METHODS AND SYSTEMS FOR PATCH-BY-PATCH CHARACTERIZATION OF QUANTOM PROCESSORS

Abstract

The present disclosure provides a method for characterizing a quantum processor including a plurality of qubits. The method comprising applying a characterization protocol to a qubit patch including a subset of qubits. The characterization protocol includes the reduction of ‘patching errors’—systematic characterization errors occurring due to interactions between qubits inside the patch and qubits outside the patch.

Description

TECHNOLOGICAL FIELD

The present disclosure relates to the field of quantum computing, specifically to the field of quantum processor characterization.

BACKGROUND

The development of useful quantum computers relies crucially on the reduction of implementation errors—differences between the actual and ideal implementations of quantum logic operations. The long-term strategy for dealing with such errors is quantum error correction, which, however, requires very low error rates to begin with, as well as an excessive overhead in the number of qubits, relative to existing hardware. Complementary strategies, suited for dealing with errors in small circuits on existing hardware, are collectively termed quantum error suppression and mitigation. Both the improvement of hardware to meet the conditions for error correction, and the majority of error suppression and mitigation protocols, require a detailed and high-resolution characterization of implementation errors. Though the characterization of implementation errors in single qubits or qubit pairs is common practice, it is widely understood as computationally intractable for more than just a few qubits—unless various simplifying and often un-realistic assumptions are made. This state of affairs prevents the detailed and high-resolution characterization of state-of-the-art quantum processors, currently numbering tens to hundreds of qubits.

As an example, given a Markovian model for implementation errors to be characterized, the most detailed characterization protocol known to date is termed gate set tomography (GST) as described for example in Nielsen et al. “Gate Set Tomography” Quantum 5, 557 (2021), which is hereby incorporated by reference. Here ‘gate set’ corresponds to a set of quantum logic operations, generally including quantum states, gates and measurements, which are characterized simultaneously, relative to one another. High resolution is achieved in GST by using long periodic sequences of gates, such that the estimation error scales as the inverse of the sequence length. The flexibility in choosing the implementation errors to be characterized, and the fact that gates, states and measurements are all being characterized, means that GST can be viewed as generalizing and unifying various pre-existing protocols, such as state, measurement, process and Hamiltonian tomography. But the detailed output of GST naturally makes it resource extensive, in terms of both quantum and classical run times, as well as classical memory. As a result, the open-source implementation of GST, which is arguably the industry standard for detailed characterization, currently only supports single- or two-qubit GST as a standard use case, see Nielsen et al. “Probing quantum processor performance with pyGSTi,” Quantum Science and Technology 5, 044002 (2020).

A common approach to bypass the computational barrier of many-qubit characterization is to characterize a small patch (i.e. a small subset) of qubits in a large device, simply ignoring all qubits outside the patch. As an example, all published experimental uses of GST to date have used either single- or two-qubit patches. The above approach is also utilized by fast GST-like characterization protocols, including the single-qubit ‘robust phase estimation’ and variations thereof, Kimmel et al. “Robust calibration of a universal single-qubit gate set via robust phase estimation” Phys. Rev. A 92, 06235 (2015) and Landa et al. “Experimental Bayesian estimation of quantum state preparation, measurement, and gate errors in multi-qubit devices” Phys. Rev. Research 4, 013199 (2022), and the two-qubit ‘Floquet calibration’, Arute et al. “Observation of separated dynamics of charge and spin in the Fermi-Hubbard model” arXiv:2010.07965 (2020), and ‘Hamiltonian error amplifying tomography’, Sundaresan et al. “Reducing Unitary and Spectator Errors in Cross Resonance with Optimized Rotary Echoes” PRX Quantum 1, 020318 (2020). Similarly, most randomized characterization protocols are commonly applied to single- or two-qubit patches. However, gates intended to act within a particular patch also act unintentionally on additional, typically neighboring, qubits. This important phenomenon is known as cross-talk, and such cross-talk errors between the patch and its neighborhood cannot be captured by single-patch characterization protocols.

GENERAL DESCRIPTION

The Applicant has found that cross-talk between the characterized patch and its neighborhood leads to systematic estimation errors of implementation errors occurring within the patch, essentially since qubits out of the patch play the role of an unmodeled environment. Such systematic errors are herein referred to as patching errors. It has been found that patching errors are, in many realistic scenarios, a limiting factor in the performance of single-patch characterization protocols, as well as more general patch-based characterization protocols utilizing data from multiple patches.

Patch-based characterization protocols suffer systematic patching errors, as defined above. The present disclosure provides a method for overcoming this difficulty, enabling an efficient, detailed and high-resolution characterization of implementation errors in single patches, which is not limited by patching errors. As a result, the characterizations of multiple patches can be combined into a reliable full device characterization. The method involves novel complementary ‘neighboring gate sequences’ acting on neighboring qubits in addition to standard ‘characterization gate sequences’ in a given patch. While the characterization gate sequences are devised to produce measurement outcomes which are sensitive to implementation errors in the patch, the neighboring gate sequences are chosen to decrease patching errors, by reducing sensitivity to implementation errors involving qubits outside the patch (i.e. cross-talk). This results in single-patch and full-device estimates carrying small patching errors, which are demonstrated to be below statistical errors in practical regimes of interest.

In general, the characterization method disclosed herein may be usable for ranking a plurality of quantum processors based on the fitted model parameters (e.g. by computing a distance measure such as fidelity or diamond distance between ideal and actual implementations of at least some or each quantum operation). In other embodiments, the characterization method may be usable for providing optimal control over the quantum processor by modifying how pulses implement the quantum logic operations on the quantum processor based on the fitted model parameters. In some embodiments, the characterization method may be usable for providing error suppression and mitigation by recompiling a quantum algorithm based on the fitted model parameters. For example, the recompiling may be performed to minimize the use of the noisiest gates. Quantum algorithms and their applications are generally described in the literature (see e.g. www.quantumalgorithimzoo.org). For example, the quantum algorithm may be a factoring algorithm to find a prime factorization of an n-bit integer for use in cryptography.

In accordance with a first aspect of the presently disclosed subject matter, there is provided a computer implemented method, for characterizing implementation errors in a set of quantum logic operations of a quantum processor. The set of quantum operations is acting on a subset of qubits, the subset of qubits is defining a qubit patch. The implementation errors involve the qubit patch and are modeled by a model. The model includes a plurality of model parameters. The method comprises applying a set of quantum circuits to the quantum processor, wherein at least some of the quantum circuits comprise a characterization gate sequence and a neighboring gate sequence. The characterization gate sequence is applicable to the qubit patch and is configured to provide measurement outcomes that are sensitive to at least some of the model's parameters. That is, a small change in said model parameters leads to a large change in (the probability distribution of) measurement outcomes obtained after applying a characterization sequence. The neighboring gate sequence is applicable to at least some neighboring qubits that are outside the patch and interacting with the qubit patch. The neighboring gate sequence is configured to reduce the sensitivity of measurement outcomes to implementation errors involving environment qubits outside the qubit patch. The method further comprises: measuring the patch qubits using a measurement apparatus of the quantum processor; repeating the aforementioned actions of application of a quantum circuit and of measurement, to collect a set of frequencies, where the frequencies are associated with a measurement outcome and a quantum circuit; computing a value of the model parameters by fitting the model to the set of frequencies.

In addition to the above features, a computer implemented method, for characterizing implementation errors in a set of quantum logic operations of a quantum processor, according to this aspect of the presently disclosed subject matter can optionally comprise one or more of features (i) to (xxxiv) below, in any technically possible combination or permutation:

• • i. the neighboring qubits comprise qubits interacting with the qubit patch based on an interaction hypergraph.
  • ii. the interaction hypergraph describes subsets of multiple qubits in the quantum processor interacting via multi-qubit implementation errors.
  • iii. the hypergraph describes subsets of multiple qubits in the quantum processor on which native multi-qubit gates can be applied.
  • iv. the characterization gate sequence is a gate set tomography sequence.
  • v. the set of quantum logic operations includes initialization and measurement.
  • vi. the gates in said set of quantum logic operations are configured to operate on single qubits or pairs of qubits.
  • vii. measuring patch qubits is performed by measuring all qubits in the quantum processor and further comprising computing reduced frequencies from the set of frequencies to aggregate outcomes with same patch qubits measurement outcomes.
  • viii. the neighboring gate sequence is configured to reduce a patching error.
  • ix. the patching error is estimated using perturbation expansion and the neighboring gate sequences are configured to cancel or at least reduce the leading orders in said perturbation expansion.
  • x. the leading orders in the perturbation expansion include perturbation expansion orders up to a given order.
  • xi. the leading orders in the perturbation expansion are up to the second order.
  • xii. the neighboring gate sequence comprises initializing the environment qubits in a state configured to cancel or at least reduce leading orders in said perturbation expansion.
  • xiii. the neighboring gate sequence comprises initializing the environment qubits in a neighborhood-mixed state, the neighborhood-mixed state being defined as a state wherein the reduced density operator of each neighboring qubit is proportional to the unit-operator. In some embodiments, the neighborhood-mixed state may not be the maximally mixed state (i.e. the state where the density operator of all environment qubits is proportional the unit-operator).
  • xiv. the neighborhood-mixed state is an even mixture (|0^⊗n+|1^⊗n)/2⊗|0^⊗(m-n), wherein m and n are the numbers of environment and neighboring qubits, respectively.
  • xv. the set of quantum circuits consists of quantum circuit clusters, each quantum circuit cluster comprising a plurality of quantum circuits having a same characterization gate sequence and a distinct neighboring gate sequence, the method further comprising combining frequencies collected on quantum circuits of the same cluster.
  • xvi. the set of quantum circuits comprises quantum circuit clusters each consisting of two quantum circuits having a same characterization gate sequence and respectively a first neighboring gate sequence starting with an idle gate and a second neighboring gate sequence starting with a π rotation, both first and second neighboring gate sequences being applied to each neighboring qubit, and wherein the method further comprises averaging frequencies collected on the two quantum circuits of each quantum circuit cluster.
  • xvii. at least one of the quantum circuits further comprises a context gate sequence applicable to one or more environment qubits outside the qubit patch, wherein context gates of said context gate sequence are applied synchronously to at least some characterization gates of the characterization gate sequence of said at least one quantum circuit.
  • xviii. the neighboring gate sequence and the context gate sequence are temporally intertwined.
  • xix. the neighboring gate sequence of at least one of the quantum circuits comprises at least some neighboring gates applied synchronously with at least some characterization gates of the characterization gate sequence of said at least one quantum circuit.
  • xx. the neighboring gate sequence of at least one of the quantum circuits comprises at least some neighboring gates applied successively to at least some characterization gates of the characterization gate sequence of said at least one quantum circuit.
  • xxi. some of the context gates are configured to form circuit layers together with the characterization gate sequences.
  • xxii. at least one of the quantum circuits includes a neighboring gate sequence comprising neighboring gates selected randomly.
  • xxiii. at least one of the quantum circuits includes a neighboring gate sequence comprising dynamical decoupling sequence configured to reduce patching errors.
  • xxiv. the dynamical decoupling sequence are configured to reduce patching errors estimated by a perturbative expansion.
  • xxv. the dynamical decoupling sequences are configured to eliminate leading orders in said perturbation expansion.
  • xxvi. the leading orders include perturbation expansion orders up to a given order.
  • xxvii. the leading orders in the perturbation expansion are up to the second order.
  • xxviii. the dynamical decoupling sequence is coordinated with said characterization gate sequences of said at least one quantum circuit.
  • xxix. at least some gates of the dynamical decoupling sequence are applied synchronously with at least some gates of the characterization gate sequence of said at least one quantum circuit.
  • xxx. at least some gates in the dynamical decoupling sequence are applied successively to at least some gates of the characterization gate sequence of said at least one quantum circuit.
  • xxxi. a qubit patch includes at least 3 qubits.
  • xxxii. gauge optimization.
  • xxxiii. the quantum processor is described in terms of qudits in place of qubits.
  • xxxiv. the gate set to be characterized contains at least one state preparation and measurement operation, or at least one non-idle gate.

In accordance with a second aspect of the presently disclosed subject matter, there is provided a computer implemented method for characterizing a quantum processor. The method comprises successively applying the method according to the first aspect of the presently disclosed subject matter to a plurality of patches of qubits of said quantum processor.

In addition to the above features, a computer implemented method for characterizing a quantum processor according to this aspect of the presently disclosed subject matter can optionally comprise one or more of features (i) to (iv) below, in any technically possible combination or permutation:

• • i. the plurality of patches are overlapping.
  • ii. the plurality of patches cover all of the qubits of the quantum processor.
  • iii. sequences corresponding to different patches, including both characterization and neighboring gate sequences, may be applied in parallel, in order to reduce the overall QPU time.
  • iv. gauge optimization.

In accordance with a third aspect of the presently disclosed subject matter, there is provided a non-transient computer-readable storage-medium storing computer instructions, wherein the computer instructions are used for causing a computer to execute the method according to the first or second aspects of the presently disclosed subject matter.

In accordance with a fourth aspect of the presently disclosed subject matter, there is provided a computer program product, comprising a computer program, wherein the computer program, when executed by a computer, implements the method according to the first or second aspects of the presently disclosed subject matter.

In accordance with a fifth aspect of the presently disclosed subject matter, there is provided a method for characterizing a quantum processor including a plurality of qubits, the method comprising applying a characterization protocol to a qubit patch including a subset of qubits, wherein the characterization protocol includes reducing a patching error due to interactions between qubits inside the qubit patch and qubits outside the qubit patch.

In addition to the above features, a method for characterizing a quantum processor including a plurality of qubits, according to this aspect of the presently disclosed subject matter can optionally comprise one or more of features (i) to (vi) below, in any technically possible combination or permutation:

• • i. successively applying the characterization protocol to a plurality of patches.
  • ii. the plurality of patches cover altogether the plurality of qubits of the quantum processor.
  • iii. at least some patches of the plurality of patches are overlapping.
  • iv. the characterization protocol includes applying neighboring gate sequences to neighboring qubits outside of the qubit patch and interacting with the qubit patch, wherein the neighboring gate sequences are configured to reduce the sensitivity of measurement outcomes to implementation errors involving qubits outside the qubit patch.
  • v. the characterization gate sequences in said characterization protocol are configured to provide measurement outcomes which are both sensitive to implementation errors in the patch and have a reduced sensitivity to implementation errors involving qubits outside the patch.
  • vi. the quantum processor is described in terms of qudits in place of qubits.

In accordance with a sixth aspect of the presently disclosed subject matter, there is provided a quantum circuit for use in characterizing implementation errors in a set of quantum logic operations of a quantum processor. The set of quantum operations is acting on a subset of qubits, the subset of qubits is defining a qubit patch. The implementation errors involve the qubit patch and are modeled by a model. The model includes a plurality of model parameters. The quantum circuit comprises: a characterization gate sequence applicable to the qubit patch and configured to provide measurement outcomes sensitive to at least some of the model parameters; a neighboring gate sequence applicable to at least some neighboring qubits interacting with the qubit patch, wherein the neighboring gate sequence is configured to reduce a sensitivity of measurement outcomes to implementation errors involving environment qubits outside the qubit patch.

In addition to the above features, a quantum circuit for use in characterizing implementation errors in a set of quantum logic operations of a quantum processor, according to this aspect of the presently disclosed subject matter can optionally comprise one or more of features (i) to (iv) below, in any technically possible combination or permutation:

• • i. the characterization gate sequence is a gate set tomography gate sequence.
  • ii. the neighboring gate sequence includes initializing the neighboring qubits in a neighborhood-mixed state.
  • iii. the neighboring gate sequence includes dynamical decoupling sequences configured to reduce patching errors.
  • iv. the neighboring gate sequence includes neighboring gates selected randomly.

In accordance with a seventh aspect of the presently disclosed subject matter, there is provided a system comprising a computer and a quantum processor, the computer having a gate-level access to the quantum processor, the system being configured to perform the method according to the first, second, fifth, or sixth aspects of the presently disclosed subject matter.

In the present disclosure, the following terms and their derivatives may be understood according to the below explanations:

The term “qubit” may refer to a 2-level quantum system.

The term “qudit” may refer to a M-level quantum system, with M≥2. This includes the qubit case M=2, and may include the case M=∞ corresponding to a bosonic mode.

The term ‘axis’ may refer to a geometrical axis of rotation in the Bloch-sphere, such as the x-axis, representation of a qubit.

The term ‘Hamiltonian’ may refer to, depending on the context, a Hamiltonian operator or any term thereof, or may refer to having the Hamiltonian property—absence of decoherence.

The term “multi-qubit interaction” may refer to an implementation error involving (m>1) multiple qubits.

The term “quantum logic operation” may refer to any of the following operations applied to a quantum processor: initialization, measurement, gate application and reset operation. Additionally, it may also refer to pulse application, mid-circuit measurement and/or adaptive measurement.

The term ‘model’ may refer to a mathematical model for the implementation of a set of quantum logic operations in a quantum processor, i.e. a function θ(θ) mapping a plurality of model parameters θ to a particular implementation (θ).

The term “fitting” may refer to any form of selecting or updating a model based on measurement outcomes obtained from a quantum processor. This includes e.g. the computation of a maximum likelihood point in a given model parameter space, or the Bayesian update of a probability distribution over said parameter space.

Both model and fitting procedure may incorporate machine learning methods, such as artificial neural networks and optimization algorithms used for fitting such models to data.

The term “hypergraph” may refer to a generalization of a graph in which edges can join any number of vertices.

The term “patch” may refer to a subset of qubits in a quantum processor.

The term “patching error” may refer to systematic estimation errors occurring in a characterization protocol applied to a patch in a quantum processor, due to unwanted interactions between the characterized patch and additional qubits in said quantum processor.

The term “environment qubit” may refer to a qubit outside of a qubit patch defined in a characterization method according to the present disclosure.

The term “neighboring qubit” may refer to an environment qubit (i.e. a qubit outside of said patch) interacting with the qubit patch.

The term “gate sequence” may refer to a plurality of gates (i.e. quantum logic operations) applicable to given qubits in a given temporal arrangement. In particular, the term “characterization gate sequence” may refer to a plurality of gates applicable to (patch) qubits in the patch and the term “neighboring gate sequence” may refer to a plurality of gates applicable to (neighboring) qubits outside the qubit patch. In particular, a gate sequence may comprise a successive plurality of bursts of gates in a given temporal arrangement, wherein the gates of each burst are applied directly successively. As will be described herein, a gate sequence (e.g. the neighboring gate sequence) may be composed of a plurality of intertwined gate sequences (e.g. the context gate sequence and the dynamical decoupling gate sequence).

It is understood that the terms “at least one” and “at least some” are used herein to provide a general description of the method. For the sake of brevity, the description is not repeated by interchanging these terms by the term “each”. However, an embodiment for which the terms “at least some” and/or “at least one” is replaced by the term “each” or “substantially each” is also herein disclosed.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to better understand the subject matter that is disclosed herein and to exemplify how it may be carried out in practice, embodiments will now be described, by way of non-limiting example only, with reference to the accompanying drawings, in which:

FIG. 1A illustrates a standard Bloch sphere representation of a two-level-system and key pure states, or kets.

FIG. 1B illustrates a density operator representation of a pure state, and a corresponding super-ket representation.

FIG. 2 illustrates π and

$\frac{\pi }{2}$

rotations or a quantum state around the x-axis of the Bloch sphere.

FIG. 3A-3D illustrates different types of implementation errors in quantum processors.

FIG. 4 illustrates an error amplification phenomenon.

FIG. 5A-5B schematically depict qubits in quantum processors and illustrates patch qubits, neighboring qubits and environment qubits.

FIG. 6 shows a flowchart illustrating a broad aspect of a method according to embodiments of the present disclosure.

FIG. 7A illustrates application of a quantum circuit, comprising a characterization gate sequence and a neighboring gate sequence, according to embodiments of the present disclosure.

FIG. 7B illustrates application of a quantum circuit, comprising a context gate sequence, according to embodiments of the present disclosure.

FIG. 7C illustrates application of a quantum circuit cluster according to embodiments of the present disclosure.

FIG. 8 illustrates a flowchart diagram of a method according to embodiments of the present disclosure.

FIG. 9 illustrates a scaling of a patching error, with and without initial twirl for a simulated quantum processor comprising two qubits.

FIG. 10 illustrates a ratio of the patching error to the statistical error, with different neighboring-sequence settings for a simulated quantum processor comprising three qubits.

FIG. 11 schematically illustrates a computer implementing a method according to embodiments of the present disclosure.

FIG. 12 schematically illustrates a system implementing a method according to embodiments of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS

Single-Qubit States and Rotations

Standard representations of single-qubit states and the meaning of rotation operations used in the following description are reviewed hereinbelow.

FIG. 1A shows the Bloch-sphere 100 representation of a two-level quantum system, i.e. a qubit. Key quantum states are indicated. Any pure state of a qubit may be represented as a point on the surface of the Bloch-sphere. A general pure state 155 is shown, along with its vector representation 150. The angle parameters correspond to the polar angle θ 160 and the azimuthal angle ϕ 170. For quantum computation, two antipodal pure states |0 and |1 replace the classical logical states 0 and 1. It is customary to represent the logical-0 state 110 and the logical-1 state 120 as intersection points between the Bloch sphere and the z-axis, corresponding the standard-basis vectors. FIG. 1A also shows states corresponding to the intersection of the Bloch sphere with the positive x-axis 130 and y-axis 155, along with their vector representations.

States which are not pure are termed mixed, and may be represented as points in the interior of the Bloch sphere—the Bloch-ball. In this case, the vector representation is replaced with a density operator representation. In FIG. 1B the density operator 180 corresponding to the state 150 is shown, along with an explicit matrix expression 185. The density operator 180 has a unique representation as a linear combination 190 of the Pauli operators and the identity operator. Finally, a vector whose elements are the coefficients of said linear combination is defined as the super-ket representation 195 of a density operator.

FIG. 2 illustrates a rotation around the x-axis of an initial state 1|ψ210 by

$\frac{\pi }{2}$

state X_π/2|ψ220 and by π to the state X_π|ψ230.

Quantum Logic Operations and Implementation Errors

The mathematical formalism used in the following description is reviewed hereinbelow.

A quantum processor, or quantum processing unit (QPU), may generally include a plurality of qubits and the following types of quantum logic operations:

• • 1. A set of initial states supported, e.g., a quantum processor might support initializing only to a single predefined state or to a plurality of predefined states. The set of initial states is denoted {|ρ}.
  • 2. A set of gates that may act on the qubits, e.g., a processor might support some operations involving one, two or more qubits. The set of gates is denoted {G}.
  • 3. A set of measurements, e.g., a processor might only support measuring all the qubits at once or may support measuring any single qubit. The set of measurements is denoted {E|}.

The union of these three sets may be referred to as a set of quantum logic operations, or “gate set”,

={E|}∪{G}∪{|ρ}.  (1)

The term “supported” means available as a native capability. Non-native gates may be replaced by equivalent combination of native gates. Similarly, non-native initial states or measurements may be obtained by applying native gates after applying a natively available initial state, or before the application of a natively available measurement. It is customary to collectively refer to the state preparations and measurements {|ρ}∪{E|} as SPAM.

Mathematically, each initial state p is a density operator, i.e. a positive semi-definite and unit-trace operator on a d-dimensional Hilbert space, where d=2ⁿ and n is the number of qubits. By choosing a Hilbert-Schmidt-orthonormal basis for Hermitian operators, each initial state can be represented as a vector |ρ∈^d2, a so-called super-ket. The operators E are positive semi-definite and satisfy ΣE=I. Each of these may be represented as a co-vector E|∈(^d2)* , or a super-bra. The gates G are quantum channels, mapping density operators to density operators. These are represented as super-operators, i.e., d²×d² matrices, G∈^d2×d2. We note that additional quantum logic operations may be considered and characterized using the method presented in this disclosure, such as reset gates, adaptive measurements, and mid-circuit measurements.

Each element in the gate set will have an ideal implementation, as intended by the hardware manufacturer. The ideal gate set may be denoted by

₀ ={ E ₀ |}∪{G ₀}∪{|ρ₀}.  (2)

As described in more detail below, we do not restrict the ideal gates G₀ to act non-trivially only on a small subset of qubits, e.g. to be 1- or 2-qubit gates. The ideal gates may act non-trivially on all qubits in the QPU, and can therefore be viewed as quantum-circuit-layers. Accordingly, we refer to a product of gates as a ‘sequence’ or a ‘circuit’ interchangeably. In some embodiments of the disclosed method, the gate set to be characterized contains at least one SPAM operation, or at least one non-idle gate, i.e. a gate G such that G o is not the identity matrix on all qubits.

The set of implementation errors may be defined as the difference between the actual implementation in Equation (1) and the ideal implementation in Equation (2). A Markovian error model may be used for simplicity without limitation of the scope of the present disclosure. In particular, this means that each gate set element may have a unique implementation error, independent of other circumstances. Such circumstances can include quantum operations applied previously, the time of day, etc. Additionally, it is noteworthy that given a Markovian error model on the full device, restricting attention to a particular patch generally leads to non-Markovian errors on that patch.

Each of the ideal gates G₀=U₀⊗U₀* corresponds to a unitary operator U₀∈SU(d), which acts on density operators as ρU₀ρU₀⁵⁵⁴. A coherent/unitary/Hamiltonian implementation error corresponds to an inaccurate unitary implementation, i.e., G=⊗U* with U≠U₀. An incoherent/dissipative implementation error corresponds to a slightly non-unitary implementation, due to un-wanted interactions between the QPU and the rest of the universe.

FIGS. 3A-3D illustrate examples of implementation errors in single- and two-qubit gates. In FIG. 3A an intended X_π/2 rotation gate over-rotates a state 310 into the state 330 (drawn with a coarsely dashed line) instead of the correct state 320 (drawn with a finely dashed line). FIG. 3B illustrates an error 375 in the rotation axis of an X₉₀ gate. The actual rotation axis is 370, and the state 310′ is rotated into the state 365 instead of 360. In FIG. 3C amplitude damping, or T₁-decay, is illustrated. Pure states residing on the Bloch sphere 350 are pushed towards the |0 state, eventually losing any information stored in the qubit. In FIG. 3D dephasing, or T₂-decay, is illustrated. Pure states residing on the Bloch sphere 350′ are pushed towards the z axis, losing the phases θ 160 and ϕ 170, eventually mapping the qubit to a random classical bit.

Commonly, quantum processors admit a unique native initial state, ideally given by |ρ₀=|0^⊗n, where |0 corresponds to the state |00|, or simply |0, on each qubit. |ρ₀ may also be written as |0, where 0 is now understood as an n-bit, rather than a single-bit, number. Similarly, E_i═=i| may correspond to a parallel measurement of all qubits in the 0-1 basis, each resulting in a binary outcome. These n bits may then be collected into an n-bit number i∈{0, . . . , d−1}. SPAM implementation errors can be understood as unwanted gates acting on the SPAM, e.g., |ρ=G_ρ═ρ₀.

Given the basic quantum logic operations in the gate set, each quantum circuit which can be implemented on the QPU corresponds to the application of a particular sequence of gates, labeled by j=(j₁, . . . , j_t) to the initial state, and a subsequent measurement which outputs one of the possible outcomes i=0, . . . , d−1. The symbol t may denote the number of gates in a sequence. The probability for the i^th outcome for the sequence j in the gate set is given by

p _i =p _i,j()=E_i|G_jt . . . G_j1|ρ,  (3)

By repeating the same circuit N times, or shots, the frequencies f_i=N_i/N may be collected, where N_i is the number of times the outcome i was obtained. The frequencies are multinomial random variables, with means p_i, and variances p_i(1−p_i)/N, and therefore correspond to experimentally available approximations for the probabilities p_i, with statistical errors ˜1/√{square root over (N)}. The ability to send quantum circuits and shot numbers to the QPU and receive the resulting measurement outcomes may be referred to as gate-level access to the QPU. This is the standard level of access provided to users by hardware manufacturers. A more detailed level of access may be referred to as ‘pulse-level’, and corresponds to users which can define and manipulate the time-dependent laser pulses generating gates. Such users may be interested in a direct characterization of pulses, rather than of the corresponding gates, and the method presented in the present disclosure is applicable to such pulse-level characterization protocols.

Finally, an important implication of equation (3) is that outcome probabilities are invariant under basis changes A∈GL(d²),

E _i | E _i ′|= E _i |A  (4)

G _j G _j ′=A ⁻¹ G _j A

|ρ|ρ=A⁻¹|ρ

which are therefore referred to as “gauge transformations”. It follows that gate sets related by gauge transformations, ˜′, are physically indistinguishable, and a given QPU is actually described, at gate-level, by an equivalence class of such gate sets. A similar gauge-equivalence occurs also in a pulse-level description of QPUs.

Gate Set Tomography (GST)

Gate set tomography (GST) is a well-known protocol for the characterization of quantum logic operations. It is described herein as an example of a detailed characterization protocol. Its main drawback is the exponential scaling of required resources with the number of qubits, currently presenting difficulty in applying it to quantum processors numbering more than a few qubits.

The GST protocol is reviewed hereinbelow as a basis for the later description of embodiments of the present disclosure.

The goal of GST is to characterize implementation errors, given gate-level access to the QPU. This goal may be achieved in a few steps:

• • 1. Model specification: a parameterization of the gate set θ(θ), is chosen by the user. This choice may be based on physical considerations, for example so as to lead to a small number of parameters providing a useful description of the device.
  • 2. Sequence selection: A set of quantum circuits {j} is chosen, such that, collectively, the outcomes probabilities p_i,j(θ)=p_i,j((θ)) of all sequences are highly sensitive to the model parameters θ. The sequence selection is further explained below.
  • 3. Data acquisition: The chosen sequences {j} are inputted to the QPU, along with a chosen number N of shots per sequence, and the resulting frequencies f_i,j may be collected.
  • 4. Model fitting: The parameterized gate set (θ) may be fitted to the obtained frequencies, by maximizing the likelihood of p_i,j(θ) given the data f_i,j. This produces a maximum likelihood estimate θ_ML.
  • 5. Gauge optimization: The estimate θ_ML represents a gauge orbit of parameters, corresponding to gauge-equivalent gate sets. A particular representative θ_est may be chosen by minimizing the distance to the ideal gate set over the gauge orbit. The distance function is usually the Frobenius distance, weighted and summed over all gate set elements.
  • 6. Error bars: An estimate for the statistical error in θ_est, due to statistical errors in the data f_i,j, may be computed. The Applicant has found that this may be advantageously done using a Jacobian J=∂p/∂θ evaluated at θ_est, where p runs over all parameterized GST probabilities, labeled by j and the measurement outcome i, and θ runs over model parameters.

Characterization protocols, and GST in particular, require both quantum and classical resources. A practical way to quantify the quantum resources is to measure the QPU time, which is the amount of time the data acquisition step takes. This is generally a device-dependent function T_QPU(N, {j}) of the number of shots per sequence, and the set of characterization sequences. As an example, on superconducting-qubit devices, the length t=t_j of each sequence may not be a substantial factor influencing the QPU time because gate times are short i.e., of the order of about 10 ns. However, the number of distinct sequences N_seq=|{j}| may be a substantial factor influencing the QPU time. First, the number of distinct sequences controls the total number of shots N_seq×N which is significant because reset times may be long, i.e., in the order of about 100 μs. Second, T_QPU can depend directly on N_seq, depending on how sequences are compiled into pulses acting on qubits.

Herein more details are provided on the sequence selection (step 2 of the GST protocol), which are relevant for the subsequent description of embodiments in the present disclosure.

In order to achieve a high sensitivity to model parameters, GST uses mostly-periodic long gate sequences, of the form Fg^qF′, with logarithmically-space powers q between 1 and q_max>>1, e.g. q∈{2^k}_k=1^log2(qmax). Here F, F′, g are short gate sequences, e.g., g=G_jl . . . G_j1. The sequence time t is therefore t˜ql, and the maximal power q_max may be chosen such that the corresponding time is t_max˜1/ϵ_max, where ϵ_max is an estimate for the largest implementation error expected on the QPU. The period g is referred to as a germ. The germ may be configured to amplify errors in the parameters of the gates from which said germ is comprised. Error amplification is illustrated in FIG. 4. An X_π/2 gate has an over-rotation error 440, and rotates the state 410 to the state 420. If the rotation is performed 5 times 450a-450e the state 430 (drawn with the coarsest dashed line) is obtained, and the accumulated error 445 is 5 times larger.

The parameter variations δθ that may be amplified are those for which δg=(∂g/∂θ)∂θ commutes with the germ, [g, δg]=0. This defines a linear space of amplified variations for each germ, amp(g). A first goal of a sequence selection algorithm may be to choose enough germs g_α such that the space of all amplified variations Σ_αamp(g_α) has the maximal rank.

The initial and final circuits F, F′ are referred to as fiducial circuits and may be configured such as to prepare a set of initial states and measurements, starting from the natively available |ρ and E_i|. A second goal of a sequence selection algorithm may be to choose, for each germ, a set of fiducial circuits which, collectively, have a large overlap with all amplified variations.

Both goals of sequence selection may be obtained by maximizing the Jacobian J=∂p/∂θ, evaluated at the ideal point in parameter space θ₀, defined such that (θ₀)=₀. This is done by minimizing the (properly normalized and regularized) score function Tr(J^TJ)⁻¹), which approximates the statistical estimation error (measured by a Frobenius distance) of a GST protocol using the sequences from which J is constructed. In this manner, the sequence selection algorithm chooses sequences which minimize the error bars of estimated model parameters. At the end of a successful sequence selection, each amplified parameter will have a statistical error δθ˜1/(t_max√{square root over (N)})«ϵ_max, so the largest amplified errors on the QPU can be characterized with a small relative error.

Neighboring Gate Sequences

As described in the Background section, detailed characterization methods known in the art, such as GST, require computational resources which quickly become infeasible as the number of qubits grows. A standard approach to dealing with this problem is to segment the plurality of qubits into “patches”—small subsets of qubits—and characterize each patch by itself, while ignoring the remaining qubits.

Naive segmentation is problematic due to crosstalk—unwanted interactions—between qubits which are in the patch and qubits of the QPU which are not in the patch, referred to as ‘environment qubits’. As demonstrated below, patch-environment interactions lead to an inaccurate characterization of implementation errors in the patch, and may be the limiting factor in the performance of single-patch characterization protocols, and multi-patch protocols constructed from these. The characterization error in single-patch protocols introduced by patch-environment interactions may be referred to as ‘patching error’.

The present disclosure addresses the above problem by adding ‘neighboring gate sequences’ in parallel to the characterization sequences (such as GST sequences described above), said neighboring gate sequences being applied to a subset of environment qubits, referred to as ‘neighboring qubits’, or as the ‘patch neighborhood’. The neighboring qubits may comprise qubits outside the patch (i.e., environment qubits) which are interacting with the qubit patch for example as defined by an interaction graph, or more generally, a hypergraph (which may also be referred to as a coupling map). The vertices of the interaction hypergraph may represent the qubits in the QPU, while the hyperedges may correspond to subsets of multiple qubits suspected to be interacting via multi-qubit implementation errors, or crosstalk. In some embodiments, the interaction hypergraph may be specified by a user. In some embodiments, the hypergraph may be defined such that hyperedges correspond to subsets of multiple qubits on which native multi-qubit gates exist in the QPU. The qubits in such subsets must be physically coupled in order to allow for the application of said native gates, and it is therefore natural to expect that these qubits will also suffer significant crosstalk. While characterization sequences are configured to produce measurement outcomes which are highly sensitive to implementation errors in the characterized patch, the neighboring gate sequences are configured to reduce the sensitivity of measurement outcomes to implementation errors involving the environment qubits, and more specifically, the neighboring qubits. Such a manipulation of the neighboring qubits results in reduced patching errors, enabling a reliable characterization of individual patches, and as a result, of full quantum processors.

FIG. 5A illustrates an interaction graph in a quantum processor 500 including a plurality of qubits. The quantum processor 500 has eight qubits, labeled Q1 to Q8. The interaction graph shows qubits that are suspected to be interacting via 2-qubit implementation errors as connected with a thin line. A patch 510 may be defined and comprise two qubits Q1 and Q2. The qubits outside of the patch may be referred to as environment qubits. The environment qubits interacting with the patch qubits may be referred to as neighboring qubits. In FIG. 5A, the patch 510 is encircled with a thick solid line. The patch qubits directly interact with qubits Q3, Q4, Q5, and Q6. These four qubits constitute the neighboring qubits 520 in an embodiment in which the neighboring qubits consist of the qubits outside the patch directly coupled with qubits in the patch. The neighboring qubits are encircled with a thick dashed line. Qubits Q7 and Q8 are not neighboring qubits. Based on the interaction graph, these environment qubits do not directly interact with the patch qubits Q1, Q2, only indirectly through the neighboring qubits. More generally, interactions may involve not only pairs of qubits, but also triplets, quadruplets, or more generally subsets of multiple qubits, in which case the interaction graph used to define the neighboring qubits is replaced by a hypergraph.

FIG. 5B illustrates a different interaction graph 500′ including a plurality of qubits. Each qubit is represented by a circle, and interacting qubits are connected via a line. In this embodiment, interactions are assumed only between geometrical nearest neighbors, i.e., a qubit may interact only with the qubits directly to the left, right, above or below. Different embodiments may model the interactions differently, e.g., between geometrical next-nearest neighbors. A patch 510′ consists of the qubits marked with a checker-board pattern, such as qubit 512′, with a total of four qubits. The neighboring qubits are marked with diagonal stripes pattern, such as qubit 520′, with a total of eight qubits. Four qubits, such as qubit 530′, are outside both the patch and the neighboring qubits and are marked plain white.

FIG. 6 is a flowchart illustrating steps of a computer implemented method for characterizing implementation errors in a set of quantum logic operations of a quantum processor according to embodiments of the present disclosure. As explained above, the quantum processor may include a plurality of qubits and a gate set. The gate set may include initialization and measurement operations. The gate set may be configured to operate on single qubits or pairs of qubits. The method may include defining a subset of qubits also referred to as a qubit patch. The gate set may act on said qubit patch and implementation errors involving the qubit patch may be modeled by a model including a plurality of model parameters.

The method may include a step 610 of applying a set of quantum circuits on the quantum processor. Applying said set of quantum circuits may include applying characterization gate sequences 613 to said qubit patch and applying neighboring gate sequences 617 to neighboring qubits. The characterization gate sequence may for example be a GST sequence. The neighboring qubits—as explained hereinabove with reference to FIG. 5—are qubits which are not in the patch but are assumed to be interacting with qubits in the patch. The neighboring gate sequences are configured to reduce the sensitivity of measurement outcomes to implementation errors involving environment qubits outside the qubit patch.

The method may further include a step of measuring 620 said qubits. The measuring of said qubits may be performed using a measurement apparatus of said quantum processor. In some embodiments, the measuring may include measuring some or all of the qubits of the quantum processor. In some embodiments, the measuring may include measuring the patch qubits.

The method may further include a step of repeating 630 steps 610-620 to collect a set of frequencies associated with a measurement outcome and a quantum circuit. In some embodiments, measuring said qubits may be performed on patch qubits and environment qubits. The method may then may further comprise computing reduced frequencies from the set of all frequencies by summing over frequencies corresponding to outcomes that differ only in the value of environment qubits. In some other embodiments, measurement of the patch qubits may be performed separately from the environment qubits, and the step of computing reduced frequencies may not be required.

The method may further include a step of fitting 640 the model parameters to said set of frequencies. The fitting principle may be similar to the fitting principles used for the GST protocol as explained hereinbelow.

In some embodiments, the method may be successively applied to a series of different qubit patches. This may provide a full characterization of the quantum processor.

In some embodiments, the method may be repeatedly applied to a same qubit patch, and fitted parameters obtained in each repetition may be used to construct both neighboring gate sequences and characterization gate sequences for the next repetition.

This may provide a characterization of the patch over time.

In some embodiments, the neighboring gate sequence may be configured to reduce a patching error estimated by a perturbative expansion.

In some embodiments, the neighboring gate sequence may be configured to cancel or reduce the leading orders in said perturbative expansion.

Patching Errors: Mathematical Definition

Hereinbelow, a detailed example, including a precise mathematical definition for patching errors is provided. Within this example, a perturbative computation of patching errors is subsequently performed, followed by examples of neighboring gate sequences motivated by the perturbative expansion, as well as numerical simulations demonstrating the performance of said neighboring gate sequences in realistic scenarios.

An n-qubit QPU, with native gate set _QPU={G_j} including 1- and 2-qubit gates is provided. It is assumed that each gate may be represented as G_j=e^hj+Lj where h_j is the ideal Hamiltonian generator of the gate, and L_j is an ‘error generator’ , which is mathematically a ‘Lindbladian’, accounting for both coherent and dissipative implementation errors occurring during the ideal gate operation. Implantation errors occurring before or after the ideal gate may easily be incorporated. More generally, time-dependent generators h_j+L_j may be considered, and may additionally be viewed as the basic objects to be characterized, in place of the corresponding gates G_j. This leads to the pulse-level characterization protocols mentioned above. The description of implementation errors in terms of error generators leads to a physically motivated definition of locality. For example, it may be assumed that interactions only exist between pairs of qubits on which native 2-qubit gates exist in _QPU, i.e., each L_j is 2-local with respect to the interaction graph defined by native 2-qubit gates. This means that L_j=Σ_x,yL_j,x,y is a sum of up to 2-qubit terms, accounting for implementation errors on each qubit x=y, and for unwanted interactions, or cross-talk, between each pair x≠y on which a native 2-qubit gate exists. The usage of error generators also leads to a physically motivated separation of coherent and dissipative implementation errors. For example, we may further assume that dissipative implementation errors affect the qubits individually, so L_j,x,y=H_j,x,y is a Hamiltonian super-operator for each pair x≠y. In different embodiments, these assumptions might not be assumed.

An artificial separation of the n qubits into n_patch qubits, to be considered as the ‘patch’, and the n_env=n−n_patch complementary qubits, to be considered as the ‘environment’ may be made. In terms of error generators, this corresponds to a splitting

L _j =L _j,patch +L _j,env +H _j,int

into patch, environment, and (Hamiltonian) interaction parts,

$L_{j,\mathrm{patch}}=\sum _{x,y\in \mathrm{patch}}{L}_{j,x,y},L_{j,\mathrm{env}}=\sum _{x,y\in \mathrm{env}}{L}_{j,x,y},H_{j,\mathrm{int}}=\sum _{\begin{array}{c}x\in \mathrm{patch}\\ y\in \mathrm{env}\end{array}}{H}_{j,x,y}.$

Note that an operator, or super-operator, A is said to ‘act only on the patch’ if it can be written as A=A_patch⊗I_env, where I_env is the identity operator, or super-operator, on the environment. In our notation we use A_patch to denote both the full A and its restriction to the patch, and the precise meaning should be clear from context. An analogous statement holds for operators that ‘act only on the environment’, of the form A=I_patch⊗A_env.

The goal of a characterization protocol applied to a single patch (‘single-patch protocol’) may be to characterize implementation errors L_j,patch involving only the patch qubits, and occurring in a subset of gates {G_j}_j∈jpatch∪_QPU which are configured to ideally act in the patch, i.e. h_j=h_j,patch. In other words, the goal of a single-patch characterization protocol may be to characterize the ‘patch gate set’

_patch ={G _j,patch =e ^{h j +L j,patch} }_j∈Jpatch.

In order to achieve this goal, a single-patch protocol will send quantum circuits j=j₁, . . . , j_t), with j_k∈J_patch, to the QPU, and receive N random samples per-sequence from the ‘reduced probabilities’

p _i,j = E _i,patch | I _env |G _{j t} . . . G _{j 2} G _{j 1} |ρ_env|ρ_patch.  (5)

These correspond to the preparation of an ideally tensor-product initial state |ρ_env|ρ_patch, the application of the chosen quantum circuit, and finally the ideal partial trace operation on the environment I_env|, and a measurement of the patch qubits E_i,patch|. Note that we implicitly account for SPAM implementation errors as the first and last gates in each gate sequence.

A single-patch characterization protocol does not account for implementation errors involving qubits out of the patch, and in particular, assumes H_int=0. Under this assumption, the action of each gate G_j with j∈J_patch becomes

G _j =e ^{h j +L j,patch +L j,env} =e ^{h j +L j,patch} ⊗e ^{L j,env} =G _j,patch ⊗e ^{L j,env} ,

and the reduced probabilities simplify to

p _i,j ⁽⁰⁾ = E _i,patch |G _{j t ,patch} . . . G _{j 1 ,patch}|ρ_patchƒ,  (6)

as would be the case if the environment qubits did not exist. Since the single-patch protocol assumes it receives samples from p_i,j⁽⁰⁾, but instead receives samples from p_i,j, it will output an incorrect estimate for implementation errors in the patch, {L_j,patch}_{j∈Jpatch′} or equivalently, for the gate set _patch. We define the ‘probabilities patching error’ as

δp=p−p⁽⁰⁾

The corresponding systematic (i.e., infinite statistics, N=∞) deviation in the estimate of L_j,patch, or alternatively _patch, may be defined the patching error itself.

As an example for a single-patch characterization protocol, consider GST applied to a single patch. A parameterization, or model,

θ_patch(θ)={G_j,patch(θ)}_j∈Jpatch

is first chosen. This parameterization may be of the error generator form G_j,patch(θ)=e^{hj+Lj,patch(θ)}, and may be assumed to be ‘large enough’, in the sense that there exists a point θ_real∈{θ} in parameter space such that L_j,patch(θ_real)=L_j,patch. Once GST sequences {j} for the model _patch(θ) are selected, they are fed to the QPU, and the resulting ‘reduced frequencies’ f_i,j, where i∈{1, . . . , d_sys=2^nsys}, are collected. These may be obtained from the full frequencies by summing over the d_env=2^nenv measurement outcomes of the environment qubits. The reduced frequencies are random samples from the reduced probabilities p_i,j discussed above, with mean [f_i,j]=p_i,j, and variance [f_i,j]=p_i,j(1−p_i,j)/N . Equation (6) and the assumption of a ‘large enough’ parameterization imply that the N=∞ and H_int=0 GST estimate θ_est⁽⁰⁾ satisfies

θ_est⁽⁰⁾=θ_real,

or, more accurately, that the left- and right-hand sides of this equation lead to gauge equivalent gate sets, _patch(θ_est⁽⁰⁾)˜_patch(θ_real). For H_int≠0, the probabilities patching error δp_i,j=[f_i,j]−p_i,j⁽⁰⁾ leads to a systematic (N=∞) estimation error

δθ_est=θ_est−θ_real,

which may be referred to as a ‘parameter patching error’. A quantitative relation between δθ_est and δp is given by

δθ_est=J⁻¹δp+0(δp²),  (7)

where J=∂p/∂θ is the Jacobian of parameterized probabilities, computed at θ_est, and J⁻¹ is its suitably defined pseudo-inverse, already mentioned above.

Layer Characterization, Context Gates and Context Dependence

Above we considered a subset of gates {G_j}_j∈Jpatch∪_QPU configured to ideally act in the patch, h_j=h_j,patch. More generally, gates which (ideally) act separably on the patch and environment, for which h_j=h_j,patch+h_j,env, may be considered. In this case the gate G_j is ideally a tensor product e^hj,patch⊗e^hj,env with a possibly non-idle operation on the environment qubits. This includes the case where a gate G_j corresponds to a ‘circuit-layer’—it is ideally a tensor product of few-qubit gates acting on non-overlapping subsets of qubits, which may cover some or all of the qubits in the QPU. The characterization of circuit-layers is useful because these frequently appear in quantum circuits implementing quantum algorithms Generally speaking, few-qubit gate operations in quantum algorithms are performed in parallel as much as possible, in order to reduce the overall effect of dissipative processes such as dephasing or amplitude damping.

The above generalization suggests a possible refinement of the goal of a single-patch characterization protocol: characterize the patch gate set _patch acting on the patch qubits, in the ‘context gate set’

_context ={G _j,context =e ^{h j,env} }_{j∈Jpatch′}

acting on the environment qubits. That is, the gate G_j,context should in principle be applied to the environment qubits whenever the gate G_j,patch is applied to the patch qubits. A gate sequence that consists of context gates may be referred to as context gate sequence. Note that the above refinement is relevant even if all context gates are idle gates h_j,env=0, and that we omit L_j,env from G_j,context to stress that these implementation errors are not modeled or characterized by a single-patch protocol.

In order to quantify the extent to which applying context gates in parallel to corresponding patch gates is important, the notion of ‘context dependence’ may be defined. Consider two different patch gates G_j1,patch, G_j2,patch, with the same ideal patch generator h_j1,patch=h_j2,patch, but possibly different ideal environment generators, or context gates G_j1,context≠G_j2,context. The context dependence between G_j1,patch and G_j2,patch may be defined as the difference L_j1,patch−L_j2,patch. If L_j1,patch=L_j2,patch, the context gates G_j1,context, G_j2,context are equivalent for the purpose of single-patch characterization, and may be freely interchanged.

Context dependence leads to a tension between reducing patching errors, and minimizing the time spent in incorrect contexts. This translates to a tension between neighboring gates sequences and context gate sequences, both acting on neighboring qubits during the application of characterization gate sequences. Generally speaking, strong context dependence favors ‘dilute’ neighboring gate sequences, which suppress patching errors efficiently with only few gates.

Below we demonstrate a perturbative analysis of patching errors. For simplicity, the analysis is carried out assuming all patch gates are meant to act in the idle context, and while ignoring context dependence. We then demonstrate in simulations that neighboring gate sequences motivated by the perturbative analysis can reduce patching errors efficiently even in context dependent scenarios.

Patching Errors: Perturbative Computation

A perturbative computation of the probabilities patching error δp_i,j, subsequently used to construct examples of neighboring gate sequences is detailed hereinbelow. The perturbative expansion may require that L_patch, L_env, H_int are bounded, in a suitable operator-norm, by ϵ_patch, ϵ_env, ϵ_int≤ϵ_max«1. It may further be required that tϵ_max«1, where t=t_j is the length of the sequence j.

For each sequence, it is convenient to express the perturbative expansion of δp_i,j using an ‘effective error generator’ _j^eff, which is a super-operator on the patch such that

δp_i,j=E_i,patch|G_jt,patch . . . G_j1,patchj^eff|ρ_patch.

The effective error generator is conveniently obtained by first expanding I_env|G_jt . . . G_j2 G_j1|ρ_env of Eq. (5) in H_int,

_j ^eff=_j,1^eff+_j,2^eff+O(t³ϵ_int³),

where _j,m^eff=O(t^mϵ_int^m). We then further expand _j,m^eff in L_patch and L_env,

$\begin{array}{cc}{ℒ}_{j,1}^{\mathrm{eff}}={\int }_uI_{\mathrm{env}}❘H_{\mathrm{int}}^{\prime }\left(u\right)❘{\rho }_{\mathrm{env}}+{\int }_{v<u}\left[I_{\mathrm{env}}❘H_{\mathrm{int}}^{\prime }\left(u\right)❘{\rho }_{\mathrm{env}},L_{\mathrm{patch}}\right]+{\int }_{v<u}I_{\mathrm{env}}❘H_{\mathrm{int}}^{\prime }\left(u\right)L_{\mathrm{env}}❘{\rho }_{\mathrm{env}}+O\left(t^3{ϵ}_{\max }^3\right),& \left(8\right)\end{array}$ $\begin{array}{cc}{ℒ}_{j,2}^{\mathrm{eff}}=\frac{1}{2}I_{\mathrm{env}}❘{\left({\int }_u{H}_{\mathrm{int}}^{\prime }\left(u\right)\right)}^2❘{\rho }_{\mathrm{env}}+\frac{1}{2}{\int }_{v<u}I_{\mathrm{env}}❘\left[H_{\mathrm{int}}^{\prime }\left(u\right),H_{\mathrm{int}}^{\prime }\left(v\right)\right]❘{\rho }_{\mathrm{env}}+O\left(t^3{ϵ}_{\max }^3\right).& \left(9\right)\end{array}$

The first line in _j,1^eff corresponds to the first and leading order in the perturbative expansion, while the other explicit parts of _j,1^eff and _j,2^eff makeup the second order. The integral ∫_u may refer to ∫₀^tdu and ∫_v<u may refer to ∫₀^t du ∫_o^u dv. We also defined the piecewise constant H_int(u)=H_jk,int for u∈[k−1, k] and k∈{1, . . . , t}, and similarly for L_patch(u), L_env(u) and h(u). Primes denote the ‘interaction picture’ A′(u)=U_j⁻¹(u)A(u)U_j(u), where U_j(u)=e^{(u−k+1)hjk . . .} e^hj1, for u∈[k−1, k], is the ideal time evolution. Equations (8)-(9) provide a complete second-order perturbative description of the probabilities patching error δp, and form an analytical basis for the construction of neighboring gate sequences in some embodiments.

Initial States, Neighborhood-Mixed States and Initial Twirl

The Applicant has found that for particular initial states of the environment qubits ρ_env, the expression «I_env|H′_int(u)|ρ_env» appearing in the first two lines of Eq. (9) vanishes. For such states, Equations (8)-(10) reduce to

$\begin{array}{cc}{ℒ}_j^{\mathrm{eff}}={\int }_{v<u}I_{\mathrm{env}}❘H_{\mathrm{int}}^{\prime }\left(u\right)L_{\mathrm{env}}\left(v\right)❘{\rho }_{\mathrm{env}}+\frac{1}{2}I_{\mathrm{env}}❘{\left({\int }_u{H}_{\mathrm{int}}^{\prime }\left(u\right)\right)}^2❘{\rho }_{\mathrm{env}}+\frac{1}{2}{\int }_{v<u}I_{\mathrm{env}}❘\left[H_{\mathrm{int}}^{\prime }\left(u\right),H_{\mathrm{int}}^{\prime }\left(v\right)\right]{\rho }_{\mathrm{env}}+O\left(t^3{ϵ}_{\max }^3\right).& \left(11\right)\end{array}$

As a first example, assume H_int is a sum of terms H_x,y acting as ρ_x⊗ρ_yi[h_x⊗Z_y, ρ_x⊗ρ_y], where h_x is a Hamiltonian operator on a patch qubit x, and Z_y is the Pauli-Z operator in an environment qubit y. Here ρ_x, ρ_y are density matrices on qubits x, y. Under the above assumption, the expression «I_env|H_int′(u)|ρ_env» vanishes if ρ_env is such that its reduced density matrix ρ_y on each neighboring qubit y (as defined above) is given by |+=|++|, where |+=|+X=(|0+|1)/√{square root over (2)}. As an example, the state

|ρ_env=|+^⊗nneigh⊗|0^{(nenv-nneigh)},

where n_neigh is the number of neighboring qubits, satisfies this condition. The above state can be efficiently prepared from the native state |ρ_env=|0^⊗nenv by applying a Y_π/2 rotation, or a Hadamard gate, to each neighboring qubit.

More generally, the Applicant has found that if the initial state of the environment qubits ρ_env is such that its reduced density matrix ρ_y on each neighboring qubit y is maximally mixed, i.e., ρ_x=I_b/2, the expression «I_env|H_int′(u)|ρ_env» appearing in the first two lines of Eq. (9) vanishes, for any H_int. Such a state may be referred to as ‘neighborhood-mixed’, and denoted ρ_env^(NM). Relevant examples for neighborhood-mixed states are

$\begin{array}{c}{{❘{\rho }_{\mathrm{env}}^{\left(\mathrm{NM}\right)}=(❘0}^{{\otimes }_{n_{\mathrm{env}}}}+❘1}^{{\otimes }_{n_{\mathrm{env}}}})/2\\ {{=(❘0\rangle \langle 0❘}^{{\otimes }_{n_{\mathrm{env}}}}+❘1\rangle \langle 1❘}^{{\otimes }_{n_{\mathrm{env}}}})/2\end{array},$

and the closely related

|ρ_env^(NM)=(|0^⊗nneigh+1^⊗nneigh)/2⊗|0^{⊗(nenv-nneigh)}.  (12)

The above states are mixed but ‘separable’, or ‘un-entangled’, and can therefore be prepared efficiently, as described below. It should be noted that a neighborhood-mixed state can be pure, but must then be entangled, requiring a non-trivial quantum circuit to prepare. An example for such a neighborhood-mixed state is the “GHZ state” |GHZ»=|GHZGHZ|, where |GHZ=(|0^⊗nenv+|1^⊗nenv)/√{square root over (2)}. Compared to the state above Eq. (12), the classical probabilistic mixture is now replaced by a quantum superposition.

In some embodiments, a patching error may be estimated using a perturbative expansion and the neighboring gate sequences may comprise initializing the environment qubits in a state configured to cancel or at least reduces leading orders in said perturbative expansion.

In some embodiments, the neighboring gate sequences may comprise initializing the environment qubits in a neighborhood-mixed state.

In some embodiments, the neighboring gate sequences may comprise initializing (only) the neighborhood qubits in a neighborhood-mixed state.

In some embodiments, the definition of a neighborhood-mixed state does not include the maximally mixed state (i.e. the state where the density operator of all environment qubits is proportional the unit-operator).

Based on the interpretation of mixed states as probability distributions over pure states, the Applicant has found that the preparation of some neighborhood-mixed states can be obtained by including in the neighboring gate sequences an ‘initial twirl’. The latter corresponds to assigning to a given characterization sequence a number of distinct neighboring gate sequences, and averaging over the corresponding frequencies. These distinct neighboring sequences may differ in their initial gate, such that the averaging corresponds to the preparation of a mixed (and neighborhood-mixed) initial state. In other words, the set of quantum circuits may include (or may consist of a series of) quantum circuit clusters wherein each quantum circuit cluster comprises a plurality of quantum circuits having a same characterization gate sequence and a distinct neighboring sequence. In these embodiments, the method may include combining (e.g. averaging) the frequencies collected on quantum circuits of the same cluster. The distinct neighboring sequences may differ in their initial gate, such that the combining (e.g. averaging) corresponds to the preparation of a mixed (and neighborhood-mixed) initial state. The initial twirl thus eliminates the first-order in the perturbative expansion of the patching error, while requiring no complex initialization procedures.

As an example, the initial twirl corresponding to the state in Eq. (12) may be such that, for each characterization sequence, two distinct neighboring sequences are assigned, the first beginning with the idle gate and the second with an X_n gate, both applied in parallel to all neighboring qubits, and a subsequent averaging of the two distinct frequencies for each characterization sequence. A characterization method including the above initial twirl may include for example the following steps:

• • 1. Providing an input comprising:
    • a. a set {j′} of characterization gate sequences (such as GST gate sequences) that may (ideally) act within a patch; and
    • b. a number of shots N for repeating the applying and measuring steps (a)-(b) of the characterization method according to the present disclosure and in particular as presented in the appended claims;
  • 2. For each characterization gate sequence j′: modifying the the first (ideal) gate in the sequence to obtain a new sequence j″. The modification may correspond to the replacement of the first gate G_j1 in the sequence j′ with G_j1⊗X_π^neighboring, where X_π^neighboring is the tensor product of X_π on all neighboing qubits.
  • 3. Applying each j′ and j″ to the QPU, with N/2 shots per sequence.
  • 4. Collecting the corresponding reduced frequencies f_i,j′ and f_i,j″.
  • 5. Computing the average f_i,j=(f_i,j′+f_i,j″)/2 for each measurement outcome i and sequence j.

The QPU time required may be T_QPU(N/2, {j′}∪{j″}). Compared to the QPU time T_QPU(N, {j′}) of the bare sequences, the number of sequences doubled, but the total number of shots is unchanged.

Reiterating, the patching error with initial twirl (or neighborhood-mixing) may be of second-order only −[f_i,j]−p_i,j⁽⁰⁾=O(t_j²ϵ_max²)—and not first order, as may be without initial-twirl/neighborhood-mixing −[f_i,j′]−p_i,j⁽⁰⁾=O(t_jϵ_max). The statistical error (in the frequencies) may not change: [f_i,j]=[f_i,j′]=O(1/N).

As explained below and shown for example in FIG. 7A, the neighboring gate sequences may include an initial twirl followed by additional gates.

FIG. 7A illustrates a quantum circuit applied in a computer implemented method for characterizing implementation errors in a set of quantum logic operations of a quantum processor according to embodiments of the present disclosure. For example, a characterization sequence 740 (such as a GST sequence) may be applied to patch qubits 710 and a neighboring gate sequence 750 may be applied to the neighboring qubits 720. In this embodiment, no gates (i.e. idle gates) may be applied to qubits that are outside the patch qubits 710 and are not part of the neighboring qubits 720. After application of the characterization and neighboring gate sequences 740, 750 the qubits are measured 760.

The characterization sequence 740 applied on the patch qubits 710 may be a GST sequence determined in accordance with the GST protocol. The characterization sequence may comprise an initial fiducial sequence F 742, a final fiducial sequence F′ 746 to prepare the qubits to measurement, and multiple applications of a germ g 744. The neighboring sequence 750 applied on the neighboring qubits 720 may comprise an initial twirl 753 and a dynamical decoupling sequence 756.

As explained above, the neighboring gate sequence 750 may include initializing the environment qubits in a neighborhood-mixed state. In particular, the neighboring sequence 750 may include an initial twirl 753 that may include applying a π rotation along the x-axis for each neighboring qubit only for half of the repetitions of each characterization sequence. For the other half, no rotation may be applied, symbolized as applying identity gates.

In some embodiments, at least some of the dynamical decoupling gates (or sets of gates) 756 are applied synchronously with germs, i.e. a dynamical decoupling (“DD”) operation may begin simultaneously with the beginning of a germ.

In the present disclosure, the term ‘coordination’ may refer to a relative timing in the application of the gates. The timing may be affected by the characteristics of the processor. For example, implementation errors dictate that even simple idling of any qubit may be sensitive to the time spent idling. Examples of relative timing between gates may include simultaneous beginning of one or more gate(s) applied on one or more qubit(s) and of one or more other gate(s) applied on one or more other qubit(s), simultaneous ending of one or more gate(s) applied on one or more qubit(s) and of one or more other gate(s) applied on one or more other qubit(s), or a predetermined delay between one or more gate(s) applied on one or more qubit(s) and one or more other gates applied to the same or one or more other qubits.

FIG. 7B illustrates a computer implemented method for characterizing implementation errors in a set of quantum logic operations of a quantum processor according to embodiments of the present disclosure including a context gate sequence. In embodiments as illustrated on FIG. 7B, the quantum circuit may comprise a characterization gate sequence 740B, a context gate sequence 745 and a neighboring gate sequence 750B. As explained above, gates that ideally act separably on the patch and the environment may be characterized using embodiments of the present disclosure. Such gates may be described as a gate applied to patch qubits synchronously with a corresponding (possibly non-idle) context gate applied to environment qubits. The characterization gate sequence 740B may include characterization gates (or short sequences of such gates, i.e. germs) g 744B that act on the qubits of the patch 710B. The context gate sequence 745 may comprise context gates (or short sequences of such gates) g′ that act synchronously with the characterization gates on the neighboring qubits 720B or more generally on the environment qubits. The combination of characterization gates and context gates may form circuit layers 743. In some embodiments, the neighboring gate sequence 750B may include dynamical decoupling gates DD 751 and an initial twirl 753B applied to the neighboring qubits 720. The neighboring gate sequence 750B and the context gate sequence 745 may be temporally intertwined. In some embodiments, the dynamical decoupling gates 751 may be coordinated with the context gates 745 so as to start (shortly or immediately) after the context gates ends while the context gates may be applied synchronously with their corresponding characterization gates. This implies that idle gates are added to the characterization sequence, shown as spaces between consecutive germs g 744B, synchronously with dynamical decoupling gates DD 751. In some embodiments, these idle gates can be replaced by other gates in order to further reduce patching errors or increase sensitivity to model parameters. Depending on an alleged influence of context dependence on the characterization, some context gates may be omitted and/or replaced by neighboring gates, e.g. dynamical decoupling gates. In some embodiments including a context gate sequence, at least one neighboring gate may be applied synchronously with a corresponding characterization gate.

The quantum circuit may further include an initial fiducial sequence F 742B, a final fiducial sequence F′ 746B, and measurement 760B. The germs 744B and the fiducial sequences 742B 746B may form the characterization gate sequence 740B.

In some embodiments, initial and final fiducial sequences F 742B, F′ 746B may also be performed together with corresponding context gates (not shown) and the initial twirl may be performed before the context gate corresponding to the (initial) fiducial sequence F 742B.

It is noteworthy that a difference in the application of dynamical decoupling, between the embodiments shown in FIG. 7A and FIG. 7B, may arise due to an alleged influence of context dependence. In the embodiment shown in FIG. 7A context dependence may not be considered significant, and the dynamical decoupling may be performed synchronously with the characterization gates. In contrast, in the embodiment shown in FIG. 7B context dependence may be considered significant, and the dynamical decoupling may be performed successively to the characterization gates.

FIG. 7C illustrates a quantum circuit cluster applied in a computer implemented method for characterizing implementation errors in a set of quantum logic operations of a quantum processor according to embodiments of the present disclosure. In this embodiment, clustering of quantum circuits is shown. A quantum circuit cluster may consist of quantum circuit 704 and quantum circuit 706. Quantum circuits 704 and 706 may comprise a same characterization sequence 740C that may be applied to the qubits of the patch 710C and two different neighboring gate sequences 751751′ that may be applied to the neighboring qubits 720C. In some embodiments, the two neighboring gate sequences may have a different initialization while the rest of the neighboring gate sequences may be similar. As shown on FIG. 7C, the two neighboring sequences 751, 751′ may include a different initialization gate of the neighboring qubits. In quantum circuit 704, the neighboring qubits 720C may be initialized with application of 7E rotation along the x-axis X 757. In quantum circuit 706, the initialization of the neighboring qubits 720C may be performed by applying the idle gates I 757′. The differences between two neighboring sequences 751751′ in the illustrated embodiment may further include different dynamical decoupling sequences D 759 and D 759′. The characterization sequence 740C may be the same regardless of the differences in the neighboring sequences 751751′. After each repetition, the qubits may be measured 760C. For each 30 quantum circuit, a predefined number of repetitions N 770 may be performed. From each set of N repetitions, a corresponding frequency is obtained. Frequency f_i,j′ 780 may correspond to the quantum circuit comprising the neighboring sequence 751 and frequency f_i,j″ 780′ may correspond to the quantum circuit comprising the neighboring sequence 751′. The two frequencies 780780′ may be averaged 790. This average may be the frequency f_i,j 795 that is assigned to the characterization sequence 740C.

In some other embodiments, the averaging of frequencies f_i,j′, f_i,j″ may be replaced by an averaging of the corresponding fitted model parameters. That is, the model is fitted once to the frequencies f_i,j′, resulting in the estimate θ_est′ of model parameters, and once to the frequencies f_i,j″, resulting in the estimate θ_est″. These estimates are then averaged, θ_est=(θ_est′+θ_est″)/2.

In some other embodiments, the clustering may include more than two quantum circuits per cluster, wherein each quantum circuit of a given cluster may comprise different neighboring sequences and the same characterization sequence.

In some other embodiments, the averaging may be weighted.

In some other embodiments, the averaging may be replaced by a non-linear function.

Accidental Amplification and De-amplification Sequences

As explained above, for every quantum circuit j the patching error satisfies δp_i,j=O(t_jϵ_max) for a generic initial state ρ_env, but δp_i,j=O(t_j²ϵ_max²) for a neighborhood-mixed initial state ρ_env^(NM). The dependence of these bounds on t_j is generically saturated for mostly-periodic long characterization sequences G_jt . . . G_j1=Fg^qF′, such as GST sequences discussed above. Conceptually, single-patch characterization protocols may include carefully-selected mostly-periodic characterization sequences in order to amplify some of the implementation errors to be characterized, L_patch. The Applicant has found that this mostly-periodic structure also generically leads to an amplification of some of the out-of-model errors H_int, L_env as well. The amplification of H_int, L_env may be referred to as “accidental amplification”. The accidental amplification can be easily shown for single-gate sequences G_jt . . . G_j1=G^t, with effective error generator _j^eff=G_patch^−tI_env|G^t|ρ_env−I_patch. If the environment qubits are not in a neighborhood-mixed state, the leading order integral in the expansion of _j^eff (see Eq. (8)) is given by

${\int }_uI_{\mathrm{env}}❘H_{\mathrm{int}}^{\prime }\left(u\right)❘{\rho }_{\mathrm{env}}=tI_{\mathrm{env}}❘H_{\mathrm{int},0}❘{\rho }_{\mathrm{env}}+O\left({ϵ}_{\mathrm{int}}\right),$

where H_int,0 denotes the projection of H_int onto the commutant of the ideal gate G_ideal=e^h (recall the commutation requirement presented in the overview of GST). Similarly, when the environment qubits are in a neighborhood-mixed state, the leading order integrals in the expansion of _j^eff (see Eq. (11)) are given by:

${\int }_{v<u}I_{\mathrm{env}}❘H_{\mathrm{int}}^{\prime }\left(u\right)L_{\mathrm{env}}❘{\rho }_{\mathrm{env}}^{\left(\mathrm{NM}\right)}=t^2I_{\mathrm{env}}❘H_{\mathrm{int},0}{L}_{\mathrm{env}}❘{\rho }_{\mathrm{env}}^{\left(\mathrm{NM}\right)}+O\left(t{ϵ}_{\mathrm{int}}{ϵ}_{\mathrm{env}}\right),$ $I_{\mathrm{env}}❘{\left({\int }_u{H}_{\mathrm{int}}^{\prime }\left(u\right)\right)}^2❘{\rho }_{\mathrm{env}}^{\left(\mathrm{NM}\right)}+{\int }_{v<u}I_{\mathrm{env}}❘\left[H_{\mathrm{int}}^{\prime }\left(u\right),H_{\mathrm{int}}^{\prime }\left(v\right)\right]❘{\rho }_{\mathrm{env}}^{\left(\mathrm{NM}\right)}=t^2I_{\mathrm{env}}❘H_{\mathrm{int},0}^2❘{\rho }_{\mathrm{env}}^{\left(\mathrm{NM}\right)}+O\left(t{ϵ}_{\mathrm{int}}^2\right).$

As long as the commutant projection H_int,0 does not vanish, which is the generic case, the leading t=t_j scaling in the effective error generator is indeed saturated. This holds also for general mostly-periodic sequences.

As demonstrated above, accidental amplification implies an un-favorable scaling of patching errors with sequence length, in mostly-periodic characterization sequences. The Applicant has found that in order to eliminate the accidental amplification of out-of-model implementation errors, while maintaining the carefully constructed amplification of in-model implementation errors, neighboring gate sequences including de-amplification sequences may be used. In some embodiments, such de-amplification sequences may be non-periodic. In some embodiments, such sequences may be periodic and have a de-amplification period differing from a period of the characterization sequence. In some embodiments, such sequences may be un-structured and/or random, or carefully structured for de-amplification. In particular, structured ‘dynamical decoupling’ sequences may be used. Dynamical decoupling (DD) is well known in the art for other applications. As a basic definition, an m′th order DD sequence applied to a given set of qubits is one which, for ideal and instantaneous DD gates, eliminates coherent errors involving said set of qubits up to (and including) the m′th order in perturbation theory. A standard example for a first order DD sequence is given by the so called XY4 sequence, which may be written as X-Y-X-Y. Applied to a qubit subjected to an unwanted Hamiltonian evolution e^tH=I+O(tH), the XY4 sequence (with ideal and instantaneous gates) modifies the evolution to Y_πe^tH/4X_πe^tH/4Y_πe^tH/4X_πe^tH/4=I+O(t²H²). DD is used, in some embodiments of this disclosure, to eliminate the systematic patching errors that arise in patch-based characterization protocols.

In some embodiments of this disclosure, the gates of the DD neighboring sequences may be coordinated with the gates of the characterization sequence. In some embodiments, the characterization sequence may have a periodic part (e.g. a GST sequence), and the DD sequence may be applied in parallel to the periodic part of the sequences (i.e., the GST germs). The synchronization may be such that each dynamical decoupling gate may be applied exactly with the first gate of the germ that may correspond it. The DD gates may be evenly spread over the germ repetitions. An algorithm for generating a quantum circuit including an almost-periodic characterization gate sequence and a neighboring gate sequence incorporating DD, for use in a characterization method according to the present disclosure, may for example include the following steps:

• • 1. Providing an input comprising:
    • a. a characterization gate sequence j that may (ideally) act within a patch. The sequence may have the structure G_jt . . . G_j1=F g^qF′ where the circuits F, g, F′ may have lengths l_F, l_g, l_F′, such that the total sequence time may be t=t_j=l_F+ql_g+l_F′; and
    • b. a dynamical decoupling sequence k of length m≤g that may correspond to gates D_k1, . . . , D_km∈_QPU that may (ideally) act on the neighboring qubits.
  • 2. For i∈{1, . . . , m}, replace the gate G_{jlF+1+lg(i−1)└q/m┘} (in the sequence j) with the tensor product G_{jF+1+lg(i−1)└q/m┘}⊗D_ki. Denote the resulting sequence j^DD
  • 3. Output j^DD.Note that, in the above algorithm, the dynamical decoupling sequence may have a number of gates equal to or smaller than the number of germs. Further, Step 2 is a combination step of the characterization and DD sequences. The dynamical decoupling gates may be applied on neighboring qubits, so mathematically, this may be represented as a tensor-product of each dynamical decoupling gate with a corresponding gate of the characterization sequence. The expression l_F+1+l_g(i−1)└q/m┘ may be the time step where a dynamical decoupling gate is combined. The terms l_F+1 may express the time step after the initial fiducial circuit. The dynamical decoupling sequences may be spread as evenly as possible over the germs. For example, if there are half as much dynamical decoupling gates than there are germ repetitions, then a dynamical decoupling gate may be combined with every second germ repetition. This may be expressed by the factor └q/m┘. The factor l_g may transform a germ repetition number (in reference to the sequence of germs) to the initial gate of that germ. The factor (i−1) may index the specific dynamical decoupling gate.

Furthermore, the QPU time is unchanged by the additional of DD neighboring sequences, T_QPU(N, {j^DD})=T_QPU(N , {j}), as the DD gates may be applied in parallel to the characterization sequence.

The accidental amplification may thus be eliminated, reducing patching errors as follows. For a first-order (or higher) DD sequence, the patching error may be [f_i,j^DD]−p_i,j⁽⁰⁾=O(ϵ_max). Neighborhood-mixing (or initial twirl) used in conjunction with a second-order) (or higher) DD sequence may result in a patching error [f_i,j^DD+NM]−p_i,j⁽⁰⁾=O(t_jϵ_max²). The statistical error may be unchanged.

FIG. 8 is a flowchart illustrating steps of a method for characterizing implementation errors in a set of quantum logic operations of a quantum processor according to embodiments of the present disclosure. As explained above, the quantum processor may include a plurality of qubits and a gate set. The gate set may be acting on a subset of qubits defining a qubit patch and the implementation errors involving the qubit patch may be modeled by a model including a plurality of model parameters. In a first initialization step 811, qubits of the quantum processor are initialized in an initial state |ρ₀. Further, the method comprises applying a set of quantum circuits to the quantum processor wherein said set of gate sequences includes characterization sequences (such as GST sequences) applied to a qubit patch and neighboring gate sequences applied to neighboring qubits. In particular, an initial fiducial circuit 812 may be applied to the patch, in parallel to an initial twirl 815 being applied to the neighboring qubits. Next, germs 813 may be applied to the patch, in parallel to a dynamical decoupling sequence 816 being applied to the neighboring qubits. Then, a final fiducial circuit 814 may be applied to the patch in parallel to idle gates 817 being applied to the environment qubits. It is noteworthy that the tensor product symbols 819 are to emphasis that branches in the flowchart are parallel processes and not different options. In a measuring step 818, a measurement outcome is determined by measuring the qubits. A measurement outcome counter 821 corresponding to the measurement outcome measured may be incremented. In a further step 822, a repetition test is performed to verify if a predefined number of repetitions N has been achieved. If not, in a reset step 828 the quantum processor is reset and a new repetition of the routine is started again at the initialization step 811. If N iterations were performed, the routine proceeds to a frequency computation step 824 in which the values of the measurement outcome counters may be divided by N 824 to compute the frequencies of measurement outcomes. In a further frequency reduction step 825, some frequencies may be summed in order to aggregate measurement outcomes with same patch qubits measurement and provide to reduced frequencies. In a further step 826, a test is performed to verify if all the characterization gate sequences of said set of quantum circuits were applied to the quantum processor. If not, the next quantum circuit is selected (an implicit counter is incremented) and the quantum processor is reset, and a new set of repetitions starts again at the initialization step 811. If all the quantum circuits of the set of quantum circuits were applied to the quantum processor, the next steps are a fitting step 831 comprising fitting of the model parameters, a gauge optimization step 833 of performing gauge optimization, a statistical error computation step 834 comprising computing the statistical error in the parameters, and finally, a result representation step 835 comprising representing the results for example by displaying said results using a bar graph.

In some embodiments, the characterization method described herein may be repeated using a plurality of different (distinct) patches. i.e. the characterization method may be successively applied to a plurality of subsets of qubits of a quantum processor.

In some embodiments, the plurality of distinct patches may cover all of the qubits of the quantum processor. In some embodiments, the plurality of distinct patches may overlap, i.e. different patches may have at least one qubit in common. Overlapping patches may be desired, for example, in order to collectively characterize all interactions specified by the interaction hypergraph. In some embodiments, some of the patches may contain more than two qubits (e.g., three, four, five or more qubits). In some embodiments, the sequences corresponding to different patches (including both characterization and neighboring gate sequences) may be applied in parallel, in order to reduce the overall QPU time.

While the method set forth hereinabove presumed qubits, the method is more generally applicable to QPUs utilizing quantum systems with M≥2 logical states. Quantum systems with two or more logical states are known as ‘qudits’. The mathematical framework of the presently disclosed characterization method has no dependency on the binary nature of qubits. In qudit embodiments, the neighborhood-mixed state (|0^⊗nenv+|1^⊗nenv)/2 changes into

${\frac{1}{M}{\sum }_m❘m}^{{\otimes }_{n_{\mathrm{env}}}}$

where m=1, . . . , M. Using initial twirl, the N repetitions (shots per sequence) are split to M ‘sub-repetitions’, with initializing in each sub-repetition the environment's qudits to one of each of the M states. For example, in a processor built using three-level systems (‘qutrits’), the neighborhood mixed state is (|0^⊗nenv+|1^⊗nenv+|2^⊗nenv)/3. Using initial twirl, the N repetitions are split to three instead of two.

Note that the disclosed method is based on a logical description of the QPU, and may be applied if qubits are ‘encoded’ or ‘logical’ rather than ‘physical’—i.e. corresponding to general 2-dimensional subspaces in the space of physical states of the QPU. The same statement holds for a QPU described by logical qudits, and is relevant in particular for the application of the disclosed method to QPUs implementing quantum error correction, computation in decoherence free subspaces, and topological quantum computation.

Simulation results

FIG. 9 shows simulation results of applying a characterization method according to some embodiments of the present disclosure. In particular, FIG. 9 provides an example of canceling the first order error in the perturbation expansion of a patching error. A quantum processor 905 consisting of two qubits is simulated. The patch 907 consists of one qubit and the environment 909 consists of the other qubit, which is a neighboring qubit. The graphs 910, 920 show the patching-error δp_i,j as a function of the strength of the interaction between the patch and the environment ϵ_int. Graph 910 shows the patching error without using initial twirl—the neighboring/environment qubit is initialized to zero. Graph 920 shows the patching error when using the initial twirl. The graphs' axes are logarithmic in both the x-axes and the y-axes.

The simulated characterization sequences j correspond to the application of X_π/2^t to the patch qubit, with time t=2, . . . , 9, corresponding to the different curves, and the measurement outcome is i=0. Each ϵ_int corresponds to the standard deviation of a normal distribution from which the interaction Hamiltonian H_int is sampled. Similarly, L_patch=H_patch is sampled, with a fixed ϵ_patch=10⁻³. The y-axis corresponds to the standard deviation of δp_i,j over many samples from these distributions, scaled by t²(t) for t even (odd). (The mean of δp_i,j over randomly-sampled coherent implementation errors vanishes).

Visible in the graphs are eye-guides for linear scaling 931931′ and for quadratic scaling 932932′ with ϵ_int. Without the initial twirl, the patching error of the odd-length sequences, marked with solid circles, scales as tϵ_int 913 and the patching error of the even-length sequences, marked with hollow squares, scales as t²ϵ_int² 914 for ϵ_int»ϵ_patch, and as t²ϵ_intϵ_patch for ϵ_int«ϵ_patch 915. With initial twirl, the tϵ_int and t²ϵ_intϵ_patch contributions vanish, and the patching is reduced significantly. Both with initial twirl and without initial twirl, the patching error saturates at high ϵ_int values 933933′.

FIG. 10 shows simulation results of applying a characterization method according to some embodiments of the present disclosure. In particular, a ratio of the patching error to the statistical error in GST model parameters, δθ_patching/δθ_stat, is illustrated. A quantum processor 1010 consisting of three qubits is simulated, including interactions between the characterized patch and its environment, and context dependence. The patch 1013 consists of two qubits, labeled q0 and q1. The environment 1017 consists of one qubit, labeled q2. The environment qubit q2 interacts only with patch qubit q1. Graphs 1020, 1030, 1040 show the ratio δθ_patching/δθ_stat for coherent parameters (e.g., rotation angles), supported on qubit q0, qubit q1 and q0-q1 interactions, respectively. Graphs 1050, 1060 show the ratio δθ_patching/δθ_stat for incoherent parameters (e.g., dissipation rates) supported on qubit q0 and qubit q1, respectively. The graphs' axes are logarithmic in both the x-axes and the y-axes.

For simplicity, the errors for different parameters are averaged, and this average (y-axis) is plotted as a function of the number of germs in the characterization sequence (x-axis). Only significant implementation errors, with magnitude larger than 10⁻⁴ are included in the average. Every sequence was repeated for N=300 shots. Four datasets are plotted. The first data set corresponds to ‘bare’ characterization sequences, with no added neighboring gate sequences (shown as circle). The remaining data sets correspond to characterization sequences with the following types of neighboring gate sequences: initial twirl (shown as square), initial twirl followed by structured second order dynamical-decoupling sequences (shown as triangle) and initial twirl followed by an equally spaced random sequence (shown as crosses).

A significant reduction of patching error (compared to the statistical error) for the coherent parameters of qubit q1 is achieved as shown in graph 1030, as predicted by the perturbative expansion. In general, it is noteworthy that without initial twirl, for some parameters

$\frac{\delta {\theta }_{\mathrm{patching}}}{\delta {\theta }_{\mathrm{stat}}}1$

for some parameters while with the initial twirl,

$\frac{\delta {\theta }_{\mathrm{patching}}}{\delta {\theta }_{\mathrm{stat}}}\lesssim 1$

for all parameters.

FIG. 11 and the following discussion are intended to provide a brief, general description of an exemplary computing environment in which the disclosed technology may be implemented. Although not required, the disclosed technology is described in the general context of computer executable instructions, such as program modules, being executed by a personal computer (PC). Generally, program modules include routines, programs, objects, components, data structures, etc., that perform particular tasks or implement particular abstract data types. Moreover, the disclosed technology may be implemented with other computer system configurations, including handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, network PCs, minicomputers, mainframe computers, and the like. The disclosed technology may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

With reference to FIG. 11, an exemplary system for implementing the disclosed technology includes a general purpose (classical) computing device in the form of an exemplary conventional PC 1100, including one or more processing units 1110, a system memory 1120, and a system bus 1130 that couples various system components including the system memory 1120 to the one or more processing units 1110. The system bus 1130 may be any of several types of bus structures including a memory bus or memory controller, a peripheral bus, and/or a local bus using any of a variety of bus architectures. The exemplary system memory 1120 includes read only memory (ROM) 1122 and random-access memory (RAM) 1127. A basic input/output system (BIOS) 1125, containing the basic routines that help with the transfer of information between elements within the PC 1100, is stored in ROM 1122. As shown in FIG. 11, the system memory 1120 may store computer-executable instructions for performing any of the disclosed techniques (e.g., sending instructions to quantum computer for applying characterization gate sequences and neighboring gate sequences to a subset of qubits, measuring outcomes, collecting frequencies, computing model parameters) in respective memory portions (shown generally as executable software 1129 for performing any embodiment of the disclosed synthesis techniques)).

The exemplary PC 1200 further includes one or more storage devices 1140, such as a hard disk drive for reading from and writing to a hard disk, a magnetic disk drive for reading from or writing to a removable magnetic disk, and/or an optical disk drive for reading from or writing to a removable optical disk (such as a CD-ROM or other optical media). Such storage devices can be connected to the system bus 1130 by a hard disk drive interface, a magnetic disk drive interface, and/or an optical drive interface, respectively. The drives and their associated computer readable media provide nonvolatile storage of computer-readable instructions, data structures, program modules, and other data for the PC 1200. Other types of computer-readable media which can store data that is accessible by a PC, such as magnetic cassettes, flash memory, digital video disks, CDs, DVDs, RAMs, NVRAMs, ROMs, and the like, may also be used in the exemplary operating environment. As used herein, the terms storage, memory, and computer-readable media may not include or encompass propagating carrier waves or signals per se.

A number of program modules may be stored in the storage devices 1140, including an operating system, one or more application programs, other program modules, and program data. Storage of results of quantum measurements and instructions for obtaining such measurements (and/or instructions for performing any embodiment of the disclosed technology) can be stored in the storage devices 1140. A user may enter commands and information into the PC 1100 through one or more input devices 1150 such as a keyboard and a pointing device such as a mouse. Other input devices may include a digital camera, microphone, joystick, game pad, satellite dish, scanner, or the like. These and other input devices are often connected to the one or more processing units 1110 through a serial port interface that is coupled to the system bus 1130, but may be connected by other interfaces such as a parallel port, game port, or universal serial bus (USB). A monitor 1180 or other type of display device is also connected to the system bus 1130 via an interface, such as a video adapter. Other peripheral output devices 1160, such as speakers and printers (not shown), may be included. In some cases, a user interface is displayed so that a user can input a circuit for synthesis, and verify successful synthesis.

The PC 1200 may operate in a networked environment using logical connections to one or more remote computers, such as a remote computer 1190. In some examples, one or more network or communication connections 1170 are included. The remote computer 1190 may be another PC, a server, a router, a network PC, or a peer device or other common network node, and typically includes many or all of the elements described above relative to the PC 1100, although only a memory storage device 1195 has been illustrated in FIG. 11. The personal computer 1100 and/or the remote computer 1190 can be connected to a local area network (LAN) and a wide area network (WAN). Such networking environments are commonplace in offices, enterprise wide computer networks, intranets, and the Internet.

When used in a LAN networking environment, the PC 1100 is connected to the LAN through a network interface. When used in a WAN networking environment, the PC 1100 typically includes a modem or other means for establishing communications over the WAN, such as the Internet. In a networked environment, program modules depicted relative to the personal computer 1100, or portions thereof, may be stored in the remote memory storage device or other locations on the LAN or WAN. The network connections shown are exemplary, and other means of establishing a communications link between the computers may be used.

With reference to FIG. 12, an exemplary system for implementing the disclosed technology includes computing environment 1200, The environment includes one or more quantum processing unit(s) 1210 including one or more monitoring/measuring device(s). The quantum processing unit(s) execute quantum circuits that are provided by a classical processing unit 1220. The quantum circuits are downloaded into or used to program or configure the quantum processing unit(s) 1210 (e.g., via control lines (quantum bus) 1270). Procedures according to any of the disclosed embodiments (e.g. a high-level description of the set of quantum circuits to be applied to a qubit patch and neighboring qubits) may be stored in a memory 1230.

With reference to FIG. 12, the high-level description of a quantum software may be translated into quantum circuits (e.g., sequences of quantum gates, or layers of gates acting in parallel on different qubits). Such high-level descriptions may be stored, as the case may be, on one or more external computers 1260 outside the computing environment 1200 utilizing one or more memory and/or storage device(s) 1265, then downloaded as necessary into the computing environment 1200 via one or more communication connection(s) 1240. Quantum circuits (according to any of the disclosed embodiments) are coupled to the quantum processor 1210.

The quantum processing unit(s) can be one or more of, but are not limited to: (a) a superconducting quantum computer; (b) an ion trap quantum computer; (c) a topological quantum computer using e.g. Majorana zero modes; (d) a photonic quantum computer; or (e) a neutral atom quantum computer. The sets of gates (e.g., using any of the disclosed embodiments) can be sent into (or otherwise applied to) the quantum processing unit(s) via control lines 1270 at a controller 1250. In the illustrated example, the desired quantum computing process is implemented with the aid of one or more controllers 1250 that are specially adapted to control a corresponding one of the quantum processor(s) 1210. The classical processor 1220 can further interact with measuring/monitoring devices (e.g., readout devices) 1280 to help control and implement the desired quantum computing process (e.g., by reading or measuring out data results from the quantum processing units once available, etc.).

Having described and illustrated the principles of the disclosed technology with reference to the illustrated embodiments, it will be recognized that the illustrated embodiments can be modified in arrangement and detail without departing from such principles. For instance, elements of the illustrated embodiments shown in software may be implemented in hardware and vice-versa. Also, the technologies from any example can be combined with the technologies described in any one or more of the other examples. It will be appreciated that procedures and functions such as those described with reference to the illustrated examples can be implemented in a single hardware or software module, or separate modules can be provided. The particular arrangements above are provided for convenient illustration, and other arrangements can be used.

Claims

1. A computer implemented method for characterizing implementation errors in a set of quantum logic operations of a quantum processor said set of quantum logic operations acting on a subset of qubits defining a qubit patch, wherein implementation errors involving the qubit patch are modeled by a model including a plurality of model parameters, the method comprising: (a) applying a set of quantum circuits to the quantum processor, wherein at least some of said quantum circuits comprise: i) a characterization gate sequence applicable to said qubit patch configured to provide measurement outcomes sensitive to at least some of the model parameters; andii) a neighboring gate sequence applicable to at least some neighboring qubits outside the qubit patch and interacting with the qubit patch, wherein said neighboring gate sequence is configured to reduce the sensitivity of measurement outcomes to implementation errors involving environment qubits outside the qubit patch;(b) measuring the patch qubits using a measurement apparatus of said quantum processor;(c) repeating (a)-(b) to collect a set of frequencies, said frequencies being associated with a measurement outcome and a quantum circuit;(d) computing a value of the model parameters by fitting the model to said set of frequencies.

2. The computer implemented method according to claim 1, wherein the neighboring qubits comprise qubits interacting with the qubit patch based on an interaction hypergraph.

3. The computer implemented method according to claim 1, wherein the characterization gate sequence is a gate set tomography sequence, and wherein the method further comprises gauge optimization.

4. The computer implemented method according to claim 1, wherein the neighboring gate sequence comprises initializing the environment qubits in a neighborhood-mixed state wherein the reduced density operator of each neighboring qubit is proportional to the unit-operator.

5. The computer implemented method according to claim 4, wherein the neighborhood-mixed state is an even mixture (|0⊗n+|1⊗n)/2⊗|0⊗(m-n), wherein m and n are the numbers of environment and neighboring qubits, respectively.

6. The computer implemented method according to claim 1, wherein the set of quantum circuits comprises of quantum circuit clusters, each quantum circuit cluster comprising a plurality of quantum circuits having a same characterization gate sequence and a distinct neighboring gate sequence, the method further comprising combining frequencies collected on quantum circuits of the same cluster.

7. The computer implemented method according to claim 6, wherein the set of quantum circuits comprises quantum circuit clusters each consisting of two quantum circuits having a same characterization gate sequence and respectively a first neighboring gate sequence starting with an idle gate and a second neighboring sequence starting with a π rotation, both first and second neighboring gate sequences being applied to each neighboring qubit, and wherein the method further comprises averaging frequencies collected on the two quantum circuits of each quantum circuit cluster.

8. The computer implemented method according to claim 1, wherein the neighboring gate sequence is configured to reduce a patching error.

9. The computer implemented method according to claim 8, wherein said patching error is estimated using perturbation expansion and the neighboring gate sequences are configured to cancel or at least reduce leading orders in said perturbation expansion.

10. The computer implemented method according to claim 9, wherein the leading orders include perturbation expansion orders up to a given order.

11. The computer implemented method according to claim 10, wherein said given order is the second order.

12. The computer implemented method according to claim 1, wherein at least one of said quantum circuits includes a neighboring gate sequence comprising a dynamical decoupling sequence configured to reduce patching errors.

13. The computer implemented method according to claim 12, wherein said dynamical decoupling sequence is coordinated with the characterization gate sequence of said at least one quantum circuit, so that at least some gates of said dynamical decoupling sequence are applied synchronously with at least some gates of the characterization gate sequence of said at least one quantum circuit, or at least some gates in said dynamical decoupling sequence are applied successively to at least some gates of the characterization gate sequence of said at least one quantum circuit.

14. The computer implemented method according to claim 1, wherein at least one of said quantum circuits further comprises a context gate sequence applicable to one or more environment qubits outside the qubit patch, wherein context gates of said context gate sequence are applied synchronously to at least some characterization gates of the characterization gate sequence of said at least one quantum circuit.

15. The computer implemented method according to claim 14, wherein the neighboring gate sequence of at least one of said quantum circuits comprises at least some neighboring gates applied synchronously with at least some characterization gates of the characterization gate sequence of said at least one quantum circuit.

16. The computer implemented method according to claim 14, wherein the neighboring gate sequence of at least one of said quantum circuits comprises at least some neighboring gates applied successively to at least some characterization gates of the characterization gate sequence of said at least one quantum circuit.

17. The computer implemented method according to claim 1, wherein the quantum processor is described in terms of qudits in place of qubits.

18. A computer implemented method for characterizing a quantum processor, the method comprising successively applying the computer implemented method of claim 1 to a plurality of patches of qubits of said quantum processor.

19. A computer program product, comprising a computer program, wherein the computer program, when executed by a computer, implements the computer implemented method according to claim 1.

20. A system comprising a computer and a quantum processor, the computer having gate-level or pulse-level access to the quantum processor, the system being configured to perform the computer implemented method according to claim 1.