Exact Grover’s Algorithm#

In this example we will showcase how the exact keyword for grovers_alg can be applied. This keyword allows to produce states, which are exact solutions to the given oracle (ie. they have 0% theoretical failure rate). This feature is based on this paper. For this to work the amount of solutions to the oracle has to be known beforehand. Furthermore the oracle has to support the phase keyword which indicates how much of a phaseshift the winner states receive. For standard Grover oracles, this phaseshift is always \(\pi\).

We demonstrate this functionality by preparing a state which is a superposition of all states that contain a fixed number of “ones”:

To count the amount of ones we use quantum phase estimation on the operator

\[U = \text{exp}\left(\frac{i 2 \pi}{2^k} \sum_{i = 0}^{n-1} ( 1 - \sigma_{z}^i )\right)\]
from qrisp import QPE, p, QuantumVariable, lifted
from qrisp.grover import grovers_alg, tag_state
import numpy as np

def U(qv, prec = None, iter = 1):
    for i in range(qv.size):
        p(iter*2*np.pi/2**prec, qv[i])

@lifted
def count_ones(qv):
    prec = int(np.ceil(np.log2(qv.size+1)))
    res = QPE(qv, U, precision = prec, iter_spec = True, kwargs = {"prec" : prec})
    return res<<prec

Quick test:

>>> qv = QuantumVariable(5)
>>> qv[:] = {"11000" : 1, "11010" : 1, "11110" : 1}
>>> count_qf = count_ones(qv)
>>> count_qf.qs.statevector()
sqrt(3)*(|11000>*|2> + |11010>*|3> + |11110>*|4>)/3

We now define the oracle

def counting_oracle(qv, phase = np.pi, k = 1):

    count_qf = count_ones(qv)

    tag_state({count_qf : k}, phase = phase)

    count_qf.uncompute()

And evaluate Grover’s algorithm

n = 5
k = 3
qv = QuantumVariable(n)

import math

grovers_alg(qv, counting_oracle, exact = True, winner_state_amount = math.comb(n,k), kwargs = {"k" : k})
>>> print(qv)
{'11100': 0.1,
 '11010': 0.1,
 '10110': 0.1,
 '01110': 0.1,
 '11001': 0.1,
 '10101': 0.1,
 '01101': 0.1,
 '10011': 0.1,
 '01011': 0.1,
 '00111': 0.1}

We see that contrary to regular Grover’s algorithm, the states which have not been tagged by the oracle have 0 percent measurement probability.