qp.Select

class Select(ops, control, work_wires=None, partial=False)[source]

Bases: Operator2

The Select operator, also available as Multiplexer and Multiplexor, applies different operations depending on the state of designated control wires.

\[Select|i\rangle \otimes |\psi\rangle = |i\rangle \otimes U_i |\psi\rangle\]
../../_images/select.png

If the applied operations \(\{U_i\}\) are all single-qubit Pauli rotations about the same axis, with the angle determined by the control wires, this is also called a uniformly controlled rotation gate.

See also

SelectPauliRot

Parameters:
  • ops (list[Operator]) – operations to apply

  • control (Sequence[int]) – the wires controlling which operation is applied. At least \(\lceil \log_2 K\rceil\) wires are required for \(K\) operations.

  • work_wires (Union[Wires, Sequence[int], or int]) – auxiliary wire(s) that may be utilized during the decomposition of the operator into native operations. For details, see the section on the unary iterator decomposition below.

  • partial (bool) – Whether the state on the wires provided in control are compatible with a partial Select decomposition. See the note below for details.

Note

The position of the operation in the list determines which qubit state implements that operation. For example, when the qubit register is in the state \(|00\rangle\), we will apply ops[0]. When the qubit register is in the state \(|10\rangle\), we will apply ops[2]. To obtain the list position index for a given binary bitstring representing the control state we can use the following relationship: index = int(state_string, 2). For example, 2 = int('10', 2).

Note

Using partial=True assumes that the quantum state \(|\psi\rangle\) on the control wires satisfies \(\langle j|\psi\rangle=0\) for all \(j\in [K, 2^c)\), where \(K\) is the number of operators (len(ops)) and \(c\) is the number of control wires (len(control)). If you are unsure whether this condition is satisfied, set partial=False to guarantee a correct, even though more expensive, decomposition. For more details on the partial Select decomposition, see its compilation page.

Example

>>> dev = qp.device('default.qubit', wires=4)
>>> ops = [qp.X(2), qp.X(3), qp.Y(2), qp.SWAP([2, 3])]
>>> @qp.qnode(dev)
... def circuit():
...     qp.Select(ops, control=[0,1])
...     return qp.state()
...
>>> print(qp.draw(circuit, level='device')())
0: ─╭○─╭○─╭●─╭●────┤ ╭State
1: ─├○─├●─├○─├●────┤ ├State
2: ─╰X─│──╰Y─├SWAP─┤ ├State
3: ────╰X────╰SWAP─┤ ╰State

If there are fewer operators to be applied than possible for the given number of control wires, we call the Select operator a partial Select. In this case, the control structure can be simplified if the state on the control wires does not have overlap with the unused computational basis states (\(|j\rangle\) with \(j>K-1\)). Passing partial=True tells Select that this criterion is satisfied, and allows the decomposition to make use of the simplification:

>>> ops = [qp.X(2), qp.X(3), qp.SWAP([2, 3])]
>>> @qp.qnode(dev)
... def circuit():
...     qp.Select(ops, control=[0, 1], partial=True)
...     return qp.state()
...
>>> print(qp.draw(circuit, level='device')())
0: ─╭○────╭●────┤ ╭State
1: ─├○─╭●─│─────┤ ├State
2: ─╰X─│──├SWAP─┤ ├State
3: ────╰X─╰SWAP─┤ ╰State

Note how the first (second) control node of the second (third) operator was skipped.

Generically, Select is decomposed into one multi-controlled operator for each target operator. However, if auxiliary wires are available, a decomposition using a “unary iterator” can be applied. It was introduced by Babbush et al. (2018).

Principle

Unary iteration leverages auxiliary wires to store intermediate values for reuse between the different multi-controlled operators, avoiding unnecessary recomputation. In addition to this caching functionality, unary iteration reduces the cost of the computation directly, because the involved reversible AND (or Toffoli) gates can be implemented at lower cost if the target is known to be in the \(|0\rangle\) state (see TemporaryAND).

For \(K\) operators to be Select-applied, \(c=\lceil\log_2 K\rceil\) control wires are required. Unary iteration demands an additional \(c-1\) auxiliary wires. Below we first show an example for \(K\) being a power of two, i.e., \(K=2^c\). Then we elaborate on implementation details for the case \(K<2^c\), which we call a partial Select operator.

Example

Assume that we want to Select-apply \(K=8=2^3\) operators to two target wires, which requires \(c=\lceil \log_2 K\rceil=3\) control wires. The generic decomposition for this takes the form

0: ─╭○─────╭○─────╭○─────╭○─────╭●─────╭●─────╭●─────╭●─────┤
1: ─├○─────├○─────├●─────├●─────├○─────├○─────├●─────├●─────┤
2: ─├○─────├●─────├○─────├●─────├○─────├●─────├○─────├●─────┤
3: ─├U(M0)─├U(M1)─├U(M2)─├U(M3)─├U(M4)─├U(M5)─├U(M6)─├U(M7)─┤
4: ─╰U(M0)─╰U(M1)─╰U(M2)─╰U(M3)─╰U(M4)─╰U(M5)─╰U(M6)─╰U(M7)─┤.

Unary iteration then uses \(c-1=2\) auxiliary wires, denoted aux0 and aux1 below, to first rewrite the control structure:

0:    ─╭○───────○╮─╭○───────○╮─╭○───────○╮─╭○───────○╮─╭●───────●╮─╭●───────●╮─╭●───────●╮─╭●───────●╮─┤
1:    ─├○───────○┤─├○───────○┤─├●───────●┤─├●───────●┤─├○───────○┤─├○───────○┤─├●───────●┤─├●───────●┤─┤
aux0:  ╰─╭●───●╮─╯ ╰─╭●───●╮─╯ ╰─╭●───●╮─╯ ╰─╭●───●╮─╯ ╰─╭●───●╮─╯ ╰─╭●───●╮─╯ ╰─╭●───●╮─╯ ╰─╭●───●╮─╯ │
2:    ───├○───○┤─────├●───●┤─────├○───○┤─────├●───●┤─────├○───○┤─────├●───●┤─────├○───○┤─────├●───●┤───┤
aux1:    ╰─╭●──╯     ╰─╭●──╯     ╰─╭●──╯     ╰─╭●──╯     ╰─╭●──╯     ╰─╭●──╯     ╰─╭●──╯     ╰─╭●──╯   │
3:    ─────├U(M0)──────├U(M1)──────├U(M2)──────├U(M3)──────├U(M4)──────├U(M5)──────├U(M6)──────├U(M7)──┤
4:    ─────╰U(M0)──────╰U(M1)──────╰U(M2)──────╰U(M3)──────╰U(M4)──────╰U(M5)──────╰U(M6)──────╰U(M7)──┤

Here, we used the symbols

0: ─╭●──       ─●─╮─
1: ─├●──  and  ─●─┤─
2:  ╰───       ───╯

for TemporaryAND and its adjoint, respectively, and skipped drawing the auxiliary wires in areas where they are guaranteed to be in the state \(|0\rangle\). We will need three simplification rules for pairs of TemporaryAND gates:

─○─╮─╭○──   ──     ─○─╮─╭○──   ─╭○─       ─○─╮─╭●──   ─╭●────
─○─┤─├○── = ──,    ─○─┤─├●── = ─│──, and  ─●─┤─├○── = ─│──╭●─.
───╯ ╰───   ──     ───╯ ╰───   ─╰X─       ───╯ ╰───   ─╰X─╰X─

Applying these simplifications reduces the computational cost of the Select template:

0:    ─╭○────────────────╭○──────────────────╭●─────────────────────╭●─────────────────●╮─┤
1:    ─├○────────────────│───────────────────│──╭●──────────────────│──────────────────●┤─┤
aux0:  ╰─╭●─────╭●────●╮─╰X─╭●─────╭●─────●╮─╰X─╰X─╭●─────╭●─────●╮─╰X─╭●─────╭●─────●╮─╯ │
2:    ───├○─────│─────●┤────├○─────│──────●┤───────├○─────│──────●┤────├○─────│──────●┤───┤
aux1:    ╰─╭●───╰X─╭●──╯    ╰─╭●───╰X──╭●──╯       ╰─╭●───╰X──╭●──╯    ╰─╭●───╰X──╭●──╯   │
3:    ─────├U(M0)──├U(M1)─────├U(M2)───├U(M3)────────├U(M4)───├U(M5)─────├U(M6)───├U(M7)──┤
4:    ─────╰U(M0)──╰U(M1)─────╰U(M2)───╰U(M3)────────╰U(M4)───╰U(M5)─────╰U(M6)───╰U(M7)──┤

An additional cost reduction then results from the fact that the TemporaryAND gate and its adjoint require four and zero T gates, respectively, in contrast to the seven T gates required by a decomposition of Toffoli.

For general \(c\) and \(K=2^c\), the decomposition takes a similar form, with alternating control and auxiliary wires.

An implementation of the unary iterator is achieved in the following steps: We first define a recursive sub-circuit R; given \(L\) operators and \(2 \lceil\log_2(L)\rceil + 1\) control and auxiliary wires, there are three cases that R distinguishes. First, if L>1, it applies the circuit

aux_j:   ╭R   ─╭●────╭●────●─╮─
j+1:     ├R = ─├○────│─────●─┤─
aux_j+1: ╰R    ╰──R──╰X─R────╯ ,

where each label R symbolizes a call to R itself, on the next recursion level. These next-level calls use \(L' = 2^{\lceil\log_2(L)\rceil-1}\) (i.e. half of \(L\), rounded up to the next power of two) and \(L-L'\) (i.e. the rest) operators, respectively.

Second, if L=1, the single operator is applied, controlled on the first control wire. Finally, if L=0, R does not apply any operators.

With R defined, we are ready to outline the main circuit structure:

  1. Apply the left-most TemporaryAND controlled on qubits 0 and 1.

  2. Split the target operators into four “quarters” (often with varying sizes) and apply the first quarter using R.

  3. Apply [X(0), CNOT([0, "aux0"]), X(0)].

  4. Apply the second quarter using R.

  5. Apply [CNOT([0, "aux0"]), CNOT([1, "aux0"])].

  6. Apply the third quarter using R.

  7. Apply [CNOT([0, "aux0"])].

  8. Apply the last quarter using R.

  9. Apply the right-most adjoint(TemporaryAND) controlled on qubits 0 and 1.

Partial Select decomposition

The unary iterator decomposition of the Select template can be simplified further if both of the following criteria are met:

  1. There are fewer target operators than would maximally be possible for the given number of control wires, i.e. \(K<2^c\).

  2. The state \(|\psi\rangle\) of the control wires satisfies \(\langle j | \psi\rangle=0\) for all computational basis states with \(j\geq K\).

We do not derive this reduction here but discuss the modifications to the implementation above that result from it.

Given \(K=2^c-b\) operators, where \(c\) is defined as above and we have \(0\leq b<2^{c-1}\), the nine steps above are modified into one of three variants. In each variant, the first \(2^{c-1}\) operators are applied in two equal portions, containing \(2^{c-2}\) operators each. After this, \(\ell=2^{c-1} -b\) operators remain and the three circuit variants are distinguished, based on \(\ell\):

  • if \(\ell \geq 2^{c-2}\), the following, rather generic, circuit is applied:

    0:    ─╭○─────╭○─────╭●────────╭●─────●─╮─
    1:    ─├○─────│──────│──╭●─────│──────●─┤─
    aux0:  ╰──╭R──╰X─╭R──╰X─╰X─╭R──╰X─╭R────╯
    2:    ────├R─────├R────────├R─────├R──────
    aux1:     ╰R     ╰R        ╰R     ╰R      .
    

    Here, each operator with three R labels symbolizes a call to R. The first call in the second half applies \(2^{\lceil\log_2(\ell)\rceil-1}\) operators. Note that this case is triggered if \(K\) is larger than or equal to \(\tfrac{3}{4}\) of the maximal capacity for \(c\) control wires. Also note how the two middle TemporaryAND gates were merged into two CNOTs, like for the non-partial Select operator.

  • if \(1<\ell < 2^{c-2}\), the following circuit is applied:

    0:    ─╭○─────╭○─────○─╮╭●─────╭●─────●─╮─
    1:    ─├○─────│──────●─┤│──────│────────│─
    aux0:  ╰──╭R──╰X─╭R────╯│      │        │
    2:    ────├R─────├R─────├○─────│──────●─┤─
    aux1:     ╰R     ╰R     ╰───R──╰X──R────╯
    

    where the second half may skip more than one control and auxiliary wire each. In this diagram, both the operators with three and one R labels represent calls to R, with single-label instances applying fewer operators. The first call to R in the second half applies \(2^{\lceil\log_2(\ell)\rceil-1}\) operators. The middle elbows act on distinct wire triples and can not be merged as above.

  • if \(\ell=1\), the following circuit is applied:

    0:    ─╭○─────╭○─────○─╮╭●──
    1:    ─├○─────│──────●─┤│───
    aux0:  ╰──╭R──╰X─╭R────╯│───
    2:    ────├R─────├R─────│───
    aux1:     ╰R     ╰R     ╰U  .
    

    Here, the three connected R labels symbolize a call to R and apply \(2^{c-2}\) operators each. The controlled gate on the right applies the single remaining operator.

arg_specs

The expected types for the arguments of an operator.

arguments

Dictionary mapping argument names to their values.

arithmetic_depth

Arithmetic depth of the operator.

batch_size

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

compilable_argnames

The names of arguments that are treated as compilable static arguments.

compilable_args

Dictionary mapping compilable argument names to their values.

control

partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.

data

The trainable parameters of the target operators.

dynamic_argnames

The names of arguments that are treated as dynamic.

dynamic_args

Dictionary mapping dynamic argument names to their values.

grad_method

Gradient computation method.

grad_recipe

Compute 'grad_recipe' lazily.

has_adjoint

has_decomposition

has_diagonalizing_gates

has_fixed_sig

Whether the expected signature of an operator is fixed.

has_generator

has_matrix

has_sparse_matrix

hybrid_argnames

The names of arguments that represent dynamic data wrapped in static structures (known as Pytrees).

hybrid_args

Dictionary mapping hybrid argument names to their values.

hyperparameters

Legacy Operator compatibility view of non-dynamic, non-wire constructor arguments.

is_fully_abstract

Whether this operator contains only abstract data.

is_verified_hermitian

This property determines if an operator is verified to be Hermitian.

name

Operator name.

ndim_params

Number of dimensions per trainable parameter of the operator.

num_params

The number of trainable parameters of the target operators.

ops

partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.

parameters

Legacy Operator compatibility view of dynamic numerical arguments as a list.

partial

partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.

pauli_rep

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

static_argnames

The names of arguments that are treated as static.

static_args

Dictionary mapping static argument names to their values.

target_wires

The wires of the target operators.

wire_argnames

The names of arguments corresponding to wires.

wire_args

Dictionary mapping wire argument names to their values.

wire_sizes

The expected number of wire labels for each wire argument.

wires

Wires that the operator acts on.

work_wires

partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.

arg_specs = {'control': AbstractWires(-1), 'work_wires': AbstractWires(-1)}

The expected types for the arguments of an operator. This attribute is optional—not setting it has no loss of functionality. If set, it can be used to perform automatic validation of an operators inputs during construction. Additionally, when defining decomposition rules for an operator, operator types with arg_specs that spans all the arguments with static types can be placed in the rules’ resources without needing to fully construct abstract operators.

Note

A type that is listed in ‘arg_specs’ says what an argument is allowed to be, not what it actually is. For example, if arg_specs contains Complex[-1, -1], the Operator can still be instantiated with a real float64 array, which will then be reported as complex even though it holds real data.

The decomposition graph goes by the reported type, so real and complex inputs will look like the same operator and share one rule. To let them decompose differently, leave the argument out of arg_specs and give each rule a register_condition that checks the type. For a concrete example see BasisRotation.

arguments

Dictionary mapping argument names to their values.

arithmetic_depth

Arithmetic depth of the operator.

batch_size

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

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

Returns:

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

Return type:

int or None

compilable_argnames = ()

The names of arguments that are treated as compilable static arguments. Compilable static arguments are a subset of static arguments—arguments whose concrete values are known when compiling the program. But, unlike static_argnames, they are lowered to the compiler intermediate representation, making their values visible to the compiler, which may be useful if the compiler needs to interact with such values. Such values may include numbers, strings, and lists, tuples, or dictionaries thereof. This feature is opt-in; if any static arguments are not guaranteed to be compilable, it is safer to place them in static_argnames.

Note

An operator can only specify static_argnames and hybrid_argnames, or compilable_argnames, but not both; if any static or hybrid arguments are not or cannot be lowered to the IR, then all static and hybrid arguments are assumed to not be lowerable.

compilable_args

Dictionary mapping compilable argument names to their values.

control

partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.

data

The trainable parameters of the target operators.

dynamic_argnames = ()

The names of arguments that are treated as dynamic. Dynamic arguments are numerical arguments whose concrete values may not be known at compile-time, such as rotation angles, matrices, etc., and may include trainable data. Additionally, dynamic arguments are the only ones used to infer batch_size, necessary for broadcasting support.

Inputs for these arguments must be scalars, arrays, or castable to arrays (e.g. multi- dimensional lists with homogenous shapes), and operator constructors will be responsible for “canonicalizing” them (such as casting a list input to an array). At compile-time, the concrete values of these arguments are assumed to be unknown—only their shapes and data types are known.

dynamic_args

Dictionary mapping dynamic argument names to their values.

grad_method

Gradient computation method.

  • 'A': analytic differentiation using the parameter-shift method.

  • 'F': finite difference numerical differentiation.

  • None: the operation may not be differentiated.

Default is 'F', or None if the Operation has zero parameters.

grad_recipe

Compute ‘grad_recipe’ lazily.

has_adjoint = False
has_decomposition = True
has_diagonalizing_gates = False
has_fixed_sig = False

Whether the expected signature of an operator is fixed. If True, then the operator’s signature will always be fully known. When defining decomposition rules for an operator, operator types with fixed signatures can be placed in the rules’ resources without needing to fully construct abstract operators. This is set automatically when arg_specs covers every dynamic and wire argument, there are no hybrid, static, or compilable arguments, and every declared type is fully fixed (no unknown array shapes or wire counts).

has_generator = False
has_matrix = False
has_sparse_matrix = False
hybrid_argnames = ('ops',)

The names of arguments that represent dynamic data wrapped in static structures (known as Pytrees). Names in this category must be disjoint from dynamic_argnames, static_argnames, and compilable_argnames, but may overlap with wire_argnames when those arguments contain nested structures of wires. Examples of hybrid arguments include collections of wires or dynamic arrays, operators, etc.

Note

An operator can only specify static_argnames and hybrid_argnames, or compilable_argnames, but not both; if any static or hybrid arguments are not or cannot be lowered to the IR, then all static and hybrid arguments are assumed to not be lowerable.

hybrid_args

Dictionary mapping hybrid argument names to their values.

hyperparameters

Legacy Operator compatibility view of non-dynamic, non-wire constructor arguments.

is_fully_abstract

Whether this operator contains only abstract data.

is_verified_hermitian

This property determines if an operator is verified to be Hermitian.

Note

This property provides a fast, non-exhaustive check used for internal optimizations. It relies on quick, provable shortcuts (e.g., operator properties) rather than a full, computationally expensive check.

For a definitive check, use the pennylane.is_hermitian() function. Please note that this comes with increased computational cost.

Returns:

The property will return True if the operator is guaranteed to be Hermitian and False if the check is inconclusive and the operator may or may not be Hermitian.

Return type:

bool

Consider this operator,

>>> op = (qp.X(0) @ qp.Y(0) - qp.X(0) @ qp.Z(0)) * 1j

In this case, Hermicity cannot be verified and leads to an inconclusive result:

>>> op.is_verified_hermitian # inconclusive
False

However, using pennylane.is_hermitian() will give the correct answer:

>>> qp.is_hermitian(op) # definitive
True
name

Operator name.

ndim_params

Number of dimensions per trainable parameter of the operator.

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

Returns:

Number of dimensions for each trainable parameter.

Return type:

tuple

num_params

The number of trainable parameters of the target operators.

ops

partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.

parameters

Legacy Operator compatibility view of dynamic numerical arguments as a list.

partial

partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.

pauli_rep

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

static_argnames = ('partial',)

The names of arguments that are treated as static. Static arguments are those whose concrete values are known when compiling the program. Arguments specified here are not lowered to a compiler intermediate representation (IR). Alternatively, if an argument is of a type that can be lowered, it can be moved to compilable_argnames.

Note

An operator can only specify static_argnames and hybrid_argnames, or compilable_argnames, but not both; if any static or hybrid arguments are not or cannot be lowered to the IR, then all static and hybrid arguments are assumed to not be lowerable.

static_args

Dictionary mapping static argument names to their values.

target_wires

The wires of the target operators.

wire_argnames = ('control', 'work_wires')

The names of arguments corresponding to wires. Values for these arguments are automatically wrapped in Wires objects by the Operator2 constructor. If an argument is expected to be a structure that wraps wires (known as pytrees), such as a list of wire registers, make sure to include its name in hybrid_argnames as well. Additionally, Operator2 does not descend into the pytree structure of such arguments, so subclasses must ensure every wire leaf inside a hybrid wire argument is already a Wires object before forwarding it to super().__init__.

The order in which names appear in wire_argnames determines the order in which their wires appear in op.wires (see Operator2.wires). For hybrid wire arguments, the contained Wires leaves are ordered by pytree traversal order. Wires contributed by arguments which are themselves Operator2 objects are appended after all wire_argnames wires. The special names "work_wires" and "work_wire" may be included in wire_argnames but their values are excluded from op.wires.

wire_args

Dictionary mapping wire argument names to their values.

wire_sizes = (None, None)

The expected number of wire labels for each wire argument. If any wire arguments support an arbitrary size, None must be used. By default, all wire arguments are assumed to support arbitrary sizes. For hybrid wire arguments, the wire size must always be None. Specifying wire sizes allows for additional automatic validation to be implemented, but, specifying it is optional if such validation is not needed.

wires

Wires that the operator acts on.

By default, the returned Wires are collected from the operator’s arguments in the following order:

  1. For each name in wire_argnames (in declaration order):

    • If the name is not in hybrid_argnames, the canonical value of that argument is added.

    • If the name is in hybrid_argnames, every Wires leaf of the argument’s pytree is added in pytree traversal order.

  2. After all wire_argnames have been processed, for each name in hybrid_argnames that is not in wire_argnames, the wires of any Operator2 leaves inside that argument are appended to the end in pytree traversal order.

Duplicate wires are removed while preserving the first occurrence, so the final result is the ordered union of all wires above.

Note

By default, work wires are not included in op.wires. In particular, wire arguments named work_wires or work_wire are excluded.

Note

This property may be overridden by developers if the default behaviour is not desired, for reasons such as including work wires or changing the order of the wires.

Returns:

wires

Return type:

Wires

work_wires

partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.

adjoint()

Create an operation that is the adjoint of this one.

compute_decomposition(*args, **kwargs)

Representation of the operator as a product of other operators (static method).

compute_diagonalizing_gates(*args, **kwargs)

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

compute_eigvals(*args, **kwargs)

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

compute_matrix(*args, **kwargs)

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

compute_sparse_matrix(*args[, format])

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

decomposition()

Representation of the operator as a product of other operators.

diagonalizing_gates()

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

eigvals()

Eigenvalues of the operator in the computational basis.

generator()

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

label([decimals, base_label, cache])

A customizable string representation of the operator.

map_wires(wire_map)

Returns a new operator with its wires changed according to the given wire map.

matrix([wire_order])

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

pow(z)

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

queue([context])

Append the operator to the Operator queue.

simplify()

Reduce the depth of nested operators to the minimum.

sparse_matrix([wire_order, format])

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

terms()

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

adjoint()

Create an operation that is the adjoint of this one. Used to simplify Adjoint operators constructed by adjoint(). TODO: [sc-120844] Fix docstring after Adjoint is added

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

Operator2.adjoint can be optionally defined by Operator developers, while adjoint() is the entry point for constructing generic adjoint representations.

Returns:

The adjointed operation.

>>> from pennylane.core.operator import Operator2
>>> class MyClass(Operator2):
...     dynamic_argnames = ("phi",)
...
...     def __init__(self, phi, wires):
...         super().__init__(phi, wires)
...
...     def adjoint(self):
...         return self
...
>>> op = MyClass(0.5, wires=0).adjoint()
>>> op
MyClass(0.5, wires=[0])
classmethod compute_decomposition(*args, **kwargs)

Representation of the operator as a product of other operators (static method).

\[O = O_1 O_2 \dots O_n.\]

Note

Operations making up the decomposition should be queued within the compute_decomposition method.

See also

decomposition().

Parameters:
  • *args (tuple) – Positional arguments of the operator

  • **kwargs (dict) – Keyword arguments of the operator

Returns:

decomposition of the operator

Return type:

list[Operator2]

static compute_diagonalizing_gates(*args, **kwargs)

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

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

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

Parameters:
  • *args (tuple) – Positional arguments of the operator

  • **kwargs (dict) – Keyword arguments of the operator

Returns:

list of diagonalizing gates

Return type:

list[.Operator2]

static compute_eigvals(*args, **kwargs)

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

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

\[O = U \Sigma U^{\dagger},\]

where \(\Sigma\) is the diagonal matrix containing the eigenvalues.

Otherwise, no particular order for the eigenvalues is guaranteed.

Parameters:
  • *args (tuple) – Positional arguments of the operator

  • **kwargs (dict) – Keyword arguments of the operator

Returns:

eigenvalues

Return type:

tensor_like

static compute_matrix(*args, **kwargs)

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

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

Parameters:
  • *args (tuple) – Positional arguments of the operator

  • **kwargs (dict) – Keyword arguments of the operator

Returns:

matrix representation

Return type:

tensor_like

static compute_sparse_matrix(*args, format='csr', **kwargs)

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

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

See also

sparse_matrix()

Parameters:
  • *args (tuple) – Positional arguments of the operator

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

  • **kwargs (dict) – Keyword arguments of the operator

Returns:

sparse matrix representation

Return type:

scipy.sparse._csr.csr_matrix

decomposition()

Representation of the operator as a product of other operators.

\[O = O_1 O_2 \dots O_n\]

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

Returns:

decomposition of the operator

Return type:

list[Operator2]

diagonalizing_gates()

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

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

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

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

Returns:

a list of operators

Return type:

list[.Operator2] or None

eigvals()

Eigenvalues of the operator in the computational basis.

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

\[O = U \Sigma U^{\dagger},\]

where \(\Sigma\) is the diagonal matrix containing the eigenvalues.

Otherwise, no particular order for the eigenvalues is guaranteed.

Note

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

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

Returns:

eigenvalues

Return type:

tensor_like

generator()

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

For example, for operator

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

we get the generator

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

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

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

A customizable string representation of the operator.

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

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

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

Returns:

label to use in drawings

Return type:

str

Example:

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

Note that by default, operator wires and static inputs are not included in the label. To override this behaviour, Operator2 subclasses must override the label method.

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

>>> op2 = qp.QubitUnitary(np.eye(2), wires=0)
>>> cache = {'matrices': []}
>>> op2.label(cache=cache)
'U\n(M0)'
>>> cache['matrices']
[array([[1., 0.],
       [0., 1.]])]
>>> op3 = qp.QubitUnitary(np.eye(4), wires=(0,1))
>>> op3.label(cache=cache)
'U\n(M1)'
>>> cache['matrices']
[array([[1., 0.],
       [0., 1.]]), array([[1., 0., 0., 0.],
       [0., 1., 0., 0.],
       [0., 0., 1., 0.],
       [0., 0., 0., 1.]])]
map_wires(wire_map)

Returns a new operator with its wires changed according to the given wire map.

Parameters:

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

Returns:

new operator

Return type:

.Operator2

matrix(wire_order=None)

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

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

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

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

See also

compute_matrix()

Parameters:

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

Returns:

matrix representation

Return type:

tensor_like

pow(z)

A list of new operators equal to this one raised to the given power. This method is used to simplify Pow instances created by pow() or op ** power. TODO: [sc-120843] Fix docstring after Pow is added

Operator2.pow can be optionally defined by Operator developers, while pow() or op ** power are the entry point for constructing generic powers to exponents.

Parameters:

z (float) – exponent for the operator

Returns:

list[Operator2]

>>> from pennylane.core.operator import Operator2
>>> class MyClass(Operator2):
...     dynamic_argnames = ("phi",)
...
...     def __init__(self, phi, wires):
...         super().__init__(phi, wires)
...
...     def pow(self, z):
...         return [MyClass(self.phi*z, self.wires)]
...
>>> MyClass(0.5, 0).pow(2)
[MyClass(1.0, wires=[0])]
queue(context=<class 'pennylane.core.queuing.QueuingManager'>)

Append the operator to the Operator queue.

simplify()

Reduce the depth of nested operators to the minimum.

Returns:

simplified operator

Return type:

.Operator2

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

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

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

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

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

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

Returns:

sparse matrix representation

Return type:

scipy.sparse._csr.csr_matrix

terms()

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

\[O = \sum_i c_i O_i\]

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

Returns:

list of coefficients \(c_i\) and list of operations \(O_i\)

Return type:

tuple[list[tensor_like or float], list[.Operator2]]

Contents

Using PennyLane

Release news

Development

API

Internals