github PennyLaneAI/pennylane v0.40.0
Release 0.40.0
on GitHub

14 hours ago

New features since last release

Efficient state preparation methods 🦾

  • State preparation tailored for matrix product states (MPS) is now supported with :class:qml.MPSPrep <pennylane.MPSPrep> on the lightning.tensor device. (#6431)

    Given a list of $n$ tensors that represents an MPS, $[A^{(0)}, ..., A^{(n-1)}]$, :class:qml.MPSPrep <pennylane.MPSPrep> lets you directly inject the MPS into a QNode as the initial state of the circuit without any need for pre-processing. The first and last tensors in the list must be rank-2, while all intermediate tensors should be rank-3. 

    import pennylane as qml
import numpy as np

mps = [
    np.array([[0.0, 0.107], [0.994, 0.0]]),
    np.array(
        [
            [[0.0, 0.0, 0.0, -0.0], [1.0, 0.0, 0.0, -0.0]],
            [[0.0, 1.0, 0.0, -0.0], [0.0, 0.0, 0.0, -0.0]],
        ]
    ),
    np.array(
        [
            [[-1.0, 0.0], [0.0, 0.0]],
            [[0.0, 0.0], [0.0, 1.0]],
            [[0.0, -1.0], [0.0, 0.0]],
            [[0.0, 0.0], [1.0, 0.0]],
        ]
    ),
    np.array([[-1.0, -0.0], [-0.0, -1.0]]),
]

dev = qml.device("lightning.tensor", wires = [0, 1, 2, 3])
@qml.qnode(dev)
def circuit():
    qml.MPSPrep(mps, wires = [0,1,2])
    return qml.state()
    >>> print(circuit())
[ 0.    +0.j  0.    +0.j  0.    +0.j -0.1066+0.j  0.    +0.j  0.    +0.j
  0.    +0.j  0.    +0.j  0.    +0.j  0.    +0.j  0.    +0.j  0.    +0.j
  0.9943+0.j  0.    +0.j  0.    +0.j  0.    +0.j]

    At this time, :class:qml.MPSPrep <pennylane.MPSPrep> is only supported on the lightning.tensor device.

  • Custom-made state preparation for linear combinations of quantum states is now available with :class:qml.Superposition <pennylane.Superposition>. (#6670)

    Given a list of $m$ coefficients $c_i$ and basic states $|b_i\rangle$, :class:qml.Superposition <pennylane.Superposition> prepares $|\phi\rangle = \sum_i^m c_i |b_i\rangle$. Here is a simple example showing how to use :class:qml.Superposition <pennylane.Superposition> to prepare $\tfrac{1}{\sqrt{2}} |00\rangle + \tfrac{1}{\sqrt{2}} |10\rangle$. 

    coeffs = np.array([0.70710678, 0.70710678])
basis =  np.array([[0, 0], [1, 0]])

@qml.qnode(qml.device('default.qubit'))
def circuit():
    qml.Superposition(coeffs, basis, wires=[0, 1], work_wire=[2])
    return qml.state()
    >>> circuit()
Array([0.7071068 +0.j, 0.        +0.j, 0.        +0.j, 0.        +0.j,
       0.70710677+0.j, 0.        +0.j, 0.        +0.j, 0.        +0.j],      dtype=complex64)

    Note that specification of one work_wire is required.

Enhanced QSVT functionality 🤩

  • New functionality to calculate and convert phase angles for QSP and QSVT has been added with :func:qml.poly_to_angles <pennylane.poly_to_angles> and :func:qml.transform_angles <pennylane.transform_angles>. (#6483)

    The :func:qml.poly_to_angles <pennylane.poly_to_angles> function calculates phase angles directly given polynomial coefficients and the routine in which the angles will be used ("QSVT", "QSP", or "GQSP"): 

    >>> poly = [0, 1.0, 0, -1/2, 0, 1/3]
>>> qsvt_angles = qml.poly_to_angles(poly, "QSVT")
>>> print(qsvt_angles)
[-5.49778714  1.57079633  1.57079633  0.5833829   1.61095884  0.74753829]

    The :func:qml.transform_angles <pennylane.transform_angles> function can be used to convert angles from one routine to another: 

    >>> qsp_angles = np.array([0.2, 0.3, 0.5])
>>> qsvt_angles = qml.transform_angles(qsp_angles, "QSP", "QSVT")
>>> print(qsvt_angles)
[-6.86858347  1.87079633 -0.28539816]

  • The :func:qml.qsvt <pennylane.qsvt> function has been improved to be more user-friendly, with enhanced capabilities. (#6520) (#6693)

    Block encoding and phase angle computation are now handled automatically, given a matrix to encode, polynomial coefficients, and a block encoding method ("prepselprep", "qubitization", "embedding", or "fable", all implemented with their corresponding operators in PennyLane). 

    # P(x) = -x + 0.5 x^3 + 0.5 x^5
poly = np.array([0, -1, 0, 0.5, 0, 0.5])
hamiltonian = qml.dot([0.3, 0.7], [qml.Z(1), qml.X(1) @ qml.Z(2)])

dev = qml.device("default.qubit")
@qml.qnode(dev)
def circuit():
    qml.qsvt(hamiltonian, poly, encoding_wires=[0], block_encoding="prepselprep")
    return qml.state()

matrix = qml.matrix(circuit, wire_order=[0, 1, 2])()
    >>> print(matrix[:4, :4].real)
[[-0.1625  0.     -0.3793  0.    ]
 [ 0.     -0.1625  0.      0.3793]
 [-0.3793  0.      0.1625  0.    ]
 [ 0.      0.3793  0.      0.1625]]

    The old functionality can still be accessed with :func:qml.qsvt_legacy <pennylane.qsvt_legacy>.

  • A new :class:qml.GQSP <pennylane.GQSP> template has been added to perform Generalized Quantum Signal Processing (GQSP). (#6565) Similar to QSVT, GQSP is an algorithm that polynomially transforms an input unitary operator, but with fewer restrictions on the chosen polynomial.

    You can also use :func:qml.poly_to_angles <pennylane.poly_to_angles> to obtain angles for GQSP! 

    # P(x) = 0.1 + 0.2j x + 0.3 x^2
poly = [0.1, 0.2j, 0.3]
angles = qml.poly_to_angles(poly, "GQSP")

@qml.prod # transforms the qfunc into an Operator
def unitary(wires):
    qml.RX(0.3, wires)

dev = qml.device("default.qubit")
@qml.qnode(dev)
def circuit(angles):
    qml.GQSP(unitary(wires = 1), angles, control = 0)
    return qml.state()

matrix = qml.matrix(circuit, wire_order=[0, 1])(angles)
``` ```pycon
>>> print(np.round(matrix,3)[:2, :2])
[[0.387+0.198j 0.03 -0.089j]
[0.03 -0.089j 0.387+0.198j]]

Generalized Trotter products 🐖

  • Trotter products that work on exponentiated operators directly instead of full system hamiltonians can now be encoded into circuits with the addition of :class:qml.TrotterizedQfunc <pennylane.TrotterizedQfunc> and :func:qml.trotterize <pennylane.trotterize>. This allows for custom specification of the first-order expansion of the Suzuki-Trotter product formula and extrapolating it to the $n^{\text{th}}$ order. (#6627)

    If the first-order of the Suzuki-Trotter product formula for a given problem is known, :class:qml.TrotterizedQfunc <pennylane.TrotterizedQfunc> and :func:qml.trotterize <pennylane.trotterize> let you implement the $n^{\text{th}}$-order product formula while only specifying the first-order term as a quantum function. 

    def my_custom_first_order_expansion(time, theta, phi, wires, flip):
  qml.RX(time * theta, wires[0])
  qml.RY(time * phi, wires[1])
  if flip:
      qml.CNOT(wires=wires[:2])

    :func:qml.trotterize <pennylane.trotterize> requires the quantum function representing the first-order product formula, the number of Trotter steps, and the desired order. It returns a function with the same call signature as the first-order product formula quantum function: 

    @qml.qnode(qml.device("default.qubit"))
def my_circuit(time, theta, phi, num_trotter_steps):
    qml.trotterize(
        first_order_expansion,
        n=num_trotter_steps,
        order=2,
    )(time, theta, phi, wires=['a', 'b'], flip=True)
    return qml.state()

    Alternatively, :class:qml.TrotterizedQfunc <pennylane.TrotterizedQfunc> can be used as follows: 

    @qml.qnode(qml.device("default.qubit"))
def my_circuit(time, theta, phi, num_trotter_steps):
    qml.TrotterizedQfunc(
        time,
        theta,
        phi,
        qfunc=my_custom_first_order_expansion,
        n=num_trotter_steps,
        order=2,
        wires=['a', 'b'],
        flip=True,
    )
    return qml.state()
    >>> time = 0.1
>>> theta, phi = (0.12, -3.45)
>>> print(qml.draw(my_circuit, level="device")(time, theta, phi, num_trotter_steps=1))
a: ──RX(0.01)──╭●─╭●──RX(0.01)──┤  State
b: ──RY(-0.17)─╰X─╰X──RY(-0.17)─┤  State

    Both methods produce the same results, but offer different interfaces based on the application or overall preference.

Bosonic operators 🎈

A new module, :mod:qml.bose <pennylane.bose>, has been added to PennyLane that includes support for constructing and manipulating Bosonic operators and converting between Bosonic operators and qubit operators.

  • Bosonic operators analogous to qml.FermiWord and qml.FermiSentence are now available with :class:qml.BoseWord <pennylane.BoseWord> and :class:qml.BoseSentence <pennylane.BoseSentence>. (#6518)

    :class:qml.BoseWord <pennylane.BoseWord> and :class:qml.BoseSentence <pennylane.BoseSentence> operate similarly to their fermionic counterparts. To create a Bose word, a dictionary is required as input, where the keys are tuples of boson indices and values are '+/-' (denoting the bosonic creation/annihilation operators). For example, the $b^{\dagger}_0 b_1$ can be constructed as follows. 

    >>> w = qml.BoseWord({(0, 0) : '+', (1, 1) : '-'})
>>> print(w)
b⁺(0) b(1)

    Multiple Bose words can then be combined to form a Bose sentence: 

    >>> w1 = qml.BoseWord({(0, 0) : '+', (1, 1) : '-'})
>>> w2 = qml.BoseWord({(0, 1) : '+', (1, 2) : '-'})
>>> s = qml.BoseSentence({w1 : 1.2, w2: 3.1})
>>> print(s)
1.2 * b⁺(0) b(1)
+ 3.1 * b⁺(1) b(2)

  • Functionality for converting bosonic operators to qubit operators is available with :func:qml.unary_mapping <pennylane.unary_mapping>, :func:qml.binary_mapping <pennylane.binary_mapping>, and :func:qml.christiansen_mapping <pennylane.christiansen_mapping>. (#6623) (#6576) (#6564)

    All three mappings follow the same syntax, where a :class:qml.BoseWord <pennylane.BoseWord> or :class:qml.BoseSentence <pennylane.BoseSentence> is required as input. 

    >>> w = qml.BoseWord({(0, 0): "+"})
>>> qml.binary_mapping(w, n_states=4)
0.6830127018922193 * X(0)
+ -0.1830127018922193 * X(0) @ Z(1)
+ -0.6830127018922193j * Y(0)
+ 0.1830127018922193j * Y(0) @ Z(1)
+ 0.3535533905932738 * X(0) @ X(1)
+ -0.3535533905932738j * X(0) @ Y(1)
+ 0.3535533905932738j * Y(0) @ X(1)
+ (0.3535533905932738+0j) * Y(0) @ Y(1)

    Additional fine-tuning is available within each function, such as the maximum number of allowed bosonic states and a tolerance for discarding imaginary parts of the coefficients.

Construct vibrational Hamiltonians 🔨

  • Several new features are available in the :mod:qml.qchem <pennylane.qchem> module to help with the construction of vibrational Hamiltonians. This includes:

    • The :class:~.qchem.VibrationalPES class to store potential energy surface information. (#6652) 
    pes_onemode = np.array([[0.309, 0.115, 0.038, 0.008, 0.000, 0.006, 0.020, 0.041, 0.070]])
pes_twomode = np.zeros((1, 1, 9, 9))
dipole_onemode = np.zeros((1, 9, 3))
gauss_weights = np.array([3.96e-05, 4.94e-03, 8.85e-02,
                                4.33e-01, 7.20e-01, 4.33e-01,
                                8.85e-02, 4.94e-03, 3.96e-05])
grid = np.array([-3.19, -2.27, -1.47, -0.72,  0.0,  0.72,  1.47,  2.27,  3.19])
pes_object = qml.qchem.VibrationalPES(
        freqs=np.array([0.025]),
        grid=grid,
        uloc=np.array([[1.0]]),
        gauss_weights=gauss_weights,
        pes_data=[pes_onemode, pes_twomode],
        dipole_data=[dipole_onemode],
        localized=False,
        dipole_level=1,
    )
    • The :func:~.qchem.taylor_hamiltonian function to build a Taylor Hamiltonian from a :class:~.VibrationalPES object. (#6523) 
    >>> qml.qchem.taylor_hamiltonian(pes_object, 4, 2)
(
    0.016867926879358452 * I(0)
  + -0.007078617919572303 * Z(0)
  + 0.0008679410939323631 * X(0)
)
    • The :func:~.qchem.taylor_bosonic function to build a Taylor Hamiltonian in terms of Bosonic operators. (#6523) 
    >>> coeffs_arr = qml.qchem.taylor_coeffs(pes_object)
>>> bose_op = qml.qchem.taylor_bosonic(coeffs_arr, pes_object.freqs, is_local=pes_object.localized, uloc=pes_object.uloc)
>>> type(bose_op)
pennylane.bose.bosonic.BoseSentence

    Additional functionality is also available to optimize molecular geometries and convert between representations:

    • Convert Christiansen Hamiltonian integrals in the harmonic oscillator basis to integrals in the vibrational self-consistent field (VSCF) basis with the :func:~.qchem.vscf_integrals function. (#6688)

    • Find the lowest energy configuration of molecules with :func:~.qchem.optimize_geometry. (#6453) (#6666)

    • Separate normal mode frequencies and localize them with :func:~.qchem.localize_normal_modes. (#6453)

Labs: a place for unified and rapid prototyping of research software 🧑‍🔬

The new :mod:qml.labs <pennylane.labs> module will house experimental research software 🔬. Features here may be useful for state-of-the-art research, beta testing, or getting a sneak peek into potential new features before they are added to PennyLane.

The experimental nature of this module means features may not integrate well with other PennyLane staples like differentiability, JAX, or JIT compatibility. There may also be unexpected sharp bits 🔪 and errors ❌.

.. warning:: This module is experimental! Breaking changes and removals will happen without warning. Please use these features carefully and let us know your thoughts. Your feedback will inform how these features become a part of mainline PennyLane.

Resource estimation

  • Resource estimation functionality in Labs is focused on being light-weight and flexible. The Labs :mod:qml.labs.resource_estimation <pennylane.labs.resource_estimation> module involves modifications to core PennyLane that reduce the memory requirements and computational time of resource estimation. These include new or modified base classes and one new function:

    • :class:~.labs.resource_estimation.Resources - This class is simplified in labs, removing the arguments: gate_sizes, depth, and shots. (#6428)
    • :class:~.labs.resource_estimation.ResourceOperator - Replaces :class:~.resource.ResourceOperation, expanded to include decompositions. (#6428)
    • :class:~.labs.resource_estimation.CompressedResourceOp - A new class with the minimum information to estimate resources: the operator type and the parameters needed to decompose it. (#6428)
    • :class:~.labs.resource_estimation.ResourceOperator - versions of many existing PennyLane operations, like Pauli operators, :class:~.labs.resource_estimation.ResourceHadamard, and :class:~.labs.resource_estimation.ResourceCNOT. (#6447) (#6579) (#6538) (#6592)
    • :func:~.labs.resource_estimation.get_resources() - The new entry point to efficiently obtain the resources of quantum circuits. (#6500)

    Using new Resource versions of existing operations and :func:~.labs.resource_estimation.get_resources, we can estimate resources quickly: 

    import pennylane.labs.resource_estimation as re

def my_circuit():
    for w in range(2):
        re.ResourceHadamard(w)
    re.ResourceCNOT([0, 1])
    re.ResourceRX(1.23, 0)
    re.ResourceRY(-4.56, 1)
    re.ResourceQFT(wires=[0, 1, 2])
    return qml.expval(re.ResourceHadamard(2))
    >>> res = re.get_resources(my_circuit)()
>>> print(res)
wires: 3
gates: 202
gate_types:
{'Hadamard': 5, 'CNOT': 10, 'T': 187}

    We can also set custom gate sets for decompositions: 

    >>> gate_set={"Hadamard","CNOT","RZ", "RX", "RY", "SWAP"}
>>> res = re.get_resources(my_circuit, gate_set=gate_set)()
>>> print(res)
wires: 3
gates: 24
gate_types:
{'Hadamard': 5, 'CNOT': 7, 'RX': 1, 'RY': 1, 'SWAP': 1, 'RZ': 9}

    Alternatively, it is possible to manually substitute associated resources: 

    >>> new_resources = re.substitute(res, "SWAP", re.Resources(2, 3, {"CNOT":3}))
>>> print(new_resources)
{'Hadamard': 5, 'CNOT': 10, 'RX': 1, 'RY': 1, 'RZ': 9}

Experimental functionality for handling dynamical Lie algebras (DLAs)

  • Use the :mod:qml.labs.dla <pennylane.labs.dla> module to perform the KAK decomposition:

    • :func:~.labs.dla.cartan_decomp: obtain a Cartan decomposition of an input Lie algebra via an involution. (#6392)
    • We provide a variety of involutions like :func:~.labs.dla.concurrence_involution, :func:~.labs.dla.even_odd_involution and canonical Cartan involutions. (#6392) (#6396)
    • :func:~.labs.dla.cartan_subalgebra: compute a horizontal Cartan subalgebra. (#6403)
    • :func:~.labs.dla.variational_kak_adj : compute a variational KAK decomposition of a Hermitian operator using a Cartan decomposition and the adjoint representation of a horizontal Cartan subalgebra. (#6446)

    To use this functionality we start with a set of Hermitian operators. 

    >>> n = 3
>>> gens = [qml.X(i) @ qml.X(i + 1) for i in range(n - 1)]
>>> gens += [qml.Z(i) for i in range(n)]
>>> H = qml.sum(*gens)

    We then generate its Lie algebra by computing the Lie closure. 

    >>> g = qml.lie_closure(gens)
>>> g = [op.pauli_rep for op in g]
>>> print(g)
[1 * X(0) @ X(1), 1 * X(1) @ X(2), 1.0 * Z(0), ...]

    We then choose an involution (e.g. :func:~.labs.dla.concurrence_involution) that defines a Cartan decomposition g = k + m. k is the vertical subalgebra, and m its horizontal complement (not a subalgebra). 

    >>> from pennylane.labs.dla import concurrence_involution, cartan_decomp
>>> involution = concurrence_involution
>>> k, m = cartan_decomp(g, involution=involution)

    The next step is just re-ordering the basis elements in g and computing its structure_constants. 

    >>> g = k + m
>>> adj = qml.structure_constants(g)

    We can then compute a (horizontal) Cartan subalgebra a, that is, a maximal Abelian subalgebra of m. 

    >>> from pennylane.labs.dla import cartan_subalgebra
>>> g, k, mtilde, a, adj = cartan_subalgebra(g, k, m, adj)

    Having determined both subalgebras k and a, we can compute the KAK decomposition variationally like in 2104.00728, see our demo on KAK decomposition in practice. 

    >>> from pennylane.labs.dla import variational_kak_adj
>>> dims = (len(k), len(mtilde), len(a))
>>> adjvec_a, theta_opt = variational_kak_adj(H, g, dims, adj, opt_kwargs={"n_epochs": 3000})

  • We also provide some additional features that are useful for handling dynamical Lie algebras.

    • :func:~.labs.dla.recursive_cartan_decomp: perform consecutive recursive Cartan decompositions. (#6396)
    • :func:~.labs.dla.lie_closure_dense: extension of qml.lie_closure using dense matrices. (#6371) (#6695)
    • :func:~.labs.dla.structure_constants_dense: extension of qml.structure_constants using dense matrices. (#6396) (#6376)

Vibrational Hamiltonians

  • New functionality in labs helps with the construction of vibrational Hamiltonians.

    • Generate potential energy surfaces (PES) with qml.labs.vibrational.vibrational_pes. (#6616) (#6676) 
    >>> symbols  = ['H', 'F']
>>> geometry = np.array([[0.0, 0.0, 0.0], [0.0, 0.0, 1.0]])
>>> mol = qml.qchem.Molecule(symbols, geometry)
>>> pes = vibrational_pes(mol)
    • Use the qml.labs.vibrational.christiansen_hamiltonian function and potential energy surfaces to generate Hamiltonians in the Christiansen form. (#6560)

Improvements 🛠

QChem improvements

  • The qml.qchem.factorize function now supports new methods for double factorization: Cholesky decomposition (cholesky=True) and compressed double factorization (compressed=True). (#6573) (#6611)

  • A new function for performing the block-invariant symmetry shift on electronic integrals has been added with qml.qchem.symmetry_shift. (#6574)

  • The differentiable Hartree-Fock workflow is now compatible with JAX. (#6096) (#6707)

Transform for combining GlobalPhase instances

  • A new transform called qml.transforms.combine_global_phases has been added. It combines all qml.GlobalPhase gates in a circuit into a single one applied at the end. This can be useful for circuits that include a lot of qml.GlobalPhase gates that are introduced directly during circuit creation, decompositions that include qml.GlobalPhase gates, etc. (#6686)

Better drawing functionality

  • qml.draw_mpl now has a wire_options keyword argument, which allows for global- and per-wire customization with options like color, linestyle, and linewidth. (#6486)

    Here is an example that would make all wires cyan and bold except for wires 2 and 6, which are dashed and a different colour. 

    @qml.qnode(qml.device("default.qubit"))
def circuit(x):
    for w in range(5):
        qml.Hadamard(w) 
    return qml.expval(qml.PauliZ(0) @ qml.PauliY(1))

wire_options = {"color": "cyan", 
                "linewidth": 5, 
                2: {"linestyle": "--", "color": "red"}, 
                6: {"linestyle": "--", "color": "orange"}
            }
print(qml.draw_mpl(circuit, wire_options=wire_options)(0.52))

New device capabilities 💾

  • Two new methods, setup_execution_config and preprocess_transforms, have been added to the Device class. Device developers are encouraged to override these two methods separately instead of the preprocess method. For now, to avoid ambiguity, a device is allowed to override either these two methods or preprocess, but not both. In the long term, we will slowly phase out the use of preprocess in favour of these two methods for better separation of concerns. (#6617)

  • Developers of plugin devices now have the option of providing a TOML-formatted configuration file to declare the capabilities of the device. See Device Capabilities for details.

  • An internal module called qml.devices.capabilities has been added that defines a new DeviceCapabilites data class, as well as functions that load and parse the TOML-formatted configuration files. (#6407) 

    >>> from pennylane.devices.capabilities import DeviceCapabilities
>>> capabilities = DeviceCapabilities.from_toml_file("my_device.toml")
>>> isinstance(capabilities, DeviceCapabilities)
True

  • Devices that extend qml.devices.Device now have an optional class attribute called capabilities, which is an instance of the DeviceCapabilities data class constructed from the configuration file if it exists. Otherwise, it is set to None. (#6433) 

    from pennylane.devices import Device

class MyDevice(Device):
    config_filepath = "path/to/config.toml"
    ...
    >>> isinstance(MyDevice.capabilities, DeviceCapabilities)
True

  • Default implementations of Device.setup_execution_config and Device.preprocess_transforms have been added to the device API for devices that provide a TOML configuration file and, thus, have a capabilities property. (#6632) (#6653)

Capturing and representing hybrid programs

  • Support has been added for if/else statements and for and while loops in circuits executed with qml.capture.enabled, via Autograph. Autograph conversion is now used by default in make_plxpr, but can be skipped with autograph=False. (#6406) (#6413) (#6426) (#6645) (#6685)

  • qml.transform now accepts a plxpr_transform argument. This argument must be a function that can transform plxpr. Note that executing a transformed function will currently raise a NotImplementedError. To see more details, check out the :func:documentation of qml.transform <pennylane.transform>. (#6633) (#6722)

  • Users can now apply transforms with program capture enabled. Transformed functions cannot be executed by default. To apply the transforms (and to be able to execute the function), it must be decorated with the new qml.capture.expand_plxpr_transforms function, which accepts a callable as input and returns a new function for which all present transforms have been applied. (#6722) 

    from functools import partial

qml.capture.enable()
wire_map = {0: 3, 1: 6, 2: 9}

@partial(qml.map_wires, wire_map=wire_map)
def circuit(x, y):
    qml.RX(x, 0)
    qml.CNOT([0, 1])
    qml.CRY(y, [1, 2])
    return qml.expval(qml.Z(2))
    >>> qml.capture.make_plxpr(circuit)(1.2, 3.4)
{ lambda ; a:f32[] b:f32[]. let
    c:AbstractMeasurement(n_wires=None) = _map_wires_transform_transform[
    args_slice=slice(0, 2, None)
    consts_slice=slice(2, 2, None)
    inner_jaxpr={ lambda ; d:f32[] e:f32[]. let
        _:AbstractOperator() = RX[n_wires=1] d 0
        _:AbstractOperator() = CNOT[n_wires=2] 0 1
        _:AbstractOperator() = CRY[n_wires=2] e 1 2
        f:AbstractOperator() = PauliZ[n_wires=1] 2
        g:AbstractMeasurement(n_wires=None) = expval_obs f
      in (g,) }
    targs_slice=slice(2, None, None)
    tkwargs={'wire_map': {0: 3, 1: 6, 2: 9}, 'queue': False}
    ] a b
  in (c,) }
>>> transformed_circuit = qml.capture.expand_plxpr_transforms(circuit)
>>> jax.make_jaxpr(transformed_circuit)(1.2, 3.4)
{ lambda ; a:f32[] b:f32[]. let
    _:AbstractOperator() = RX[n_wires=1] a 3
    _:AbstractOperator() = CNOT[n_wires=2] 3 6
    _:AbstractOperator() = CRY[n_wires=2] b 6 9
    c:AbstractOperator() = PauliZ[n_wires=1] 9
    d:AbstractMeasurement(n_wires=None) = expval_obs c
  in (d,) }

  • The qml.iterative_qpe function can now be compactly captured into plxpr. (#6680)

  • Three new plxpr interpreters have been added that allow for functions and plxpr to be natively transformed with the same API as the corresponding existing transforms in PennyLane when program capture is enabled:

    • qml.capture.transforms.CancelInterpreter:this class cancels operators appearing consecutively that are adjoints of each other following the same API as qml.transforms.cancel_inverses. (#6692)

    • qml.capture.transforms.DecomposeInterpreter: this class decomposes pennylane operators following the same API as qml.transforms.decompose. (#6691)

    • qml.capture.transforms.MapWiresInterpreter: this class maps wires to new values following the same API as qml.map_wires. (#6697)

  • A qml.tape.plxpr_to_tape function is now available that converts plxpr to a tape. (#6343)

  • Execution with capture enabled now follows a new execution pipeline and natively passes the captured plxpr to the device. Since it no longer falls back to the old pipeline, execution only works with a reduced feature set. (#6655) (#6596)

  • PennyLane transforms can now be captured as primitives with experimental program capture enabled. (#6633)

  • jax.vmap can be captured with qml.capture.make_plxpr and is compatible with quantum circuits. (#6349) (#6422) (#6668)

  • A qml.capture.PlxprInterpreter base class has been added for easy transformation and execution of plxpr. (#6141)

  • A DefaultQubitInterpreter class has been added to provide plxpr execution using python based tools, and the DefaultQubit.eval_jaxpr method has been implemented. (#6594) (#6328)

  • An optional method, eval_jaxpr, has been added to the device API for native execution of plxpr programs. (#6580)

  • qml.capture.qnode_call has been made private and moved to the workflow module. (#6620)

Other Improvements

  • qml.math.grad and qml.math.jacobian have been added to differentiate a function with inputs of any interface in a JAX-like manner. (#6741)

  • qml.GroverOperator now has a work_wires property. (#6738)

  • The Wires object's usage across Pennylane source code has been tidied up for internal consistency. (#6689)

  • qml.equal now supports qml.PauliWord and qml.PauliSentence instances. (#6703)

  • Redundant commutator computations from qml.lie_closure have been removed. (#6724)

  • A comprehensive error is now raised when using qml.fourier.qnode_spectrum with standard Numpy arguments and interface="auto". (#6622)

  • Pauli string representations for the gates {X, Y, Z, S, T, SX, SWAP, ISWAP, ECR, SISWAP} have been added, and a shape error in the matrix conversion of qml.PauliSentences with list or array inputs has been fixed. (#6562) (#6587)

  • qml.QNode and qml.execute now forbid certain keyword arguments from being passed positionally. (#6610)

  • The string representations for the qml.S, qml.T, and qml.SX have been shortened. (#6542)

  • Internal class functions and dunder methods have been added to allow for multiplying Resources objects in series and in parallel. (#6567)

  • The diagonalize_measurements transform no longer raises an error for unknown observables. Instead, they are left un-diagonalized, with the expectation that observable validation will catch any un-diagonalized observables that are also unsupported by the device. (#6653)

  • A qml.wires.Wires object can now be converted to a JAX array, if all wire labels are supported as JAX array elements. (#6699)

  • PennyLane is compatible with quimb 1.10.0. (#6630) (#6736)

  • A developer focused run function has been added to the qml.workflow module for a cleaner and standardized approach to executing tapes on an ML interface. (#6657)

  • Internal changes have been made to standardize execution interfaces, which resolves ambiguities in how the interface value is handled during execution. (#6643)

  • All interface handling logic has been moved to interface_utils.py in the qml.math module. (#6649)

  • qml.execute can now be used with diff_method="best". Classical cotransform information is now handled lazily by the workflow. Gradient method validation and program setup are now handled inside of qml.execute, instead of in QNode. (#6716)

  • PyTree support for measurements in a circuit has been added. (#6378) 

    @qml.qnode(qml.device("default.qubit"))
def circuit():
    qml.Hadamard(0)
    qml.CNOT([0,1])
    return {"Probabilities": qml.probs(), "State": qml.state()}
    >>> circuit()
{'Probabilities': array([0.5, 0. , 0. , 0.5]), 'State': array([0.70710678+0.j, 0.        +0.j, 0.        +0.j, 0.70710678+0.j])}

  • The _cache_transform transform has been moved to its own file located in pennylane/workflow/_cache_transform.py. (#6624)

  • The qml.BasisRotation template is now JIT compatible. (#6019) (#6779)

  • The Jaxpr primitives for for_loop, while_loop and cond now store slices instead of numbers of arguments. This helps with keeping track of what order the arguments come in. (#6521)

  • The ExecutionConfig.gradient_method function has been expanded to store TransformDispatcher type. (#6455)

  • The string representation of Resources instances has been improved to match the attribute names. (#6581)

  • The documentation for the dynamic_one_shot transform has been improved, and a warning is raised when a user-applied dynamic_one_shot transform is ignored in favour of the existing transform in a device's preprocessing transform program. (#6701)

  • A qml.devices.qubit_mixed module has been added for mixed-state qubit device support. This module introduces an apply_operation helper function that features:

    • Two density matrix contraction methods using einsum and tensordot
    • Optimized handling of special cases including: Diagonal operators, Identity operators, CX (controlled-X), Multi-controlled X gates, Grover operators (#6379)

  • A function called create_initial_state has been added to allow for initializing a circuit with a density matrix using qml.StatePrep or qml.QubitDensityMatrix. (#6503)

  • Several additions have been made to eventually migrate the "default.mixed" device to the new device API:

    • A preprocess method has been added to the QubitMixed device class to preprocess the quantum circuit before execution. (#6601)
    • A new class called DefaultMixedNewAPI has been added to the qml.devices.qubit_mixed module, which will replace the legacy DefaultMixed. (#6607)
    • A new submodule called devices.qubit_mixed.measure has been added, featuring a measure function for measuring qubits in mixed-state devices. (#6637)
    • A new submodule called devices.qubit_mixed.simulate has been added, featuring a simulate function for simulating mixed states in analytic mode. (#6618)
    • A new submodule called devices.qubit_mixed.sampling has been added, featuring functions sample_state, measure_with_samples and sample_probs for sampling qubits in mixed-state devices. (#6639)
    • The finite-shot branch of devices.qubit_mixed.simulate has been added, which allows for accepting stochastic arguments such as shots, rng and prng_key. (#6665)
    • Support for qml.Snapshot has been added. (#6659)

  • Reporting of test warnings as failures has been added. (#6217)

  • A warning message in the Gradients and training documentation has been added that pertains to ComplexWarnings. (#6543)

  • A new figure was added to the landing page of the PennyLane website. (#6696)

Breaking changes 💔

  • The default graph coloring method of qml.dot, qml.sum, and qml.pauli.optimize_measurements for grouping observables was changed from "rlf" to "lf". Internally, qml.pauli.group_observables has been replaced with qml.pauli.compute_partition_indices in several places to improve efficiency. (#6706)

  • qml.fourier.qnode_spectrum no longer automatically converts pure Numpy parameters to the Autograd framework. As the function uses automatic differentiation for validation, parameters from such a framework have to be used. (#6622)

  • qml.math.jax_argnums_to_tape_trainable has been moved and made private to avoid an unnecessary QNode dependency in the qml.math module. (#6609)

  • Gradient transforms are now applied after the user's transform program. This ensures user transforms work as expected on initial structures (e.g., embeddings or entangling layers), guarantees that gradient transforms only process compatible operations, aligns transform order with user expectations, and avoids confusion. (#6590)

  • Legacy operator arithmetic has been removed. This includes qml.ops.Hamiltonian, qml.operation.Tensor, qml.operation.enable_new_opmath, qml.operation.disable_new_opmath, and qml.operation.convert_to_legacy_H. Note that qml.Hamiltonian will continue to dispatch to qml.ops.LinearCombination. For more information, check out the updated operator troubleshooting page. (#6548) (#6602) (#6589)

  • The developer-facing qml.utils module has been removed. (#6588):

    Specifically, the following 4 sets of functions have been either moved or removed:

    • qml.utils._flatten, qml.utils.unflatten has been moved and renamed to qml.optimize.qng._flatten_np and qml.optimize.qng._unflatten_np respectively.
    • qml.utils._inv_dict and qml._get_default_args have been removed.
    • qml.utils.pauli_eigs has been moved to qml.pauli.utils.
    • qml.utils.expand_vector has been moved to qml.math.expand_vector.

  • The qml.qinfo module has been removed. Please use the corresponding functions in the qml.math and qml.measurements modules instead. (#6584)

  • Top level access to Device, QubitDevice, and QutritDevice have been removed. Instead, they are available as qml.devices.LegacyDevice, qml.devices.QubitDevice, and qml.devices.QutritDevice, respectively. (#6537)

  • The 'ancilla' argument for qml.iterative_qpe has been removed. Instead, use the 'aux_wire' argument. (#6532)

  • The qml.BasisStatePreparation template has been removed. Instead, use qml.BasisState. (#6528)

  • The qml.workflow.set_shots helper function has been removed. We no longer interact with the legacy device interface in our code. Instead, shots should be specified on the tape, and the device should use these shots. (#6534)

  • QNode.gradient_fn has been removed. Please use QNode.diff_method instead. QNode.get_gradient_fn can also be used to process the differentiation method. (#6535)

  • The qml.QubitStateVector template has been removed. Instead, use qml.StatePrep. (#6525)

  • qml.broadcast has been removed. Users should use for loops instead. (#6527)

  • The max_expansion argument for qml.transforms.clifford_t_decomposition has been removed. (#6571)

  • The expand_depth argument for qml.compile has been removed. (#6531)

  • The qml.shadows.shadow_expval transform has been removed. Instead, please use the qml.shadow_expval measurement process. (#6530) (#6561)

  • The developer-facing qml.drawer.MPLDrawer argument n_wires has been replaced with wire_map, which contains more complete information about wire labels and order. This allows the new functionality to specify wire_options for specific wires when using string wire labels or non-sequential wire ordering. (#6805)

Deprecations 👋

  • The tape and qtape properties of QNode have been deprecated. Instead, use the qml.workflow.construct_tape function. (#6583) (#6650)

  • The max_expansion argument in qml.devices.preprocess.decompose is deprecated and will be removed in v0.41. (#6400)

  • The decomp_depth argument in qml.transforms.set_decomposition is deprecated and will be removed in v0.41. (#6400)

  • The output_dim property of qml.tape.QuantumScript has been deprecated. Instead, use method shape of QuantumScript or MeasurementProcess to get the same information. (#6577)

  • The QNode.get_best_method and QNode.best_method_str methods have been deprecated. Instead, use the qml.workflow.get_best_diff_method function. (#6418)

  • The qml.execute gradient_fn keyword argument has been renamed to diff_method to better align with the termionology used by the QNode. gradient_fn will be removed in v0.41. (#6549)

  • The old qml.qsvt functionality is moved to qml.qsvt_legacy and is now deprecated. It will be removed in v0.41. (#6520)

Documentation 📝

  • The docstrings for qml.qchem.Molecule and qml.qchem.molecular_hamiltonian have been updated to include a note that says that they are not compatible with qjit or jit. (#6702)

  • The documentation of TrotterProduct has been updated to include the impact of the operands in the Hamiltonian on the structure of the created circuit. (#6629)

  • The documentation of QSVT has been updated to include examples for different block encodings. (#6673)

  • The link to qml.ops.one_qubit_transform was fixed in the QubitUnitary docstring. (#6745)

Bug fixes 🐛

  • Validation has been added to ensure that the device vjp is only used when the device actually supports it. (#6755)

  • qml.counts now returns all outcomes when the all_outcomes argument is True and mid-circuit measurements are present. (#6732)

  • qml.ControlledQubitUnitary now has consistent behaviour with program capture enabled. (#6719)

  • The Wires object now throws a TypeError if wires=None. (#6713) (#6720)

  • The qml.Hermitian class no longer checks that the provided matrix is hermitian. The reason for this removal is to allow for faster execution and avoid incompatibilities with jax.jit. (#6642)

  • Subclasses of qml.ops.Controlled no longer bind the primitives of their base operators when program capture is enabled. (#6672)

  • The qml.HilbertSchmidt and qml.LocalHilbertSchmidt templates now apply the complex conjugate of the unitaries instead of the adjoint, providing the correct result. (#6604)

  • QNode return behaviour is now consistent for lists and tuples. (#6568)

  • QNodes now accept arguments with types defined in libraries that are not necessarily in the list of supported interfaces, such as the Graph class defined in networkx. (#6600)

  • qml.math.get_deep_interface now works properly for Autograd arrays. (#6557)

  • Printing instances of qml.Identity now returns the correct wires list. (#6506)

Contributors ✍️

This release contains contributions from (in alphabetical order):

Guillermo Alonso, Shiwen An, Utkarsh Azad, Astral Cai, Yushao Chen, Isaac De Vlugt, Diksha Dhawan, Lasse Dierich, Lillian Frederiksen, Pietropaolo Frisoni, Simone Gasperini, Diego Guala, Austin Huang, Korbinian Kottmann, Christina Lee, Alan Martin, William Maxwell, Anton Naim Ibrahim, Andrija Paurevic, Justin Pickering, Jay Soni, David Wierichs.

Check out latest releases or
releases around PennyLaneAI/pennylane v0.40.0

Don't miss a new pennylane release

NewReleases is sending notifications on new releases.

Get notifications