qp.Select¶
- class Select(ops, control, work_wires=None, partial=False)[source]¶
Bases:
Operator2The
Selectoperator, also available asMultiplexerandMultiplexor, applies different operations depending on the state of designated control wires.\[Select|i\rangle \otimes |\psi\rangle = |i\rangle \otimes U_i |\psi\rangle\]
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
- 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
controlare 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 applyops[2]. To obtain the list positionindexfor 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=Trueassumes that the quantum state \(|\psi\rangle\) on thecontrolwires 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, setpartial=Falseto 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
Selectoperator 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 \(j>K-1\)). Passingpartial=TruetellsSelectthat 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.
Available decompositions
Resources are general, for \(K\) operators \(U_j=\)
ops[j], \(c\geq 2\) control wires, \(w\) work wires andpartial=False. Each circuit shows one example of the rule. Use a rule when decomposing withqp.decompose:qp.decompose(gate_set=..., fixed_decomps={qp.Select: "<name>"})
multi_controlExample: \(K=4\), \(c=2\)
Gate
Count
Example
\(U_j\) with \(c\) controls
\(K\) (one per \(U_j\))
4
unaryExample: \(K=4\), \(c=2\)
Requires \(w \geq c - 1\) work wires
Gate
Count
Example
\(N\)
1
\(N\)
1
\(K - 1\) \(+ [K > 2^{c-1}]\) \(- [K > 2^{c-2}]\)
3
Xwith 1 control on \(|0\rangle\)\([K > 2^{c-2}]\)
1
\(U_j\) with 1 control
\(K\) (one per \(U_j\))
4
where \(N = c + K - 2 - \mathrm{wt}(K-1) - [K > 2^{c-1}]\), with \(\mathrm{wt}\) the number of ones in the binary representation, and \([\cdot]\) equal to 1 if the condition holds and 0 otherwise
multi_control_work_wireExample: \(K=4\), \(c=2\)
Requires \(w \geq 1\) work wire
Gate
Count
Example
MultiControlledXwith \(c\) controls\(2K\)
8
\(U_j\) with 1 control
\(K\) (one per \(U_j\))
4
Unary iterator decomposition
Generically,
Selectis 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
aux0andaux1below, 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
TemporaryANDand 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 ofTemporaryANDgates:─○─╮─╭○── ── ─○─╮─╭○── ─╭○─ ─○─╮─╭●── ─╭●──── ─○─┤─├○── = ──, ─○─┤─├●── = ─│──, and ─●─┤─├○── = ─│──╭●─. ───╯ ╰─── ── ───╯ ╰─── ─╰X─ ───╯ ╰─── ─╰X─╰X─
Applying these simplifications reduces the computational cost of the
Selecttemplate: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
TemporaryANDgate and its adjoint require four and zeroTgates, respectively, in contrast to the sevenTgates required by a decomposition ofToffoli.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 thatRdistinguishes. First, ifL>1, it applies the circuitaux_j: ╭R ─╭●────╭●────●─╮─ j+1: ├R = ─├○────│─────●─┤─ aux_j+1: ╰R ╰──R──╰X─R────╯ ,
where each label
Rsymbolizes a call toRitself, 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, ifL=0,Rdoes not apply any operators.With
Rdefined, we are ready to outline the main circuit structure:Apply the left-most
TemporaryANDcontrolled on qubits0and1.Split the target operators into four “quarters” (often with varying sizes) and apply the first quarter using
R.Apply
[X(0), CNOT([0, "aux0"]), X(0)].Apply the second quarter using
R.Apply
[CNOT([0, "aux0"]), CNOT([1, "aux0"])].Apply the third quarter using
R.Apply
[CNOT([0, "aux0"])].Apply the last quarter using
R.Apply the right-most
adjoint(TemporaryAND)controlled on qubits0and1.
Partial Select decomposition
The unary iterator decomposition of the
Selecttemplate can be simplified further if both of the following criteria are met:There are fewer target operators than would maximally be possible for the given number of control wires, i.e. \(K<2^c\).
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
Rlabels symbolizes a call toR. 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 middleTemporaryANDgates 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
Rlabels represent calls toR, with single-label instances applying fewer operators. The first call toRin 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
Rlabels symbolize a call toRand apply \(2^{c-2}\) operators each. The controlled gate on the right applies the single remaining operator.
Attributes
The expected types for the arguments of an operator.
Dictionary mapping argument names to their values.
Arithmetic depth of the operator.
Batch size of the operator if it is used with broadcasted parameters.
The names of arguments that are treated as compilable static arguments.
Dictionary mapping compilable argument names to their values.
partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.
The trainable parameters of the target operators.
The names of arguments that are treated as dynamic.
Dictionary mapping dynamic argument names to their values.
Gradient computation method.
Compute 'grad_recipe' lazily.
Whether the expected signature of an operator is fixed.
The names of arguments that represent dynamic data wrapped in static structures (known as Pytrees).
Dictionary mapping hybrid argument names to their values.
Legacy Operator compatibility view of non-dynamic, non-wire constructor arguments.
Whether this operator contains only abstract data.
This property determines if an operator is verified to be Hermitian.
Operator name.
Number of dimensions per trainable parameter of the operator.
The number of trainable parameters of the target operators.
partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.
Legacy Operator compatibility view of dynamic numerical arguments as a list.
partial(func, *args, **keywords) - new function with partial application of the given arguments and keywords.
A
PauliSentencerepresentation of the Operator, orNoneif it doesn't have one.The names of arguments that are treated as static.
Dictionary mapping static argument names to their values.
The wires of the target operators.
The names of arguments corresponding to wires.
Dictionary mapping wire argument names to their values.
The expected number of wire labels for each wire argument.
Wires that the operator acts on.
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_specsthat 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_specsand give each rule aregister_conditionthat checks the type. For a concrete example seeBasisRotation.
- 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_sizeis determined based onndim_paramsand 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 isNone.- 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 instatic_argnames.Note
An operator can only specify
static_argnamesandhybrid_argnames, orcompilable_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', orNoneif 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 whenarg_specscovers 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, andcompilable_argnames, but may overlap withwire_argnameswhen 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_argnamesandhybrid_argnames, orcompilable_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
Trueif the operator is guaranteed to be Hermitian andFalseif 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
PauliSentencerepresentation of the Operator, orNoneif 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_argnamesandhybrid_argnames, orcompilable_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
Wiresobjects by theOperator2constructor. 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 inhybrid_argnamesas well. Additionally,Operator2does not descend into the pytree structure of such arguments, so subclasses must ensure every wire leaf inside a hybrid wire argument is already aWiresobject before forwarding it tosuper().__init__.The order in which names appear in
wire_argnamesdetermines the order in which their wires appear inop.wires(seeOperator2.wires). For hybrid wire arguments, the containedWiresleaves are ordered by pytree traversal order. Wires contributed by arguments which are themselvesOperator2objects are appended after allwire_argnameswires. The special names"work_wires"and"work_wire"may be included inwire_argnamesbut their values are excluded fromop.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,
Nonemust be used. By default, all wire arguments are assumed to support arbitrary sizes. For hybrid wire arguments, the wire size must always beNone. 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
Wiresare collected from the operator’s arguments in the following order: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, everyWiresleaf of the argument’s pytree is added in pytree traversal order.
After all
wire_argnameshave been processed, for each name inhybrid_argnamesthat is not inwire_argnames, the wires of anyOperator2leaves 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 namedwork_wiresorwork_wireare 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:
Methods
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).
Representation of the operator as a product of other operators.
Sequence of gates that diagonalize the operator in the computational basis.
eigvals()Eigenvalues of the operator in the computational basis.
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
Adjointoperators constructed byadjoint(). TODO: [sc-120844] Fix docstring after Adjoint is addedAdjointed operations are the conjugated and transposed version of the original operation. Adjointed ops are equivalent to the inverted operation for unitary gates.
Operator2.adjointcan be optionally defined by Operator developers, whileadjoint()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_decompositionmethod.See also
- 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.
See also
- 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_gatesare specified and implement a unitary \(U^{\dagger}\), the operator can be reconstructed as\[O = U \Sigma U^{\dagger},\]where \(\Sigma\) is the diagonal matrix containing the eigenvalues.
Otherwise, no particular order for the eigenvalues is guaranteed.
See also
- 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.
See also
- 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
- 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
DecompositionUndefinedErroris raised if no representation by decomposition is defined.See also
- 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
DiagGatesUndefinedErroris raised if no representation by decomposition is defined.See also
- Returns:
a list of operators
- Return type:
list[.Operator2] or None
- eigvals()¶
Eigenvalues of the operator in the computational basis.
If
diagonalizing_gatesare 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
EigvalsUndefinedErroris raised if the eigenvalues have not been defined and cannot be inferred from the matrix representation.See also
- 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
LinearCombinationandSparseHamiltonianrespectively).
- 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,
Operator2subclasses must override thelabelmethod.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_orderis 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
MatrixUndefinedErroris raised if the matrix representation has not been defined.See also
- 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
Powinstances created bypow()orop ** power. TODO: [sc-120843] Fix docstring after Pow is addedOperator2.powcan be optionally defined by Operator developers, whilepow()orop ** powerare 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_orderis 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
SparseMatrixUndefinedErroris raised if the sparse matrix representation has not been defined.See also
- 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
TermsUndefinedErroris 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]]