pyqpanda_alg.QAOA.qaoa¶

Classes¶

QAOA

This class provides the quantum alternative operator ansatz algorithm optimizer. It assumes a polynomial problem

Functions¶

`p_1`(n)	Transfer binary variable \(x_n\) to pauli operator \(\frac{I-Z_n}{2}\)
`p_0`(n)	Transfer binary variable \(x_n\) to pauli operator \(\frac{I+Z_n}{2}\)
`problem_to_z_operator`(problem[, norm])	Transfer polynomial function with binary variables \(f(x_0, \cdots, x_n)\) to
`parameter_interpolate`(pm)	Use INTERP heuristic strategy to guess the initial parameter of \(p+1\) layer QAOA
`pauli_z_operator_to_circuit`(operator, qlist[, gamma])	Circuit of simulation diagonal Hamiltonian \(e^{-iH heta}\).

Module Contents¶

pyqpanda_alg.QAOA.qaoa.p_1(n)¶

Transfer binary variable \(x_n\) to pauli operator \(\frac{I-Z_n}{2}\)

Parameters: n : int

index of the variable, start with 0.
Return: operator : PauliOperator

Pauli operator \(\frac{I-Z_n}{2}\)
Examples: Transfer \(x_0\) into pauli operator \(\frac{I-Z_0}{2}\)

>>> from pyqpanda_alg.QAOA import qaoa
>>> operator_0 = qaoa.p_1(0)
>>> print(operator_0)
    { qbit_total = 1, pauli_with_coef_s = { '':0.5 + 0j, 'Z0 ':-0.5 + 0j, } }

pyqpanda_alg.QAOA.qaoa.p_0(n)¶

Transfer binary variable \(x_n\) to pauli operator \(\frac{I+Z_n}{2}\)

Parameters: n : integer

index of the variable, start with 0.
Return: operator : PauliOperator

Pauli operator \(\frac{I+Z_n}{2}\)
Examples: Transfer \(x_0\) into pauli operator \(\frac{I+Z_0}{2}\)

>>> from pyqpanda_alg.QAOA import qaoa
>>> operator_0 = qaoa.p_0(0)
>>> print(operator_0)
    { qbit_total = 1, pauli_with_coef_s = { '':0.5 + 0j, 'Z0 ':0.5 + 0j, } }

pyqpanda_alg.QAOA.qaoa.problem_to_z_operator(problem, norm=False)¶

Transfer polynomial function with binary variables \(f(x_0, \cdots, x_n)\) to pauli operator \(f(\frac{I-Z_0}{2}, \cdots, \frac{I-Z_n}{2})\)

Parameters: problem : expression in sympy
Return: hamiltonian : PauliOperator

Pauli operators \(f(\frac{I-Z_n}{2})\) in list form.
Examples: Transfer \(2 x_0 x_1 + 3 x_2 - 1\) into pauli operators

>>> import sympy as sp
>>> from pyqpanda_alg.QAOA import qaoa
>>> vars = sp.symbols('x0:3')
>>> f = 2*vars[0]*vars[1] + 3*vars[2] - 1
>>> print(f)
    2*x0*x1 + 3*x2 - 1
>>> hamiltonian = qaoa.problem_to_z_operator(f)
>>> print(hamiltonian)
    { qbit_total = 3, pauli_with_coef_s = { '':1 + 0j, 'Z2 ':-1.5 + 0j, 'Z1 ':-0.5 + 0j, 'Z0 ':-0.5 + 0j, 'Z0 Z1 ':0.5 + 0j, } }

pyqpanda_alg.QAOA.qaoa.parameter_interpolate(pm)¶

Use INTERP heuristic strategy to guess the initial parameter of \(p+1\) layer QAOA from the optimal parameter found from \(p\) layer QAOA circuit.

Parameters

pm : array-like

Optimal parameters of \(p\) layer QAOA circuit, with length \(2 imes p\)

Return

operator : array-like

A guess for the initial parameter of \(p+1\) layer QAOA, with length \(2 imes (p+1)\)

References

[1] ZHOU L, WANG S T, CHOI S, et. Quantum Approximate Optimization Algorithm: Performance, Mechanism, and Implementation on Near-Term Devices[J/OL]. Physical Review X, 2020, 10(2): 021067. :DOI: 10.1103/PhysRevX.10.021067.

Examples

Transfer \(x_0\) into pauli operator \(\frac{I-Z_0}{2}\)

>>> import numpy as np
>>> from pyqpanda_alg.QAOA import qaoa
>>> initial_parameter = np.array([0.1, 0.2, 0.2, 0.1])
>>> new_parameter = qaoa.parameter_interpolate(initial_parameter)
>>> print(new_parameter)
    [0.1  0.15 0.25 0.2  0.15 0.05]

pyqpanda_alg.QAOA.qaoa.pauli_z_operator_to_circuit(operator, qlist, gamma=np.pi)¶

Circuit of simulation diagonal Hamiltonian \(e^{-iH heta}\).

Parameters

operator : list

Pauli Operator in list form. (By method operator.toHamiltonian(1))

qlist : qubit list

gamma : float

Value of theta in \(e^{-iH heta}\).

Return

circuit : pq.QCircuit

Circuit of simulation diagonal Hamiltonian \(e^{-iH heta}\).

constant : float

Constant number in the hamiltonian.

Example

Print a circuit of hamiltonian for problem \(f(\vec{x})=2x_0x_1 + 3x_2 - 1\) with \(\theta=0\)

import sympy as sp
from pyqpanda_alg.QAOA import qaoa
vars = sp.symbols('x0:3')
f = 2*vars[0]*vars[1] + 3*vars[2] - 1

hamiltonian = qaoa.problem_to_z_operator(f)
circuit, constant = qaoa.pauli_z_operator_to_circuit(hamiltonian, list(range(3)), 0)
print(circuit)

          ┌────────────┐                              ┌─┐
q_0:  |0>─┤RZ(0.000000)├ ───■── ────────────── ───■── ┤I├
          ├────────────┤ ┌──┴─┐ ┌────────────┐ ┌──┴─┐ ├─┤
q_1:  |0>─┤RZ(0.000000)├ ┤CNOT├ ┤RZ(0.000000)├ ┤CNOT├ ┤I├
          ├────────────┤ ├─┬──┘ └────────────┘ └────┘ └─┘
q_2:  |0>─┤RZ(0.000000)├ ┤I├─── ────────────── ────── ───
          └────────────┘ └─┘

class pyqpanda_alg.QAOA.qaoa.QAOA(problem, init_circuit=None, mixer_circuit=None, norm=False)¶

This class provides the quantum alternative operator ansatz algorithm optimizer. It assumes a polynomial problem consisting only of binary variables. The problem is then translated into an Ising Hamiltonian whose minimal eigen vector and corresponding eigenstate correspond to the optimal solution of the original optimization problem. The provided solver is then used to approximate the ground state of the Hamiltonian to find a good solution for the optimization problem.

Parameters

problem : expression in sympy or pq.PauliOperator

A polynomial function with binary variables to be optimized. Support an expression in sympy. Next version will support an object from pypanda PauliOperator.

init_circuit : function, optional

The quantum circuit to create the initial state of QAOA algorithm. Default is Hadamard circuit to create an equal superposition state \(\ket{\psi} = 2^{-n/2}\sum_{i=0}^{2^n-1}\ket{i}\).

mixer_circuit : function, optional

The function which returns a mixer quantum circuit \(U_M(\beta)=\exp(-i\beta H_M)\). The function should accept two parameters (qubit list, array-like angles) as input, and return a quantum circuit as output. Default is X mixer circuit \(\exp(-i\beta \sum_i X_i)=RX(2\beta)^{\otimes n}\)

norm : bool

Attributes

energy_dict : dict

The dict which stores the function value for solutions being sampled during the optimization.

problem_dimension : integer

The problem dimension, and also the qubit number.

circuit iter : integer

The number of times the quantum circuit being called during optimization.

Methods

calculate_energy : Calculate the function value for one solution.

run_qaoa_circuit : Given parameters, run the qaoa circuit and get the theoretical probability distribution.

run : run the optimization process

Reference

[1] FARHI E, GOLDSTONE J, GUTMANN S. A Quantum Approximate Optimization Algorithm[J/OL]. 2014[2022-03-09]. https://arxiv.org/abs/1411.4028v1. DOI:10.48550/arXiv.1411.4028.

[2] ZHOU L, WANG S T, CHOI S, et. Quantum Approximate Optimization Algorithm: Performance, Mechanism, and Implementation on Near-Term Devices[J/OL]. Physical Review X, 2020, 10(2): 021067. DOI:10.1103/PhysRevX.10.021067.

measure_type = None¶

optimizer = None¶

optimize_option = None¶

shots = None¶

optimize_type = None¶

temperature = None¶

alpha = None¶

loss_type = None¶

problem¶

operator = None¶

problem_dimension = 0¶

init_circuit = None¶

mixer_circuit = None¶

layer = None¶

iter = 0¶

circuit_iter = 0¶

energy_dict¶

calculate_energy(x)¶

Calculate the function value for one solution. TODO: using new method to acccelrate the calculation.

Parameter

x : array-like

one binary variables solution in vector form.

Return

float

function value of the solution \(f(x)\).

Example

Let \(f(\vec{x})=2x_0x_1 + 3x_2 - 1\), calculate \(f(1,0,0)\)

>>> import sympy as sp
>>> from pyqpanda_alg.QAOA.qaoa import *
>>> vars = sp.symbols('x0:3')
>>> f = 2*vars[0]*vars[1] + 3*vars[2] - 1
>>> print(f)
    2*x0*x1 + 3*x2 - 1
>>> qaoa_f = QAOA(f)
>>> solution_1 = [1, 0, 0]
>>> print(qaoa_f.calculate_energy(solution_1))
    -1
>>> solution_2 = [0, 1, 1]
>>> print(qaoa_f.calculate_energy(solution_2))
    2
>>> ham_f = 2 * p_1(0) * p_1(1) + 3 * p_1(2)- PauliOperator('I')*1
>>> qaoa_ham = QAOA(ham_f)
>>> print(qaoa_ham.calculate_energy(solution_1))
    -1.0

run_qaoa_circuit(gammas, betas, shots=-1)¶

Given parameters, run the qaoa circuit and get the theoretical probability distribution.

Parameters

gammas : array-like

Parameter gamma for QAOA phase circuit

betas : array-like

Parameter beta for QAOA mixer circuit

shots : integer, optional

Times of running the same circuit. Must be positive integer or -1. If it is -1, the results are given as amplitudes of all state vectors, which can be viewed as running the circuit infinite times. Default is -1.

Return

prob_result : dict

Probability of each computational basis state. The keys are binary form of qubits where the first qubit sits at the right-most position and the items are the corresponding probability (if shots = -1) or frequency (if shots > 0).

Example

Run a two-layer QAOA algorithm circuit of problem \(f(\vec{x})=2x_0x_1 + 3x_2 - 1\) with parameters \([0, 0, 0, 1, 1, 1]\)

import pyqpanda as pq
import sympy as sp
from pyqpanda_alg.QAOA.qaoa import *

vars = sp.symbols('x0:3')
f = 2*vars[0]*vars[1] + 3*vars[2] - 1
qaoa_f = QAOA(f)

gammas = [0, 0]
betas = [1, 1]

qaoa_result = qaoa_f.run_qaoa_circuit(gammas, betas, -1)
print(qaoa_result)
qaoa_result = qaoa_f.run_qaoa_circuit(gammas, betas, 500)
print(qaoa_result)

The codes above would give results like (with errors due to randomness):

{'000': 0.12500000000000008, '001': 0.12500000000000008, '010': 0.12500000000000008,
'011': 0.12500000000000008, '100': 0.12500000000000008, '101': 0.12500000000000008,
'110': 0.12500000000000008, '111': 0.12500000000000008}

{'000': 0.132, '001': 0.134, '010': 0.112, '011': 0.136, '100': 0.094, '101': 0.13,
'110': 0.122, '111': 0.14}

run(layer=1, initial_para=None, shots=-1, loss_type=None, optimize_type=None, optimizer=None, optimizer_option=None, **loss_option)¶

Optimize the function by QAOA algorithm.

Parameters

layer : integer, optional

Layers number of QAOA circuit. Default is 1. If optimize type is interp, then it represents the final layer of the optimization progress.

initial_para : array-like, optional

initial parameters of \(p\) layer QAOA circuit, with length \(2\times p\). If not given, a random distribution from \(U(0, \pi)\) of size \(2p\) is generated.

shots : integer, optional

Circuit measured times. If shots takes -1, then use theoretical probability (by state vector) instead. Default is -1

loss_type : string, optional

The loss function used by the optimizer. Should be one of

default : Given a result, calculate the energy expectation.

See Note Energy expectation

Gibbs : Given a result and argument temperature \(T\), calculate the Gibbs energy expectation.

See Note Gibbs energy

CVaR : Given a result and argument \(\alpha\), calculate the CVaR loss function.

See Note CVaR loss functio

If not given, default by default.

optimize_type : string, optional

The method to optimize the QAOA circuit. Should be one of

default: Directly optimize the \(p\) layer QAOA circuit.

interp: Use interpolate method to train a big QAOA circuit.

See Note interp method

If not given, default by default.

optimizer : string, optional

Type of solver. Should be one of

SPSA : See spsa.spsa_minimize

one of Nelder-Mead, Powell, CG, BFGS, Newton-CG, TNC, COBYLA, SLSQP,

trust-constr, dogleg, trust-ncg, trust-exact, trust-krylov. See scipy.optimize.minimize.

If not given, default by SLSQP.

optimizer_option : dict, optional

A dictionary of solver options. Accept the following generic options:

bounds : List[tuple], optional

Bounds for the variables. Sequence of (min, max) pairs for each element in x. If specified, variables are clipped to fit inside the bounds after each iteration. None is used to specify no bound.

options : integer

Maximum number of iterations to perform. Depending on the method each iteration may use several function evaluations.

For TNC use maxfun instead of maxiter.

loss_option :

temperature : float, optional

parameter calculated in _loss_function_Gibbs. Default is 1. See Note Gibbs energy.

alpha : float, optional

parameter calculated in _loss_function_cvar. Default is 1. See Note Gibbs energy.

Return

qaoa_result : dict

dict of all possible solutions with corresponding probabilities. The elements are arranged in descending order of probability.

para_result : array-like

Array of the optimized QAOA parameters.

loss_result : float

Loss function value of the optimized QAOA parameters.

Example

Run a two-layer QAOA algorithm circuit of problem \(f(\vec{x})=2x_0 + x_1 + 3x_2 - 1\) with parameters [0, 0, 0, 1, 1, 1]

import sympy as sp
from pyqpanda_alg.QAOA import qaoa
vars = sp.symbols('x0:3')
f = 2*vars[0]*vars[1] + 3*vars[2] - 1
qaoa_f = qaoa.QAOA(f)

qaoa_result = qaoa_f.run(layer=3)
sorted_result = sorted(qaoa_result[0].items(), key=lambda k:k[1], reverse=True)[:5]
print(sorted_result[:5])

[('000', 0.6848946054168573),
 ('010', 0.1575526972909123),
 ('001', 0.15755269729091226),
 ('100', 7.957749518311524e-13),
 ('110', 1.8305953815081342e-13)]

Notes

Energy expectation:

In traditional QAOA algorithm, the parameter is optimized by minimize the energy expectation

\(\bra{\psi(\gamma, \beta}H\ket{\psi(\gamma, \beta)}\)

If measure type is sample, it is calculated by

\(E=\frac{1}{N_{\rm{shots}}}\sum_{i=0}^{2^n-1} n_iE_i\).

If measure type is theoretical, it is calculated by

\(E=\sum_{i=0}^{2^n-1} p_iE_i\).
Gibbs energy:

Inspired by Ref[1]. Instead of the traditional energy expectation value, using the Gibbs function as the object function. The function is

\(f_G=-\ln \langle e^{-E/T}\rangle\)

Here \(T\) is the hyperparameter of temperature, as \(T\) decreases, the weight of the lower energy states in the loss function then becomes larger. When \(T=1\), it turns back to energy expectation function.

If measure type is sample, it is calculated by

\(G=-\log (\frac{1}{N_{\rm{shots}}}\sum_{i=0}^{2^n-1} n_i \exp(-E_i/T))\).

If measure type is theoretical, it is calculated by

\(G=-\log (\sum_{i=0}^{2^n-1} p_i \exp(-E_i/T))\).
CVaR loss function:

Inspired by Ref[3].Instead of the traditional energy expectation value, using the Conditional Value at Risk function as the object function. The function is

\(CVaR_\alpha(X) = \mathbb{E}[X|X\leq F_X^{-1}(alpha)]\)

Here \(\alpha\) is the confidence level. CVaR is the expected value of the lower α-tail of the distribution of X. \(\alpha=0\) corresponds to the minimum, and \(\alpha=1\) corresponds to the expectation value.

If measure type is sample, it is calculated by

\(E=\frac{1}{\alpha N}(\sum_{i=0}^{k} n_iE_i + (\alpha N - n_{k+1})E_{k+1}),\sum_{i=0}^k n_i < \alpha N\)

If measure type is theoretical, it is calculated by

\(E=\sum_{i=0}^{k} p_iE_i + (\alpha - p_{k+1})E_{k+1}, \sum_{i=0}^k p_i < \alpha\)
Interpolate method:

Inspired by Ref[2].

Reference

[1] LI L, FAN M, CORAM M, et. Quantum Optimization with a Novel Gibbs Objective Function and Ansatz Architecture Search[J/OL]. Physical Review Research, 2020, 2(2): 023074. DOI:10.1103/PhysRevResearch.2.023074.

[3] BARKOUTSOS P K, NANNICINI G, ROBERT A, et. Improving Variational Quantum Optimization using CVaR[J/OL]. Quantum, 2020, 4: 256. DOI:10.22331/q-2020-04-20-256.