Skip to main content
IBM Quantum Platform

Quantum algorithms: Grover Search and applications

Note

Atsushi Matsuo (May 10, 2024)

PDFs of lecture notes will be available soon. Note that some code snippets may become deprecated since these are static images.

Approximate QPU time to run this experiment is 2 seconds.

1. Introduction to Grover's algorithm

This notebook is the fourth in a series of lectures on the Path to Utility in Quantum Computing. In this notebook, we will learn about Grover's algorithm.

Grover's algorithm is one of the most well-known quantum algorithms due to its quadratic speedup over classical search methods. In classical computing, searching an unsorted database of NN items requires O(N)O(N) time complexity, meaning that in the worst case, one might have to examine each item individually. However, Grover's algorithm allows us to achieve this search in O(N)O(\sqrt{N}) time, leveraging the principles of quantum mechanics to identify the target item more efficiently.

The algorithm uses amplitude amplification, a process that increases the probability amplitude of the correct answer state in a quantum superposition, allowing it to be measured with higher probability. This speedup makes Grover's algorithm valuable in various applications beyond simple database search, especially when the dataset size is large. Detailed explanations of the algorithm is provided in the Grover's algorithm notebook.

The Basic Structure of Grover's Algorithm

Grover's algorithm comprises four main components:

  1. Initialization: Setting up the superposition over all possible states.
  2. Oracle: Applying an oracle function that marks the target state by flipping its phase.
  3. Diffusion Operator: Applying a series of operations to amplify the probability of the marked state.

Each of these steps plays a critical role in making the algorithm work efficiently. Detailed explanations for each step are provided later.

2. Implementing Grover's Algorithm

2.1 Preparation

Import the necessary libraries and set up the environment for running the quantum circuit.

%config InlineBackend.figure_format = 'svg' # Makes the images look nice
# importing Qiskit
from qiskit import QuantumCircuit, ClassicalRegister, QuantumRegister
 
# import basic plot tools
from qiskit.visualization import plot_histogram

Step 1: Map problem to quantum circuits and operators

Consider a list of 4 elements, where our goal is to identify the index of an element that meets a specific condition. For instance, we want to find the index of the element equal to 2. In this example, the quantum state ∣01⟩|01\rangle represents the index of the element that satisfies this condition, as it points to the position where the value 2 is located.

Step 2: Optimize for target hardware

1: Initialization

In the initialization step, we create a superposition of all possible states. This is achieved by applying a Hadamard gate to each qubit in an n-qubit register, which will result in an equal superposition of 2n2^n states. Mathematically, this can be represented as:

1N∑x=0N−1∣x⟩\frac{1}{\sqrt{N}} \sum_{x=0}^{N-1} |x\rangle

where N=2nN = 2^n is the total number of possible states. We also change the state of the ancilla bit to ∣−⟩|-\rangle.

def initialization(circuit):
    # Initialization
    n = circuit.num_qubits
    # For input qubits
    for qubit in range(n - 1):
        circuit.h(qubit)
    # For the ancilla bit
    circuit.x(n - 1)
    circuit.h(n - 1)
    circuit.barrier()
    return circuit
n = 2
qr = QuantumRegister(n, "q")
anc = QuantumRegister(1, "ancillary")
initialization_circuit = QuantumCircuit(qr, anc)
 
initialization(initialization_circuit)
initialization_circuit.draw(output="mpl", idle_wires=False)

Output:

Output of the previous code cell

2: Oracle

The oracle is a key part of Grover's algorithm. It marks the target state by applying a phase shift, typically flipping the sign of the amplitude associated with that state. The oracle is often problem-specific and constructed based on the criteria for identifying the target state. In mathematical terms, the oracle applies the following transformation:

f(x)={1,if x=xtarget0,otherwisef(x) = \begin{cases} 1, & \text{if } x = x_{\text{target}} \\ 0, & \text{otherwise} \end{cases}

This phase flip is achieved by applying a negative sign to the amplitude of the target state via phase kickback.

def oracle(circuit):
    circuit.x(1)
    circuit.ccx(0, 1, 2)
    circuit.x(1)
    circuit.barrier()
n = 2
qr = QuantumRegister(n, "q")
anc = QuantumRegister(1, "ancillary")
oracle_circuit = QuantumCircuit(qr, anc)
 
oracle(oracle_circuit)
oracle_circuit.draw(output="mpl", idle_wires=False)

Output:

Output of the previous code cell

3: Diffusion Operator

The amplitude amplification process is what differentiates Grover's algorithm from a classical search. After the oracle has marked the target state, we apply a series of operations that increase the amplitude of this marked state, making it more likely to be observed upon measurement. This process is achieved through the Diffusion operator, which effectively performs an inversion about the average amplitude. The mathematical operation is as follows:

D=2∣ψ⟩⟨ψ∣−ID = 2|\psi\rangle\langle\psi| - I

where DD is the diffusion operator, II is the identity matrix, and ∣ψ⟩|\psi\rangle is the equal superposition state. The combination of the oracle and the diffusion operator is applied approximately N\sqrt{N} times to achieve maximum probability for measuring the marked state.

def diffusion(circuit):
    input_qubits = circuit.num_qubits - 1
    circuit.h(range(0, input_qubits))
    circuit.x(range(0, input_qubits))
    circuit.h(input_qubits - 1)
    circuit.mcx([i for i in range(0, input_qubits - 1)], input_qubits - 1)
    circuit.h(input_qubits - 1)
    circuit.x(range(0, input_qubits))
    circuit.h(range(0, input_qubits))
    circuit.barrier()
n = 2
qr = QuantumRegister(n, "q")
anc = QuantumRegister(1, "ancillary")
diffusion_circuit = QuantumCircuit(qr, anc)
 
diffusion(diffusion_circuit)
diffusion_circuit.draw(output="mpl", idle_wires=False)

Output:

Output of the previous code cell

2.2 A 2-qubit Grover search example.

n = 2
qr = QuantumRegister(n, "q")
anc = QuantumRegister(1, "ancillary")
meas = ClassicalRegister(3, "meas")
grover_circuit = QuantumCircuit(qr, anc, meas)
# the number of iterations
num_iterations = 1
# Let's do Grover search
initialization(grover_circuit)
 
for i in range(0, num_iterations):
    oracle(grover_circuit)
    diffusion(grover_circuit)
 
# Clear the ancilla bit
grover_circuit.h(n)
grover_circuit.x(n)
grover_circuit.measure_all(add_bits=False)
 
grover_circuit.draw(output="mpl", idle_wires=False)

Output:

Output of the previous code cell

2.3 Experiment with Simulators

Step 3: Executing the circuit.

from qiskit_aer import AerSimulator
from qiskit_ibm_runtime import Sampler
from qiskit.transpiler.preset_passmanagers import generate_preset_pass_manager
 
# Define backend
backend = AerSimulator()
 
# Transpile to backend
pm = generate_preset_pass_manager(backend=backend, optimization_level=1)
isa_qc = pm.run(grover_circuit)
 
# Run the job
sampler = Sampler(mode=backend)
job = sampler.run([isa_qc])
result = job.result()

Step 4: Post-processing the results.

# Print the results
counts = result[0].data.meas.get_counts()
print(counts)
 
# Plot the counts in a histogram
plot_histogram(counts)

Output:

{'001': 1024}
Output of the previous code cell

We got the correct answer ∣01⟩|01\rangle. Note that be careful with the order of qubits

3. Experiment with Real Devices

Step 2: Optimize for target hardware

from qiskit_ibm_runtime import QiskitRuntimeService
 
service = QiskitRuntimeService()
real_backend = service.backend("ENTER-QPU-NAME-HERE")
# You can also identify the least busy device
 
real_backend = service.least_busy(simulator=False, operational=True)
print("The least busy device is ", real_backend)
# Transpile the circuit into basis gates executable on the hardware
from qiskit.transpiler.preset_passmanagers import generate_preset_pass_manager
 
pm = generate_preset_pass_manager(backend=real_backend, optimization_level=1)
target_circuit = pm.run(grover_circuit)
 
target_circuit.draw(output="mpl", idle_wires=False)

Output:

Output of the previous code cell

By transpiling the circuit, it was converted to a circuit using the native basis gates of the device.

Step 3: Executing the circuit.

sampler = Sampler(real_backend)
job_real = sampler.run([target_circuit])
job_id = job_real.job_id()
print("job id:", job_id)

Output:

job id: cw69csv19rzg0080yfkg

# Check the job status
job_real.status()

Output:

'QUEUED'

# If the Notebook session got disconnected you can also check your job status by running the following code
job_real = service.job(job_id)  # Input your job-id between the quotations
job_real.status()

Output:

'DONE'

# Execute after job has successfully run
result_real = job_real.result()
print(result_real[0].data.meas.get_counts())

Output:

{'101': 540, '001': 2253, '011': 476, '000': 251, '110': 105, '100': 100, '010': 168, '111': 203}

Step 4: Post-processing the results.

plot_histogram(result_real[0].data.meas.get_counts())

Output:

Output of the previous code cell

4. A 3-qubit Grover Search

Now, let's try a 3-qubit Grover search example.

n = 3
qr = QuantumRegister(n, "q")
anc = QuantumRegister(1, "ancilla")
grover_circuit = QuantumCircuit(qr, anc)
# the number of iterations
num_iterations = 2
def oracle(circuit):
    circuit.mcx([0, 1, 2], 3)
    circuit.barrier()

This time, ∣111⟩|111\rangle is the "good" state.

# Let's do Grover search
initialization(grover_circuit)
 
for i in range(0, num_iterations):
    oracle(grover_circuit)
    diffusion(grover_circuit)
 
# Clear the ancilla bit
grover_circuit.h(n)
grover_circuit.x(n)
 
 
grover_circuit.draw(output="mpl", idle_wires=False)

Output:

Output of the previous code cell 
grover_circuit.measure_all()
 
# Define backend
backend = AerSimulator()
 
# Transpile to backend
pm = generate_preset_pass_manager(backend=backend, optimization_level=1)
isa_qc = pm.run(grover_circuit)
 
# Run the job
sampler = Sampler(backend)
job = sampler.run([isa_qc], shots=1024)
result = job.result()
 
# Print the results
counts = result[0].data.meas.get_counts()
print(counts)
 
# Plot the counts in a histogram
plot_histogram(counts)

Output:

{'0111': 977, '0100': 11, '0001': 8, '0000': 8, '0011': 5, '0010': 12, '0110': 3}
Output of the previous code cell

∣0111⟩|0111\rangle is observed with the highest probability, as expected. Note that two iterations are optimal in this case. However, the probability of obtaining the correct answer is not 100%, which is usual in Grover's search.

What happens if we iterate 3 times?

Now, let's try iterating 3 times.

n = 3
qr = QuantumRegister(n, "q")
anc = QuantumRegister(1, "ancillary")
grover_circuit = QuantumCircuit(qr, anc)
# the number of iterations
num_iterations = 3
# Let's do Grover search
initialization(grover_circuit)
 
for i in range(0, num_iterations):
    oracle(grover_circuit)
    diffusion(grover_circuit)
 
# Clear the ancilla bit
grover_circuit.h(n)
grover_circuit.x(n)
 
 
grover_circuit.draw(output="mpl", idle_wires=False, fold=-1, scale=0.5)

Output:

Output of the previous code cell 
grover_circuit.measure_all()
 
# Define backend
backend = AerSimulator()
 
# Transpile to backend
pm = generate_preset_pass_manager(backend=backend, optimization_level=1)
isa_qc = pm.run(grover_circuit)
 
# Run the job
sampler = Sampler(backend)
job = sampler.run([isa_qc], shots=1024)
result = job.result()
 
# Print the results
counts = result[0].data.meas.get_counts()
print(counts)
 
# Plot the counts in a histogram
plot_histogram(counts)

Output:

{'0010': 88, '0001': 103, '0000': 94, '0111': 334, '0100': 112, '0110': 106, '0101': 99, '0011': 88}
Output of the previous code cell

∣0111⟩|0111\rangle is still observed with the highest probability, though the probability of obtaining the correct answer has slightly decreased.

How about 4 times?

Now, let's try iterating 4 times.

n = 3
qr = QuantumRegister(n, "q")
anc = QuantumRegister(1, "ancillary")
grover_circuit = QuantumCircuit(qr, anc)
# the number of iterations
num_iterations = 4
# Let's do Grover search
initialization(grover_circuit)
 
for i in range(0, num_iterations):
    oracle(grover_circuit)
    diffusion(grover_circuit)
 
# Clear the ancilla bit
grover_circuit.h(n)
grover_circuit.x(n)
 
 
grover_circuit.draw(output="mpl", idle_wires=False, fold=-1, scale=0.5)

Output:

Output of the previous code cell 
grover_circuit.measure_all()
 
# Define backend
backend = AerSimulator()
 
# Transpile to backend
pm = generate_preset_pass_manager(backend=backend, optimization_level=1)
isa_qc = pm.run(grover_circuit)
 
# Run the job
sampler = Sampler(backend)
job = sampler.run([isa_qc])
result = job.result()
 
# Print the results
counts = result[0].data.meas.get_counts()
print(counts)
 
# Plot the counts in a histogram
plot_histogram(counts)

Output:

{'0110': 127, '0000': 135, '0001': 150, '0101': 164, '0010': 153, '0011': 131, '0100': 150, '0111': 14}
Output of the previous code cell

∣0111⟩|0111\rangle is observed with the lowest probability, and the probability of obtaining the correct answer has decreased further. This demonstrates the importance of choosing the optimal number of iterations for Grover's algorithm to achieve the best results.

# See the version of Qiskit
import qiskit
 
qiskit.__version__

Output:

'2.0.2'
Was this page helpful?
Report a bug or request content on GitHub.