]

You can also define a function or operation using qsharp.eval() and then reference it from dump_operation. The single qubit represented earlier can also be represented as

qsharp.eval("operation SingleQ(qs : Qubit[]) : Unit { }")

res = dump_operation("SingleQ", 1)
print(res)

[[(1+0j), 0j], [0j, (1+0j)]]

This example uses a Controlled Ry gate to apply a rotation to the second qubit

qsharp.eval ("operation ControlRy(qs : Qubit[]) : Unit {qs[0]; Controlled Ry([qs[0]], (0.5, qs[1]));}")

res = dump_operation("ControlRy", 2)
print(res)

[[(1+0j), 0j, 0j, 0j], [0j, (1+0j), 0j, 0j], [0j, 0j, (0.968912+0j), (-0.247404+0j)], [0j, 0j, (0.247404+0j), (0.968912+0j)]]

The following code defines Q# operation ApplySWAP and prints its matrix alongside that of the two-qubit identity operation.

qsharp.eval("operation ApplySWAP(qs : Qubit[]) : Unit is Ctl + Adj { SWAP(qs[0], qs[1]); }")

res = dump_operation("qs => ()", 2)
print(res)
res = dump_operation("ApplySWAP", 2)
print(res)

[[(1+0j), 0j, 0j, 0j], [0j, (1+0j), 0j, 0j], [0j, 0j, (1+0j), 0j], [0j, 0j, 0j, (1+0j)]]
[[(1+0j), 0j, 0j, 0j], [0j, 0j, (1+0j), 0j], [0j, (1+0j), 0j, 0j], [0j, 0j, 0j, (1+0j)]]

More examples of testing operations using dump_operation() can be found on the samples page Testing Operations in the QDK.