qml.GateFabric

class GateFabric(weights, wires, init_state, include_pi=False, id=None)

Bases: Operation

Implements a local, expressive, and quantum-number-preserving ansatz proposed by Anselmetti et al. (2021).

This template prepares the $N$-qubit trial state by applying $D$ layers of gate-fabric blocks $\hat{U}_{GF}(\vec{\theta},\vec{\phi})$ to the Hartree-Fock state in the Jordan-Wigner basis

$$\vert \Psi(\vec{\theta},\vec{\phi})\rangle = \hat{U}_{GF}^{(D)}(\vec{\theta}_{D},\vec{\phi}_{D}) \ldots \hat{U}_{GF}^{(2)}(\vec{\theta}_{2},\vec{\phi}_{2}) \hat{U}_{GF}^{(1)}(\vec{\theta}_{1},\vec{\phi}_{1}) \vert HF \rangle,$$

where each of the gate fabric blocks $\hat{U}_{GF}(\vec{\theta},\vec{\phi})$ is comprised of two-parameter four-qubit gates $\hat{Q}(\theta, \phi)$ that act on four nearest-neighbour qubits. The circuit implementing a single layer of the gate fabric block for $N = 8$ is shown in the figure below:

The gate element $\hat{Q}(\theta, \phi)$ (Anselmetti et al. (2021)) is composed of a four-qubit spin-adapted spatial orbital rotation gate, which is implemented by the OrbitalRotation() operation and a four-qubit diagonal pair-exchange gate, which is equivalent to the DoubleExcitation() operation. In addition to these two gates, the gate element $\hat{Q}(\theta, \phi)$ can also include an optional constant $\hat{\Pi} \in \{\hat{I}, \text{OrbitalRotation}(\pi)\}$ gate.

The four-qubit DoubleExcitation() and OrbitalRotation() gates given here are equivalent to the $\text{QNP}_{PX}(\theta)$ and $\text{QNP}_{OR}(\phi)$ gates presented in Anselmetti et al. (2021), respectively. Moreover, regardless of the choice of $\hat{\Pi}$, this gate fabric will exactly preserve the number of particles and total spin of the state.

Parameters:

• weights (tensor_like) – Array of weights of shape (D, L, 2), where D is the number of gate fabric layers and L = N/2-1 is the number of $\hat{Q}(\theta, \phi)$ gates per layer with N being the total number of qubits.

• wires (Iterable) – wires that the template acts on.

• init_state (tensor_like) – iterable of shape (len(wires),), representing the input Hartree-Fock state in the Jordan-Wigner representation.

• include_pi (boolean) – If True, the optional constant $\hat{\Pi}$ gate is set to $\text{OrbitalRotation}(\pi)$. Default value is $\hat{I}$.

Usage Details

1. The number of wires $N$ has to be equal to the number of spin-orbitals included in the active space, and should be even.

2. The number of trainable parameters scales linearly with the number of layers as $2 D (N/2-1)$.

An example of how to use this template is shown below:

import pennylane as qml
from pennylane import numpy as np

# Build the electronic Hamiltonian
symbols = ["H", "H"]
coordinates = np.array([0.0, 0.0, -0.6614, 0.0, 0.0, 0.6614])
H, qubits = qml.qchem.molecular_hamiltonian(symbols, coordinates)

# Define the Hartree-Fock state
electrons = 2
ref_state = qml.qchem.hf_state(electrons, qubits)

# Define the device
dev = qml.device('default.qubit', wires=qubits)

# Define the ansatz
@qml.qnode(dev)
def ansatz(weights):
    qml.GateFabric(weights, wires=[0,1,2,3],
                init_state=ref_state, include_pi=True)
    return qml.expval(H)

# Get the shape of the weights for this template
layers = 2
shape = qml.GateFabric.shape(n_layers=layers, n_wires=qubits)

# Initialize the weight tensors
np.random.seed(42)
weights = np.random.random(size=shape)

# Define the optimizer
opt = qml.GradientDescentOptimizer(stepsize=0.4)

# Store the values of the cost function
energy = [ansatz(weights)]

# Store the values of the circuit weights
angle = [weights]

max_iterations = 100
conv_tol = 1e-06

for n in range(max_iterations):
    weights, prev_energy = opt.step_and_cost(ansatz, weights)
    energy.append(ansatz(weights))
    angle.append(weights)
    conv = np.abs(energy[-1] - prev_energy)

    if n % 2 == 0:
        print(f"Step = {n},  Energy = {energy[-1]:.8f} Ha")

    if conv <= conv_tol:
        break

print("\n" f"Final value of the ground-state energy = {energy[-1]:.8f} Ha")
print("\n" f"Optimal value of the circuit parameters = {angle[-1]}")

Step = 0,  Energy = -0.87007254 Ha
Step = 2,  Energy = -1.13107530 Ha
Step = 4,  Energy = -1.13611971 Ha
Step = 6,  Energy = -1.13618810 Ha

Final value of the ground-state energy = -1.13618903 Ha

Optimal value of the circuit parameters = [[[ 0.60328427  0.41850407]]
[[ 0.85581129 -0.24522642]]]

Parameter shape

The shape of the weights argument can be computed by the static method shape() and used when creating randomly initialised weight tensors:

shape = GateFabric.shape(n_layers=2, n_wires=4)
weights = np.random.random(size=shape)

>>> weights.shape
(2, 1, 2)

Attributes

arithmetic_depth	Arithmetic depth of the operator.
basis	The basis of an operation, or for controlled gates, of the target operation.
batch_size	Batch size of the operator if it is used with broadcasted parameters.
control_wires	Control wires of the operator.
grad_method
grad_recipe	Gradient recipe for the parameter-shift method.
has_adjoint
has_decomposition
has_diagonalizing_gates
has_generator
has_matrix
has_qfunc_decomposition
has_sparse_matrix
hash	Integer hash that uniquely represents the operator.
hyperparameters	Dictionary of non-trainable variables that this operation depends on.
id	Custom string to label a specific operator instance.
is_hermitian	This property determines if an operator is hermitian.
name	String for the name of the operator.
ndim_params	Number of dimensions per trainable parameter of the operator.
num_params	Number of trainable parameters that the operator depends on.
num_wires	Number of wires the operator acts on.
parameter_frequencies	Returns the frequencies for each operator parameter with respect to an expectation value of the form $\langle \psi | U(\mathbf{p})^\dagger \hat{O} U(\mathbf{p})|\psi\rangle$.
parameters	Trainable parameters that the operator depends on.
pauli_rep	A PauliSentence representation of the Operator, or None if it doesn't have one.
resource_keys
resource_params	A dictionary containing the minimal information needed to compute a resource estimate of the operator's decomposition.
wires	Wires that the operator acts on.

arithmetic_depth

Arithmetic depth of the operator.

basis

The basis of an operation, or for controlled gates, of the target operation. If not None, should take a value of "X", "Y", or "Z".

For example, X and CNOT have basis = "X", whereas ControlledPhaseShift and RZ have basis = "Z".

Type:

str or None

batch_size

Batch size of the operator if it is used with broadcasted parameters.

The batch_size is determined based on ndim_params and the provided parameters for the operator. If (some of) the latter have an additional dimension, and this dimension has the same size for all parameters, its size is the batch size of the operator. If no parameter has an additional dimension, the batch size is None.

Returns:

Size of the parameter broadcasting dimension if present, else None.

Return type:

int or None

control_wires

Control wires of the operator.

For operations that are not controlled, this is an empty Wires object of length 0.

Returns:

The control wires of the operation.

Return type:

Wires

grad_method = None

grad_recipe = None

Gradient recipe for the parameter-shift method.

This is a tuple with one nested list per operation parameter. For parameter $\phi_k$, the nested list contains elements of the form $[c_i, a_i, s_i]$ where $i$ is the index of the term, resulting in a gradient recipe of

$$\frac{\partial}{\partial\phi_k}f = \sum_{i} c_i f(a_i \phi_k + s_i).$$

If None, the default gradient recipe containing the two terms $[c_0, a_0, s_0]=[1/2, 1, \pi/2]$ and $[c_1, a_1, s_1]=[-1/2, 1, -\pi/2]$ is assumed for every parameter.

Type:

tuple(Union(list[list[float]], None)) or None

has_adjoint = False

has_decomposition = True

has_diagonalizing_gates = False

has_generator = False

has_matrix = False

has_qfunc_decomposition = False

has_sparse_matrix = False

hash

Integer hash that uniquely represents the operator.

Type:

int

hyperparameters

Dictionary of non-trainable variables that this operation depends on.

Type:

dict

id

Custom string to label a specific operator instance.

is_hermitian

This property determines if an operator is hermitian.

name

String for the name of the operator.

ndim_params

Number of dimensions per trainable parameter of the operator.

By default, this property returns the numbers of dimensions of the parameters used for the operator creation. If the parameter sizes for an operator subclass are fixed, this property can be overwritten to return the fixed value.

Returns:

Number of dimensions for each trainable parameter.

Return type:

tuple

num_params

num_wires = None

Number of wires the operator acts on.

parameter_frequencies

Returns the frequencies for each operator parameter with respect to an expectation value of the form $\langle \psi | U(\mathbf{p})^\dagger \hat{O} U(\mathbf{p})|\psi\rangle$.

These frequencies encode the behaviour of the operator $U(\mathbf{p})$ on the value of the expectation value as the parameters are modified. For more details, please see the pennylane.fourier module.

Returns:

Tuple of frequencies for each parameter. Note that only non-negative frequency values are returned.

Return type:

list[tuple[int or float]]

Example

>>> op = qml.CRot(0.4, 0.1, 0.3, wires=[0, 1])
>>> op.parameter_frequencies
[(0.5, 1), (0.5, 1), (0.5, 1)]

For operators that define a generator, the parameter frequencies are directly related to the eigenvalues of the generator:

>>> op = qml.ControlledPhaseShift(0.1, wires=[0, 1])
>>> op.parameter_frequencies
[(1,)]
>>> gen = qml.generator(op, format="observable")
>>> gen_eigvals = qml.eigvals(gen)
>>> qml.gradients.eigvals_to_frequencies(tuple(gen_eigvals))
(1.0,)

For more details on this relationship, see eigvals_to_frequencies().

parameters

Trainable parameters that the operator depends on.

pauli_rep

A PauliSentence representation of the Operator, or None if it doesn’t have one.

resource_keys = {}

resource_params

A dictionary containing the minimal information needed to compute a resource estimate of the operator’s decomposition.

The keys of this dictionary should match the resource_keys attribute of the operator class. Two instances of the same operator type should have identical resource_params iff their decompositions exhibit the same counts for each gate type, even if the individual gate parameters differ.

Examples

The MultiRZ has non-empty resource_keys:

>>> qml.MultiRZ.resource_keys
{"num_wires"}

The resource_params of an instance of MultiRZ will contain the number of wires:

>>> op = qml.MultiRZ(0.5, wires=[0, 1])
>>> op.resource_params
{"num_wires": 2}

Note that another MultiRZ may have different parameters but the same resource_params:

>>> op2 = qml.MultiRZ(0.7, wires=[1, 2])
>>> op2.resource_params
{"num_wires": 2}

wires

Wires that the operator acts on.

Returns:

wires

Return type:

Wires

Methods

adjoint()	Create an operation that is the adjoint of this one.
compute_decomposition(weights, wires, ...)	Representation of the operator as a product of other operators.
compute_diagonalizing_gates(*params, wires, ...)	Sequence of gates that diagonalize the operator in the computational basis (static method).
compute_eigvals(*params, **hyperparams)	Eigenvalues of the operator in the computational basis (static method).
compute_matrix(*params, **hyperparams)	Representation of the operator as a canonical matrix in the computational basis (static method).
compute_qfunc_decomposition(*args, ...)	Experimental method to compute the dynamic decomposition of the operator with program capture enabled.
compute_sparse_matrix(*params[, format])	Representation of the operator as a sparse matrix in the computational basis (static method).
decomposition()	Representation of the operator as a product of other operators.
diagonalizing_gates()	Sequence of gates that diagonalize the operator in the computational basis.
eigvals()	Eigenvalues of the operator in the computational basis.
generator()	Generator of an operator that is in single-parameter-form.
label([decimals, base_label, cache])	A customizable string representation of the operator.
map_wires(wire_map)	Returns a copy of the current operator with its wires changed according to the given wire map.
matrix([wire_order])	Representation of the operator as a matrix in the computational basis.
pow(z)	A list of new operators equal to this one raised to the given power.
queue([context])	Append the operator to the Operator queue.
shape(n_layers, n_wires)	Returns the shape of the weight tensor required for this template.
simplify()	Reduce the depth of nested operators to the minimum.
single_qubit_rot_angles()	The parameters required to implement a single-qubit gate as an equivalent Rot gate, up to a global phase.
sparse_matrix([wire_order, format])	Representation of the operator as a sparse matrix in the computational basis.
terms()	Representation of the operator as a linear combination of other operators.

adjoint()

Create an operation that is the adjoint of this one.

Adjointed operations are the conjugated and transposed version of the original operation. Adjointed ops are equivalent to the inverted operation for unitary gates.

Returns:

The adjointed operation.

static compute_decomposition(weights, wires, init_state, include_pi)

Representation of the operator as a product of other operators.

$$O = O_1 O_2 \dots O_n.$$

See also

decomposition().

Parameters:

• weights (tensor_like) – Array of weights of shape (D, L, 2), where D is the number of gate fabric layers and L = N/2-1 is the number of $\hat{Q}(\theta, \phi)$ gates per layer with N being the total number of qubits.

• wires (Any or Iterable[Any]) – wires that the operator acts on.

• init_state (tensor_like) – iterable of shape (len(wires),), representing the input Hartree-Fock state in the Jordan-Wigner representation.

• include_pi (boolean) – If True, the optional constant $\hat{\Pi}$ gate is set to $\text{OrbitalRotation}(\pi)$. Default value is $\hat{I}$.

Returns:

decomposition of the operator

Return type:

list[.Operator]

Example

>>> weights = torch.tensor([[[0.3, 1.]]])
>>> qml.GateFabric.compute_decomposition(weights, wires=["a", "b", "c", "d"], init_state=[0, 1, 0, 1], include_pi=False)
[BasisEmbedding(wires=['a', 'b', 'c', 'd']),
DoubleExcitation(tensor(0.3000), wires=['a', 'b', 'c', 'd']),
OrbitalRotation(tensor(1.), wires=['a', 'b', 'c', 'd'])]

static compute_diagonalizing_gates(*params, wires, **hyperparams)

Sequence of gates that diagonalize the operator in the computational basis (static method).

Given the eigendecomposition $O = U \Sigma U^{\dagger}$ where $\Sigma$ is a diagonal matrix containing the eigenvalues, the sequence of diagonalizing gates implements the unitary $U^{\dagger}$.

The diagonalizing gates rotate the state into the eigenbasis of the operator.

See also

diagonalizing_gates().

Parameters:

• params (list) – trainable parameters of the operator, as stored in the parameters attribute

• wires (Iterable[Any], Wires) – wires that the operator acts on

• hyperparams (dict) – non-trainable hyperparameters of the operator, as stored in the hyperparameters attribute

Returns:

list of diagonalizing gates

Return type:

list[.Operator]

static compute_eigvals(*params, **hyperparams)

Eigenvalues of the operator in the computational basis (static method).

If diagonalizing_gates are specified and implement a unitary $U^{\dagger}$, the operator can be reconstructed as

$$O = U \Sigma U^{\dagger},$$

where $\Sigma$ is the diagonal matrix containing the eigenvalues.

Otherwise, no particular order for the eigenvalues is guaranteed.

See also

Operator.eigvals() and qml.eigvals()

Parameters:

• *params (list) – trainable parameters of the operator, as stored in the parameters attribute

• **hyperparams (dict) – non-trainable hyperparameters of the operator, as stored in the hyperparameters attribute

Returns:

eigenvalues

Return type:

tensor_like

static compute_matrix(*params, **hyperparams)

Representation of the operator as a canonical matrix in the computational basis (static method).

The canonical matrix is the textbook matrix representation that does not consider wires. Implicitly, this assumes that the wires of the operator correspond to the global wire order.

See also

Operator.matrix() and qml.matrix()

Parameters:

• *params (list) – trainable parameters of the operator, as stored in the parameters attribute

• **hyperparams (dict) – non-trainable hyperparameters of the operator, as stored in the hyperparameters attribute

Returns:

matrix representation

Return type:

tensor_like

static compute_qfunc_decomposition(*args, **hyperparameters)

Experimental method to compute the dynamic decomposition of the operator with program capture enabled.

When the program capture feature is enabled with qml.capture.enable(), the decomposition of the operator is computed with this method if it is defined. Otherwise, the compute_decomposition() method is used.

The exception to this rule is when the operator is returned from the compute_decomposition() method of another operator, in which case the decomposition is performed with compute_decomposition() (even if this method is defined), and not with this method.

When compute_qfunc_decomposition is defined for an operator, the control flow operations within the method (specifying the decomposition of the operator) are recorded in the JAX representation.

Note

This method is experimental and subject to change.

See also

compute_decomposition().

Parameters:

• *args (list) – positional arguments passed to the operator, including trainable parameters and wires

• **hyperparameters (dict) – non-trainable hyperparameters of the operator, as stored in the hyperparameters attribute

static compute_sparse_matrix(*params, format='csr', **hyperparams)

Representation of the operator as a sparse matrix in the computational basis (static method).

The canonical matrix is the textbook matrix representation that does not consider wires. Implicitly, this assumes that the wires of the operator correspond to the global wire order.

See also

sparse_matrix()

Parameters:

• *params (list) – trainable parameters of the operator, as stored in the parameters attribute

• format (str) – format of the returned scipy sparse matrix, for example ‘csr’

• **hyperparams (dict) – non-trainable hyperparameters of the operator, as stored in the hyperparameters attribute

Returns:

sparse matrix representation

Return type:

scipy.sparse._csr.csr_matrix

decomposition()

Representation of the operator as a product of other operators.

$$O = O_1 O_2 \dots O_n$$

A DecompositionUndefinedError is raised if no representation by decomposition is defined.

See also

compute_decomposition().

Returns:

decomposition of the operator

Return type:

list[Operator]

diagonalizing_gates()

Sequence of gates that diagonalize the operator in the computational basis.

Given the eigendecomposition $O = U \Sigma U^{\dagger}$ where $\Sigma$ is a diagonal matrix containing the eigenvalues, the sequence of diagonalizing gates implements the unitary $U^{\dagger}$.

The diagonalizing gates rotate the state into the eigenbasis of the operator.

A DiagGatesUndefinedError is raised if no representation by decomposition is defined.

See also

compute_diagonalizing_gates().

Returns:

a list of operators

Return type:

list[.Operator] or None

eigvals()

Eigenvalues of the operator in the computational basis.

If diagonalizing_gates are specified and implement a unitary $U^{\dagger}$, the operator can be reconstructed as

$$O = U \Sigma U^{\dagger},$$

where $\Sigma$ is the diagonal matrix containing the eigenvalues.

Otherwise, no particular order for the eigenvalues is guaranteed.

Note

When eigenvalues are not explicitly defined, they are computed automatically from the matrix representation. Currently, this computation is not differentiable.

A EigvalsUndefinedError is raised if the eigenvalues have not been defined and cannot be inferred from the matrix representation.

See also

compute_eigvals()

Returns:

eigenvalues

Return type:

tensor_like

generator()

Generator of an operator that is in single-parameter-form.

For example, for operator

$$U(\phi) = e^{i\phi (0.5 Y + Z\otimes X)}$$

we get the generator

>>> U.generator()
0.5 * Y(0) + Z(0) @ X(1)

The generator may also be provided in the form of a dense or sparse Hamiltonian (using LinearCombination and SparseHamiltonian respectively).

label(decimals=None, base_label=None, cache=None)

A customizable string representation of the operator.

Parameters:

• decimals=None (int) – If None, no parameters are included. Else, specifies how to round the parameters.

• base_label=None (str) – overwrite the non-parameter component of the label

• cache=None (dict) – dictionary that carries information between label calls in the same drawing

Returns:

label to use in drawings

Return type:

str

Example:

>>> op = qml.RX(1.23456, wires=0)
>>> op.label()
"RX"
>>> op.label(base_label="my_label")
"my_label"
>>> op = qml.RX(1.23456, wires=0, id="test_data")
>>> op.label()
"RX("test_data")"
>>> op.label(decimals=2)
"RX\n(1.23,"test_data")"
>>> op.label(base_label="my_label")
"my_label("test_data")"
>>> op.label(decimals=2, base_label="my_label")
"my_label\n(1.23,"test_data")"

If the operation has a matrix-valued parameter and a cache dictionary is provided, unique matrices will be cached in the 'matrices' key list. The label will contain the index of the matrix in the 'matrices' list.

>>> op2 = qml.QubitUnitary(np.eye(2), wires=0)
>>> cache = {'matrices': []}
>>> op2.label(cache=cache)
'U(M0)'
>>> cache['matrices']
[tensor([[1., 0.],
 [0., 1.]], requires_grad=True)]
>>> op3 = qml.QubitUnitary(np.eye(4), wires=(0,1))
>>> op3.label(cache=cache)
'U(M1)'
>>> cache['matrices']
[tensor([[1., 0.],
        [0., 1.]], requires_grad=True),
tensor([[1., 0., 0., 0.],
        [0., 1., 0., 0.],
        [0., 0., 1., 0.],
        [0., 0., 0., 1.]], requires_grad=True)]

map_wires(wire_map)

Returns a copy of the current operator with its wires changed according to the given wire map.

Parameters:

wire_map (dict) – dictionary containing the old wires as keys and the new wires as values

Returns:

new operator

Return type:

.Operator

matrix(wire_order=None)

Representation of the operator as a matrix in the computational basis.

If wire_order is provided, the numerical representation considers the position of the operator’s wires in the global wire order. Otherwise, the wire order defaults to the operator’s wires.

If the matrix depends on trainable parameters, the result will be cast in the same autodifferentiation framework as the parameters.

A MatrixUndefinedError is raised if the matrix representation has not been defined.

See also

compute_matrix()

Parameters:

wire_order (Iterable) – global wire order, must contain all wire labels from the operator’s wires

Returns:

matrix representation

Return type:

tensor_like

pow(z)

A list of new operators equal to this one raised to the given power.

Parameters:

z (float) – exponent for the operator

Returns:

list[Operator]

queue(context=<class 'pennylane.queuing.QueuingManager'>)

Append the operator to the Operator queue.

static shape(n_layers, n_wires)

Returns the shape of the weight tensor required for this template.

Parameters:

• n_layers (int) – number of layers

• n_wires (int) – number of qubits

Returns:

shape

Return type:

tuple[int]

simplify()

Reduce the depth of nested operators to the minimum.

Returns:

simplified operator

Return type:

.Operator

single_qubit_rot_angles()

The parameters required to implement a single-qubit gate as an equivalent Rot gate, up to a global phase.

Returns:

A list of values $[\phi, \theta, \omega]$ such that $RZ(\omega) RY(\theta) RZ(\phi)$ is equivalent to the original operation.

Return type:

tuple[float, float, float]

sparse_matrix(wire_order=None, format='csr')

Representation of the operator as a sparse matrix in the computational basis.

If wire_order is provided, the numerical representation considers the position of the operator’s wires in the global wire order. Otherwise, the wire order defaults to the operator’s wires.

A SparseMatrixUndefinedError is raised if the sparse matrix representation has not been defined.

See also

compute_sparse_matrix()

Parameters:

• wire_order (Iterable) – global wire order, must contain all wire labels from the operator’s wires

• format (str) – format of the returned scipy sparse matrix, for example ‘csr’

Returns:

sparse matrix representation

Return type:

scipy.sparse._csr.csr_matrix

terms()

Representation of the operator as a linear combination of other operators.

$$O = \sum_i c_i O_i$$

A TermsUndefinedError is raised if no representation by terms is defined.

Returns:

list of coefficients $c_i$ and list of operations $O_i$

Return type:

tuple[list[tensor_like or float], list[.Operation]]

code/api/pennylane.GateFabric

 

Download Python script

 

Download Notebook

 

View on GitHub