-
Custom gradient transforms can now be created using the new
@qml.gradients.gradient_transform
decorator on a batch-tape transform. (#1589)Quantum gradient transforms are a specific case of
qml.batch_transform
. All quantum gradient transforms accept a tape, and outputs a batch of tapes to be independently executed on a quantum device, alongside a post-processing function that returns the result.Furthermore, a smart default expansion function is provided, which automatically expands tape operations which are not differentiable prior to applying the quantum gradient. All gradient transforms in
qml.gradients
are now decorated with this decorator.Supported gradient transforms must be of the following form:
@qml.gradients.gradient_transform def my_custom_gradient(tape, argnum=None, **kwargs): ... return gradient_tapes, processing_fn
Once defined, quantum gradient transforms can be applied directly to QNodes:
>>> @qml.qnode(dev) ... def circuit(x): ... qml.RX(x, wires=0) ... qml.CNOT(wires=[0, 1]) ... return qml.expval(qml.PauliZ(0)) >>> circuit(0.3) tensor(0.95533649, requires_grad=True) >>> qml.gradients.param_shift(circuit)(0.5) array([[-0.47942554]])
Quantum gradient transforms are fully differentiable, allowing higher order derivatives to be accessed:
>>> qml.grad(qml.gradients.param_shift(circuit))(0.5) tensor(-0.87758256, requires_grad=True)
-
The ability to define batch transforms has been added via the new
@qml.batch_transform
decorator. (#1493)A batch transform is a transform that takes a single tape or QNode as input, and executes multiple tapes or QNodes independently. The results may then be post-processed before being returned.
For example, consider the following batch transform:
@qml.batch_transform def my_transform(tape, a, b): """Generates two tapes, one with all RX replaced with RY, and the other with all RX replaced with RZ.""" tape1 = qml.tape.JacobianTape() tape2 = qml.tape.JacobianTape() # loop through all operations on the input tape for op in tape.operations + tape.measurements: if op.name == "RX": with tape1: qml.RY(a * qml.math.abs(op.parameters[0]), wires=op.wires) with tape2: qml.RZ(b * qml.math.abs(op.parameters[0]), wires=op.wires) else: for t in [tape1, tape2]: with t: qml.apply(op) def processing_fn(results): return qml.math.sum(qml.math.stack(results)) return [tape1, tape2], processing_fn
We can transform a QNode directly using decorator syntax:
>>> @my_transform(0.65, 2.5) ... @qml.qnode(dev) ... def circuit(x): ... qml.Hadamard(wires=0) ... qml.RX(x, wires=0) ... return qml.expval(qml.PauliX(0)) >>> print(circuit(-0.5)) 1.2629730888100839
Batch tape transforms are fully differentiable:
>>> gradient = qml.grad(circuit)(-0.5) >>> print(gradient) 2.5800122591960153
Batch transforms can also be applied to existing QNodes,
>>> new_qnode = my_transform(existing_qnode, *transform_weights) >>> new_qnode(weights)
or to tapes (in which case, the processed tapes and classical post-processing functions are returned):
>>> tapes, fn = my_transform(tape, 0.65, 2.5) >>> from pennylane.interfaces.batch import execute >>> dev = qml.device("default.qubit", wires=1) >>> res = execute(tapes, dev, interface="autograd", gradient_fn=qml.gradients.param_shift) 1.2629730888100839
-
Added a new
SISWAP
operation and aSQISW
alias with support to thedefault_qubit
device. #1563 -
The
RotosolveOptimizer
now can tackle general parametrized circuits, and is no longer restricted to single-qubit Pauli rotations. (#1489)This includes:
- layers of gates controlled by the same parameter,
- controlled variants of parametrized gates, and
- Hamiltonian time evolution.
Note that the eigenvalue spectrum of the gate generator needs to be known to use
RotosolveOptimizer
for a general gate, and it is required to produce equidistant frequencies. For details see Vidal and Theis, 2018 and Wierichs, Izaac, Wang, Lin 2021.Consider a circuit with a mixture of Pauli rotation gates, controlled Pauli rotations, and single-parameter layers of Pauli rotations:
dev = qml.device('default.qubit', wires=3, shots=None) @qml.qnode(dev) def cost_function(rot_param, layer_par, crot_param): for i, par in enumerate(rot_param): qml.RX(par, wires=i) for w in dev.wires: qml.RX(layer_par, wires=w) for i, par in enumerate(crot_param): qml.CRY(par, wires=[i, (i+1) % 3]) return qml.expval(qml.PauliZ(0) @ qml.PauliZ(1) @ qml.PauliZ(2))
This cost function has one frequency for each of the first
RX
rotation angles, three frequencies for the layer ofRX
gates that depend onlayer_par
, and two frequencies for each of theCRY
gate parameters. Rotosolve can then be used to minimize thecost_function
:# Initial parameters init_param = [ np.array([0.3, 0.2, 0.67], requires_grad=True), np.array(1.1, requires_grad=True), np.array([-0.2, 0.1, -2.5], requires_grad=True), ] # Numbers of frequencies per parameter num_freqs = [[1, 1, 1], 3, [2, 2, 2]] opt = qml.RotosolveOptimizer() param = init_param.copy()
In addition, the optimization technique for the Rotosolve substeps can be chosen via the
optimizer
andoptimizer_kwargs
keyword arguments and the minimized cost of the intermediate univariate reconstructions can be read out viafull_output
, including the cost after the full Rotosolve step:for step in range(3): param, cost, sub_cost = opt.step_and_cost( cost_function, *param, num_freqs=num_freqs, full_output=True, optimizer="brute", ) print(f"Cost before step: {cost}") print(f"Minimization substeps: {np.round(sub_cost, 6)}")
Cost before step: 0.042008210392535605 Minimization substeps: [-0.230905 -0.863336 -0.980072 -0.980072 -1. -1. -1. ] Cost before step: -0.999999999068121 Minimization substeps: [-1. -1. -1. -1. -1. -1. -1.] Cost before step: -1.0 Minimization substeps: [-1. -1. -1. -1. -1. -1. -1.]
For usage details please consider the docstring.
-
The
frobenius_inner_product
function has been moved to theqml.math
module, and is now differentiable using all autodiff frameworks. (#1388) -
Vector-Jacobian product transforms have been added to the
qml.gradients
package. (#1494)The new transforms include:
qml.gradients.vjp
qml.gradients.batch_vjp
-
The Hamiltonian can now store grouping information, which can be accessed by a device to speed up computations of the expectation value of a Hamiltonian. (#1515)
obs = [qml.PauliX(0), qml.PauliX(1), qml.PauliZ(0)] coeffs = np.array([1., 2., 3.]) H = qml.Hamiltonian(coeffs, obs, grouping_type='qwc')
Initialization with a
grouping_type
other thanNone
stores the indices required to make groups of commuting observables and their coefficients.>>> H.grouping_indices [[0, 1], [2]]
-
Hamiltonians are now trainable with respect to their coefficients. (#1483)
from pennylane import numpy as np dev = qml.device("default.qubit", wires=2) @qml.qnode(dev) def circuit(coeffs, param): qml.RX(param, wires=0) qml.RY(param, wires=0) return qml.expval( qml.Hamiltonian(coeffs, [qml.PauliX(0), qml.PauliZ(0)], simplify=True) ) coeffs = np.array([-0.05, 0.17]) param = np.array(1.7) grad_fn = qml.grad(circuit)
>>> grad_fn(coeffs, param) (array([-0.12777055, 0.0166009 ]), array(0.0917819))
-
Support for differentiable execution of batches of circuits has been added, via the beta
pennylane.interfaces.batch
module. (#1501) (#1508) (#1542) (#1549)For example:
from pennylane.interfaces.batch import execute def cost_fn(x): with qml.tape.JacobianTape() as tape1: qml.RX(x[0], wires=[0]) qml.RY(x[1], wires=[1]) qml.CNOT(wires=[0, 1]) qml.var(qml.PauliZ(0) @ qml.PauliX(1)) with qml.tape.JacobianTape() as tape2: qml.RX(x[0], wires=0) qml.RY(x[0], wires=1) qml.CNOT(wires=[0, 1]) qml.probs(wires=1) result = execute( [tape1, tape2], dev, gradient_fn=qml.gradients.param_shift, interface="autograd" ) return result[0] + result[1][0, 0] res = qml.grad(cost_fn)(params)
-
The device test suite has been expanded to cover more qubit operations and observables. (#1510)
-
The
MultiControlledX
class now inherits fromOperation
instead ofControlledQubitUnitary
which makes theMultiControlledX
gate a non-parameterized gate. (#1557) -
The
utils.sparse_hamiltonian
function can now deal with non-integer wire labels, and it throws an error for the edge case of observables that are created from multi-qubit operations. (#1550) -
Added the matrix attribute to
qml.templates.subroutines.GroverOperator
(#1553) -
The
tape.to_openqasm()
method now has ameasure_all
argument that specifies whether the serialized OpenQASM script includes computational basis measurements on all of the qubits or just those specified by the tape. (#1559) -
An error is raised when no arguments are passed to a
qml.operation.Observable
to inform the user about specifying wires. (#1547) -
The Hamiltonian class was moved to the
ops/qubit
folder from thevqe
module, since it is now an observable. (#1534) -
The
group_observables
transform is now differentiable. (#1483)For example:
import jax from jax import numpy as jnp coeffs = jnp.array([1., 2., 3.]) obs = [PauliX(wires=0), PauliX(wires=1), PauliZ(wires=1)] def group(coeffs, select=None): _, grouped_coeffs = qml.grouping.group_observables(obs, coeffs) # in this example, grouped_coeffs is a list of two jax tensors # [DeviceArray([1., 2.], dtype=float32), DeviceArray([3.], dtype=float32)] return grouped_coeffs[select] jac_fn = jax.jacobian(group)
>>> jac_fn(coeffs, select=0) [[1. 0. 0.] [0. 1. 0.]] >>> jac_fn(coeffs, select=1) [[0., 0., 1.]]
-
The tape does not verify any more that all Observables have owners in the annotated queue. (#1505)
This allows manipulation of Observables inside a tape context. An example is
expval(Tensor(qml.PauliX(0), qml.Identity(1)).prune())
which makes the expval an owner of the pruned tensor and its constituent observables, but leaves the original tensor in the queue without an owner. -
Create a separate requirements file for the CI issue , to have a separate requirements.txt (pinned) and requirements-ci.txt (unpinned). This latter would be used by the CI. (#1535)
-
The QFT operation is moved to template (#1548)
-
The
qml.ResetError
is now supported fordefault.mixed
device. (#1541)
- The class
qml.Interferometer
is deprecated and will be renamedqml.InterferometerUnitary
after one release cycle. (#1546)
-
Fix bug when computing expectations of Hamiltonians using TensorFlow. (#1586)
-
Fix bug when computing the specs of a circuit with a Hamiltonian observable. (#1533)
-
The
qml.Identity
operation is placed under the sections Qubit observables and CV observables. (#1576) -
Updated the documentation of
qml.grouping
,qml.kernels
andqml.qaoa
modules to present the list of functions first followed by the technical details of the module. (#1581) -
Recategorized Qubit operations into new and existing categories so that code for each operation is easier to locate. (#1566)
This release contains contributions from (in alphabetical order):
Vishnu Ajith, Akash Narayanan B, Thomas Bromley, Tanya Garg, Josh Izaac, Prateek Jain, Johannes Jakob Meyer, Pratul Saini, Maria Schuld, Ingrid Strandberg, David Wierichs, Vincent Wong.
-
PennyLane can now perform quantum circuit optimization using the top-level transform
qml.compile
. Thecompile
transform allows you to chain together sequences of tape and quantum function transforms into custom circuit optimization pipelines. (#1475)For example, take the following decorated quantum function:
dev = qml.device('default.qubit', wires=[0, 1, 2]) @qml.qnode(dev) @qml.compile() def qfunc(x, y, z): qml.Hadamard(wires=0) qml.Hadamard(wires=1) qml.Hadamard(wires=2) qml.RZ(z, wires=2) qml.CNOT(wires=[2, 1]) qml.RX(z, wires=0) qml.CNOT(wires=[1, 0]) qml.RX(x, wires=0) qml.CNOT(wires=[1, 0]) qml.RZ(-z, wires=2) qml.RX(y, wires=2) qml.PauliY(wires=2) qml.CZ(wires=[1, 2]) return qml.expval(qml.PauliZ(wires=0))
The default behaviour of
qml.compile
is to apply a sequence of three transforms:commute_controlled
,cancel_inverses
, and thenmerge_rotations
.>>> print(qml.draw(qfunc)(0.2, 0.3, 0.4)) 0: ──H───RX(0.6)──────────────────┤ ⟨Z⟩ 1: ──H──╭X────────────────────╭C──┤ 2: ──H──╰C────────RX(0.3)──Y──╰Z──┤
The
qml.compile
transform is flexible and accepts a custom pipeline of tape and quantum function transforms (you can even write your own!). For example, if we wanted to only push single-qubit gates through controlled gates and cancel adjacent inverses, we could do:from pennylane.transforms import commute_controlled, cancel_inverses pipeline = [commute_controlled, cancel_inverses] @qml.qnode(dev) @qml.compile(pipeline=pipeline) def qfunc(x, y, z): qml.Hadamard(wires=0) qml.Hadamard(wires=1) qml.Hadamard(wires=2) qml.RZ(z, wires=2) qml.CNOT(wires=[2, 1]) qml.RX(z, wires=0) qml.CNOT(wires=[1, 0]) qml.RX(x, wires=0) qml.CNOT(wires=[1, 0]) qml.RZ(-z, wires=2) qml.RX(y, wires=2) qml.PauliY(wires=2) qml.CZ(wires=[1, 2]) return qml.expval(qml.PauliZ(wires=0))
>>> print(qml.draw(qfunc)(0.2, 0.3, 0.4)) 0: ──H───RX(0.4)──RX(0.2)────────────────────────────┤ ⟨Z⟩ 1: ──H──╭X───────────────────────────────────────╭C──┤ 2: ──H──╰C────────RZ(0.4)──RZ(-0.4)──RX(0.3)──Y──╰Z──┤
The following compilation transforms have been added and are also available to use, either independently, or within a
qml.compile
pipeline:-
commute_controlled
: push commuting single-qubit gates through controlled operations. (#1464) -
cancel_inverses
: removes adjacent pairs of operations that cancel out. (#1455) -
merge_rotations
: combines adjacent rotation gates of the same type into a single gate, including controlled rotations. (#1455) -
single_qubit_fusion
: acts on all sequences of single-qubit operations in a quantum function, and converts each sequence to a singleRot
gate. (#1458)
For more details on
qml.compile
and the available compilation transforms, see the compilation documentation. -
-
Computational basis samples directly from the underlying device can now be returned directly from QNodes via
qml.sample()
. (#1441)dev = qml.device("default.qubit", wires=3, shots=5) @qml.qnode(dev) def circuit_1(): qml.Hadamard(wires=0) qml.Hadamard(wires=1) return qml.sample() @qml.qnode(dev) def circuit_2(): qml.Hadamard(wires=0) qml.Hadamard(wires=1) return qml.sample(wires=[0,2]) # no observable provided and wires specified
>>> print(circuit_1()) [[1, 0, 0], [1, 1, 0], [1, 0, 0], [0, 0, 0], [0, 1, 0]] >>> print(circuit_2()) [[1, 0], [1, 0], [1, 0], [0, 0], [0, 0]] >>> print(qml.draw(circuit_2)()) 0: ──H──╭┤ Sample[basis] 1: ──H──│┤ 2: ─────╰┤ Sample[basis]
-
The new
qml.apply
function can be used to add operations that might have already been instantiated elsewhere to the QNode and other queuing contexts: (#1433)op = qml.RX(0.4, wires=0) dev = qml.device("default.qubit", wires=2) @qml.qnode(dev) def circuit(x): qml.RY(x, wires=0) qml.apply(op) return qml.expval(qml.PauliZ(0))
>>> print(qml.draw(circuit)(0.6)) 0: ──RY(0.6)──RX(0.4)──┤ ⟨Z⟩
Previously instantiated measurements can also be applied to QNodes.
-
The new Device Tracker capabilities allows for flexible and versatile tracking of executions, even inside parameter-shift gradients. This functionality will improve the ease of monitoring large batches and remote jobs. (#1355)
dev = qml.device('default.qubit', wires=1, shots=100) @qml.qnode(dev, diff_method="parameter-shift") def circuit(x): qml.RX(x, wires=0) return qml.expval(qml.PauliZ(0)) x = np.array(0.1) with qml.Tracker(circuit.device) as tracker: qml.grad(circuit)(x)
>>> tracker.totals {'executions': 3, 'shots': 300, 'batches': 1, 'batch_len': 2} >>> tracker.history {'executions': [1, 1, 1], 'shots': [100, 100, 100], 'batches': [1], 'batch_len': [2]} >>> tracker.latest {'batches': 1, 'batch_len': 2}
Users can also provide a custom function to the
callback
keyword that gets called each time the information is updated. This functionality allows users to monitor remote jobs or large parameter-shift batches.>>> def shots_info(totals, history, latest): ... print("Total shots: ", totals['shots']) >>> with qml.Tracker(circuit.device, callback=shots_info) as tracker: ... qml.grad(circuit)(0.1) Total shots: 100 Total shots: 200 Total shots: 300 Total shots: 300
-
Docker support for building PennyLane with support for all interfaces (TensorFlow, Torch, and Jax), as well as device plugins and QChem, for GPUs and CPUs, has been added. (#1391)
The build process using Docker and
make
requires that the repository source code is cloned or downloaded from GitHub. Visit the the detailed description for an extended list of options.
-
Added a sparse Hamiltonian observable and the functionality to support computing its expectation value with
default.qubit
. (#1398)For example, the following QNode returns the expectation value of a sparse Hamiltonian:
dev = qml.device("default.qubit", wires=2) @qml.qnode(dev, diff_method="parameter-shift") def circuit(param, H): qml.PauliX(0) qml.SingleExcitation(param, wires=[0, 1]) return qml.expval(qml.SparseHamiltonian(H, [0, 1]))
We can execute this QNode, passing in a sparse identity matrix:
>>> print(circuit([0.5], scipy.sparse.eye(4).tocoo())) 0.9999999999999999
The expectation value of the sparse Hamiltonian is computed directly, which leads to executions that are faster by orders of magnitude. Note that "parameter-shift" is the only differentiation method that is currently supported when the observable is a sparse Hamiltonian.
-
VQE problems can now be intuitively set up by passing the Hamiltonian as an observable. (#1474)
dev = qml.device("default.qubit", wires=2) H = qml.Hamiltonian([1., 2., 3.], [qml.PauliZ(0), qml.PauliY(0), qml.PauliZ(1)]) w = qml.init.strong_ent_layers_uniform(1, 2, seed=1967) @qml.qnode(dev) def circuit(w): qml.templates.StronglyEntanglingLayers(w, wires=range(2)) return qml.expval(H)
>>> print(circuit(w)) -1.5133943637878295 >>> print(qml.grad(circuit)(w)) [[[-8.32667268e-17 1.39122955e+00 -9.12462052e-02] [ 1.02348685e-16 -7.77143238e-01 -1.74708049e-01]]]
Note that other measurement types like
var(H)
orsample(H)
, as well as multiple expectations likeexpval(H1), expval(H2)
are not supported. -
Added functionality to compute the sparse matrix representation of a
qml.Hamiltonian
object. (#1394)
-
A new gradients module
qml.gradients
has been added, which provides differentiable quantum gradient transforms. (#1476) (#1479) (#1486)Available quantum gradient transforms include:
qml.gradients.finite_diff
qml.gradients.param_shift
qml.gradients.param_shift_cv
For example,
>>> params = np.array([0.3,0.4,0.5], requires_grad=True) >>> with qml.tape.JacobianTape() as tape: ... qml.RX(params[0], wires=0) ... qml.RY(params[1], wires=0) ... qml.RX(params[2], wires=0) ... qml.expval(qml.PauliZ(0)) ... qml.var(qml.PauliZ(0)) >>> tape.trainable_params = {0, 1, 2} >>> gradient_tapes, fn = qml.gradients.finite_diff(tape) >>> res = dev.batch_execute(gradient_tapes) >>> fn(res) array([[-0.69688381, -0.32648317, -0.68120105], [ 0.8788057 , 0.41171179, 0.85902895]])
-
Grover Diffusion Operator template added. (#1442)
For example, if we have an oracle that marks the "all ones" state with a negative sign:
n_wires = 3 wires = list(range(n_wires)) def oracle(): qml.Hadamard(wires[-1]) qml.Toffoli(wires=wires) qml.Hadamard(wires[-1])
We can perform Grover's Search Algorithm:
dev = qml.device('default.qubit', wires=wires) @qml.qnode(dev) def GroverSearch(num_iterations=1): for wire in wires: qml.Hadamard(wire) for _ in range(num_iterations): oracle() qml.templates.GroverOperator(wires=wires) return qml.probs(wires)
We can see this circuit yields the marked state with high probability:
>>> GroverSearch(num_iterations=1) tensor([0.03125, 0.03125, 0.03125, 0.03125, 0.03125, 0.03125, 0.03125, 0.78125], requires_grad=True) >>> GroverSearch(num_iterations=2) tensor([0.0078125, 0.0078125, 0.0078125, 0.0078125, 0.0078125, 0.0078125, 0.0078125, 0.9453125], requires_grad=True)
-
A decomposition has been added to
QubitUnitary
that makes the single-qubit case fully differentiable in all interfaces. Furthermore, a quantum function transform,unitary_to_rot()
, has been added to decompose all single-qubit instances ofQubitUnitary
in a quantum circuit. (#1427)Instances of
QubitUnitary
may now be decomposed directly toRot
operations, orRZ
operations if the input matrix is diagonal. For example, let>>> U = np.array([ [-0.28829348-0.78829734j, 0.30364367+0.45085995j], [ 0.53396245-0.10177564j, 0.76279558-0.35024096j] ])
Then, we can compute the decomposition as:
>>> qml.QubitUnitary.decomposition(U, wires=0) [Rot(-0.24209530281458358, 1.1493817777199102, 1.733058145303424, wires=[0])]
We can also apply the transform directly to a quantum function, and compute the gradients of parameters used to construct the unitary matrices.
def qfunc_with_qubit_unitary(angles): z, x = angles[0], angles[1] Z_mat = np.array([[np.exp(-1j * z / 2), 0.0], [0.0, np.exp(1j * z / 2)]]) c = np.cos(x / 2) s = np.sin(x / 2) * 1j X_mat = np.array([[c, -s], [-s, c]]) qml.Hadamard(wires="a") qml.QubitUnitary(Z_mat, wires="a") qml.QubitUnitary(X_mat, wires="b") qml.CNOT(wires=["b", "a"]) return qml.expval(qml.PauliX(wires="a"))
>>> dev = qml.device("default.qubit", wires=["a", "b"]) >>> transformed_qfunc = qml.transforms.unitary_to_rot(qfunc_with_qubit_unitary) >>> transformed_qnode = qml.QNode(transformed_qfunc, dev) >>> input = np.array([0.3, 0.4], requires_grad=True) >>> transformed_qnode(input) tensor(0.95533649, requires_grad=True) >>> qml.grad(transformed_qnode)(input) array([-0.29552021, 0. ])
-
Ising YY gate functionality added. (#1358)
-
The tape does not verify any more that all Observables have owners in the annotated queue. (#1505)
This allows manipulation of Observables inside a tape context. An example is
expval(Tensor(qml.PauliX(0), qml.Identity(1)).prune())
which makes the expval an owner of the pruned tensor and its constituent observables, but leaves the original tensor in the queue without an owner. -
The
step
andstep_and_cost
methods ofQNGOptimizer
now accept a customgrad_fn
keyword argument to use for gradient computations. (#1487) -
The precision used by
default.qubit.jax
now matches the float precision indicated byfrom jax.config import config config.read('jax_enable_x64')
where
True
meansfloat64
/complex128
andFalse
meansfloat32
/complex64
. (#1485) -
The
./pennylane/ops/qubit.py
file is broken up into a folder of six separate files. (#1467) -
Changed to using commas as the separator of wires in the string representation of
qml.Hamiltonian
objects for multi-qubit terms. (#1465) -
Changed to using
np.object_
instead ofnp.object
as per the NumPy deprecations starting version 1.20. (#1466) -
Change the order of the covariance matrix and the vector of means internally in
default.gaussian
. (#1331) -
Added the
id
attribute to templates. (#1438) -
The
qml.math
module, for framework-agnostic tensor manipulation, has two new functions available: (#1490)-
qml.math.get_trainable_indices(sequence_of_tensors)
: returns the indices corresponding to trainable tensors in the input sequence. -
qml.math.unwrap(sequence_of_tensors)
: unwraps a sequence of tensor-like objects to NumPy arrays.
In addition, the behaviour of
qml.math.requires_grad
has been improved in order to correctly determine trainability during Autograd and JAX backwards passes. -
-
A new tape method,
tape.unwrap()
is added. This method is a context manager; inside the context, the tape's parameters are unwrapped to NumPy arrays and floats, and the trainable parameter indices are set. (#1491)These changes are temporary, and reverted on exiting the context.
>>> with tf.GradientTape(): ... with qml.tape.QuantumTape() as tape: ... qml.RX(tf.Variable(0.1), wires=0) ... qml.RY(tf.constant(0.2), wires=0) ... qml.RZ(tf.Variable(0.3), wires=0) ... with tape.unwrap(): ... print("Trainable params:", tape.trainable_params) ... print("Unwrapped params:", tape.get_parameters()) Trainable params: {0, 2} Unwrapped params: [0.1, 0.3] >>> print("Original parameters:", tape.get_parameters()) Original parameters: [<tf.Variable 'Variable:0' shape=() dtype=float32, numpy=0.1>, <tf.Variable 'Variable:0' shape=() dtype=float32, numpy=0.3>]
In addition,
qml.tape.Unwrap
is a context manager that unwraps multiple tapes:>>> with qml.tape.Unwrap(tape1, tape2):
-
Removed the deprecated tape methods
get_resources
andget_depth
as they are superseded by thespecs
tape attribute. (#1522) -
Specifying
shots=None
withqml.sample
was previously deprecated. From this release onwards, settingshots=None
when sampling will raise an error. (#1522) -
The existing
pennylane.collections.apply
function is no longer accessible viaqml.apply
, and needs to be imported directly from thecollections
package. (#1358)
-
Fixes a bug in
qml.adjoint
andqml.ctrl
where the adjoint of operations outside of aQNode
or aQuantumTape
could not be obtained. (#1532) -
Fixes a bug in
GradientDescentOptimizer
andNesterovMomentumOptimizer
where a cost function with one trainable parameter and non-trainable parameters raised an error. (#1495) -
Fixed an example in the documentation's introduction to numpy gradients, where the wires were a non-differentiable argument to the QNode. (#1499)
-
Fixed a bug where the adjoint of
qml.QFT
when using theqml.adjoint
function was not correctly computed. (#1451) -
Fixed the differentiability of the operation
IsingYY
for Autograd, Jax and Tensorflow. (#1425) -
Fixed a bug in the
torch
interface that prevented gradients from being computed on a GPU. (#1426) -
Quantum function transforms now preserve the format of the measurement results, so that a single measurement returns a single value rather than an array with a single element. (#1434)
-
Fixed a bug in the parameter-shift Hessian implementation, which resulted in the incorrect Hessian being returned for a cost function that performed post-processing on a vector-valued QNode. (#1436)
-
Fixed a bug in the initialization of
QubitUnitary
where the size of the matrix was not checked against the number of wires. (#1439)
-
Improved Contribution Guide and Pull Requests Guide. (#1461)
-
Examples have been added to clarify use of the continuous-variable
FockStateVector
operation in the multi-mode case. (#1472)
This release contains contributions from (in alphabetical order):
Juan Miguel Arrazola, Olivia Di Matteo, Anthony Hayes, Theodor Isacsson, Josh Izaac, Soran Jahangiri, Nathan Killoran, Arshpreet Singh Khangura, Leonhard Kunczik, Christina Lee, Romain Moyard, Lee James O'Riordan, Ashish Panigrahi, Nahum Sá, Maria Schuld, Jay Soni, Antal Száva, David Wierichs.
-
The new
qml.kernels
module provides basic functionalities for working with quantum kernels as well as post-processing methods to mitigate sampling errors and device noise: (#1102)num_wires = 6 wires = range(num_wires) dev = qml.device('default.qubit', wires=wires) @qml.qnode(dev) def kernel_circuit(x1, x2): qml.templates.AngleEmbedding(x1, wires=wires) qml.adjoint(qml.templates.AngleEmbedding)(x2, wires=wires) return qml.probs(wires) kernel = lambda x1, x2: kernel_circuit(x1, x2)[0] X_train = np.random.random((10, 6)) X_test = np.random.random((5, 6)) # Create symmetric square kernel matrix (for training) K = qml.kernels.square_kernel_matrix(X_train, kernel) # Compute kernel between test and training data. K_test = qml.kernels.kernel_matrix(X_train, X_test, kernel) K1 = qml.kernels.mitigate_depolarizing_noise(K, num_wires, method='single')
-
PennyLane now has a
fourier
module, which hosts a growing library of methods that help with investigating the Fourier representation of functions implemented by quantum circuits. The Fourier representation can be used to examine and characterize the expressivity of the quantum circuit. (#1160) (#1378)For example, one can plot distributions over Fourier series coefficients like this one:
-
Added functionality for constructing and manipulating the Pauli group (#1181).
The function
qml.grouping.pauli_group
provides a generator to easily loop over the group, or construct and store it in its entirety. For example, we can construct the single-qubit Pauli group like so:>>> from pennylane.grouping import pauli_group >>> pauli_group_1_qubit = list(pauli_group(1)) >>> pauli_group_1_qubit [Identity(wires=[0]), PauliZ(wires=[0]), PauliX(wires=[0]), PauliY(wires=[0])]
We can multiply together its members at the level of Pauli words using the
pauli_mult
andpauli_multi_with_phase
functions. This can be done on arbitrarily-labeled wires as well, by defining a wire map.>>> from pennylane.grouping import pauli_group, pauli_mult >>> wire_map = {'a' : 0, 'b' : 1, 'c' : 2} >>> pg = list(pauli_group(3, wire_map=wire_map)) >>> pg[3] PauliZ(wires=['b']) @ PauliZ(wires=['c']) >>> pg[55] PauliY(wires=['a']) @ PauliY(wires=['b']) @ PauliZ(wires=['c']) >>> pauli_mult(pg[3], pg[55], wire_map=wire_map) PauliY(wires=['a']) @ PauliX(wires=['b'])
Functions for conversion of Pauli observables to strings (and back), are included.
>>> from pennylane.grouping import pauli_word_to_string, string_to_pauli_word >>> pauli_word_to_string(pg[55], wire_map=wire_map) 'YYZ' >>> string_to_pauli_word('ZXY', wire_map=wire_map) PauliZ(wires=['a']) @ PauliX(wires=['b']) @ PauliY(wires=['c'])
Calculation of the matrix representation for arbitrary Paulis and wire maps is now also supported.
>>> from pennylane.grouping import pauli_word_to_matrix >>> wire_map = {'a' : 0, 'b' : 1} >>> pauli_word = qml.PauliZ('b') # corresponds to Pauli 'IZ' >>> pauli_word_to_matrix(pauli_word, wire_map=wire_map) array([[ 1., 0., 0., 0.], [ 0., -1., 0., -0.], [ 0., 0., 1., 0.], [ 0., -0., 0., -1.]])
-
The
qml.specs
QNode transform creates a function that returns specifications or details about the QNode, including depth, number of gates, and number of gradient executions required. (#1245)For example:
dev = qml.device('default.qubit', wires=4) @qml.qnode(dev, diff_method='parameter-shift') def circuit(x, y): qml.RX(x[0], wires=0) qml.Toffoli(wires=(0, 1, 2)) qml.CRY(x[1], wires=(0, 1)) qml.Rot(x[2], x[3], y, wires=0) return qml.expval(qml.PauliZ(0)), qml.expval(qml.PauliX(1))
We can now use the
qml.specs
transform to generate a function that returns details and resource information:>>> x = np.array([0.05, 0.1, 0.2, 0.3], requires_grad=True) >>> y = np.array(0.4, requires_grad=False) >>> specs_func = qml.specs(circuit) >>> specs_func(x, y) {'gate_sizes': defaultdict(int, {1: 2, 3: 1, 2: 1}), 'gate_types': defaultdict(int, {'RX': 1, 'Toffoli': 1, 'CRY': 1, 'Rot': 1}), 'num_operations': 4, 'num_observables': 2, 'num_diagonalizing_gates': 1, 'num_used_wires': 3, 'depth': 4, 'num_trainable_params': 4, 'num_parameter_shift_executions': 11, 'num_device_wires': 4, 'device_name': 'default.qubit', 'diff_method': 'parameter-shift'}
The tape methods
get_resources
andget_depth
are superseded byspecs
and will be deprecated after one release cycle. -
Adds a decorator
@qml.qfunc_transform
to easily create a transformation that modifies the behaviour of a quantum function. (#1315)For example, consider the following transform, which scales the parameter of all
RX
gates by :math:x \rightarrow \sin(a) \sqrt{x}
, and the parameters of allRY
gates by :math:y \rightarrow \cos(a * b) y
:@qml.qfunc_transform def my_transform(tape, a, b): for op in tape.operations + tape.measurements: if op.name == "RX": x = op.parameters[0] qml.RX(qml.math.sin(a) * qml.math.sqrt(x), wires=op.wires) elif op.name == "RY": y = op.parameters[0] qml.RX(qml.math.cos(a * b) * y, wires=op.wires) else: op.queue()
We can now apply this transform to any quantum function:
dev = qml.device("default.qubit", wires=2) def ansatz(x): qml.Hadamard(wires=0) qml.RX(x[0], wires=0) qml.RY(x[1], wires=1) qml.CNOT(wires=[0, 1]) @qml.qnode(dev) def circuit(params, transform_weights): qml.RX(0.1, wires=0) # apply the transform to the ansatz my_transform(*transform_weights)(ansatz)(params) return qml.expval(qml.PauliZ(1))
We can print this QNode to show that the qfunc transform is taking place:
>>> x = np.array([0.5, 0.3], requires_grad=True) >>> transform_weights = np.array([0.1, 0.6], requires_grad=True) >>> print(qml.draw(circuit)(x, transform_weights)) 0: ──RX(0.1)────H──RX(0.0706)──╭C──┤ 1: ──RX(0.299)─────────────────╰X──┤ ⟨Z⟩
Evaluating the QNode, as well as the derivative, with respect to the gate parameter and the transform weights:
>>> circuit(x, transform_weights) tensor(0.00672829, requires_grad=True) >>> qml.grad(circuit)(x, transform_weights) (array([ 0.00671711, -0.00207359]), array([6.69695008e-02, 3.73694364e-06]))
-
Adds a
hamiltonian_expand
tape transform. This takes a tape ending inqml.expval(H)
, whereH
is a Hamiltonian, and maps it to a collection of tapes which can be executed and passed into a post-processing function yielding the expectation value. (#1142)Example use:
H = qml.PauliZ(0) + 3 * qml.PauliZ(0) @ qml.PauliX(1) with qml.tape.QuantumTape() as tape: qml.Hadamard(wires=1) qml.expval(H) tapes, fn = qml.transforms.hamiltonian_expand(tape)
We can now evaluate the transformed tapes, and apply the post-processing function:
>>> dev = qml.device("default.qubit", wires=3) >>> res = dev.batch_execute(tapes) >>> fn(res) 3.999999999999999
-
The
quantum_monte_carlo
transform has been added, allowing an input circuit to be transformed into the full quantum Monte Carlo algorithm. (#1316)Suppose we want to measure the expectation value of the sine squared function according to a standard normal distribution. We can calculate the expectation value analytically as
0.432332
, but we can also estimate using the quantum Monte Carlo algorithm. The first step is to discretize the problem:from scipy.stats import norm m = 5 M = 2 ** m xmax = np.pi # bound to region [-pi, pi] xs = np.linspace(-xmax, xmax, M) probs = np.array([norm().pdf(x) for x in xs]) probs /= np.sum(probs) func = lambda i: np.sin(xs[i]) ** 2 r_rotations = np.array([2 * np.arcsin(np.sqrt(func(i))) for i in range(M)])
The
quantum_monte_carlo
transform can then be used:from pennylane.templates.state_preparations.mottonen import ( _uniform_rotation_dagger as r_unitary, ) n = 6 N = 2 ** n a_wires = range(m) wires = range(m + 1) target_wire = m estimation_wires = range(m + 1, n + m + 1) dev = qml.device("default.qubit", wires=(n + m + 1)) def fn(): qml.templates.MottonenStatePreparation(np.sqrt(probs), wires=a_wires) r_unitary(qml.RY, r_rotations, control_wires=a_wires[::-1], target_wire=target_wire) @qml.qnode(dev) def qmc(): qml.quantum_monte_carlo(fn, wires, target_wire, estimation_wires)() return qml.probs(estimation_wires) phase_estimated = np.argmax(qmc()[:int(N / 2)]) / N
The estimated value can be retrieved using:
>>> (1 - np.cos(np.pi * phase_estimated)) / 2 0.42663476277231915
The resources required to perform the quantum Monte Carlo algorithm can also be inspected using the
specs
transform.
-
Functionality to support solving the maximum-weighted cycle problem has been added to the
qaoa
module. (#1207) (#1209) (#1251) (#1213) (#1220) (#1214) (#1283) (#1297) (#1396) (#1403)The
max_weight_cycle
function returns the appropriate cost and mixer Hamiltonians:>>> a = np.random.random((3, 3)) >>> np.fill_diagonal(a, 0) >>> g = nx.DiGraph(a) # create a random directed graph >>> cost, mixer, mapping = qml.qaoa.max_weight_cycle(g) >>> print(cost) (-0.9775906842165344) [Z2] + (-0.9027248603361988) [Z3] + (-0.8722207409852838) [Z0] + (-0.6426184210832898) [Z5] + (-0.2832594164291379) [Z1] + (-0.0778133996933755) [Z4] >>> print(mapping) {0: (0, 1), 1: (0, 2), 2: (1, 0), 3: (1, 2), 4: (2, 0), 5: (2, 1)}
Additional functionality can be found in the qml.qaoa.cycle module.
-
Added functionality to compute the sparse matrix representation of a
qml.Hamiltonian
object. (#1394)coeffs = [1, -0.45] obs = [qml.PauliZ(0) @ qml.PauliZ(1), qml.PauliY(0) @ qml.PauliZ(1)] H = qml.Hamiltonian(coeffs, obs) H_sparse = qml.utils.sparse_hamiltonian(H)
The resulting matrix is a sparse matrix in scipy coordinate list (COO) format:
>>> H_sparse <4x4 sparse matrix of type '<class 'numpy.complex128'>' with 8 stored elements in COOrdinate format>
The sparse matrix can be converted to an array as:
>>> H_sparse.toarray() array([[ 1.+0.j , 0.+0.j , 0.+0.45j, 0.+0.j ], [ 0.+0.j , -1.+0.j , 0.+0.j , 0.-0.45j], [ 0.-0.45j, 0.+0.j , -1.+0.j , 0.+0.j ], [ 0.+0.j , 0.+0.45j, 0.+0.j , 1.+0.j ]])
-
Adds the new template
AllSinglesDoubles
to prepare quantum states of molecules using theSingleExcitation
andDoubleExcitation
operations. The new template reduces significantly the number of operations and the depth of the quantum circuit with respect to the traditional UCCSD unitary. (#1383)For example, consider the case of two particles and four qubits. First, we define the Hartree-Fock initial state and generate all possible single and double excitations.
import pennylane as qml from pennylane import numpy as np electrons = 2 qubits = 4 hf_state = qml.qchem.hf_state(electrons, qubits) singles, doubles = qml.qchem.excitations(electrons, qubits)
Now we can use the template
AllSinglesDoubles
to define the quantum circuit,from pennylane.templates import AllSinglesDoubles wires = range(qubits) dev = qml.device('default.qubit', wires=wires) @qml.qnode(dev) def circuit(weights, hf_state, singles, doubles): AllSinglesDoubles(weights, wires, hf_state, singles, doubles) return qml.expval(qml.PauliZ(0)) params = np.random.normal(0, np.pi, len(singles) + len(doubles))
and execute it:
>>> circuit(params, hf_state, singles=singles, doubles=doubles) tensor(-0.73772194, requires_grad=True)
-
Adds
QubitCarry
andQubitSum
operations for basic arithmetic. (#1169)The following example adds two 1-bit numbers, returning a 2-bit answer:
dev = qml.device('default.qubit', wires = 4) a = 0 b = 1 @qml.qnode(dev) def circuit(): qml.BasisState(np.array([a, b]), wires=[1, 2]) qml.QubitCarry(wires=[0, 1, 2, 3]) qml.CNOT(wires=[1, 2]) qml.QubitSum(wires=[0, 1, 2]) return qml.probs(wires=[3, 2]) probs = circuit() bitstrings = tuple(itertools.product([0, 1], repeat = 2)) indx = np.argwhere(probs == 1).flatten()[0] output = bitstrings[indx]
>>> print(output) (0, 1)
-
Added the
qml.Projector
observable, which is available on all devices inheriting from theQubitDevice
class. (#1356) (#1368)Using
qml.Projector
, we can define the basis state projectors to use when computing expectation values. Let us take for example a circuit that prepares Bell states:dev = qml.device("default.qubit", wires=2) @qml.qnode(dev) def circuit(basis_state): qml.Hadamard(wires=[0]) qml.CNOT(wires=[0, 1]) return qml.expval(qml.Projector(basis_state, wires=[0, 1]))
We can then specify the
|00>
basis state to construct the|00><00|
projector and compute the expectation value:>>> basis_state = [0, 0] >>> circuit(basis_state) tensor(0.5, requires_grad=True)
As expected, we get similar results when specifying the
|11>
basis state:>>> basis_state = [1, 1] >>> circuit(basis_state) tensor(0.5, requires_grad=True)
-
The following new operations have been added:
-
The
argnum
keyword argument can now be specified for a QNode to define a subset of trainable parameters used to estimate the Jacobian. (#1371)For example, consider two trainable parameters and a quantum function:
dev = qml.device("default.qubit", wires=2) x = np.array(0.543, requires_grad=True) y = np.array(-0.654, requires_grad=True) def circuit(x,y): qml.RX(x, wires=[0]) qml.RY(y, wires=[1]) qml.CNOT(wires=[0, 1]) return qml.expval(qml.PauliZ(0) @ qml.PauliX(1))
When computing the gradient of the QNode, we can specify the trainable parameters to consider by passing the
argnum
keyword argument:>>> qnode1 = qml.QNode(circuit, dev, diff_method="parameter-shift", argnum=[0,1]) >>> print(qml.grad(qnode1)(x,y)) (array(0.31434679), array(0.67949903))
Specifying a proper subset of the trainable parameters will estimate the Jacobian:
>>> qnode2 = qml.QNode(circuit, dev, diff_method="parameter-shift", argnum=[0]) >>> print(qml.grad(qnode2)(x,y)) (array(0.31434679), array(0.))
-
Allows creating differentiable observables that return custom objects such that the observable is supported by devices. (1291)
As an example, first we define
NewObservable
class:from pennylane.devices import DefaultQubit class NewObservable(qml.operation.Observable): """NewObservable""" num_wires = qml.operation.AnyWires num_params = 0 par_domain = None def diagonalizing_gates(self): """Diagonalizing gates""" return []
Once we have this new observable class, we define a
SpecialObject
class that can be used to encode data in an observable and a new device that supports our new observable and returns aSpecialObject
as the expectation value (the code is shortened for brevity, the extended example can be found as a test in the previously referenced pull request):class SpecialObject: def __init__(self, val): self.val = val def __mul__(self, other): new = SpecialObject(self.val) new *= other return new ... class DeviceSupportingNewObservable(DefaultQubit): name = "Device supporting NewObservable" short_name = "default.qubit.newobservable" observables = DefaultQubit.observables.union({"NewObservable"}) def expval(self, observable, **kwargs): if self.shots is None and isinstance(observable, NewObservable): val = super().expval(qml.PauliZ(wires=0), **kwargs) return SpecialObject(val) return super().expval(observable, **kwargs)
At this point, we can create a device that will support the differentiation of a
NewObservable
object:dev = DeviceSupportingNewObservable(wires=1, shots=None) @qml.qnode(dev, diff_method="parameter-shift") def qnode(x): qml.RY(x, wires=0) return qml.expval(NewObservable(wires=0))
We can then compute the jacobian of this object:
>>> result = qml.jacobian(qnode)(0.2) >>> print(result) <__main__.SpecialObject object at 0x7fd2c54721f0> >>> print(result.item().val) -0.19866933079506116
-
PennyLane NumPy now includes the random module's
Generator
objects, the recommended way of random number generation. This allows for random number generation using a local, rather than global seed. (#1267)from pennylane import numpy as np rng = np.random.default_rng() random_mat1 = rng.random((3,2)) random_mat2 = rng.standard_normal(3, requires_grad=False)
-
The performance of adjoint jacobian differentiation was significantly improved as the method now reuses the state computed on the forward pass. This can be turned off to save memory with the Torch and TensorFlow interfaces by passing
adjoint_cache=False
during QNode creation. (#1341) -
The
Operator
(and by inheritance, theOperation
andObservable
class and their children) now have anid
attribute, which can mark an operator in a circuit, for example to identify it on the tape by a tape transform. (#1377) -
The
benchmark
module was deleted, since it was outdated and is superseded by the new separate benchmark repository. (#1343) -
Decompositions in terms of elementary gates has been added for:
-
qml.CSWAP
(#1306) -
qml.SWAP
(#1329) -
qml.SingleExcitation
(#1303) -
qml.SingleExcitationPlus
andqml.SingleExcitationMinus
(#1278) -
qml.DoubleExcitation
(#1303) -
qml.Toffoli
(#1320) -
qml.MultiControlledX
. (#1287) When controlling on three or more wires, an ancilla register of worker wires is required to support the decomposition.ctrl_wires = [f"c{i}" for i in range(5)] work_wires = [f"w{i}" for i in range(3)] target_wires = ["t0"] all_wires = ctrl_wires + work_wires + target_wires dev = qml.device("default.qubit", wires=all_wires) with qml.tape.QuantumTape() as tape: qml.MultiControlledX(control_wires=ctrl_wires, wires=target_wires, work_wires=work_wires)
>>> tape = tape.expand(depth=1) >>> print(tape.draw(wire_order=qml.wires.Wires(all_wires))) c0: ──────────────╭C──────────────────────╭C──────────┤ c1: ──────────────├C──────────────────────├C──────────┤ c2: ──────────╭C──│───╭C──────────────╭C──│───╭C──────┤ c3: ──────╭C──│───│───│───╭C──────╭C──│───│───│───╭C──┤ c4: ──╭C──│───│───│───│───│───╭C──│───│───│───│───│───┤ w0: ──│───│───├C──╰X──├C──│───│───│───├C──╰X──├C──│───┤ w1: ──│───├C──╰X──────╰X──├C──│───├C──╰X──────╰X──├C──┤ w2: ──├C──╰X──────────────╰X──├C──╰X──────────────╰X──┤ t0: ──╰X──────────────────────╰X──────────────────────┤
-
-
Added
qml.CPhase
as an alias for the existingqml.ControlledPhaseShift
operation. (#1319). -
The
Device
class now uses caching when mapping wires. (#1270) -
The
Wires
class now uses caching for computing itshash
. (#1270) -
Added custom gate application for Toffoli in
default.qubit
. (#1249) -
Added validation for noise channel parameters. Invalid noise parameters now raise a
ValueError
. (#1357) -
The device test suite now provides test cases for checking gates by comparing expectation values. (#1212)
-
PennyLane's test suite is now code-formatted using
black -l 100
. (#1222) -
PennyLane's
qchem
package and tests are now code-formatted usingblack -l 100
. (#1311)
-
The
qml.inv()
function is now deprecated with a warning to use the more generalqml.adjoint()
. (#1325) -
Removes support for Python 3.6 and adds support for Python 3.9. (#1228)
-
The tape methods
get_resources
andget_depth
are superseded byspecs
and will be deprecated after one release cycle. (#1245) -
Using the
qml.sample()
measurement on devices withshots=None
continue to raise a warning with this functionality being fully deprecated and raising an error after one release cycle. (#1079) (#1196)
-
QNodes now display readable information when in interactive environments or when printed. (#1359).
-
Fixes a bug with
qml.math.cast
where theMottonenStatePreparation
operation expected a float type instead of double. (#1400) -
Fixes a bug where a copy of
qml.ControlledQubitUnitary
was non-functional as it did not have all the necessary information. (#1411) -
Warns when adjoint or reversible differentiation specified or called on a device with finite shots. (#1406)
-
Fixes the differentiability of the operations
IsingXX
andIsingZZ
for Autograd, Jax and Tensorflow. (#1390) -
Fixes a bug where multiple identical Hamiltonian terms will produce a different result with
optimize=True
usingExpvalCost
. (#1405) -
Fixes bug where
shots=None
was not reset when changing shots temporarily in a QNode call likecircuit(0.1, shots=3)
. (#1392) -
Fixes floating point errors with
diff_method="finite-diff"
andorder=1
when parameters arefloat32
. (#1381) -
Fixes a bug where
qml.ctrl
would fail to transform gates that had no control defined and no decomposition defined. (#1376) -
Copying the
JacobianTape
now correctly also copies thejacobian_options
attribute. This fixes a bug allowing the JAX interface to support adjoint differentiation. (#1349) -
Fixes drawing QNodes that contain multiple measurements on a single wire. (#1353)
-
Fixes drawing QNodes with no operations. (#1354)
-
Fixes incorrect wires in the decomposition of the
ControlledPhaseShift
operation. (#1338) -
Fixed tests for the
Permute
operation that used a QNode and hence expanded tapes twice instead of once due to QNode tape expansion and an explicit tape expansion call. (#1318). -
Prevent Hamiltonians that share wires from being multiplied together. (#1273)
-
Fixed a bug where the custom range sequences could not be passed to the
StronglyEntanglingLayers
template. (#1332) -
Fixed a bug where
qml.sum()
andqml.dot()
do not support the JAX interface. (#1380)
-
Math present in the
QubitParamShiftTape
class docstring now renders correctly. (#1402) -
Fix typo in the documentation of
qml.StronglyEntanglingLayers
. (#1367) -
Fixed typo in TensorFlow interface documentation (#1312)
-
Fixed typos in the mathematical expressions in documentation of
qml.DoubleExcitation
. (#1278) -
Remove unsupported
None
option from theqml.QNode
docstrings. (#1271) -
Updated the docstring of
qml.PolyXP
to reference the new location of internal usage. (#1262) -
Removes occurrences of the deprecated device argument
analytic
from the documentation. (#1261) -
Updated PyTorch and TensorFlow interface introductions. (#1333)
-
Updates the quantum chemistry quickstart to reflect recent changes to the
qchem
module. (#1227)
This release contains contributions from (in alphabetical order):
Marius Aglitoiu, Vishnu Ajith, Juan Miguel Arrazola, Thomas Bromley, Jack Ceroni, Alaric Cheng, Miruna Daian, Olivia Di Matteo, Tanya Garg, Christian Gogolin, Alain Delgado Gran, Diego Guala, Anthony Hayes, Ryan Hill, Theodor Isacsson, Josh Izaac, Soran Jahangiri, Pavan Jayasinha, Nathan Killoran, Christina Lee, Ryan Levy, Alberto Maldonado, Johannes Jakob Meyer, Romain Moyard, Ashish Panigrahi, Nahum Sá, Maria Schuld, Brian Shi, Antal Száva, David Wierichs, Vincent Wong.
-
Fixes two bugs in the parameter-shift Hessian. (#1260)
-
Fixes a bug where having an unused parameter in the Autograd interface would result in an indexing error during backpropagation.
-
The parameter-shift Hessian only supports the two-term parameter-shift rule currently, so raises an error if asked to differentiate any unsupported gates (such as the controlled rotation gates).
-
-
A bug which resulted in
qml.adjoint()
andqml.inv()
failing to work with templates has been fixed. (#1243) -
Deprecation warning instances in PennyLane have been changed to
UserWarning
, to account for recent changes to how Python warnings are filtered in PEP565. (#1211)
- Updated the order of the parameters to the
GaussianState
operation to match the way that the PennyLane-SF plugin uses them. (#1255)
This release contains contributions from (in alphabetical order):
Thomas Bromley, Olivia Di Matteo, Diego Guala, Anthony Hayes, Ryan Hill, Josh Izaac, Christina Lee, Maria Schuld, Antal Száva.
-
Adds a new optimizer
qml.ShotAdaptiveOptimizer
, a gradient-descent optimizer where the shot rate is adaptively calculated using the variances of the parameter-shift gradient. (#1139)By keeping a running average of the parameter-shift gradient and the variance of the parameter-shift gradient, this optimizer frugally distributes a shot budget across the partial derivatives of each parameter.
In addition, if computing the expectation value of a Hamiltonian, weighted random sampling can be used to further distribute the shot budget across the local terms from which the Hamiltonian is constructed.
This optimizer is based on both the iCANS1 and Rosalin shot-adaptive optimizers.
Once constructed, the cost function can be passed directly to the optimizer's
step
method. The attributeopt.total_shots_used
can be used to track the number of shots per iteration.>>> coeffs = [2, 4, -1, 5, 2] >>> obs = [ ... qml.PauliX(1), ... qml.PauliZ(1), ... qml.PauliX(0) @ qml.PauliX(1), ... qml.PauliY(0) @ qml.PauliY(1), ... qml.PauliZ(0) @ qml.PauliZ(1) ... ] >>> H = qml.Hamiltonian(coeffs, obs) >>> dev = qml.device("default.qubit", wires=2, shots=100) >>> cost = qml.ExpvalCost(qml.templates.StronglyEntanglingLayers, H, dev) >>> params = qml.init.strong_ent_layers_uniform(n_layers=2, n_wires=2) >>> opt = qml.ShotAdaptiveOptimizer(min_shots=10) >>> for i in range(5): ... params = opt.step(cost, params) ... print(f"Step {i}: cost = {cost(params):.2f}, shots_used = {opt.total_shots_used}") Step 0: cost = -5.68, shots_used = 240 Step 1: cost = -2.98, shots_used = 336 Step 2: cost = -4.97, shots_used = 624 Step 3: cost = -5.53, shots_used = 1054 Step 4: cost = -6.50, shots_used = 1798
-
Batches of shots can now be specified as a list, allowing measurement statistics to be course-grained with a single QNode evaluation. (#1103)
>>> shots_list = [5, 10, 1000] >>> dev = qml.device("default.qubit", wires=2, shots=shots_list)
When QNodes are executed on this device, a single execution of 1015 shots will be submitted. However, three sets of measurement statistics will be returned; using the first 5 shots, second set of 10 shots, and final 1000 shots, separately.
For example, executing a circuit with two outputs will lead to a result of shape
(3, 2)
:>>> @qml.qnode(dev) ... def circuit(x): ... qml.RX(x, wires=0) ... qml.CNOT(wires=[0, 1]) ... return qml.expval(qml.PauliZ(0) @ qml.PauliX(1)), qml.expval(qml.PauliZ(0)) >>> circuit(0.5) [[0.33333333 1. ] [0.2 1. ] [0.012 0.868 ]]
This output remains fully differentiable.
-
The number of shots can now be specified on a per-call basis when evaluating a QNode. (#1075).
For this, the qnode should be called with an additional
shots
keyword argument:>>> dev = qml.device('default.qubit', wires=1, shots=10) # default is 10 >>> @qml.qnode(dev) ... def circuit(a): ... qml.RX(a, wires=0) ... return qml.sample(qml.PauliZ(wires=0)) >>> circuit(0.8) [ 1 1 1 -1 -1 1 1 1 1 1] >>> circuit(0.8, shots=3) [ 1 1 1] >>> circuit(0.8) [ 1 1 1 -1 -1 1 1 1 1 1]
A new module is available, qml.transforms, which contains differentiable quantum transforms. These are functions that act on QNodes, quantum functions, devices, and tapes, transforming them while remaining fully differentiable.
-
A new adjoint transform has been added. (#1111) (#1135)
This new method allows users to apply the adjoint of an arbitrary sequence of operations.
def subroutine(wire): qml.RX(0.123, wires=wire) qml.RY(0.456, wires=wire) dev = qml.device('default.qubit', wires=1) @qml.qnode(dev) def circuit(): subroutine(0) qml.adjoint(subroutine)(0) return qml.expval(qml.PauliZ(0))
This creates the following circuit:
>>> print(qml.draw(circuit)()) 0: --RX(0.123)--RY(0.456)--RY(-0.456)--RX(-0.123)--| <Z>
Directly applying to a gate also works as expected.
qml.adjoint(qml.RX)(0.123, wires=0) # applies RX(-0.123)
-
A new transform
qml.ctrl
is now available that adds control wires to subroutines. (#1157)def my_ansatz(params): qml.RX(params[0], wires=0) qml.RZ(params[1], wires=1) # Create a new operation that applies `my_ansatz` # controlled by the "2" wire. my_ansatz2 = qml.ctrl(my_ansatz, control=2) @qml.qnode(dev) def circuit(params): my_ansatz2(params) return qml.state()
This is equivalent to:
@qml.qnode(...) def circuit(params): qml.CRX(params[0], wires=[2, 0]) qml.CRZ(params[1], wires=[2, 1]) return qml.state()
-
The
qml.transforms.classical_jacobian
transform has been added. (#1186)This transform returns a function to extract the Jacobian matrix of the classical part of a QNode, allowing the classical dependence between the QNode arguments and the quantum gate arguments to be extracted.
For example, given the following QNode:
>>> @qml.qnode(dev) ... def circuit(weights): ... qml.RX(weights[0], wires=0) ... qml.RY(weights[0], wires=1) ... qml.RZ(weights[2] ** 2, wires=1) ... return qml.expval(qml.PauliZ(0))
We can use this transform to extract the relationship :math:
f: \mathbb{R}^n \rightarrow\mathbb{R}^m
between the input QNode arguments :math:w
and the gate arguments :math:g
, for a given value of the QNode arguments:>>> cjac_fn = qml.transforms.classical_jacobian(circuit) >>> weights = np.array([1., 1., 1.], requires_grad=True) >>> cjac = cjac_fn(weights) >>> print(cjac) [[1. 0. 0.] [1. 0. 0.] [0. 0. 2.]]
The returned Jacobian has rows corresponding to gate arguments, and columns corresponding to QNode arguments; that is, :math:
J_{ij} = \frac{\partial}{\partial g_i} f(w_j)
.
-
Added the
SingleExcitation
two-qubit operation, which is useful for quantum chemistry applications. (#1121)It can be used to perform an SO(2) rotation in the subspace spanned by the states :math:
|01\rangle
and :math:|10\rangle
. For example, the following circuit performs the transformation :math:|10\rangle \rightarrow \cos(\phi/2)|10\rangle - \sin(\phi/2)|01\rangle
:dev = qml.device('default.qubit', wires=2) @qml.qnode(dev) def circuit(phi): qml.PauliX(wires=0) qml.SingleExcitation(phi, wires=[0, 1])
The
SingleExcitation
operation supports analytic gradients on hardware using only four expectation value calculations, following results from Kottmann et al. -
Added the
DoubleExcitation
four-qubit operation, which is useful for quantum chemistry applications. (#1123)It can be used to perform an SO(2) rotation in the subspace spanned by the states :math:
|1100\rangle
and :math:|0011\rangle
. For example, the following circuit performs the transformation :math:|1100\rangle\rightarrow \cos(\phi/2)|1100\rangle - \sin(\phi/2)|0011\rangle
:dev = qml.device('default.qubit', wires=2) @qml.qnode(dev) def circuit(phi): qml.PauliX(wires=0) qml.PauliX(wires=1) qml.DoubleExcitation(phi, wires=[0, 1, 2, 3])
The
DoubleExcitation
operation supports analytic gradients on hardware using only four expectation value calculations, following results from Kottmann et al.. -
Added the
QuantumMonteCarlo
template for performing quantum Monte Carlo estimation of an expectation value on simulator. (#1130)The following example shows how the expectation value of sine squared over a standard normal distribution can be approximated:
from scipy.stats import norm m = 5 M = 2 ** m n = 10 N = 2 ** n target_wires = range(m + 1) estimation_wires = range(m + 1, n + m + 1) xmax = np.pi # bound to region [-pi, pi] xs = np.linspace(-xmax, xmax, M) probs = np.array([norm().pdf(x) for x in xs]) probs /= np.sum(probs) func = lambda i: np.sin(xs[i]) ** 2 dev = qml.device("default.qubit", wires=(n + m + 1)) @qml.qnode(dev) def circuit(): qml.templates.QuantumMonteCarlo( probs, func, target_wires=target_wires, estimation_wires=estimation_wires, ) return qml.probs(estimation_wires) phase_estimated = np.argmax(circuit()[:int(N / 2)]) / N expectation_estimated = (1 - np.cos(np.pi * phase_estimated)) / 2
-
Added the
QuantumPhaseEstimation
template for performing quantum phase estimation for an input unitary matrix. (#1095)Consider the matrix corresponding to a rotation from an
RX
gate:>>> phase = 5 >>> target_wires = [0] >>> unitary = qml.RX(phase, wires=0).matrix
The
phase
parameter can be estimated usingQuantumPhaseEstimation
. For example, using five phase-estimation qubits:n_estimation_wires = 5 estimation_wires = range(1, n_estimation_wires + 1) dev = qml.device("default.qubit", wires=n_estimation_wires + 1) @qml.qnode(dev) def circuit(): # Start in the |+> eigenstate of the unitary qml.Hadamard(wires=target_wires) QuantumPhaseEstimation( unitary, target_wires=target_wires, estimation_wires=estimation_wires, ) return qml.probs(estimation_wires) phase_estimated = np.argmax(circuit()) / 2 ** n_estimation_wires # Need to rescale phase due to convention of RX gate phase_estimated = 4 * np.pi * (1 - phase)
-
Added the
ControlledPhaseShift
gate as well as theQFT
operation for applying quantum Fourier transforms. (#1064)@qml.qnode(dev) def circuit_qft(basis_state): qml.BasisState(basis_state, wires=range(3)) qml.templates.QFT(wires=range(3)) return qml.state()
-
Added the
ControlledQubitUnitary
operation. This enables implementation of multi-qubit gates with a variable number of control qubits. It is also possible to specify a different state for the control qubits using thecontrol_values
argument (also known as a mixed-polarity multi-controlled operation). (#1069) (#1104)For example, we can create a multi-controlled T gate using:
T = qml.T._matrix() qml.ControlledQubitUnitary(T, control_wires=[0, 1, 3], wires=2, control_values="110")
Here, the T gate will be applied to wire
2
if control wires0
and1
are in state1
, and control wire3
is in state0
. If no value is passed tocontrol_values
, the gate will be applied if all control wires are in the1
state. -
Added
MultiControlledX
for multi-controlledNOT
gates. This is a special case ofControlledQubitUnitary
that applies a Pauli X gate conditioned on the state of an arbitrary number of control qubits. (#1104)
-
Computing second derivatives and Hessians of QNodes is now supported with the parameter-shift differentiation method, on all machine learning interfaces. (#1130) (#1129) (#1110)
Hessians are computed using the parameter-shift rule, and can be evaluated on both hardware and simulator devices.
dev = qml.device('default.qubit', wires=1) @qml.qnode(dev, diff_method="parameter-shift") def circuit(p): qml.RY(p[0], wires=0) qml.RX(p[1], wires=0) return qml.expval(qml.PauliZ(0)) x = np.array([1.0, 2.0], requires_grad=True)
>>> hessian_fn = qml.jacobian(qml.grad(circuit)) >>> hessian_fn(x) [[0.2248451 0.7651474] [0.7651474 0.2248451]]
-
Added the function
finite_diff()
to compute finite-difference approximations to the gradient and the second-order derivatives of arbitrary callable functions. (#1090)This is useful to compute the derivative of parametrized
pennylane.Hamiltonian
observables with respect to their parameters.For example, in quantum chemistry simulations it can be used to evaluate the derivatives of the electronic Hamiltonian with respect to the nuclear coordinates:
>>> def H(x): ... return qml.qchem.molecular_hamiltonian(['H', 'H'], x)[0] >>> x = np.array([0., 0., -0.66140414, 0., 0., 0.66140414]) >>> grad_fn = qml.finite_diff(H, N=1) >>> grad = grad_fn(x) >>> deriv2_fn = qml.finite_diff(H, N=2, idx=[0, 1]) >>> deriv2_fn(x)
-
The JAX interface now supports all devices, including hardware devices, via the parameter-shift differentiation method. (#1076)
For example, using the JAX interface with Cirq:
dev = qml.device('cirq.simulator', wires=1) @qml.qnode(dev, interface="jax", diff_method="parameter-shift") def circuit(x): qml.RX(x[1], wires=0) qml.Rot(x[0], x[1], x[2], wires=0) return qml.expval(qml.PauliZ(0)) weights = jnp.array([0.2, 0.5, 0.1]) print(circuit(weights))
Currently, when used with the parameter-shift differentiation method, only a single returned expectation value or variance is supported. Multiple expectations/variances, as well as probability and state returns, are not currently allowed.
dev = qml.device("default.qubit", wires=2)
inputstate = [np.sqrt(0.2), np.sqrt(0.3), np.sqrt(0.4), np.sqrt(0.1)]
@qml.qnode(dev)
def circuit():
mottonen.MottonenStatePreparation(inputstate,wires=[0, 1])
return qml.expval(qml.PauliZ(0))
Previously returned:
>>> print(qml.draw(circuit)())
0: ──RY(1.57)──╭C─────────────╭C──╭C──╭C──┤ ⟨Z⟩
1: ──RY(1.35)──╰X──RY(0.422)──╰X──╰X──╰X──┤
In this release, it now returns:
>>> print(qml.draw(circuit)())
0: ──RY(1.57)──╭C─────────────╭C──┤ ⟨Z⟩
1: ──RY(1.35)──╰X──RY(0.422)──╰X──┤
-
The templates are now classes inheriting from
Operation
, and define the ansatz in theirexpand()
method. This change does not affect the user interface. (#1138) (#1156) (#1163) (#1192)For convenience, some templates have a new method that returns the expected shape of the trainable parameter tensor, which can be used to create random tensors.
shape = qml.templates.BasicEntanglerLayers.shape(n_layers=2, n_wires=4) weights = np.random.random(shape) qml.templates.BasicEntanglerLayers(weights, wires=range(4))
-
QubitUnitary
now validates to ensure the input matrix is two dimensional. (#1128)
-
Most layers in Pytorch or Keras accept arbitrary dimension inputs, where each dimension barring the last (in the case where the actual weight function of the layer operates on one-dimensional vectors) is broadcast over. This is now also supported by KerasLayer and TorchLayer. (#1062).
Example use:
dev = qml.device("default.qubit", wires=4) x = tf.ones((5, 4, 4)) @qml.qnode(dev) def layer(weights, inputs): qml.templates.AngleEmbedding(inputs, wires=range(4)) qml.templates.StronglyEntanglingLayers(weights, wires=range(4)) return [qml.expval(qml.PauliZ(i)) for i in range(4)] qlayer = qml.qnn.KerasLayer(layer, {"weights": (4, 4, 3)}, output_dim=4) out = qlayer(x)
The output tensor has the following shape:
>>> out.shape (5, 4, 4)
-
If only one argument to the function
qml.grad
has therequires_grad
attribute set to True, then the returned gradient will be a NumPy array, rather than a tuple of length 1. (#1067) (#1081) -
An improvement has been made to how
QubitDevice
generates and post-processess samples, allowing QNode measurement statistics to work on devices with more than 32 qubits. (#1088) -
Due to the addition of
density_matrix()
as a return type from a QNode, tuples are now supported by theoutput_dim
parameter inqnn.KerasLayer
. (#1070) -
Two new utility methods are provided for working with quantum tapes. (#1175)
-
qml.tape.get_active_tape()
gets the currently recording tape. -
tape.stop_recording()
is a context manager that temporarily stops the currently recording tape from recording additional tapes or quantum operations.
For example:
>>> with qml.tape.QuantumTape(): ... qml.RX(0, wires=0) ... current_tape = qml.tape.get_active_tape() ... with current_tape.stop_recording(): ... qml.RY(1.0, wires=1) ... qml.RZ(2, wires=1) >>> current_tape.operations [RX(0, wires=[0]), RZ(2, wires=[1])]
-
-
When printing
qml.Hamiltonian
objects, the terms are sorted by number of wires followed by coefficients. (#981) -
Adds
qml.math.conj
to the PennyLane math module. (#1143)This new method will do elementwise conjugation to the given tensor-like object, correctly dispatching to the required tensor-manipulation framework to preserve differentiability.
>>> a = np.array([1.0 + 2.0j]) >>> qml.math.conj(a) array([1.0 - 2.0j])
-
The four-term parameter-shift rule, as used by the controlled rotation operations, has been updated to use coefficients that minimize the variance as per https://arxiv.org/abs/2104.05695. (#1206)
-
A new transform
qml.transforms.invisible
has been added, to make it easier to transform QNodes. (#1175)
-
Devices do not have an
analytic
argument or attribute anymore. Instead,shots
is the source of truth for whether a simulator estimates return values from a finite number of shots, or whether it returns analytic results (shots=None
). (#1079) (#1196)dev_analytic = qml.device('default.qubit', wires=1, shots=None) dev_finite_shots = qml.device('default.qubit', wires=1, shots=1000) def circuit(): qml.Hadamard(wires=0) return qml.expval(qml.PauliZ(wires=0)) circuit_analytic = qml.QNode(circuit, dev_analytic) circuit_finite_shots = qml.QNode(circuit, dev_finite_shots)
Devices with
shots=None
return deterministic, exact results:>>> circuit_analytic() 0.0 >>> circuit_analytic() 0.0
Devices with
shots > 0
return stochastic results estimated from samples in each run:>>> circuit_finite_shots() -0.062 >>> circuit_finite_shots() 0.034
The
qml.sample()
measurement can only be used on devices on which the number of shots is set explicitly. -
If creating a QNode from a quantum function with an argument named
shots
, aUserWarning
is raised, warning the user that this is a reserved argument to change the number of shots on a per-call basis. (#1075) -
For devices inheriting from
QubitDevice
, the methodsexpval
,var
,sample
accept two new keyword arguments ---shot_range
andbin_size
. (#1103)These new arguments allow for the statistics to be performed on only a subset of device samples. This finer level of control is accessible from the main UI by instantiating a device with a batch of shots.
For example, consider the following device:
>>> dev = qml.device("my_device", shots=[5, (10, 3), 100])
This device will execute QNodes using 135 shots, however measurement statistics will be course grained across these 135 shots:
-
All measurement statistics will first be computed using the first 5 shots --- that is,
shots_range=[0, 5]
,bin_size=5
. -
Next, the tuple
(10, 3)
indicates 10 shots, repeated 3 times. This will useshot_range=[5, 35]
, performing the expectation value in bins of size 10 (bin_size=10
). -
Finally, we repeat the measurement statistics for the final 100 shots,
shot_range=[35, 135]
,bin_size=100
.
-
-
The old PennyLane core has been removed, including the following modules: (#1100)
pennylane.variables
pennylane.qnodes
As part of this change, the location of the new core within the Python module has been moved:
- Moves
pennylane.tape.interfaces
→pennylane.interfaces
- Merges
pennylane.CircuitGraph
andpennylane.TapeCircuitGraph
→pennylane.CircuitGraph
- Merges
pennylane.OperationRecorder
andpennylane.TapeOperationRecorder
→ pennylane.tape.operation_recorder
- Merges
pennylane.measure
andpennylane.tape.measure
→pennylane.measure
- Merges
pennylane.operation
andpennylane.tape.operation
→pennylane.operation
- Merges
pennylane._queuing
andpennylane.tape.queuing
→pennylane.queuing
This has no affect on import location.
In addition,
- All tape-mode functions have been removed (
qml.enable_tape()
,qml.tape_mode_active()
), - All tape fixtures have been deleted,
- Tests specifically for non-tape mode have been deleted.
-
The device test suite no longer accepts the
analytic
keyword. (#1216)
-
Fixes a bug where using the circuit drawer with a
ControlledQubitUnitary
operation raised an error. (#1174) -
Fixes a bug and a test where the
QuantumTape.is_sampled
attribute was not being updated. (#1126) -
Fixes a bug where
BasisEmbedding
would not accept inputs whose bits are all ones or all zeros. (#1114) -
The
ExpvalCost
class raises an error if instantiated with non-expectation measurement statistics. (#1106) -
Fixes a bug where decompositions would reset the differentiation method of a QNode. (#1117)
-
Fixes a bug where the second-order CV parameter-shift rule would error if attempting to compute the gradient of a QNode with more than one second-order observable. (#1197)
-
Fixes a bug where repeated Torch interface applications after expansion caused an error. (#1223)
-
Sampling works correctly with batches of shots specified as a list. (#1232)
-
Updated the diagram used in the Architectural overview page of the Development guide such that it doesn't mention Variables. (#1235)
-
Typos addressed in templates documentation. (#1094)
-
Upgraded the documentation to use Sphinx 3.5.3 and the new m2r2 package. (#1186)
-
Added
flaky
as dependency for running tests in the documentation. (#1113)
This release contains contributions from (in alphabetical order):
Shahnawaz Ahmed, Juan Miguel Arrazola, Thomas Bromley, Olivia Di Matteo, Alain Delgado Gran, Kyle Godbey, Diego Guala, Theodor Isacsson, Josh Izaac, Soran Jahangiri, Nathan Killoran, Christina Lee, Daniel Polatajko, Chase Roberts, Sankalp Sanand, Pritish Sehzpaul, Maria Schuld, Antal Száva, David Wierichs.
-
Fixes a testing bug where tests that required JAX would fail if JAX was not installed. The tests will now instead be skipped if JAX can not be imported. (#1066)
-
Fixes a bug where inverse operations could not be differentiated using backpropagation on
default.qubit
. (#1072) -
The QNode has a new keyword argument,
max_expansion
, that determines the maximum number of times the internal circuit should be expanded when executed on a device. In addition, the default number of max expansions has been increased from 2 to 10, allowing devices that require more than two operator decompositions to be supported. (#1074) -
Fixes a bug where
Hamiltonian
objects created with non-list arguments raised an error for arithmetic operations. (#1082) -
Fixes a bug where
Hamiltonian
objects with no coefficients or operations would return a faulty result when used withExpvalCost
. (#1082)
- Updates mentions of
generate_hamiltonian
tomolecular_hamiltonian
in the docstrings of theExpvalCost
andHamiltonian
classes. (#1077)
This release contains contributions from (in alphabetical order):
Thomas Bromley, Josh Izaac, Antal Száva.
-
QNodes created with
default.qubit
now support a JAX interface, allowing JAX to be used to create, differentiate, and optimize hybrid quantum-classical models. (#947)This is supported internally via a new
default.qubit.jax
device. This device runs end to end in JAX, meaning that it supports all of the awesome JAX transformations (jax.vmap
,jax.jit
,jax.hessian
, etc).Here is an example of how to use the new JAX interface:
dev = qml.device("default.qubit", wires=1) @qml.qnode(dev, interface="jax", diff_method="backprop") def circuit(x): qml.RX(x[1], wires=0) qml.Rot(x[0], x[1], x[2], wires=0) return qml.expval(qml.PauliZ(0)) weights = jnp.array([0.2, 0.5, 0.1]) grad_fn = jax.grad(circuit) print(grad_fn(weights))
Currently, only
diff_method="backprop"
is supported, with plans to support more in the future.
-
A new differentiation method has been added for use with simulators. The
"adjoint"
method operates after a forward pass by iteratively applying inverse gates to scan backwards through the circuit. (#1032)This method is similar to the reversible method, but has a lower time overhead and a similar memory overhead. It follows the approach provided by Jones and Gacon. This method is only compatible with certain statevector-based devices such as
default.qubit
.Example use:
import pennylane as qml wires = 1 device = qml.device("default.qubit", wires=wires) @qml.qnode(device, diff_method="adjoint") def f(params): qml.RX(0.1, wires=0) qml.Rot(*params, wires=0) qml.RX(-0.3, wires=0) return qml.expval(qml.PauliZ(0)) params = [0.1, 0.2, 0.3] qml.grad(f)(params)
-
The default logic for choosing the 'best' differentiation method has been altered to improve performance. (#1008)
-
If the quantum device provides its own gradient, this is now the preferred differentiation method.
-
If the quantum device natively supports classical backpropagation, this is now preferred over the parameter-shift rule.
This will lead to marked speed improvement during optimization when using
default.qubit
, with a sight penalty on the forward-pass evaluation.
More details are available below in the 'Improvements' section for plugin developers.
-
-
PennyLane now supports analytical quantum gradients for noisy channels, in addition to its existing support for unitary operations. The noisy channels
BitFlip
,PhaseFlip
, andDepolarizingChannel
all support analytic gradients out of the box. (#968) -
A method has been added for calculating the Hessian of quantum circuits using the second-order parameter shift formula. (#961)
The following example shows the calculation of the Hessian:
n_wires = 5 weights = [2.73943676, 0.16289932, 3.4536312, 2.73521126, 2.6412488] dev = qml.device("default.qubit", wires=n_wires) with qml.tape.QubitParamShiftTape() as tape: for i in range(n_wires): qml.RX(weights[i], wires=i) qml.CNOT(wires=[0, 1]) qml.CNOT(wires=[2, 1]) qml.CNOT(wires=[3, 1]) qml.CNOT(wires=[4, 3]) qml.expval(qml.PauliZ(1)) print(tape.hessian(dev))
The Hessian is not yet supported via classical machine learning interfaces, but will be added in a future release.
-
Two new error channels,
BitFlip
andPhaseFlip
have been added. (#954)They can be used in the same manner as existing error channels:
dev = qml.device("default.mixed", wires=2) @qml.qnode(dev) def circuit(): qml.RX(0.3, wires=0) qml.RY(0.5, wires=1) qml.BitFlip(0.01, wires=0) qml.PhaseFlip(0.01, wires=1) return qml.expval(qml.PauliZ(0))
-
Apply permutations to wires using the
Permute
subroutine. (#952)import pennylane as qml dev = qml.device('default.qubit', wires=5) @qml.qnode(dev) def apply_perm(): # Send contents of wire 4 to wire 0, of wire 2 to wire 1, etc. qml.templates.Permute([4, 2, 0, 1, 3], wires=dev.wires) return qml.expval(qml.PauliZ(0))
-
The
qml.metric_tensor
function transforms a QNode to produce the Fubini-Study metric tensor with full autodifferentiation support---even on hardware. (#1014)Consider the following QNode:
dev = qml.device("default.qubit", wires=3) @qml.qnode(dev, interface="autograd") def circuit(weights): # layer 1 qml.RX(weights[0, 0], wires=0) qml.RX(weights[0, 1], wires=1) qml.CNOT(wires=[0, 1]) qml.CNOT(wires=[1, 2]) # layer 2 qml.RZ(weights[1, 0], wires=0) qml.RZ(weights[1, 1], wires=2) qml.CNOT(wires=[0, 1]) qml.CNOT(wires=[1, 2]) return qml.expval(qml.PauliZ(0) @ qml.PauliZ(1)), qml.expval(qml.PauliY(2))
We can use the
metric_tensor
function to generate a new function, that returns the metric tensor of this QNode:>>> met_fn = qml.metric_tensor(circuit) >>> weights = np.array([[0.1, 0.2, 0.3], [0.4, 0.5, 0.6]], requires_grad=True) >>> met_fn(weights) tensor([[0.25 , 0. , 0. , 0. ], [0. , 0.25 , 0. , 0. ], [0. , 0. , 0.0025, 0.0024], [0. , 0. , 0.0024, 0.0123]], requires_grad=True)
The returned metric tensor is also fully differentiable, in all interfaces. For example, differentiating the
(3, 2)
element:>>> grad_fn = qml.grad(lambda x: met_fn(x)[3, 2]) >>> grad_fn(weights) array([[ 0.04867729, -0.00049502, 0. ], [ 0. , 0. , 0. ]])
Differentiation is also supported using Torch, Jax, and TensorFlow.
-
Adds the new function
qml.math.cov_matrix()
. This function accepts a list of commuting observables, and the probability distribution in the shared observable eigenbasis after the application of an ansatz. It uses these to construct the covariance matrix in a framework independent manner, such that the output covariance matrix is autodifferentiable. (#1012)For example, consider the following ansatz and observable list:
obs_list = [qml.PauliX(0) @ qml.PauliZ(1), qml.PauliY(2)] ansatz = qml.templates.StronglyEntanglingLayers
We can construct a QNode to output the probability distribution in the shared eigenbasis of the observables:
dev = qml.device("default.qubit", wires=3) @qml.qnode(dev, interface="autograd") def circuit(weights): ansatz(weights, wires=[0, 1, 2]) # rotate into the basis of the observables for o in obs_list: o.diagonalizing_gates() return qml.probs(wires=[0, 1, 2])
We can now compute the covariance matrix:
>>> weights = qml.init.strong_ent_layers_normal(n_layers=2, n_wires=3) >>> cov = qml.math.cov_matrix(circuit(weights), obs_list) >>> cov array([[0.98707611, 0.03665537], [0.03665537, 0.99998377]])
Autodifferentiation is fully supported using all interfaces:
>>> cost_fn = lambda weights: qml.math.cov_matrix(circuit(weights), obs_list)[0, 1] >>> qml.grad(cost_fn)(weights)[0] array([[[ 4.94240914e-17, -2.33786398e-01, -1.54193959e-01], [-3.05414996e-17, 8.40072236e-04, 5.57884080e-04], [ 3.01859411e-17, 8.60411436e-03, 6.15745204e-04]], [[ 6.80309533e-04, -1.23162742e-03, 1.08729813e-03], [-1.53863193e-01, -1.38700657e-02, -1.36243323e-01], [-1.54665054e-01, -1.89018172e-02, -1.56415558e-01]]])
-
A new
qml.draw
function is available, allowing QNodes to be easily drawn without execution by providing example input. (#962)@qml.qnode(dev) def circuit(a, w): qml.Hadamard(0) qml.CRX(a, wires=[0, 1]) qml.Rot(*w, wires=[1]) qml.CRX(-a, wires=[0, 1]) return qml.expval(qml.PauliZ(0) @ qml.PauliZ(1))
The QNode circuit structure may depend on the input arguments; this is taken into account by passing example QNode arguments to the
qml.draw()
drawing function:>>> drawer = qml.draw(circuit) >>> result = drawer(a=2.3, w=[1.2, 3.2, 0.7]) >>> print(result) 0: ──H──╭C────────────────────────────╭C─────────╭┤ ⟨Z ⊗ Z⟩ 1: ─────╰RX(2.3)──Rot(1.2, 3.2, 0.7)──╰RX(-2.3)──╰┤ ⟨Z ⊗ Z⟩
-
The new core of PennyLane, rewritten from the ground up and developed over the last few release cycles, has achieved feature parity and has been made the new default in PennyLane v0.14. The old core has been marked as deprecated, and will be removed in an upcoming release. (#1046) (#1040) (#1034) (#1035) (#1027) (#1026) (#1021) (#1054) (#1049)
While high-level PennyLane code and tutorials remain unchanged, the new core provides several advantages and improvements:
-
Faster and more optimized: The new core provides various performance optimizations, reducing pre- and post-processing overhead, and reduces the number of quantum evaluations in certain cases.
-
Support for in-QNode classical processing: this allows for differentiable classical processing within the QNode.
dev = qml.device("default.qubit", wires=1) @qml.qnode(dev, interface="tf") def circuit(p): qml.RX(tf.sin(p[0])**2 + p[1], wires=0) return qml.expval(qml.PauliZ(0))
The classical processing functions used within the QNode must match the QNode interface. Here, we use TensorFlow:
>>> params = tf.Variable([0.5, 0.1], dtype=tf.float64) >>> with tf.GradientTape() as tape: ... res = circuit(params) >>> grad = tape.gradient(res, params) >>> print(res) tf.Tensor(0.9460913127754935, shape=(), dtype=float64) >>> print(grad) tf.Tensor([-0.27255248 -0.32390003], shape=(2,), dtype=float64)
As a result of this change, quantum decompositions that require classical processing are fully supported and end-to-end differentiable in tape mode.
-
No more Variable wrapping: QNode arguments no longer become
Variable
objects within the QNode.dev = qml.device("default.qubit", wires=1) @qml.qnode(dev) def circuit(x): print("Parameter value:", x) qml.RX(x, wires=0) return qml.expval(qml.PauliZ(0))
Internal QNode parameters can be easily inspected, printed, and manipulated:
>>> circuit(0.5) Parameter value: 0.5 tensor(0.87758256, requires_grad=True)
-
Less restrictive QNode signatures: There is no longer any restriction on the QNode signature; the QNode can be defined and called following the same rules as standard Python functions.
For example, the following QNode uses positional, named, and variable keyword arguments:
x = torch.tensor(0.1, requires_grad=True) y = torch.tensor([0.2, 0.3], requires_grad=True) z = torch.tensor(0.4, requires_grad=True) @qml.qnode(dev, interface="torch") def circuit(p1, p2=y, **kwargs): qml.RX(p1, wires=0) qml.RY(p2[0] * p2[1], wires=0) qml.RX(kwargs["p3"], wires=0) return qml.var(qml.PauliZ(0))
When we call the QNode, we may pass the arguments by name even if defined positionally; any argument not provided will use the default value.
>>> res = circuit(p1=x, p3=z) >>> print(res) tensor(0.2327, dtype=torch.float64, grad_fn=<SelectBackward>) >>> res.backward() >>> print(x.grad, y.grad, z.grad) tensor(0.8396) tensor([0.0289, 0.0193]) tensor(0.8387)
This extends to the
qnn
module, whereKerasLayer
andTorchLayer
modules can be created from QNodes with unrestricted signatures. -
Smarter measurements: QNodes can now measure wires more than once, as long as all observables are commuting:
@qml.qnode(dev) def circuit(x): qml.RX(x, wires=0) return [ qml.expval(qml.PauliZ(0)), qml.expval(qml.PauliZ(0) @ qml.PauliZ(1)) ]
Further, the
qml.ExpvalCost()
function allows for optimizing measurements to reduce the number of quantum evaluations required.
With the new PennyLane core, there are a few small breaking changes, detailed below in the 'Breaking Changes' section.
-
-
The built-in PennyLane optimizers allow more flexible cost functions. The cost function passed to most optimizers may accept any combination of trainable arguments, non-trainable arguments, and keyword arguments. (#959) (#1053)
The full changes apply to:
AdagradOptimizer
AdamOptimizer
GradientDescentOptimizer
MomentumOptimizer
NesterovMomentumOptimizer
RMSPropOptimizer
RotosolveOptimizer
The
requires_grad=False
property must mark any non-trainable constant argument. TheRotoselectOptimizer
allows passing only keyword arguments.Example use:
def cost(x, y, data, scale=1.0): return scale * (x[0]-data)**2 + scale * (y-data)**2 x = np.array([1.], requires_grad=True) y = np.array([1.0]) data = np.array([2.], requires_grad=False) opt = qml.GradientDescentOptimizer() # the optimizer step and step_and_cost methods can # now update multiple parameters at once x_new, y_new, data = opt.step(cost, x, y, data, scale=0.5) (x_new, y_new, data), value = opt.step_and_cost(cost, x, y, data, scale=0.5) # list and tuple unpacking is also supported params = (x, y, data) params = opt.step(cost, *params)
-
The circuit drawer has been updated to support the inclusion of unused or inactive wires, by passing the
show_all_wires
argument. (#1033)dev = qml.device('default.qubit', wires=[-1, "a", "q2", 0]) @qml.qnode(dev) def circuit(): qml.Hadamard(wires=-1) qml.CNOT(wires=[-1, "q2"]) return qml.expval(qml.PauliX(wires="q2"))
>>> print(qml.draw(circuit, show_all_wires=True)()) >>> -1: ──H──╭C──┤ a: ─────│───┤ q2: ─────╰X──┤ ⟨X⟩ 0: ─────────┤
-
The logic for choosing the 'best' differentiation method has been altered to improve performance. (#1008)
-
If the device provides its own gradient, this is now the preferred differentiation method.
-
If a device provides additional interface-specific versions that natively support classical backpropagation, this is now preferred over the parameter-shift rule.
Devices define additional interface-specific devices via their
capabilities()
dictionary. For example,default.qubit
supports supplementary devices for TensorFlow, Autograd, and JAX:{ "passthru_devices": { "tf": "default.qubit.tf", "autograd": "default.qubit.autograd", "jax": "default.qubit.jax", }, }
As a result of this change, if the QNode
diff_method
is not explicitly provided, it is possible that the QNode will run on a supplementary device of the device that was specifically provided:dev = qml.device("default.qubit", wires=2) qml.QNode(dev) # will default to backprop on default.qubit.autograd qml.QNode(dev, interface="tf") # will default to backprop on default.qubit.tf qml.QNode(dev, interface="jax") # will default to backprop on default.qubit.jax
-
-
The
default.qubit
device has been updated so that internally it applies operations in a more functional style, i.e., by accepting an input state and returning an evolved state. (#1025) -
A new test series,
pennylane/devices/tests/test_compare_default_qubit.py
, has been added, allowing to test if a chosen device gives the same result asdefault.qubit
. (#897)Three tests are added:
test_hermitian_expectation
,test_pauliz_expectation_analytic
, andtest_random_circuit
.
-
Adds the following agnostic tensor manipulation functions to the
qml.math
module:abs
,angle
,arcsin
,concatenate
,dot
,squeeze
,sqrt
,sum
,take
,where
. These functions are required to fully support end-to-end differentiable Mottonen and Amplitude embedding. (#922) (#1011) -
The
qml.math
module now supports JAX. (#985) -
Several improvements have been made to the
Wires
class to reduce overhead and simplify the logic of how wire labels are interpreted: (#1019) (#1010) (#1005) (#983) (#967)-
If the input
wires
to a wires class instantiationWires(wires)
can be iterated over, its elements are interpreted as wire labels. Otherwise,wires
is interpreted as a single wire label. The only exception to this are strings, which are always interpreted as a single wire label, so users can address wires with labels such as"ancilla"
. -
Any type can now be a wire label as long as it is hashable. The hash is used to establish the uniqueness of two labels.
-
Indexing wires objects now returns a label, instead of a new
Wires
object. For example:>>> w = Wires([0, 1, 2]) >>> w[1] >>> 1
-
The check for uniqueness of wires moved from
Wires
instantiation to theqml.wires._process
function in order to reduce overhead from repeated creation ofWires
instances. -
Calls to the
Wires
class are substantially reduced, for example by avoiding to call Wires on Wires instances onOperation
instantiation, and by using labels instead ofWires
objects inside the default qubit device.
-
-
Adds the
PauliRot
generator to theqml.operation
module. This generator is required to construct the metric tensor. (#963) -
The templates are modified to make use of the new
qml.math
module, for framework-agnostic tensor manipulation. This allows the template library to be differentiable in backpropagation mode (diff_method="backprop"
). (#873) -
The circuit drawer now allows for the wire order to be (optionally) modified: (#992)
>>> dev = qml.device('default.qubit', wires=["a", -1, "q2"]) >>> @qml.qnode(dev) ... def circuit(): ... qml.Hadamard(wires=-1) ... qml.CNOT(wires=["a", "q2"]) ... qml.RX(0.2, wires="a") ... return qml.expval(qml.PauliX(wires="q2"))
Printing with default wire order of the device:
>>> print(circuit.draw()) a: ─────╭C──RX(0.2)──┤ -1: ──H──│────────────┤ q2: ─────╰X───────────┤ ⟨X⟩
Changing the wire order:
>>> print(circuit.draw(wire_order=["q2", "a", -1])) q2: ──╭X───────────┤ ⟨X⟩ a: ──╰C──RX(0.2)──┤ -1: ───H───────────┤
-
QNodes using the new PennyLane core will no longer accept ragged arrays as inputs.
-
When using the new PennyLane core and the Autograd interface, non-differentiable data passed as a QNode argument or a gate must have the
requires_grad
property set toFalse
:@qml.qnode(dev) def circuit(weights, data): basis_state = np.array([1, 0, 1, 1], requires_grad=False) qml.BasisState(basis_state, wires=[0, 1, 2, 3]) qml.templates.AmplitudeEmbedding(data, wires=[0, 1, 2, 3]) qml.templates.BasicEntanglerLayers(weights, wires=[0, 1, 2, 3]) return qml.probs(wires=0) data = np.array(data, requires_grad=False) weights = np.array(weights, requires_grad=True) circuit(weights, data)
-
Fixes an issue where if the constituent observables of a tensor product do not exist in the queue, an error is raised. With this fix, they are first queued before annotation occurs. (#1038)
-
Fixes an issue with tape expansions where information about sampling (specifically the
is_sampled
tape attribute) was not preserved. (#1027) -
Tape expansion was not properly taking into devices that supported inverse operations, causing inverse operations to be unnecessarily decomposed. The QNode tape expansion logic, as well as the
Operation.expand()
method, has been modified to fix this. (#956) -
Fixes an issue where the Autograd interface was not unwrapping non-differentiable PennyLane tensors, which can cause issues on some devices. (#941)
-
qml.vqe.Hamiltonian
prints any observable with any number of strings. (#987) -
Fixes a bug where parameter-shift differentiation would fail if the QNode contained a single probability output. (#1007)
-
Fixes an issue when using trainable parameters that are lists/arrays with
tape.vjp
. (#1042) -
The
TensorN
observable is updated to support being copied without any parameters or wires passed. (#1047) -
Fixed deprecation warning when importing
Sequence
fromcollections
instead ofcollections.abc
invqe/vqe.py
. (#1051)
This release contains contributions from (in alphabetical order):
Juan Miguel Arrazola, Thomas Bromley, Olivia Di Matteo, Theodor Isacsson, Josh Izaac, Christina Lee, Alejandro Montanez, Steven Oud, Chase Roberts, Sankalp Sanand, Maria Schuld, Antal Száva, David Wierichs, Jiahao Yao.
-
QNodes in tape mode now support returning observables on the same wire whenever the observables are qubit-wise commuting Pauli words. Qubit-wise commuting observables can be evaluated with a single device run as they are diagonal in the same basis, via a shared set of single-qubit rotations. (#882)
The following example shows a single QNode returning the expectation values of the qubit-wise commuting Pauli words
XX
andXI
:qml.enable_tape() @qml.qnode(dev) def f(x): qml.Hadamard(wires=0) qml.Hadamard(wires=1) qml.CRot(0.1, 0.2, 0.3, wires=[1, 0]) qml.RZ(x, wires=1) return qml.expval(qml.PauliX(0) @ qml.PauliX(1)), qml.expval(qml.PauliX(0))
>>> f(0.4) tensor([0.89431013, 0.9510565 ], requires_grad=True)
-
The
ExpvalCost
class (previouslyVQECost
) now provides observable optimization using theoptimize
argument, resulting in potentially fewer device executions. (#902)This is achieved by separating the observables composing the Hamiltonian into qubit-wise commuting groups and evaluating those groups on a single QNode using functionality from the
qml.grouping
module:qml.enable_tape() commuting_obs = [qml.PauliX(0), qml.PauliX(0) @ qml.PauliZ(1)] H = qml.vqe.Hamiltonian([1, 1], commuting_obs) dev = qml.device("default.qubit", wires=2) ansatz = qml.templates.StronglyEntanglingLayers cost_opt = qml.ExpvalCost(ansatz, H, dev, optimize=True) cost_no_opt = qml.ExpvalCost(ansatz, H, dev, optimize=False) params = qml.init.strong_ent_layers_uniform(3, 2)
Grouping these commuting observables leads to fewer device executions:
>>> cost_opt(params) >>> ex_opt = dev.num_executions >>> cost_no_opt(params) >>> ex_no_opt = dev.num_executions - ex_opt >>> print("Number of executions:", ex_no_opt) Number of executions: 2 >>> print("Number of executions (optimized):", ex_opt) Number of executions (optimized): 1
-
Compute the analytic gradient of quantum circuits in parallel on supported devices. (#840)
This release introduces support for batch execution of circuits, via a new device API method
Device.batch_execute()
. Devices that implement this new API support submitting a batch of circuits for parallel evaluation simultaneously, which can significantly reduce the computation time.Furthermore, if using tape mode and a compatible device, gradient computations will automatically make use of the new batch API---providing a speedup during optimization.
-
Gradient recipes are now much more powerful, allowing for operations to define their gradient via an arbitrary linear combination of circuit evaluations. (#909) (#915)
With this change, gradient recipes can now be of the form :math:
\frac{\partial}{\partial\phi_k}f(\phi_k) = \sum_{i} c_i f(a_i \phi_k + s_i )
, and are no longer restricted to two-term shifts with identical (but opposite in sign) shift values.As a result, PennyLane now supports native analytic quantum gradients for the controlled rotation operations
CRX
,CRY
,CRZ
, andCRot
. This allows for parameter-shift analytic gradients on hardware, without decomposition.Note that this is a breaking change for developers; please see the Breaking Changes section for more details.
-
The
qnn.KerasLayer
class now supports differentiating the QNode through classical backpropagation in tape mode. (#869)qml.enable_tape() dev = qml.device("default.qubit.tf", wires=2) @qml.qnode(dev, interface="tf", diff_method="backprop") def f(inputs, weights): qml.templates.AngleEmbedding(inputs, wires=range(2)) qml.templates.StronglyEntanglingLayers(weights, wires=range(2)) return [qml.expval(qml.PauliZ(i)) for i in range(2)] weight_shapes = {"weights": (3, 2, 3)} qlayer = qml.qnn.KerasLayer(f, weight_shapes, output_dim=2) inputs = tf.constant(np.random.random((4, 2)), dtype=tf.float32) with tf.GradientTape() as tape: out = qlayer(inputs) tape.jacobian(out, qlayer.trainable_weights)
-
Adds the
qml.density_matrix
QNode return with partial trace capabilities. (#878)The density matrix over the provided wires is returned, with all other subsystems traced out.
qml.density_matrix
currently works for both thedefault.qubit
anddefault.mixed
devices.qml.enable_tape() dev = qml.device("default.qubit", wires=2) def circuit(x): qml.PauliY(wires=0) qml.Hadamard(wires=1) return qml.density_matrix(wires=[1]) # wire 0 is traced out
-
Adds the square-root X gate
SX
. (#871)dev = qml.device("default.qubit", wires=1) @qml.qnode(dev) def circuit(): qml.SX(wires=[0]) return qml.expval(qml.PauliZ(wires=[0]))
-
Two new hardware-efficient particle-conserving templates have been implemented to perform VQE-based quantum chemistry simulations. The new templates apply several layers of the particle-conserving entanglers proposed in Figs. 2a and 2b of Barkoutsos et al., arXiv:1805.04340 (#875) (#876)
-
The
QuantumTape
class now contains basic resource estimation functionality. The methodtape.get_resources()
returns a dictionary with a list of the constituent operations and the number of times they appear in the circuit. Similarly,tape.get_depth()
computes the circuit depth. (#862)>>> with qml.tape.QuantumTape() as tape: ... qml.Hadamard(wires=0) ... qml.RZ(0.26, wires=1) ... qml.CNOT(wires=[1, 0]) ... qml.Rot(1.8, -2.7, 0.2, wires=0) ... qml.Hadamard(wires=1) ... qml.CNOT(wires=[0, 1]) ... qml.expval(qml.PauliZ(0) @ qml.PauliZ(1)) >>> tape.get_resources() {'Hadamard': 2, 'RZ': 1, 'CNOT': 2, 'Rot': 1} >>> tape.get_depth() 4
-
The number of device executions over a QNode's lifetime can now be returned using
num_executions
. (#853)>>> dev = qml.device("default.qubit", wires=2) >>> @qml.qnode(dev) ... def circuit(x, y): ... qml.RX(x, wires=[0]) ... qml.RY(y, wires=[1]) ... qml.CNOT(wires=[0, 1]) ... return qml.expval(qml.PauliZ(0) @ qml.PauliX(1)) >>> for _ in range(10): ... circuit(0.432, 0.12) >>> print(dev.num_executions) 10
-
Support for tape mode has improved across PennyLane. The following features now work in tape mode:
-
A new function,
qml.refresh_devices()
, has been added, allowing PennyLane to rescan installed PennyLane plugins and refresh the device list. In addition, theqml.device
loader will attempt to refresh devices if the required plugin device cannot be found. This will result in an improved experience if installing PennyLane and plugins within a running Python session (for example, on Google Colab), and avoid the need to restart the kernel/runtime. (#907) -
When using
grad_fn = qml.grad(cost)
to compute the gradient of a cost function with the Autograd interface, the value of the intermediate forward pass is now available via thegrad_fn.forward
property (#914):def cost_fn(x, y): return 2 * np.sin(x[0]) * np.exp(-x[1]) + x[0] ** 3 + np.cos(y) params = np.array([0.1, 0.5], requires_grad=True) data = np.array(0.65, requires_grad=False) grad_fn = qml.grad(cost_fn) grad_fn(params, data) # perform backprop and evaluate the gradient grad_fn.forward # the cost function value
-
Gradient-based optimizers now have a
step_and_cost
method that returns both the next step as well as the objective (cost) function output. (#916)>>> opt = qml.GradientDescentOptimizer() >>> params, cost = opt.step_and_cost(cost_fn, params)
-
PennyLane provides a new experimental module
qml.proc
which provides framework-agnostic processing functions for array and tensor manipulations. (#886)Given the input tensor-like object, the call is dispatched to the corresponding array manipulation framework, allowing for end-to-end differentiation to be preserved.
>>> x = torch.tensor([1., 2.]) >>> qml.proc.ones_like(x) tensor([1, 1]) >>> y = tf.Variable([[0], [5]]) >>> qml.proc.ones_like(y, dtype=np.complex128) <tf.Tensor: shape=(2, 1), dtype=complex128, numpy= array([[1.+0.j], [1.+0.j]])>
Note that these functions are experimental, and only a subset of common functionality is supported. Furthermore, the names and behaviour of these functions may differ from similar functions in common frameworks; please refer to the function docstrings for more details.
-
The gradient methods in tape mode now fully separate the quantum and classical processing. Rather than returning the evaluated gradients directly, they now return a tuple containing the required quantum and classical processing steps. (#840)
def gradient_method(idx, param, **options): # generate the quantum tapes that must be computed # to determine the quantum gradient tapes = quantum_gradient_tapes(self) def processing_fn(results): # perform classical processing on the evaluated tapes # returning the evaluated quantum gradient return classical_processing(results) return tapes, processing_fn
The
JacobianTape.jacobian()
method has been similarly modified to accumulate all gradient quantum tapes and classical processing functions, evaluate all quantum tapes simultaneously, and then apply the post-processing functions to the evaluated tape results. -
The MultiRZ gate now has a defined generator, allowing it to be used in quantum natural gradient optimization. (#912)
-
The CRot gate now has a
decomposition
method, which breaks the gate down into rotations and CNOT gates. This allowsCRot
to be used on devices that do not natively support it. (#908) -
The classical processing in the
MottonenStatePreparation
template has been largely rewritten to use dense matrices and tensor manipulations wherever possible. This is in preparation to support differentiation through the template in the future. (#864) -
Device-based caching has replaced QNode caching. Caching is now accessed by passing a
cache
argument to the device. (#851)The
cache
argument should be an integer specifying the size of the cache. For example, a cache of size 10 is created using:>>> dev = qml.device("default.qubit", wires=2, cache=10)
-
The
Operation
,Tensor
, andMeasurementProcess
classes now have the__copy__
special method defined. (#840)This allows us to ensure that, when a shallow copy is performed of an operation, the mutable list storing the operation parameters is also shallow copied. Both the old operation and the copied operation will continue to share the same parameter data,
>>> import copy >>> op = qml.RX(0.2, wires=0) >>> op2 = copy.copy(op) >>> op.data[0] is op2.data[0] True
however the list container is not a reference:
>>> op.data is op2.data False
This allows the parameters of the copied operation to be modified, without mutating the parameters of the original operation.
-
The
QuantumTape.copy
method has been tweaked so that (#840):-
Optionally, the tape's operations are shallow copied in addition to the tape by passing the
copy_operations=True
boolean flag. This allows the copied tape's parameters to be mutated without affecting the original tape's parameters. (Note: the two tapes will share parameter data until one of the tapes has their parameter list modified.) -
Copied tapes can be cast to another
QuantumTape
subclass by passing thetape_cls
keyword argument.
-
-
Updated how parameter-shift gradient recipes are defined for operations, allowing for gradient recipes that are specified as an arbitrary number of terms. (#909)
Previously,
Operation.grad_recipe
was restricted to two-term parameter-shift formulas. With this change, the gradient recipe now contains elements of the form :math:[c_i, a_i, s_i]
, resulting in a gradient recipe of :math:\frac{\partial}{\partial\phi_k}f(\phi_k) = \sum_{i} c_i f(a_i \phi_k + s_i )
.As this is a breaking change, all custom operations with defined gradient recipes must be updated to continue working with PennyLane 0.13. Note though that if
grad_recipe = None
, the default gradient recipe remains unchanged, and corresponds to the two terms :math:[c_0, a_0, s_0]=[1/2, 1, \pi/2]
and :math:[c_1, a_1, s_1]=[-1/2, 1, -\pi/2]
for every parameter.
- The
VQECost
class has been renamed toExpvalCost
to reflect its general applicability beyond VQE. Use ofVQECost
is still possible but will result in a deprecation warning. (#913)
-
The
default.qubit.tf
device is updated to handle TensorFlow objects (e.g.,tf.Variable
) as gate parameters correctly when using theMultiRZ
andCRot
operations. (#921) -
PennyLane tensor objects are now unwrapped in BaseQNode when passed as a keyword argument to the quantum function. (#903) (#893)
-
The new tape mode now prevents multiple observables from being evaluated on the same wire if the observables are not qubit-wise commuting Pauli words. (#882)
-
Fixes a bug in
default.qubit
whereby inverses of common gates were not being applied via efficient gate-specific methods, instead falling back to matrix-vector multiplication. The following gates were affected:PauliX
,PauliY
,PauliZ
,Hadamard
,SWAP
,S
,T
,CNOT
,CZ
. (#872) -
The
PauliRot
operation now gracefully handles single-qubit Paulis, and all-identity Paulis (#860). -
Fixes a bug whereby binary Python operators were not properly propagating the
requires_grad
attribute to the output tensor. (#889) -
Fixes a bug which prevents
TorchLayer
from doingbackward
when CUDA is enabled. (#899) -
Fixes a bug where multi-threaded execution of
QNodeCollection
sometimes fails because of simultaneous queuing. This is fixed by adding thread locking during queuing. (#910) -
Fixes a bug in
QuantumTape.set_parameters()
. The previous implementation assumed that theself.trainable_parms
set would always be iterated over in increasing integer order. However, this is not guaranteed behaviour, and can lead to the incorrect tape parameters being set if this is not the case. (#923) -
Fixes broken error message if a QNode is instantiated with an unknown exception. (#930)
This release contains contributions from (in alphabetical order):
Juan Miguel Arrazola, Thomas Bromley, Christina Lee, Alain Delgado Gran, Olivia Di Matteo, Anthony Hayes, Theodor Isacsson, Josh Izaac, Soran Jahangiri, Nathan Killoran, Shumpei Kobayashi, Romain Moyard, Zeyue Niu, Maria Schuld, Antal Száva.
-
PennyLane now supports a new device,
default.mixed
, designed for simulating mixed-state quantum computations. This enables native support for implementing noisy channels in a circuit, which generally map pure states to mixed states. (#794) (#807) (#819)The device can be initialized as
>>> dev = qml.device("default.mixed", wires=1)
This allows the construction of QNodes that include non-unitary operations, such as noisy channels:
>>> @qml.qnode(dev) ... def circuit(params): ... qml.RX(params[0], wires=0) ... qml.RY(params[1], wires=0) ... qml.AmplitudeDamping(0.5, wires=0) ... return qml.expval(qml.PauliZ(0)) >>> print(circuit([0.54, 0.12])) 0.9257702929524184 >>> print(circuit([0, np.pi])) 0.0
-
The new
grouping
module provides functionality for grouping simultaneously measurable Pauli word observables. (#761) (#850) (#852)-
The
optimize_measurements
function will take as input a list of Pauli word observables and their corresponding coefficients (if any), and will return the partitioned Pauli terms diagonalized in the measurement basis and the corresponding diagonalizing circuits.from pennylane.grouping import optimize_measurements h, nr_qubits = qml.qchem.molecular_hamiltonian("h2", "h2.xyz") rotations, grouped_ops, grouped_coeffs = optimize_measurements(h.ops, h.coeffs, grouping="qwc")
The diagonalizing circuits of
rotations
correspond to the diagonalized Pauli word groupings ofgrouped_ops
. -
Pauli word partitioning utilities are performed by the
PauliGroupingStrategy
class. An input list of Pauli words can be partitioned into mutually commuting, qubit-wise-commuting, or anticommuting groupings.For example, partitioning Pauli words into anticommutative groupings by the Recursive Largest First (RLF) graph colouring heuristic:
from pennylane import PauliX, PauliY, PauliZ, Identity from pennylane.grouping import group_observables pauli_words = [ Identity('a') @ Identity('b'), Identity('a') @ PauliX('b'), Identity('a') @ PauliY('b'), PauliZ('a') @ PauliX('b'), PauliZ('a') @ PauliY('b'), PauliZ('a') @ PauliZ('b') ] groupings = group_observables(pauli_words, grouping_type='anticommuting', method='rlf')
-
Various utility functions are included for obtaining and manipulating Pauli words in the binary symplectic vector space representation.
For instance, two Pauli words may be converted to their binary vector representation:
>>> from pennylane.grouping import pauli_to_binary >>> from pennylane.wires import Wires >>> wire_map = {Wires('a'): 0, Wires('b'): 1} >>> pauli_vec_1 = pauli_to_binary(qml.PauliX('a') @ qml.PauliY('b')) >>> pauli_vec_2 = pauli_to_binary(qml.PauliZ('a') @ qml.PauliZ('b')) >>> pauli_vec_1 [1. 1. 0. 1.] >>> pauli_vec_2 [0. 0. 1. 1.]
Their product up to a phase may be computed by taking the sum of their binary vector representations, and returned in the operator representation.
>>> from pennylane.grouping import binary_to_pauli >>> binary_to_pauli((pauli_vec_1 + pauli_vec_2) % 2, wire_map) Tensor product ['PauliY', 'PauliX']: 0 params, wires ['a', 'b']
For more details on the grouping module, see the grouping module documentation
-
-
The quantum state of a QNode can now be returned using the
qml.state()
return function. (#818)import pennylane as qml dev = qml.device("default.qubit", wires=3) qml.enable_tape() @qml.qnode(dev) def qfunc(x, y): qml.RZ(x, wires=0) qml.CNOT(wires=[0, 1]) qml.RY(y, wires=1) qml.CNOT(wires=[0, 2]) return qml.state() >>> qfunc(0.56, 0.1) array([0.95985437-0.27601028j, 0. +0.j , 0.04803275-0.01381203j, 0. +0.j , 0. +0.j , 0. +0.j , 0. +0.j , 0. +0.j ])
Differentiating the state is currently available when using the classical backpropagation differentiation method (
diff_method="backprop"
) with a compatible device, and when using the new tape mode.
-
PennyLane now includes standard channels such as the Amplitude-damping, Phase-damping, and Depolarizing channels, as well as the ability to make custom qubit channels. (#760) (#766) (#778)
-
The controlled-Y operation is now available via
qml.CY
. For devices that do not natively support the controlled-Y operation, it will be decomposed intoqml.RY
,qml.CNOT
, andqml.S
operations. (#806)
-
The new PennyLane
tape
module provides a re-formulated QNode class, rewritten from the ground-up, that uses a newQuantumTape
object to represent the QNode's quantum circuit. Tape mode provides several advantages over the standard PennyLane QNode. (#785) (#792) (#796) (#800) (#803) (#804) (#805) (#808) (#810) (#811) (#815) (#820) (#823) (#824) (#829)-
Support for in-QNode classical processing: Tape mode allows for differentiable classical processing within the QNode.
-
No more Variable wrapping: In tape mode, QNode arguments no longer become
Variable
objects within the QNode. -
Less restrictive QNode signatures: There is no longer any restriction on the QNode signature; the QNode can be defined and called following the same rules as standard Python functions.
-
Unifying all QNodes: The tape-mode QNode merges all QNodes (including the
JacobianQNode
and thePassthruQNode
) into a single unified QNode, with identical behaviour regardless of the differentiation type. -
Optimizations: Tape mode provides various performance optimizations, reducing pre- and post-processing overhead, and reduces the number of quantum evaluations in certain cases.
Note that tape mode is experimental, and does not currently have feature-parity with the existing QNode. Feedback and bug reports are encouraged and will help improve the new tape mode.
Tape mode can be enabled globally via the
qml.enable_tape
function, without changing your PennyLane code:qml.enable_tape() dev = qml.device("default.qubit", wires=1) @qml.qnode(dev, interface="tf") def circuit(p): print("Parameter value:", p) qml.RX(tf.sin(p[0])**2 + p[1], wires=0) return qml.expval(qml.PauliZ(0))
For more details, please see the tape mode documentation.
-
-
QNode caching has been introduced, allowing the QNode to keep track of the results of previous device executions and reuse those results in subsequent calls. Note that QNode caching is only supported in the new and experimental tape-mode. (#817)
Caching is available by passing a
caching
argument to the QNode:dev = qml.device("default.qubit", wires=2) qml.enable_tape() @qml.qnode(dev, caching=10) # cache up to 10 evaluations def qfunc(x): qml.RX(x, wires=0) qml.RX(0.3, wires=1) qml.CNOT(wires=[0, 1]) return qml.expval(qml.PauliZ(1)) qfunc(0.1) # first evaluation executes on the device qfunc(0.1) # second evaluation accesses the cached result
-
Sped up the application of certain gates in
default.qubit
by using array/tensor manipulation tricks. The following gates are affected:PauliX
,PauliY
,PauliZ
,Hadamard
,SWAP
,S
,T
,CNOT
,CZ
. (#772) -
The computation of marginal probabilities has been made more efficient for devices with a large number of wires, achieving in some cases a 5x speedup. (#799)
-
Adds arithmetic operations (addition, tensor product, subtraction, and scalar multiplication) between
Hamiltonian
,Tensor
, andObservable
objects, and inline arithmetic operations between Hamiltonians and other observables. (#765)Hamiltonians can now easily be defined as sums of observables:
>>> H = 3 * qml.PauliZ(0) - (qml.PauliX(0) @ qml.PauliX(1)) + qml.Hamiltonian([4], [qml.PauliZ(0)]) >>> print(H) (7.0) [Z0] + (-1.0) [X0 X1]
-
Adds
compare()
method toObservable
andHamiltonian
classes, which allows for comparison between observable quantities. (#765)>>> H = qml.Hamiltonian([1], [qml.PauliZ(0)]) >>> obs = qml.PauliZ(0) @ qml.Identity(1) >>> print(H.compare(obs)) True
>>> H = qml.Hamiltonian([2], [qml.PauliZ(0)]) >>> obs = qml.PauliZ(1) @ qml.Identity(0) >>> print(H.compare(obs)) False
-
Adds
simplify()
method to theHamiltonian
class. (#765)>>> H = qml.Hamiltonian([1, 2], [qml.PauliZ(0), qml.PauliZ(0) @ qml.Identity(1)]) >>> H.simplify() >>> print(H) (3.0) [Z0]
-
Added a new bit-flip mixer to the
qml.qaoa
module. (#774) -
Summation of two
Wires
objects is now supported and will return aWires
object containing the set of all wires defined by the terms in the summation. (#812)
-
The PennyLane NumPy module now returns scalar (zero-dimensional) arrays where Python scalars were previously returned. (#820) (#833)
For example, this affects array element indexing, and summation:
>>> x = np.array([1, 2, 3], requires_grad=False) >>> x[0] tensor(1, requires_grad=False) >>> np.sum(x) tensor(6, requires_grad=True)
This may require small updates to user code. A convenience method,
np.tensor.unwrap()
, has been added to help ease the transition. This converts PennyLane NumPy tensors to standard NumPy arrays and Python scalars:>>> x = np.array(1.543, requires_grad=False) >>> x.unwrap() 1.543
Note, however, that information regarding array differentiability will be lost.
-
The device capabilities dictionary has been redesigned, for clarity and robustness. In particular, the capabilities dictionary is now inherited from the parent class, various keys have more expressive names, and all keys are now defined in the base device class. For more details, please refer to the developer documentation. (#781)
-
Changed to use lists for storing variable values inside
BaseQNode
allowing complex matrices to be passed toQubitUnitary
. (#773) -
Fixed a bug within
default.qubit
, resulting in greater efficiency when applying a state vector to all wires on the device. (#849)
- Equations have been added to the
qml.sample
andqml.probs
docstrings to clarify the mathematical foundation of the performed measurements. (#843)
This release contains contributions from (in alphabetical order):
Aroosa Ijaz, Juan Miguel Arrazola, Thomas Bromley, Jack Ceroni, Alain Delgado Gran, Josh Izaac, Soran Jahangiri, Nathan Killoran, Robert Lang, Cedric Lin, Olivia Di Matteo, Nicolás Quesada, Maria Schuld, Antal Száva.
-
Added a new device,
default.qubit.autograd
, a pure-state qubit simulator written using Autograd. This device supports classical backpropagation (diff_method="backprop"
); this can be faster than the parameter-shift rule for computing quantum gradients when the number of parameters to be optimized is large. (#721)>>> dev = qml.device("default.qubit.autograd", wires=1) >>> @qml.qnode(dev, diff_method="backprop") ... def circuit(x): ... qml.RX(x[1], wires=0) ... qml.Rot(x[0], x[1], x[2], wires=0) ... return qml.expval(qml.PauliZ(0)) >>> weights = np.array([0.2, 0.5, 0.1]) >>> grad_fn = qml.grad(circuit) >>> print(grad_fn(weights)) array([-2.25267173e-01, -1.00864546e+00, 6.93889390e-18])
See the device documentation for more details.
-
A new experimental C++ state-vector simulator device is now available,
lightning.qubit
. It uses the C++ Eigen library to perform fast linear algebra calculations for simulating quantum state-vector evolution.lightning.qubit
is currently in beta; it can be installed viapip
:$ pip install pennylane-lightning
Once installed, it can be used as a PennyLane device:
>>> dev = qml.device("lightning.qubit", wires=2)
For more details, please see the lightning qubit documentation.
-
Added built-in QAOA functionality via the new
qml.qaoa
module. (#712) (#718) (#741) (#720)This includes the following features:
-
New
qml.qaoa.x_mixer
andqml.qaoa.xy_mixer
functions for defining Pauli-X and XY mixer Hamiltonians. -
MaxCut: The
qml.qaoa.maxcut
function allows easy construction of the cost Hamiltonian and recommended mixer Hamiltonian for solving the MaxCut problem for a supplied graph. -
Layers:
qml.qaoa.cost_layer
andqml.qaoa.mixer_layer
take cost and mixer Hamiltonians, respectively, and apply the corresponding QAOA cost and mixer layers to the quantum circuit
For example, using PennyLane to construct and solve a MaxCut problem with QAOA:
wires = range(3) graph = Graph([(0, 1), (1, 2), (2, 0)]) cost_h, mixer_h = qaoa.maxcut(graph) def qaoa_layer(gamma, alpha): qaoa.cost_layer(gamma, cost_h) qaoa.mixer_layer(alpha, mixer_h) def antatz(params, **kwargs): for w in wires: qml.Hadamard(wires=w) # repeat the QAOA layer two times qml.layer(qaoa_layer, 2, params[0], params[1]) dev = qml.device('default.qubit', wires=len(wires)) cost_function = qml.VQECost(ansatz, cost_h, dev)
-
-
Added an
ApproxTimeEvolution
template to the PennyLane templates module, which can be used to implement Trotterized time-evolution under a Hamiltonian. (#710) -
Added a
qml.layer
template-constructing function, which takes a unitary, and repeatedly applies it on a set of wires to a given depth. (#723)def subroutine(): qml.Hadamard(wires=[0]) qml.CNOT(wires=[0, 1]) qml.PauliX(wires=[1]) dev = qml.device('default.qubit', wires=3) @qml.qnode(dev) def circuit(): qml.layer(subroutine, 3) return [qml.expval(qml.PauliZ(0)), qml.expval(qml.PauliZ(1))]
This creates the following circuit:
>>> circuit() >>> print(circuit.draw()) 0: ──H──╭C──X──H──╭C──X──H──╭C──X──┤ ⟨Z⟩ 1: ─────╰X────────╰X────────╰X─────┤ ⟨Z⟩
-
Added the
qml.utils.decompose_hamiltonian
function. This function can be used to decompose a Hamiltonian into a linear combination of Pauli operators. (#671)>>> A = np.array( ... [[-2, -2+1j, -2, -2], ... [-2-1j, 0, 0, -1], ... [-2, 0, -2, -1], ... [-2, -1, -1, 0]]) >>> coeffs, obs_list = decompose_hamiltonian(A)
-
It is now possible to specify custom wire labels, such as
['anc1', 'anc2', 0, 1, 3]
, where the labels can be strings or numbers. (#666)Custom wire labels are defined by passing a list to the
wires
argument when creating the device:>>> dev = qml.device("default.qubit", wires=['anc1', 'anc2', 0, 1, 3])
Quantum operations should then be invoked with these custom wire labels:
>>> @qml.qnode(dev) >>> def circuit(): ... qml.Hadamard(wires='anc2') ... qml.CNOT(wires=['anc1', 3]) ... ...
The existing behaviour, in which the number of wires is specified on device initialization, continues to work as usual. This gives a default behaviour where wires are labelled by consecutive integers.
>>> dev = qml.device("default.qubit", wires=5)
-
An integrated device test suite has been added, which can be used to run basic integration tests on core or external devices. (#695) (#724) (#733)
The test can be invoked against a particular device by calling the
pl-device-test
command line program:$ pl-device-test --device=default.qubit --shots=1234 --analytic=False
If the tests are run on external devices, the device and its dependencies must be installed locally. For more details, please see the plugin test documentation.
-
The functions implementing the quantum circuits building the Unitary Coupled-Cluster (UCCSD) VQE ansatz have been improved, with a more consistent naming convention and improved docstrings. (#748)
The changes include:
-
The terms 1particle-1hole (ph) and 2particle-2hole (pphh) excitations were replaced with the names single and double excitations, respectively.
-
The non-differentiable arguments in the
UCCSD
template were renamed accordingly:ph
→s_wires
,pphh
→d_wires
-
The term virtual, previously used to refer the unoccupied orbitals, was discarded.
-
The Usage Details sections were updated and improved.
-
-
Added support for TensorFlow 2.3 and PyTorch 1.6. (#725)
-
Returning probabilities is now supported from photonic QNodes. As with qubit QNodes, photonic QNodes returning probabilities are end-to-end differentiable. (#699)
>>> dev = qml.device("strawberryfields.fock", wires=2, cutoff_dim=5) >>> @qml.qnode(dev) ... def circuit(a): ... qml.Displacement(a, 0, wires=0) ... return qml.probs(wires=0) >>> print(circuit(0.5)) [7.78800783e-01 1.94700196e-01 2.43375245e-02 2.02812704e-03 1.26757940e-04]
- The
pennylane.plugins
andpennylane.beta.plugins
folders have been renamed topennylane.devices
andpennylane.beta.devices
, to reflect their content better. (#726)
- The PennyLane interface conversion functions can now convert QNodes with pre-existing interfaces. (#707)
- The interfaces section of the documentation has been renamed to 'Interfaces and training', and updated with the latest variable handling details. (#753)
This release contains contributions from (in alphabetical order):
Juan Miguel Arrazola, Thomas Bromley, Jack Ceroni, Alain Delgado Gran, Shadab Hussain, Theodor Isacsson, Josh Izaac, Nathan Killoran, Maria Schuld, Antal Száva, Nicola Vitucci.
-
Added a new device,
default.qubit.tf
, a pure-state qubit simulator written using TensorFlow. As a result, it supports classical backpropagation as a means to compute the Jacobian. This can be faster than the parameter-shift rule for computing quantum gradients when the number of parameters to be optimized is large.default.qubit.tf
is designed to be used with end-to-end classical backpropagation (diff_method="backprop"
) with the TensorFlow interface. This is the default method of differentiation when creating a QNode with this device.Using this method, the created QNode is a 'white-box' that is tightly integrated with your TensorFlow computation, including AutoGraph support:
>>> dev = qml.device("default.qubit.tf", wires=1) >>> @tf.function ... @qml.qnode(dev, interface="tf", diff_method="backprop") ... def circuit(x): ... qml.RX(x[1], wires=0) ... qml.Rot(x[0], x[1], x[2], wires=0) ... return qml.expval(qml.PauliZ(0)) >>> weights = tf.Variable([0.2, 0.5, 0.1]) >>> with tf.GradientTape() as tape: ... res = circuit(weights) >>> print(tape.gradient(res, weights)) tf.Tensor([-2.2526717e-01 -1.0086454e+00 1.3877788e-17], shape=(3,), dtype=float32)
See the
default.qubit.tf
documentation for more details. -
The default.tensor plugin has been significantly upgraded. It now allows two different tensor network representations to be used:
"exact"
and"mps"
. The former uses a exact factorized representation of quantum states, while the latter uses a matrix product state representation. (#572) (#599)
-
PennyLane QNodes can now be converted into Torch layers, allowing for creation of quantum and hybrid models using the
torch.nn
API. (#588)A PennyLane QNode can be converted into a
torch.nn
layer using theqml.qnn.TorchLayer
class:>>> @qml.qnode(dev) ... def qnode(inputs, weights_0, weight_1): ... # define the circuit ... # ... >>> weight_shapes = {"weights_0": 3, "weight_1": 1} >>> qlayer = qml.qnn.TorchLayer(qnode, weight_shapes)
A hybrid model can then be easily constructed:
>>> model = torch.nn.Sequential(qlayer, torch.nn.Linear(2, 2))
-
Added a new "reversible" differentiation method which can be used in simulators, but not hardware.
The reversible approach is similar to backpropagation, but trades off extra computation for enhanced memory efficiency. Where backpropagation caches the state tensors at each step during a simulated evolution, the reversible method only caches the final pre-measurement state.
Compared to the parameter-shift method, the reversible method can be faster or slower, depending on the density and location of parametrized gates in a circuit (circuits with higher density of parametrized gates near the end of the circuit will see a benefit). (#670)
>>> dev = qml.device("default.qubit", wires=2) ... @qml.qnode(dev, diff_method="reversible") ... def circuit(x): ... qml.RX(x, wires=0) ... qml.RX(x, wires=0) ... qml.CNOT(wires=[0,1]) ... return qml.expval(qml.PauliZ(0)) >>> qml.grad(circuit)(0.5) (array(-0.47942554),)
-
Added the new templates
UCCSD
,SingleExcitationUnitary
, andDoubleExcitationUnitary
, which together implement the Unitary Coupled-Cluster Singles and Doubles (UCCSD) ansatz to perform VQE-based quantum chemistry simulations using PennyLane-QChem. (#622) (#638) (#654) (#659) (#622) -
Added module
pennylane.qnn.cost
with classSquaredErrorLoss
. The module contains classes to calculate losses and cost functions on circuits with trainable parameters. (#642)
-
Improves the wire management by making the
Operator.wires
attribute awires
object. (#666) -
A significant improvement with respect to how QNodes and interfaces mark quantum function arguments as differentiable when using Autograd, designed to improve performance and make QNodes more intuitive. (#648) (#650)
In particular, the following changes have been made:
-
A new
ndarray
subclasspennylane.numpy.tensor
, which extends NumPy arrays with the keyword argument and attributerequires_grad
. Tensors which haverequires_grad=False
are treated as non-differentiable by the Autograd interface. -
A new subpackage
pennylane.numpy
, which wrapsautograd.numpy
such that NumPy functions accept therequires_grad
keyword argument, and allows Autograd to differentiatepennylane.numpy.tensor
objects. -
The
argnum
argument toqml.grad
is now optional; if not provided, arguments explicitly marked asrequires_grad=False
are excluded for the list of differentiable arguments. The ability to passargnum
has been retained for backwards compatibility, and if present the old behaviour persists.
-
-
The QNode Torch interface now inspects QNode positional arguments. If any argument does not have the attribute
requires_grad=True
, it is automatically excluded from quantum gradient computations. (#652) (#660) -
The QNode TF interface now inspects QNode positional arguments. If any argument is not being watched by a
tf.GradientTape()
, it is automatically excluded from quantum gradient computations. (#655) (#660) -
QNodes have two new public methods:
QNode.set_trainable_args()
andQNode.get_trainable_args()
. These are designed to be called by interfaces, to specify to the QNode which of its input arguments are differentiable. Arguments which are non-differentiable will not be converted to PennyLane Variable objects within the QNode. (#660) -
Added
decomposition
method to PauliX, PauliY, PauliZ, S, T, Hadamard, and PhaseShift gates, which decomposes each of these gates into rotation gates. (#668) -
The
CircuitGraph
class now supports serializing contained circuit operations and measurement basis rotations to an OpenQASM2.0 script via the newCircuitGraph.to_openqasm()
method. (#623)
- Removes support for Python 3.5. (#639)
- Various small typos were fixed.
This release contains contributions from (in alphabetical order):
Thomas Bromley, Jack Ceroni, Alain Delgado Gran, Theodor Isacsson, Josh Izaac, Nathan Killoran, Maria Schuld, Antal Száva, Nicola Vitucci.
-
PennyLane QNodes can now be converted into Keras layers, allowing for creation of quantum and hybrid models using the Keras API. (#529)
A PennyLane QNode can be converted into a Keras layer using the
KerasLayer
class:from pennylane.qnn import KerasLayer @qml.qnode(dev) def circuit(inputs, weights_0, weight_1): # define the circuit # ... weight_shapes = {"weights_0": 3, "weight_1": 1} qlayer = qml.qnn.KerasLayer(circuit, weight_shapes, output_dim=2)
A hybrid model can then be easily constructed:
model = tf.keras.models.Sequential([qlayer, tf.keras.layers.Dense(2)])
-
Added a new type of QNode,
qml.qnodes.PassthruQNode
. For simulators which are coded in an external library which supports automatic differentiation, PennyLane will treat a PassthruQNode as a "white box", and rely on the external library to directly provide gradients via backpropagation. This can be more efficient than the using parameter-shift rule for a large number of parameters. (#488)Currently this behaviour is supported by PennyLane's
default.tensor.tf
device backend, compatible with the'tf'
interface using TensorFlow 2:dev = qml.device('default.tensor.tf', wires=2) @qml.qnode(dev, diff_method="backprop") def circuit(params): qml.RX(params[0], wires=0) qml.RX(params[1], wires=1) qml.CNOT(wires=[0, 1]) return qml.expval(qml.PauliZ(0)) qnode = PassthruQNode(circuit, dev) params = tf.Variable([0.3, 0.1]) with tf.GradientTape() as tape: tape.watch(params) res = qnode(params) grad = tape.gradient(res, params)
-
Added the
qml.RotosolveOptimizer
, a gradient-free optimizer that minimizes the quantum function by updating each parameter, one-by-one, via a closed-form expression while keeping other parameters fixed. (#636) (#539) -
Added the
qml.RotoselectOptimizer
, which uses Rotosolve to minimizes a quantum function with respect to both the rotation operations applied and the rotation parameters. (#636) (#539)For example, given a quantum function
f
that accepts parametersx
and a list of corresponding rotation operationsgenerators
, the Rotoselect optimizer will, at each step, update both the parameter values and the list of rotation gates to minimize the loss:>>> opt = qml.optimize.RotoselectOptimizer() >>> x = [0.3, 0.7] >>> generators = [qml.RX, qml.RY] >>> for _ in range(100): ... x, generators = opt.step(f, x, generators)
-
Added the
PauliRot
gate, which performs an arbitrary Pauli rotation on multiple qubits, and theMultiRZ
gate, which performs a rotation generated by a tensor product of Pauli Z operators. (#559)dev = qml.device('default.qubit', wires=4) @qml.qnode(dev) def circuit(angle): qml.PauliRot(angle, "IXYZ", wires=[0, 1, 2, 3]) return [qml.expval(qml.PauliZ(wire)) for wire in [0, 1, 2, 3]]
>>> circuit(0.4) [1. 0.92106099 0.92106099 1. ] >>> print(circuit.draw()) 0: ──╭RI(0.4)──┤ ⟨Z⟩ 1: ──├RX(0.4)──┤ ⟨Z⟩ 2: ──├RY(0.4)──┤ ⟨Z⟩ 3: ──╰RZ(0.4)──┤ ⟨Z⟩
If the
PauliRot
gate is not supported on the target device, it will be decomposed intoHadamard
,RX
andMultiRZ
gates. Note that identity gates in the Pauli word result in untouched wires:>>> print(circuit.draw()) 0: ───────────────────────────────────┤ ⟨Z⟩ 1: ──H──────────╭RZ(0.4)──H───────────┤ ⟨Z⟩ 2: ──RX(1.571)──├RZ(0.4)──RX(-1.571)──┤ ⟨Z⟩ 3: ─────────────╰RZ(0.4)──────────────┤ ⟨Z⟩
If the
MultiRZ
gate is not supported, it will be decomposed intoCNOT
andRZ
gates:>>> print(circuit.draw()) 0: ──────────────────────────────────────────────────┤ ⟨Z⟩ 1: ──H──────────────╭X──RZ(0.4)──╭X──────H───────────┤ ⟨Z⟩ 2: ──RX(1.571)──╭X──╰C───────────╰C──╭X──RX(-1.571)──┤ ⟨Z⟩ 3: ─────────────╰C───────────────────╰C──────────────┤ ⟨Z⟩
-
PennyLane now provides
DiagonalQubitUnitary
for diagonal gates, that are e.g., encountered in IQP circuits. These kinds of gates can be evaluated much faster on a simulator device. (#567)The gate can be used, for example, to efficiently simulate oracles:
dev = qml.device('default.qubit', wires=3) # Function as a bitstring f = np.array([1, 0, 0, 1, 1, 0, 1, 0]) @qml.qnode(dev) def circuit(weights1, weights2): qml.templates.StronglyEntanglingLayers(weights1, wires=[0, 1, 2]) # Implements the function as a phase-kickback oracle qml.DiagonalQubitUnitary((-1)**f, wires=[0, 1, 2]) qml.templates.StronglyEntanglingLayers(weights2, wires=[0, 1, 2]) return [qml.expval(qml.PauliZ(w)) for w in range(3)]
-
Added the
TensorN
CVObservable that can represent the tensor product of theNumberOperator
on photonic backends. (#608)
-
Added the
ArbitraryUnitary
andArbitraryStatePreparation
templates, which usePauliRot
gates to perform an arbitrary unitary and prepare an arbitrary basis state with the minimal number of parameters. (#590)dev = qml.device('default.qubit', wires=3) @qml.qnode(dev) def circuit(weights1, weights2): qml.templates.ArbitraryStatePreparation(weights1, wires=[0, 1, 2]) qml.templates.ArbitraryUnitary(weights2, wires=[0, 1, 2]) return qml.probs(wires=[0, 1, 2])
-
Added the
IQPEmbedding
template, which encodes inputs into the diagonal gates of an IQP circuit. (#605) -
Added the
SimplifiedTwoDesign
template, which implements the circuit design of Cerezo et al. (2020). (#556) -
Added the
BasicEntanglerLayers
template, which is a simple layer architecture of rotations and CNOT nearest-neighbour entanglers. (#555) -
PennyLane now offers a broadcasting function to easily construct templates:
qml.broadcast()
takes single quantum operations or other templates and applies them to wires in a specific pattern. (#515) (#522) (#526) (#603)For example, we can use broadcast to repeat a custom template across multiple wires:
from pennylane.templates import template @template def mytemplate(pars, wires): qml.Hadamard(wires=wires) qml.RY(pars, wires=wires) dev = qml.device('default.qubit', wires=3) @qml.qnode(dev) def circuit(pars): qml.broadcast(mytemplate, pattern="single", wires=[0,1,2], parameters=pars) return qml.expval(qml.PauliZ(0))
>>> circuit([1, 1, 0.1]) -0.841470984807896 >>> print(circuit.draw()) 0: ──H──RY(1.0)──┤ ⟨Z⟩ 1: ──H──RY(1.0)──┤ 2: ──H──RY(0.1)──┤
For other available patterns, see the broadcast function documentation.
-
The
QAOAEmbedding
now uses the newMultiRZ
gate as aZZ
entangler, which changes the convention. While previously, theZZ
gate in the embedding was implemented asCNOT(wires=[wires[0], wires[1]]) RZ(2 * parameter, wires=wires[0]) CNOT(wires=[wires[0], wires[1]])
the
MultiRZ
corresponds toCNOT(wires=[wires[1], wires[0]]) RZ(parameter, wires=wires[0]) CNOT(wires=[wires[1], wires[0]])
which differs in the factor of
2
, and fixes a bug in the wires that theCNOT
was applied to. (#609) -
Probability methods are handled by
QubitDevice
and device method requirements are modified to simplify plugin development. (#573) -
The internal variables
All
andAny
to mark anOperation
as acting on all or any wires have been renamed toAllWires
andAnyWires
. (#614)
-
A new
Wires
class was introduced for the internal bookkeeping of wire indices. (#615) -
Improvements to the speed/performance of the
default.qubit
device. (#567) (#559) -
Added the
"backprop"
and"device"
differentiation methods to theqnode
decorator. (#552)-
"backprop"
: Use classical backpropagation. Default on simulator devices that are classically end-to-end differentiable. The returned QNode can only be used with the same machine learning framework (e.g.,default.tensor.tf
simulator with thetensorflow
interface). -
"device"
: Queries the device directly for the gradient.
Using the
"backprop"
differentiation method with thedefault.tensor.tf
device, the created QNode is a 'white-box', and is tightly integrated with the overall TensorFlow computation:>>> dev = qml.device("default.tensor.tf", wires=1) >>> @qml.qnode(dev, interface="tf", diff_method="backprop") >>> def circuit(x): ... qml.RX(x[1], wires=0) ... qml.Rot(x[0], x[1], x[2], wires=0) ... return qml.expval(qml.PauliZ(0)) >>> vars = tf.Variable([0.2, 0.5, 0.1]) >>> with tf.GradientTape() as tape: ... res = circuit(vars) >>> tape.gradient(res, vars) <tf.Tensor: shape=(3,), dtype=float32, numpy=array([-2.2526717e-01, -1.0086454e+00, 1.3877788e-17], dtype=float32)>
-
-
The circuit drawer now displays inverted operations, as well as wires where probabilities are returned from the device: (#540)
>>> @qml.qnode(dev) ... def circuit(theta): ... qml.RX(theta, wires=0) ... qml.CNOT(wires=[0, 1]) ... qml.S(wires=1).inv() ... return qml.probs(wires=[0, 1]) >>> circuit(0.2) array([0.99003329, 0. , 0. , 0.00996671]) >>> print(circuit.draw()) 0: ──RX(0.2)──╭C───────╭┤ Probs 1: ───────────╰X──S⁻¹──╰┤ Probs
-
You can now evaluate the metric tensor of a VQE Hamiltonian via the new
VQECost.metric_tensor
method. This allowsVQECost
objects to be directly optimized by the quantum natural gradient optimizer (qml.QNGOptimizer
). (#618) -
The input check functions in
pennylane.templates.utils
are now public and visible in the API documentation. (#566) -
Added keyword arguments for step size and order to the
qnode
decorator, as well as theQNode
andJacobianQNode
classes. This enables the user to set the step size and order when using finite difference methods. These options are also exposed when creating QNode collections. (#530) (#585) (#587) -
The decomposition for the
CRY
gate now uses the simpler formRY @ CNOT @ RY @ CNOT
(#547) -
The underlying queuing system was refactored, removing the
qml._current_context
property that held the currently activeQNode
orOperationRecorder
. Now, all objects that expose a queue for operations inherit fromQueuingContext
and register their queue globally. (#548) -
The PennyLane repository has a new benchmarking tool which supports the comparison of different git revisions. (#568) (#560) (#516)
-
Updated the development section by creating a landing page with links to sub-pages containing specific guides. (#596)
-
Extended the developer's guide by a section explaining how to add new templates. (#564)
-
tf.GradientTape().jacobian()
can now be evaluated on QNodes using the TensorFlow interface. (#626) -
RandomLayers()
is now compatible with the qiskit devices. (#597) -
DefaultQubit.probability()
now returns the correct probability when called withdevice.analytic=False
. (#563) -
Fixed a bug in the
StronglyEntanglingLayers
template, allowing it to work correctly when applied to a single wire. (544) -
Fixed a bug when inverting operations with decompositions; operations marked as inverted are now correctly inverted when the fallback decomposition is called. (#543)
-
The
QNode.print_applied()
method now correctly displays wires whereqml.prob()
is being returned. #542
This release contains contributions from (in alphabetical order):
Ville Bergholm, Lana Bozanic, Thomas Bromley, Theodor Isacsson, Josh Izaac, Nathan Killoran, Maggie Li, Johannes Jakob Meyer, Maria Schuld, Sukin Sim, Antal Száva.
- Beginning of support for Python 3.8, with the test suite now being run in a Python 3.8 environment. (#501)
- Present templates as a gallery of thumbnails showing the basic circuit architecture. (#499)
-
Fixed a bug where multiplying a QNode parameter by 0 caused a divide by zero error when calculating the parameter shift formula. (#512)
-
Fixed a bug where the shape of differentiable QNode arguments was being cached on the first construction, leading to indexing errors if the QNode was re-evaluated if the argument changed shape. (#505)
This release contains contributions from (in alphabetical order):
Ville Bergholm, Josh Izaac, Johannes Jakob Meyer, Maria Schuld, Antal Száva.
-
Added a quantum chemistry package,
pennylane.qchem
, which supports integration with OpenFermion, Psi4, PySCF, and OpenBabel. (#453)Features include:
- Generate the qubit Hamiltonians directly starting with the atomic structure of the molecule.
- Calculate the mean-field (Hartree-Fock) electronic structure of molecules.
- Allow to define an active space based on the number of active electrons and active orbitals.
- Perform the fermionic-to-qubit transformation of the electronic Hamiltonian by using different functions implemented in OpenFermion.
- Convert OpenFermion's QubitOperator to a Pennylane
Hamiltonian
class. - Perform a Variational Quantum Eigensolver (VQE) computation with this Hamiltonian in PennyLane.
Check out the quantum chemistry quickstart, as well the quantum chemistry and VQE tutorials.
-
PennyLane now has some functions and classes for creating and solving VQE problems. (#467)
-
qml.Hamiltonian
: a lightweight class for representing qubit Hamiltonians -
qml.VQECost
: a class for quickly constructing a differentiable cost function given a circuit ansatz, Hamiltonian, and one or more devices>>> H = qml.vqe.Hamiltonian(coeffs, obs) >>> cost = qml.VQECost(ansatz, hamiltonian, dev, interface="torch") >>> params = torch.rand([4, 3]) >>> cost(params) tensor(0.0245, dtype=torch.float64)
-
-
Added a circuit drawing feature that provides a text-based representation of a QNode instance. It can be invoked via
qnode.draw()
. The user can specify to display variable names instead of variable values and choose either an ASCII or Unicode charset. (#446)Consider the following circuit as an example:
@qml.qnode(dev) def qfunc(a, w): qml.Hadamard(0) qml.CRX(a, wires=[0, 1]) qml.Rot(w[0], w[1], w[2], wires=[1]) qml.CRX(-a, wires=[0, 1]) return qml.expval(qml.PauliZ(0) @ qml.PauliZ(1))
We can draw the circuit after it has been executed:
>>> result = qfunc(2.3, [1.2, 3.2, 0.7]) >>> print(qfunc.draw()) 0: ──H──╭C────────────────────────────╭C─────────╭┤ ⟨Z ⊗ Z⟩ 1: ─────╰RX(2.3)──Rot(1.2, 3.2, 0.7)──╰RX(-2.3)──╰┤ ⟨Z ⊗ Z⟩ >>> print(qfunc.draw(charset="ascii")) 0: --H--+C----------------------------+C---------+| <Z @ Z> 1: -----+RX(2.3)--Rot(1.2, 3.2, 0.7)--+RX(-2.3)--+| <Z @ Z> >>> print(qfunc.draw(show_variable_names=True)) 0: ──H──╭C─────────────────────────────╭C─────────╭┤ ⟨Z ⊗ Z⟩ 1: ─────╰RX(a)──Rot(w[0], w[1], w[2])──╰RX(-1*a)──╰┤ ⟨Z ⊗ Z⟩
-
Added
QAOAEmbedding
and its parameter initialization as a new trainable template. (#442) -
Added the
qml.probs()
measurement function, allowing QNodes to differentiate variational circuit probabilities on simulators and hardware. (#432)@qml.qnode(dev) def circuit(x): qml.Hadamard(wires=0) qml.RY(x, wires=0) qml.RX(x, wires=1) qml.CNOT(wires=[0, 1]) return qml.probs(wires=[0])
Executing this circuit gives the marginal probability of wire 1:
>>> circuit(0.2) [0.40066533 0.59933467]
QNodes that return probabilities fully support autodifferentiation.
-
Added the convenience load functions
qml.from_pyquil
,qml.from_quil
andqml.from_quil_file
that convert pyQuil objects and Quil code to PennyLane templates. This feature requires version 0.8 or above of the PennyLane-Forest plugin. (#459) -
Added a
qml.inv
method that inverts templates and sequences of Operations. Added a@qml.template
decorator that makes templates return the queued Operations. (#462)For example, using this function to invert a template inside a QNode:
@qml.template def ansatz(weights, wires): for idx, wire in enumerate(wires): qml.RX(weights[idx], wires=[wire]) for idx in range(len(wires) - 1): qml.CNOT(wires=[wires[idx], wires[idx + 1]]) dev = qml.device('default.qubit', wires=2) @qml.qnode(dev) def circuit(weights): qml.inv(ansatz(weights, wires=[0, 1])) return qml.expval(qml.PauliZ(0) @ qml.PauliZ(1))
-
Added the
QNodeCollection
container class, that allows independent QNodes to be stored and evaluated simultaneously. Experimental support for asynchronous evaluation of contained QNodes is provided with theparallel=True
keyword argument. (#466) -
Added a high level
qml.map
function, that maps a quantum circuit template over a list of observables or devices, returning aQNodeCollection
. (#466)For example:
>>> def my_template(params, wires, **kwargs): >>> qml.RX(params[0], wires=wires[0]) >>> qml.RX(params[1], wires=wires[1]) >>> qml.CNOT(wires=wires) >>> obs_list = [qml.PauliX(0) @ qml.PauliZ(1), qml.PauliZ(0) @ qml.PauliX(1)] >>> dev = qml.device("default.qubit", wires=2) >>> qnodes = qml.map(my_template, obs_list, dev, measure="expval") >>> qnodes([0.54, 0.12]) array([-0.06154835 0.99280864])
-
Added high level
qml.sum
,qml.dot
,qml.apply
functions that act on QNode collections. (#466)qml.apply
allows vectorized functions to act over the entire QNode collection:>>> qnodes = qml.map(my_template, obs_list, dev, measure="expval") >>> cost = qml.apply(np.sin, qnodes) >>> cost([0.54, 0.12]) array([-0.0615095 0.83756375])
qml.sum
andqml.dot
take the sum of a QNode collection, and a dot product of tensors/arrays/QNode collections, respectively.
- Deprecated the old-style
QNode
such that only the new-styleQNode
and its syntax can be used, moved all related files from thepennylane/beta
folder topennylane
. (#440)
-
Added the
Tensor.prune()
method and theTensor.non_identity_obs
property for extracting non-identity instances from the observables making up aTensor
instance. (#498) -
Renamed the
expt.tensornet
andexpt.tensornet.tf
devices todefault.tensor
anddefault.tensor.tf
. (#495) -
Added a serialization method to the
CircuitGraph
class that is used to create a unique hash for each quantum circuit graph. (#470) -
Added the
Observable.eigvals
method to return the eigenvalues of observables. (#449) -
Added the
Observable.diagonalizing_gates
method to return the gates that diagonalize an observable in the computational basis. (#454) -
Added the
Operator.matrix
method to return the matrix representation of an operator in the computational basis. (#454) -
Added a
QubitDevice
class which implements common functionalities of plugin devices such that plugin devices can rely on these implementations. The newQubitDevice
also includes a newexecute
method, which allows for more convenient plugin design. In addition,QubitDevice
also unifies the way samples are generated on qubit-based devices. (#452) (#473) -
Improved documentation of
AmplitudeEmbedding
andBasisEmbedding
templates. (#441) (#439) -
Codeblocks in the documentation now have a 'copy' button for easily copying examples. (#437)
- Update the developers plugin guide to use QubitDevice. (#483)
- Fixed a bug in
CVQNode._pd_analytic
, where non-descendant observables were not Heisenberg-transformed before evaluating the partial derivatives when using the order-2 parameter-shift method, resulting in an erroneous Jacobian for some circuits. (#433)
This release contains contributions from (in alphabetical order):
Juan Miguel Arrazola, Ville Bergholm, Alain Delgado Gran, Olivia Di Matteo, Theodor Isacsson, Josh Izaac, Soran Jahangiri, Nathan Killoran, Johannes Jakob Meyer, Zeyue Niu, Maria Schuld, Antal Száva.
-
Custom padding constant in
AmplitudeEmbedding
is supported (see 'Breaking changes'.) (#419) -
StronglyEntanglingLayer
andRandomLayer
now work with a single wire. (#409) (#413) -
Added support for applying the inverse of an
Operation
within a circuit. (#377) -
Added an
OperationRecorder()
context manager, that allows templates and quantum functions to be executed while recording events. The recorder can be used with and without QNodes as a debugging utility. (#388) -
Operations can now specify a decomposition that is used when the desired operation is not supported on the target device. (#396)
-
The ability to load circuits from external frameworks as templates has been added via the new
qml.load()
function. This feature requires plugin support --- this initial release provides support for Qiskit circuits and QASM files whenpennylane-qiskit
is installed, via the functionsqml.from_qiskit
andqml.from_qasm
. (#418) -
An experimental tensor network device has been added (#416) (#395) (#394) (#380)
-
An experimental tensor network device which uses TensorFlow for backpropagation has been added (#427)
-
Custom padding constant in
AmplitudeEmbedding
is supported (see 'Breaking changes'.) (#419)
-
The
pad
parameter inAmplitudeEmbedding()
is now eitherNone
(no automatic padding), or a number that is used as the padding constant. (#419) -
Initialization functions now return a single array of weights per function. Utilities for multi-weight templates
Interferometer()
andCVNeuralNetLayers()
are provided. (#412) -
The single layer templates
RandomLayer()
,CVNeuralNetLayer()
andStronglyEntanglingLayer()
have been turned into private functions_random_layer()
,_cv_neural_net_layer()
and_strongly_entangling_layer()
. Recommended use is now via the correspondingLayers()
templates. (#413)
-
Added extensive input checks in templates. (#419)
-
Templates integration tests are rewritten - now cover keyword/positional argument passing, interfaces and combinations of templates. (#409) (#419)
-
State vector preparation operations in the
default.qubit
plugin can now be applied to subsets of wires, and are restricted to being the first operation in a circuit. (#346) -
The
QNode
class is split into a hierarchy of simpler classes. (#354) (#398) (#415) (#417) (#425) -
Added the gates U1, U2 and U3 parametrizing arbitrary unitaries on 1, 2 and 3 qubits and the Toffoli gate to the set of qubit operations. (#396)
-
Changes have been made to accomodate the movement of the main function in
pytest._internal
topytest._internal.main
in pip 19.3. (#404) -
Added the templates
BasisStatePreparation
andMottonenStatePreparation
that use gates to prepare a basis state and an arbitrary state respectively. (#336) -
Added decompositions for
BasisState
andQubitStateVector
based on state preparation templates. (#414) -
Replaces the pseudo-inverse in the quantum natural gradient optimizer (which can be numerically unstable) with
np.linalg.solve
. (#428)
This release contains contributions from (in alphabetical order):
Ville Bergholm, Josh Izaac, Nathan Killoran, Angus Lowe, Johannes Jakob Meyer, Oluwatobi Ogunbayo, Maria Schuld, Antal Száva.
- Added a
print_applied
method to QNodes, allowing the operation and observable queue to be printed as last constructed. (#378)
-
A new
Operator
base class is introduced, which is inherited by both theObservable
class and theOperation
class. (#355) -
Removed deprecated
@abstractproperty
decorators in_device.py
. (#374) -
The
CircuitGraph
class is updated to deal withOperation
instances directly. (#344) -
Comprehensive gradient tests have been added for the interfaces. (#381)
-
The new restructured documentation has been polished and updated. (#387) (#375) (#372) (#370) (#369) (#367) (#364)
-
Added all modules, classes, and functions to the API section in the documentation. (#373)
- Replaces the existing
np.linalg.norm
normalization with hand-coded normalization, allowingAmplitudeEmbedding
to be used with differentiable parameters. AmplitudeEmbedding tests have been added and improved. (#376)
This release contains contributions from (in alphabetical order):
Ville Bergholm, Josh Izaac, Nathan Killoran, Maria Schuld, Antal Száva
-
The devices
default.qubit
anddefault.gaussian
have a new initialization parameteranalytic
that indicates if expectation values and variances should be calculated analytically and not be estimated from data. (#317) -
Added C-SWAP gate to the set of qubit operations (#330)
-
The TensorFlow interface has been renamed from
"tfe"
to"tf"
, and now supports TensorFlow 2.0. (#337) -
Added the S and T gates to the set of qubit operations. (#343)
-
Tensor observables are now supported within the
expval
,var
, andsample
functions, by using the@
operator. (#267)
- The argument
n
specifying the number of samples in the methodDevice.sample
was removed. Instead, the method will always returnDevice.shots
many samples. (#317)
-
The number of shots / random samples used to estimate expectation values and variances,
Device.shots
, can now be changed after device creation. (#317) -
Unified import shortcuts to be under qml in qnode.py and test_operation.py (#329)
-
The quantum natural gradient now uses
scipy.linalg.pinvh
which is more efficient for symmetric matrices than the previously usedscipy.linalg.pinv
. (#331) -
The deprecated
qml.expval.Observable
syntax has been removed. (#267) -
Remainder of the unittest-style tests were ported to pytest. (#310)
-
The
do_queue
argument for operations now only takes effect within QNodes. Outside of QNodes, operations can now be instantiated without needing to specifydo_queue
. (#359)
-
The docs are rewritten and restructured to contain a code introduction section as well as an API section. (#314)
-
Added Ising model example to the tutorials (#319)
-
Added tutorial for QAOA on MaxCut problem (#328)
-
Added QGAN flow chart figure to its tutorial (#333)
-
Added missing figures for gallery thumbnails of state-preparation and QGAN tutorials (#326)
-
Fixed typos in the state preparation tutorial (#321)
-
Fixed bug in VQE tutorial 3D plots (#327)
- Fixed typo in measurement type error message in qnode.py (#341)
This release contains contributions from (in alphabetical order):
Shahnawaz Ahmed, Ville Bergholm, Aroosa Ijaz, Josh Izaac, Nathan Killoran, Angus Lowe, Johannes Jakob Meyer, Maria Schuld, Antal Száva, Roeland Wiersema.
-
Adds a new optimizer,
qml.QNGOptimizer
, which optimizes QNodes using quantum natural gradient descent. See https://arxiv.org/abs/1909.02108 for more details. (#295) (#311) -
Adds a new QNode method,
QNode.metric_tensor()
, which returns the block-diagonal approximation to the Fubini-Study metric tensor evaluated on the attached device. (#295) -
Sampling support: QNodes can now return a specified number of samples from a given observable via the top-level
pennylane.sample()
function. To support this on plugin devices, there is a newDevice.sample
method.Calculating gradients of QNodes that involve sampling is not possible. (#256)
-
default.qubit
has been updated to provide support for sampling. (#256) -
Added controlled rotation gates to PennyLane operations and
default.qubit
plugin. (#251)
-
The method
Device.supported
was removed, and replaced with the methodsDevice.supports_observable
andDevice.supports_operation
. Both methods can be called with string arguments (dev.supports_observable('PauliX')
) and class arguments (dev.supports_observable(qml.PauliX)
). (#276) -
The following CV observables were renamed to comply with the new Operation/Observable scheme:
MeanPhoton
toNumberOperator
,Homodyne
toQuadOperator
andNumberState
toFockStateProjector
. (#254)
-
The
AmplitudeEmbedding
function now provides options to normalize and pad features to ensure a valid state vector is prepared. (#275) -
Operations can now optionally specify generators, either as existing PennyLane operations, or by providing a NumPy array. (#295) (#313)
-
Adds a
Device.parameters
property, so that devices can view a dictionary mapping free parameters to operation parameters. This will allow plugin devices to take advantage of parametric compilation. (#283) -
Introduces two enumerations:
Any
andAll
, representing any number of wires and all wires in the system respectively. They can be imported frompennylane.operation
, and can be used when defining theOperation.num_wires
class attribute of operations. (#277)As part of this change:
-
All
is equivalent to the integer 0, for backwards compatibility with the existing test suite -
Any
is equivalent to the integer -1 to allow numeric comparison operators to continue working -
An additional validation is now added to the
Operation
class, which will alert the user that an operation withnum_wires = All
is being incorrectly.
-
-
The one-qubit rotations in
pennylane.plugins.default_qubit
no longer depend on Scipy'sexpm
. Instead they are calculated with Euler's formula. (#292) -
Creates an
ObservableReturnTypes
enumeration class containingSample
,Variance
andExpectation
. These new values can be assigned to thereturn_type
attribute of anObservable
. (#290) -
Changed the signature of the
RandomLayer
andRandomLayers
templates to have a fixed seed by default. (#258) -
setup.py
has been cleaned up, removing the non-working shebang, and removing unused imports. (#262)
-
A documentation refactor to simplify the tutorials and include Sphinx-Gallery. (#291)
-
Examples and tutorials previously split across the
examples/
anddoc/tutorials/
directories, in a mixture of ReST and Jupyter notebooks, have been rewritten as Python scripts with ReST comments in a single location, theexamples/
folder. -
Sphinx-Gallery is used to automatically build and run the tutorials. Rendered output is displayed in the Sphinx documentation.
-
Links are provided at the top of every tutorial page for downloading the tutorial as an executable python script, downloading the tutorial as a Jupyter notebook, or viewing the notebook on GitHub.
-
The tutorials table of contents have been moved to a single quick start page.
-
-
Fixed a typo in
QubitStateVector
. (#296) -
Fixed a typo in the
default_gaussian.gaussian_state
function. (#293) -
Fixed a typo in the gradient recipe within the
RX
,RY
,RZ
operation docstrings. (#248) -
Fixed a broken link in the tutorial documentation, as a result of the
qml.expval.Observable
deprecation. (#246)
- Fixed a bug where a
PolyXP
observable would fail if applied to subsets of wires ondefault.gaussian
. (#277)
This release contains contributions from (in alphabetical order):
Simon Cross, Aroosa Ijaz, Josh Izaac, Nathan Killoran, Johannes Jakob Meyer, Rohit Midha, Nicolás Quesada, Maria Schuld, Antal Száva, Roeland Wiersema.
-
pennylane.expval()
is now a top-level function, and is no longer a package of classes. For now, the existingpennylane.expval.Observable
interface continues to work, but will raise a deprecation warning. (#232) -
Variance support: QNodes can now return the variance of observables, via the top-level
pennylane.var()
function. To support this on plugin devices, there is a newDevice.var
method.The following observables support analytic gradients of variances:
-
All qubit observables (requiring 3 circuit evaluations for involutory observables such as
Identity
,X
,Y
,Z
; and 5 circuit evals for non-involutary observables, currently onlyqml.Hermitian
) -
First-order CV observables (requiring 5 circuit evaluations)
Second-order CV observables support numerical variance gradients.
-
-
pennylane.about()
function added, providing details on current PennyLane version, installed plugins, Python, platform, and NumPy versions (#186) -
Removed the logic that allowed
wires
to be passed as a positional argument in quantum operations. This allows us to raise more useful error messages for the user if incorrect syntax is used. (#188) -
Adds support for multi-qubit expectation values of the
pennylane.Hermitian()
observable (#192) -
Adds support for multi-qubit expectation values in
default.qubit
. (#202) -
Organize templates into submodules (#195). This included the following improvements:
-
Distinguish embedding templates from layer templates.
-
New random initialization functions supporting the templates available in the new submodule
pennylane.init
. -
Added a random circuit template (
RandomLayers()
), in which rotations and 2-qubit gates are randomly distributed over the wires -
Add various embedding strategies
-
- The
Device
methodsexpectations
,pre_expval
, andpost_expval
have been renamed toobservables
,pre_measure
, andpost_measure
respectively. (#232)
-
default.qubit
plugin now usesnp.tensordot
when applying quantum operations and evaluating expectations, resulting in significant speedup (#239), (#241) -
PennyLane now allows division of quantum operation parameters by a constant (#179)
-
Portions of the test suite are in the process of being ported to pytest. Note: this is still a work in progress.
Ported tests include:
test_ops.py
test_about.py
test_classical_gradients.py
test_observables.py
test_measure.py
test_init.py
test_templates*.py
test_ops.py
test_variable.py
test_qnode.py
(partial)
-
Fixed a bug in
Device.supported
, which would incorrectly mark an operation as supported if it shared a name with an observable (#203) -
Fixed a bug in
Operation.wires
, by explicitly casting the type of each wire to an integer (#206) -
Removed code in PennyLane which configured the logger, as this would clash with users' configurations (#208)
-
Fixed a bug in
default.qubit
, in whichQubitStateVector
operations were accidentally being cast tonp.float
instead ofnp.complex
. (#211)
This release contains contributions from:
Shahnawaz Ahmed, riveSunder, Aroosa Ijaz, Josh Izaac, Nathan Killoran, Maria Schuld.
- Fixed a bug where the interfaces submodule was not correctly being packaged via setup.py
- PennyLane now includes a new
interfaces
submodule, which enables QNode integration with additional machine learning libraries. - Adds support for an experimental PyTorch interface for QNodes
- Adds support for an experimental TensorFlow eager execution interface for QNodes
- Adds a PyTorch+GPU+QPU tutorial to the documentation
- Documentation now includes links and tutorials including the new PennyLane-Forest plugin.
-
Printing a QNode object, via
print(qnode)
or in an interactive terminal, now displays more useful information regarding the QNode, including the device it runs on, the number of wires, it's interface, and the quantum function it uses:>>> print(qnode) <QNode: device='default.qubit', func=circuit, wires=2, interface=PyTorch>
This release contains contributions from:
Josh Izaac and Nathan Killoran.
- Added the
Identity
expectation value for both CV and qubit models (#135) - Added the
templates.py
submodule, containing some commonly used QML models to be used as ansatz in QNodes (#133) - Added the
qml.Interferometer
CV operation (#152) - Wires are now supported as free QNode parameters (#151)
- Added ability to update stepsizes of the optimizers (#159)
- Removed use of hardcoded values in the optimizers, made them parameters (see #131 and #132)
- Created the new
PlaceholderExpectation
, to be used when both CV and qubit expval modules contain expectations with the same name - Provide a way for plugins to view the operation queue before applying operations. This allows for on-the-fly modifications of the queue, allowing hardware-based plugins to support the full range of qubit expectation values. (#143)
- QNode return values now support any form of sequence, such as lists, sets, etc. (#144)
- CV analytic gradient calculation is now more robust, allowing for operations which may not themselves be differentiated, but have a
well defined
_heisenberg_rep
method, and so may succeed operations that are analytically differentiable (#152)
- Fixed a bug where the variational classifier example was not batching when learning parity (see #128 and #129)
- Fixed an inconsistency where some initial state operations were documented as accepting complex parameters - all operations now accept real values (#146)
This release contains contributions from:
Christian Gogolin, Josh Izaac, Nathan Killoran, and Maria Schuld.
Initial public release.
This release contains contributions from:Ville Bergholm, Josh Izaac, Maria Schuld, Christian Gogolin, and Nathan Killoran.