Patent Application
20250156498
Embodiments described herein relate to a method and an apparatus for performing a quantum computation. The method uses a quantum system including constituents, such as qubits. The constituents of the quantum system are acted upon by, for example, a quantum processing system, to process the information carried by the constituents. Some of the constituents are measured to reveal the information contained in the constituents. Based on the read-out obtained from the measurement, a computational problem is solved.
Quantum computing devices are computing devices which make use of quantum mechanical effects to solve computational problems. In a quantum computing device, or quantum computer, information is carried by quantum systems, such as e.g. quantum bits (“qubits”). This is in contrast to conventional computers, which operate with classical bits, i.e. 0 and 1. During a quantum computation, quantum bits can be processed by evolving the quantum system. For example, groups of qubits of the quantum system can be coupled to each other according to a specified interaction. By evolving the quantum system, the information carried by the quantum system can be processed in order to carry out a computation, i.e. in order to solve a computational problem. In many cases, a quantum computer can be assisted by a classical computer, i.e. a computer operating with classical bits. The classical computer can provide instructions to the quantum computer as to how the qubits in the system are to be processed by the quantum computer.
In many approaches to quantum computation, in order to carry out an arbitrary quantum computation it is necessary to perform long-range interactions. Long-range interactions are interactions that couple qubits which are far apart from each other in the quantum system. Such long-range interactions provide an obstacle, since their practical realization is difficult. In some set-ups, long-range interactions can be replaced by sequences of short-range interactions. Yet, these approaches have the disadvantage that the sequences of short-range interactions are inherently sequential, i.e. they cannot be parallelized, leading to an increased runtime of the quantum computation. In turn, the fact that such sequences cannot be parallelized can compromise the scalability of the quantum computers based on such principles.
Alternatively, some approaches to quantum computing use short-range interactions only, but have the disadvantage that they are not fully programmable. That is, such quantum computers are restricted in the sense that they are tailored to solve certain specific computational problems, but they are not capable of solving arbitrary computational problems.
In yet other approaches, the quantum computation can be parallelized to a certain degree, but this comes at the cost of reducing the efficiency of the quantum computation, i.e. the runtime needed by the quantum computer for solving the computational problem at hand is increased in such approaches.
Therefore, there is a need for improved methods and devices for performing a quantum computation.
According to an embodiment, a method of performing a quantum computation is provided. The method includes providing a quantum system comprising constituents. The method includes encoding a computational problem into a problem Hamiltonian of the quantum system. The problem Hamiltonian is a single-body Hamiltonian being a sum of summand problem Hamiltonians. The method includes determining a constraint Hamiltonian of the quantum system. The constraint Hamiltonian is a sum of summand constraint Hamiltonians. A ground state of a total Hamiltonian encodes a solution to the computational problem. The total Hamiltonian includes a sum of the problem Hamiltonian and the constraint Hamiltonian. The method includes determining a first subset of the summand constraint Hamiltonians of the constraint Hamiltonian and a second subset of the summand constraint Hamiltonians of the constraint Hamiltonian. The method includes performing N rounds of operations, wherein N≥2. Each round includes preparing an initial quantum state. Each round includes evolving the quantum system according to a sequence of unitary operators. The sequence includes problem-encoding unitary operators, constraint-enforcing unitary operators and unitary driver operators. Each problem-encoding unitary operator is a unitary time evolution operator of a summand problem Hamiltonian of the problem Hamiltonian or is a unitary time evolution operator of a sum of summand problem Hamiltonians of the problem Hamiltonian. Each constraint-enforcing unitary operator is a unitary time evolution operator of a summand constraint Hamiltonian taken from the first subset of the summand constraint Hamiltonians of the constraint Hamiltonian, or is a unitary time evolution operator of a sum of summand constraint Hamiltonians taken from said first subset. Each unitary driver operator is a unitary operator that commutes with every summand constraint Hamiltonian from the second subset of the summand constraint Hamiltonians of the constraint Hamiltonian. Each round includes performing a measurement of one or more constituents of the quantum system. The method includes outputting a result of the quantum computation.
According to a further embodiment, an apparatus for performing a quantum computation is provided. The apparatus includes a quantum system comprising constituents. The apparatus includes a classical computing system. The classical computing system is configured to encode a computational problem into a problem Hamiltonian of the quantum system. The problem Hamiltonian is a single-body Hamiltonian being a sum of summand problem Hamiltonians. The classical computing system is configured to determine a constraint Hamiltonian of the quantum system, the constraint Hamiltonian being a sum of summand constraint Hamiltonians. A ground state of a total Hamiltonian encodes a solution to the computational problem, wherein the total Hamiltonian includes a sum of the problem Hamiltonian and the constraint Hamiltonian. The classical computing system is configured to determine a first subset of the summand constraint Hamiltonians of the constraint Hamiltonian and a second subset of the summand constraint Hamiltonians of the constraint Hamiltonian. The apparatus includes a quantum processing system including a unitary evolution device and a measurement device. The quantum processing system is configured to perform N rounds of operations, wherein N≥2. Each round includes evolving, by the unitary evolution device, the quantum system according to a sequence of unitary operators. The sequence includes problem-encoding unitary operators, constraint-enforcing unitary operators and unitary driver operators. Each problem-encoding unitary operator is a unitary time evolution operator of a summand problem Hamiltonian of the problem Hamiltonian or is a unitary time evolution operator of a sum of summand problem Hamiltonians of the problem Hamiltonian. Each constraint-enforcing unitary operator is a unitary time evolution operator of a summand constraint Hamiltonian taken from the first subset of the summand constraint Hamiltonians of the constraint Hamiltonian, or is a unitary time evolution operator of a sum of summand constraint Hamiltonians taken from said first subset. Each unitary driver operator is a unitary operator that commutes with every summand constraint Hamiltonian from the second subset of the summand constraint Hamiltonians of the constraint Hamiltonian. Each round includes performing, by the measurement device, a measurement of one or more constituents of the quantum system. The classical computing system is further configured to output a result of the quantum computation.
Embodiments are also directed to methods for operating the systems described herein, and to the use of the systems to perform the methods according to the embodiments described herein.
Further advantages, features, aspects and details that can be combined with embodiments described herein are evident from the dependent claims, the description and the drawings.
A full and enabling disclosure to one of ordinary skill in the art is set forth more particularly in the remainder of the specification including reference to the accompanying drawings wherein:
Reference will now be made in detail to the various exemplary embodiments, one or more examples of which are illustrated in each figure. Each example is provided by way of explanation and is not meant as a limitation. For example, features illustrated or described as part of one embodiment can be used on or in conjunction with other embodiments to yield yet further embodiments. It is intended that the present disclosure includes such modifications and variations.
Within the description of the drawings, the same reference numbers refer to the same or similar components. Generally, only the differences with respect to the individual embodiments are described. The structures shown in the drawings are not necessarily depicted true to scale, and may contain details drawn in an exaggerated way to allow for a better understanding of the embodiments.
Embodiments described herein relate to methods and apparatuses for performing gate-based quantum computing. Gate-based quantum computing, or digital quantum computing, can be understood as a method of computation wherein the quantum computation is driven by sequences of unitary operators. Gate-based quantum computing is distinguished from other approaches, such as e.g. adiabatic quantum computation (quantum annealing) or measurement-based quantum computation.
A quantum system as described herein is a physical system exhibiting quantum effects. That means, the quantum system is a real-world object. The quantum system includes constituents. The constituents are physical quantum entities themselves, and can be regarded as smaller d-level quantum systems that jointly form the quantum system. Specifically, the constituents of the quantum system can be qubits. A qubit shall be understood as a physical entity that realizes a two-level quantum system. The constituents may be d-level quantum systems (“qudits”) with d>2, wherein only two levels of the d levels might be used.
The quantum system can be in different quantum states, such as an initial quantum state (in which it may be prepared at the beginning of a quantum computation) and a final quantum (in which it may end up due to the quantum computation). The final quantum state can be, or can approximate, a ground state of a Hamiltonian of the quantum system, such as the total quantum Hamiltonian as described herein. The quantum system can be evolved from an initial quantum state towards, or to, a ground state of the total quantum Hamiltonian by performing sequences of unitary operators. Such an evolution is a real-world process, and particularly a controlled technical process (quantum computation) which brings the quantum system from an initial quantum state to an apriori unknown final quantum state that contains information about the solution to a computational problem. This information can be revealed by measuring the quantum system or a part thereof, i.e., at least some of its constituents. The act of measuring is also a physical/technical process. Measurements allow to obtain a read-out of the quantum system. A read-out of a quantum system is a set of measurement values obtained by measurements of constituents of the quantum system, involving physical interactions with the constituents.
The quantum system may include K constituents, which may be qubits, wherein K may be at least 100, at least 1.000 or at least 10.000. K may be from 100 to 10.000, or from 100 to 100.000, but K may be larger than 100.000. It shall be understood that the quantum systems shown in the figures and described in examples may be much smaller for illustrative and explanatory purposes, but shall not be understood to provide any limitation.
If Ĥ is a Hamiltonian of the quantum system, the operator exp(itĤ) is a unitary operator. Therein, t is a time parameter. A unitary operator of the form exp(itĤ) shall be referred to herein as a unitary time evolution operator, or unitary time evolution for short, according to the Hamiltonian Ĥ. A quantum system may be evolved by a unitary time evolution of a Hamiltonian. The act of performing a unitary operator is a physical/technical process. Evolving the quantum system by a unitary time evolution exp(itĤ) may include switching on an interaction between subsets of the constituents of the quantum system, wherein the interaction is defined by the Hamiltonian Ĥ. The interaction may be switched on for a time period t. The interaction may be switched off after the time period t has elapsed.
In any realistic system, at least a small amount of noise is always present. Accordingly, quantum states cannot be realized with 100% accuracy. Likewise, operations performed on a quantum system, such as unitary operators and measurements, are always subject to at least some noise, and are not realized with 100% accuracy. It shall be understood that the quantum states and operations described herein encompass states and operations that are subject to small amounts of noise.
According to an embodiment, a method of performing a quantum computation is provided. The method includes providing a quantum system comprising constituents. The method includes encoding a computational problem into a problem Hamiltonian of the quantum system. The problem Hamiltonian is a single-body Hamiltonian being a sum of summand problem Hamiltonians. The method includes determining a constraint Hamiltonian of the quantum system. The constraint Hamiltonian is a sum of summand constraint Hamiltonians. A ground state of a total Hamiltonian encodes a solution to the computational problem. The total Hamiltonian includes, or is, a sum of the problem Hamiltonian and the constraint Hamiltonian. The method includes determining a first subset of the summand constraint Hamiltonians of the constraint Hamiltonian and a second subset of the summand constraint Hamiltonians of the constraint Hamiltonian. The method includes performing N rounds of operations, wherein N≥2. Each round includes preparing an initial quantum state. Each round includes evolving the quantum system according to a sequence of unitary operators. The sequence includes, or consists of, problem-encoding unitary operators, constraint-enforcing unitary operators and unitary driver operators. Each problem-encoding unitary operator is a unitary time evolution operator of a summand problem Hamiltonian of the problem Hamiltonian or is a unitary time evolution operator of a sum of summand problem Hamiltonians of the problem Hamiltonian. Each constraint-enforcing unitary operator is a unitary time evolution operator of a summand constraint Hamiltonian taken from the first subset of the summand constraint Hamiltonians of the constraint Hamiltonian, or is a unitary time evolution operator of a sum of summand constraint Hamiltonians taken from said first subset. Each unitary driver operator is a unitary operator that commutes with every summand constraint Hamiltonian from the second subset of the summand constraint Hamiltonians of the constraint Hamiltonian. Each round includes performing a measurement of one or more constituents of the quantum system. The method includes outputting a result of the quantum computation.
According to embodiments described herein, an (a priori unknown) solution of the computational problem is encoded in the ground space of the total Hamiltonian. To determine said solution, the quantum system is evolved towards a ground state of the total Hamiltonian by unitary evolution, more specifically by applying sequences of unitary operators during the N rounds of operations. The N rounds of operations provide an iterative process in which the quantum system is moved gradually closer to a ground state of the total Hamiltonian. A final measurement of the quantum system state at the end of the iterative process can then reveal the solution of the computational problem.
The total Hamiltonian may be a sum of two parts, namely a sum of the problem Hamiltonian and the constraint Hamiltonian. A ground state of the total Hamiltonian is thus characterized as a quantum state having a low energy with respect to both the problem Hamiltonian and the constraint Hamiltonian. Accordingly, the quantum system can be evolved towards a ground state of the total Hamiltonian by lowering the energy of the quantum system with respect to both the problem Hamiltonian and the constraint Hamiltonian. By applying the problem-encoding unitary operators in the sequences of unitary operators of the N rounds—the problem-encoding unitary operators being time evolutions of (sums of) summand problem Hamiltonians—the quantum system is evolved to a region of states that have a low energy with respect to the problem Hamiltonian. As regards the constraint Hamiltonian, the summand constraint Hamiltonians are split up into two groups, namely the first subset (denoted by S1) and the second subset (denoted by S2) of the summand constraint Hamiltonians. The summand constraint Hamiltonians from the first subset S1 are treated similarly to the summand problem Hamiltonians. Namely, by performing unitary time evolutions of (sums of) summand constraint Hamiltonians from the first subset S1—these time evolution operators are the constraint-enforcing unitary operators—the quantum system evolves to quantum states having a low energy with respect to each of the constraint Hamiltonians from the first subset S1. The summand constraint Hamiltonians of the first subset S1 are said to be enforced “explicitly”. In contrast, the summand constraint Hamiltonians from the second subset S2 are not enforced explicitly. Rather, a set of unitary operators is chosen—the unitary driver operators—which are such that the energy of the quantum system with respect to the summand constraint Hamiltonians of the second subset S2 is conserved when evolving the quantum system according to the unitary driver operators (that is to say, the unitary driver operators commute with every summand constraint Hamiltonian from the second subset S2). Accordingly, if the quantum system starts out in a ground state of the summand constraint Hamiltonians of the second subset S2, the quantum system will remain within the ground space of said summand constraint Hamiltonians throughout the evolution of the quantum system, and so there is no need to enforce the summand constraint Hamiltonians of the second subset S2 explicitly. The summand constraint Hamiltonians of the second subset S2 are said to be enforced “implicitly”.
The present disclosure thereby provides a “hybrid” approach where some summand constraint Hamiltonians are enforced explicitly while others are enforced implicitly. The explicit enforcement of the summand constraint Hamiltonians of the first subset S1 has the advantage that the corresponding constraint-enforcing unitary operators are highly parallelizable, i.e. these unitary operators can be implemented with small circuit depth, which greatly facilitates their practical realization. Still, not all of the summand constraint Hamiltonians of the constraint Hamiltonian are enforced explicitly, since the explicit enforcement leads to an increase in the size of the subspace of quantum states that is to be searched during the course of the iterative process described above, hence resulting in an increase in the runtime of the computation. In contrast, by its very construction, the implicit enforcement of the second subset S2 of summand constraint Hamiltonians forces the quantum system to stay within the ground space of the summand constraint Hamiltonians of the second subset S2, and thereby restricts the size of the subregion of the quantum system that is probed during the quantum computation. Embodiments described herein thus provide a combination of two benefits, namely a high degree of parallelizability (due to the explicit enforcement of the summand constraint Hamiltonians from the first subset S1) combined with a smaller search space and hence an improved runtime of the computation (due to the implicit enforcement of the summand constraint Hamiltonians from the second subset S2).
The computational problem may be a decision problem, an optimization problem, or a different kind of computational problem. The computational problem may be any one of a variety of computational problems considered in, e.g., the fields of computer science, physics, chemistry or engineering. The computational problem may be an NP-hard problem, for example an Ising spin model problem. The computational problem of the present disclosure can be any computational problem as described in EP 3 113 084 B1. The document EP 3 113 084 B1 is incorporated herein.
The size of a computational problem may be understood as a measure for the number of classical information units, e.g. the number of classical bits, required to specify the computational problem. The size of a computational problem may depend on, or be, the number of input variables of the computational problem. The size of a computational problem may increase as the number of input variables increases.
The problem Hamiltonian, denoted by ĤP, is a single-body Hamiltonian of the quantum system. A single-body Hamiltonian is a Hamiltonian wherein no interactions occur between groups of two or more constituents. A single-body Hamiltonian may represent interactions between the constituents of the quantum system and an external entity, e.g. a magnetic field or an electric field, wherein each constituent interacts individually with the external entity.
The problem Hamiltonian is a sum of summand problem Hamiltonians. Each summand problem Hamiltonian may act on a single constituent of the quantum system. The problem Hamiltonian may have the form ĤP=Σk ĤP,K wherein each ĤP,k is a summand problem Hamiltonian acting solely on the k-th constituent of the quantum system.
The problem Hamiltonian may have adjustable parameters. An adjustable parameter of the problem Hamiltonian can be a parameter representing a strength and/or a direction of an interaction between a constituent of the quantum system and an external entity. The external entity may be a field, particularly a single-body field. A single-body field may refer to a field influencing a single constituent of the quantum system. The external entity may, e.g., include: one or more magnetic fields; one or more electric fields; one or more laser fields; one or more microwaves; and one or more phase shifts from mechanical deformations; or any combination thereof. The adjustable parameters of the problem Hamiltonian may include a plurality of field strengths and/or a plurality of field directions of single-body fields acting on the constituents of the quantum system.
The problem Hamiltonian may have the form ĤP=ΣkJk{circumflex over (σ)}z(k), wherein {circumflex over (σ)}z(k) is a Pauli operator of a k-th constituent of the quantum system, and wherein each Jk is a coefficient. The coefficients Jk may form the adjustable parameters of the problem Hamiltonian. Each term Jk{circumflex over (σ)}z(k) may be a summand problem Hamiltonian as described herein.
Encoding the computational problem into the problem Hamiltonian may include determining, from the computational problem, a problem-encoding configuration of the adjustable parameters of the problem Hamiltonian. The problem-encoding configuration may contain information about the computational problem. In particular, there may be a one-to-one correspondence between the computational problem and the problem-encoding configuration. For example, for a problem Hamiltonian of the form ĤP=Σk Jk σz(k), the coefficients Jk may form the adjustable parameters, and the problem-encoding configuration may be a specific set of values of the parameters Jk that encodes the computational problem that is to be solved by the quantum computation.
Encoding the computational problem into the problem Hamiltonian may include a two-step process wherein the computational problem is first mapped to an auxiliary computational problem and the auxiliary computational problem is thereafter mapped to the problem Hamiltonian.
Encoding the computational problem into the problem Hamiltonian may include mapping the computational problem onto an auxiliary computational problem, wherein the auxiliary computational problem includes determining a ground state of a spin model, such as an Ising spin model. The auxiliary computational problem may be an Ising spin model problem. The auxiliary computational problem may be an NP-hard computational problem, such as the Ising spin model problem. Mappings from a variety of computational problems to the Ising spin model problem, or other NP-hard problems, are known in the literature.
Encoding the computational problem into the problem Hamiltonian may include determining the problem Hamiltonian from the auxiliary computational problem. Specifically, a problem-encoding configuration of the adjustable parameters of the problem Hamiltonian may be determined from the auxiliary computational problem. For example, each interaction between spins in the spin model of the auxiliary computational problem may be mapped to a summand problem Hamiltonian of the problem Hamiltonian. Specific encodings (called “parity” encodings) that allow to determine the problem Hamiltonian from the Ising spin model problem are described in EP 3 113 084 B1 and WO 2022/008057 A1. The document WO 2022/008057 A1 is incorporated herein.
The act of determining the constraint Hamiltonian can include determining a classical description of the constraint Hamiltonian. Determining can include calculating (e.g. by a classical computing system), reading (e.g. from a memory), receiving (e.g. via a communication channel), and the like. The act of determining the first subset and the second subset of the summand constraint Hamiltonians can be understood similarly.
The constraint Hamiltonian (denoted by ĤC) may be a short-range Hamiltonian. A short-range Hamiltonian may refer to a Hamiltonian representing joint interactions within groups of constituents, wherein no interactions occur between constituents which are distanced from each other by a distance greater than an interaction cut-off distance DSR. The interaction cut-off distance DSR may be a constant distance. The interaction cut-off distance DSR may be much smaller than a maximal constituent distance between the constituents in the quantum system. For example, the interaction cut-off distance may be 30% or less, 20% or less, or 10% or less of the maximal constituent distance. If the constituents are arranged in a lattice having an elementary distance (lattice constant), a short-range quantum Hamiltonian may be such that no interactions occur between constituents distanced from each other by a distance greater than r times the elementary distance (lattice constant) of the lattice. Therein, r may be from 1 to 5, e.g. r={umlaut over (2)}, 2, 3, 4 or 5. A short-range Hamiltonian Ĥ may have the form Ĥ=ΣiĤi, wherein each Ĥi is a summand Hamiltonian of Ĥ, and wherein each summand Hamiltonian Ĥi acts only within a group of constituents of the quantum system such that any two constituents in the group are distanced from each other by a distance not greater than the interaction cut-off distance DSR. Each summand Hamiltonian Ĥi may have the form Ĥi={circumflex over (K)}i ⊗I, wherein ⊗ is the tensor product, {circumflex over (K)}i is an operator acting within the group of constituents, and I is the identity operator acting on all constituents outside of said group of constituents.
The constraint Hamiltonian may be a d-body Hamiltonian, wherein d is 8 or less, particularly 4 or less. A d-body Hamiltonian may refer to a Hamiltonian representing interactions of the plurality of constituents, wherein no joint interactions occur between groups comprising d+1 or more constituents. A d-body Hamiltonian may be a sum of summand Hamiltonians, wherein each summand Hamiltonian represents a joint interaction between a group of d constituents or less.
A Z-type operator is an operator of the form Σj aj {circumflex over (Z)}j (including the case where the sum includes only one term), wherein each aj is a coefficient and each {circumflex over (Z)}j is a tensor product of Pauli {circumflex over (σ)}z operators or a single Pauli {circumflex over (σ)}z operator. The constraint Hamiltonian may be a Z-type operator. The constraint Hamiltonian may have the form ĤC=Σl Ĉl, wherein each Ĉl has the form Ĉl=al {circumflex over (Z)}l+blI, wherein {circumflex over (Z)}l is a tensor product of Pauli {circumflex over (σ)}z operators, I is the identity operator, and al and bl are coefficients. Each Ĉl may be a summand constraint Hamiltonian.
Herein, specific forms of the Hamiltonians (problem Hamiltonian, constraint Hamiltonian, driver Hamiltonian, and the like) are provided by way of example. For example, as described above, the problem Hamiltonian and the constraint Hamiltonian can employ Pauli {circumflex over (σ)}z operators. It shall be understood that this choice of types of Pauli operators is without loss of generality in that corresponding orientations (x, y, z) can be chosen freely, or else the types of Pauli operators can be permuted. The problem Hamiltonian and the constraint Hamiltonians may employ a same type of Pauli operators.
The constraint Hamiltonian ĤC has the property that the ground state of the total Hamiltonian (denoted by Ĥtotal) encodes a solution to the computational problem, wherein the total Hamiltonian is a sum of the problem Hamiltonian ĤP and the constraint Hamiltonian ĤC, i.e. Ĥtotal=ĤP+ĤC. That a ground state of the total Hamiltonian encodes a solution to the computational problem can be understood in the sense that said ground state contains information about the solution in question. The information can be revealed by performing one or more measurements on the ground state. Based on the outcome(s) of said measurement(s), a solution (e.g. a trial solution) of the computational problem can be determined.
The terminology “constraint Hamiltonian” stems from the property that the encoding of the Ising model problem (which may be the original computational problem or the auxiliary computational problem to which the original computational problem is mapped) into the problem Hamiltonian may introduce an increase in the number of degrees of freedom, in the sense that the ground space of the problem Hamiltonian alone includes quantum states that do not correspond to spin configurations of the Ising model, i.e. quantum states that cannot be “mapped back” onto the Ising model. To remove these additional degrees of freedom, the constraint Hamiltonian is introduced. The constraint Hamiltonians imposes an energy penalty, or energy constraint, to the aforementioned quantum states, such that the ground space of the sum of the problem Hamiltonian and the constraint Hamiltonian—i.e. the total Hamiltonian—only contains quantum states that correspond to the solution(s) of the computational problem. Specifically, each summand constraint Hamiltonian may impose a parity constraint on a subgroup of the constituents, such that, within said subgroup, the number of constituents that are in the quantum state |1> is even.
Specific encodings that allow to determine the problem Hamiltonian and a corresponding constraint Hamiltonian from the Ising spin model problem are described in EP 3 113 084 B1 and WO 2022/008057 A1.
As described above, the total Hamiltonian can be the sum of the problem Hamiltonian and the constraint Hamiltonian. In other embodiments, the total Hamiltonian may include additional terms, in other words the total Hamiltonian may include the sum of the problem Hamiltonian and the constraint Hamiltonian, plus optional additional terms. The additional terms may, for example, correspond to additional conditions (“side conditions”) imposed on the solution of the computational problem.
The first subset and the second subset of the summand constraint Hamiltonians of the constraint Hamiltonian are denoted herein by S1 and S2, respectively. The first subset S1 and the second subset S2 may be disjoint subsets. The union of the first subset S1 and the second subset S2 may form the entire set of summand constraint Hamiltonians of the constraint Hamiltonian. The first subset S1 and the second subset S2 may form a partition of the summand constraint Hamiltonians of the constraint Hamiltonian. Specific examples of first and second subsets of the summand constraint Hamiltonians are described below in section “Further aspects”.
The N rounds of operations may include 10 or more, particularly 100 or more, more particularly 1.000 or more rounds of operations, or even 100.000 or more rounds of operations.
Each round of the N rounds of operations includes preparing an initial quantum state for said round. The initial quantum state may be the same for all of the N rounds of operations. Alternatively, different initial quantum states may be prepared for different rounds of operations.
The initial quantum state of at least some, optionally all, of the N rounds of operations may be a ground state of a partial constraint Hamiltonian, wherein the partial constraint Hamiltonian is a sum of all summand constraint Hamiltonians taken from the second subset S2. (Said partial constraint Hamiltonian is the “second partial constraint Hamiltonian” as described further below.) The partial constraint Hamiltonian has a ground space. The ground space consists of all quantum states that are ground states of the partial Hamiltonian. The ground space has a ground space basis, or orthonormal basis, consisting of a set of quantum basis states. For example, each quantum basis state may have the form |x> where x is a bit string, i.e. each quantum basis state may be a computational basis state (standard basis state). The initial quantum state of at least some, optionally all, of the N rounds of operations may be a superposition of all quantum basis states of the ground space basis. Specific examples of initial quantum states are described below in the section “Further aspects”.
Each round of the N rounds of operations includes evolving the quantum system according to a sequence of unitary operators. The sequence includes problem-encoding unitary operators, constraint-enforcing unitary operators and unitary driver operators.
Each problem-encoding unitary operator is a unitary time evolution operator of a summand problem Hamiltonian of the problem Hamiltonian or is a unitary time evolution operator of a sum of summand problem Hamiltonians of the problem Hamiltonian. A problem-encoding unitary operator may have the form exp(itÂ) where t is a coefficient and where  is either equal to a single summand problem Hamiltonian ĤP,k or equal to a sum of two or more summand problem Hamiltonians ĤP,k (including the case where  is the sum of all summand problem Hamiltonians ĤP,k, so that  is equal to ĤP).
Each constraint-enforcing unitary operator is a unitary time evolution operator of a summand constraint Hamiltonian taken from the first subset S1 of the summand constraint Hamiltonians of the constraint Hamiltonian, or is a unitary time evolution operator of a sum of summand constraint Hamiltonians taken from said first subset S1. The constraint Hamiltonian HC may be denoted by
ĤC=ΣiĈi+ΣjĈj,
wherein the first sum runs over all summand constraint Hamiltonians Ĉi of the first subset S1 and the second sum runs over all summand constraint Hamiltonians Ĉj of the second subset S2. The first sum is equal to the first partial constraint Hamiltonian as described herein (see the section “Further aspects”). The second sum is equal to the second partial constraint Hamiltonian as described herein. A constraint-enforcing unitary operator may have the form exp(itÂ) where t is a coefficient and where  is either equal to a single summand constraint Hamiltonian Ĉi taken from the first subset S1 or equal to a sum of several summand constraint Hamiltonians Ĉi taken from the first subset S1 (including the case where  is the sum of all summand constraint Hamiltonians Ĉi taken from the first subset S1, so that  is equal to the first partial constraint Hamiltonian).
Each constraint-enforcing unitary operator may act trivially on the ground space basis of the second partial constraint Hamiltonian. Therein, an operator is considered to act trivially on the ground space basis if the operator maps each quantum basis state of the ground space basis to itself up to a proportionality factor.
Each unitary driver operator commutes with every summand constraint Hamiltonian Ĉj from the second subset S2 of the summand constraint Hamiltonians. It may be the case that each unitary driver operator does not commute with one or more summand constraint Hamiltonians Ĉi from the first subset S1. Each unitary driver operator may have the form exp(itĤ), wherein t is a coefficient and Ĥ is an operator of the form Σj bj {circumflex over (X)}j (including the case where the sum includes only one term), wherein each bj is a coefficient and each {circumflex over (X)}j is a tensor product of Pauli σX operators or a single Pauli σX operator.
Each unitary driver operator may act non-trivially on the ground space basis of the second partial constraint Hamiltonian. An operator is considered to act non-trivially on the ground space basis if the operator does not act trivially on the ground space basis. For each quantum basis state |x> of the ground space basis of the second partial constraint Hamiltonian, each unitary driver operator may map |x> to a linear combination of two or more quantum basis states of the ground space basis of the second partial constraint Hamiltonian.
A method according to embodiments described herein may include determining a driver Hamiltonian being a sum of summand driver Hamiltonians, wherein the driver Hamiltonian commutes with every summand constraint Hamiltonian of the second subset S2. Each unitary driver operator may be a unitary time evolution operator of a summand driver Hamiltonian of the driver Hamiltonian or may be a unitary time evolution operator of a sum of two or more summand driver Hamiltonians of the driver Hamiltonian (including the case where the unitary driver operator is a unitary time evolution operator of the driver Hamiltonian, the driver Hamiltonian being the sum of all summand driver Hamiltonians).
The driver Hamiltonian may be an X-type operator. An X-type operator is an operator of the form Σj bj {circumflex over (X)}j (including the case where the sum includes only one term), wherein each bj is a coefficient and each {circumflex over (X)}j is a tensor product of Pauli {circumflex over (σ)}x operators or a single Pauli {circumflex over (σ)}x operator. It shall again be understood that this choice of Pauli operators is without loss of generality in that corresponding orientations (x, y, z) can be chosen freely, or else the types of Pauli operators can be permuted. The driver Hamiltonian may employ a type of Pauli operators different from the problem Hamiltonian and the constraint Hamiltonian. In the section “Further aspects” below, the driver Hamiltonian is referred to as “hybrid driver Hamiltonian”.
The sequence of unitary operators of a round of operations may be denoted by Û1, Û2, . . . , Ûm wherein each Ûi is a unitary operator. At least some, and optionally all, of the Ûi may be problem-encoding unitary operators, constraint-enforcing unitary operators or unitary driver operators. The initial state of the round in question may be denoted |ψ>. Evolving the quantum system according to the sequence Û1, Û2, . . . , Ûm can be understood in the sense that, after the sequence is applied, the quantum state of the quantum system is Ûm . . . Û2Û1|ψ>, at least approximately.
That the quantum system is evolved according to a sequence Û1, Û2, . . . , Ûm does not necessarily imply that each operator Ûi of the sequence shall be implemented as a single unitary operator. Any operator Ûi can itself be implemented as a product, or circuit, of several unitary operators (quantum gates). This may be advantageous if the unitary operator Ûi is too complex to be implemented as a single unitary operator. Decomposing a unitary operator Ûi as a quantum circuit of several simpler unitary operators (e.g. short-range d-body unitary operators with a small constant d) may facilitate the implementation of the unitary operation Ûi.
Evolving the quantum system according to a sequence of unitary operators may include implementing at least some unitary operators of the sequence by a quantum circuit comprising a plurality of quantum gates. At least some, particularly all, of the problem-encoding unitary operators, constraint-enforcing unitary operators and unitary driver operators of the sequence may be implemented by a quantum circuit. The term “quantum circuit” refers to a logic gate circuit comprising logic gates, wherein each logic gate is a unitary operator (called “quantum gate” in this context).
Each quantum gate of a quantum circuit may be a short-range unitary operator. A short-range unitary operator is a unitary operator acting only within a subgroup of constituents of the quantum system, wherein any two constituents within the subgroup are distanced from each other by a distance of at most an interaction cut-off distance of the quantum system, as described herein. The short-range unitary operator does not act on any constituent outside of said subgroup of constituents.
Additionally or alternatively, each quantum gate of a quantum circuit may be a d-body unitary operator. Therein, d may be a small constant. For example, d may be 8 or less, or even 4 or less. A d-body unitary operator refers to a unitary operator acting only within a subgroup including at most d constituents of the quantum system. The d-body unitary operator does not act on any constituent outside of said subgroup. A single-body unitary operator is a d-body unitary operator where d=1.
The sequence of unitary operators of at least some rounds of the N rounds of operations may include K problem-encoding unitary operators, and/or L constraint-enforcing unitary operators, and/or M unitary driver operators. Therein, K, L and/or M may be 5 or more, particularly 10 or more, more particularly 200 or more.
For each round of the N rounds of operations, evolving the quantum system according to the sequence ofunitary operators of said round may include applying said sequence ofunitary operations to the initial quantum state of said round.
The sequence of unitary operators of at least some of the N rounds of operations may have the form A1 A2 . . . Ap, or may include at least a sub-sequence of said form, wherein p≥3, particularly p may be 10 or more, 100 or more, or 1000 or more. Each Ai may be a product of the form Xi Yi Zi, wherein one of Xi, Yi and Zi is a problem-encoding unitary operator, another one of Xi, Yi and Zi is constraint-enforcing unitary operator and yet another one of Xi, Yi and Zi is a unitary driver operator. Examples of possible sequences of unitary operators are described in more detail below in the section “Further aspects”.
Each round of the N rounds of operations includes performing a measurement of one or more constituents of the quantum system. For each round of the N rounds of operations, the measurement of said round may be performed on a quantum state resulting from evolving the quantum system according to the sequence of unitary operators of said round. Performing a measurement of the one or more constituents may include measuring a Pauli operator, e.g. the Pauli operator σz, for each of the one or more constituents.
In some embodiments, the method includes a feed-forward of information, wherein the sequence of unitary operators to be applied in a round of operations may depend on measurement outcomes of measurements performed in one or more, e.g. at least two, previous rounds of operations. The N rounds of operations may include one or more adaptive rounds of operations, for example, 10 or more, 100 or more, 1000 or more or even 100000 or more adaptive rounds. For each adaptive round of operations, the unitary operators of the sequence of unitary operators of the adaptive round may be determined based on at least one measurement outcome of a measurement performed in a previous round of the N rounds of operations.
The N rounds of operations may include a first round of operations. Evolving the quantum system according to the sequence of unitary operators of the first round of operations may result in a first quantum state of the quantum system. Performing the measurement in the first round may include measuring an energy of the first quantum state. Measuring the energy of a quantum state, such as the first quantum state, may include measuring a Hamiltonian, such as the total Hamiltonian as described herein.
The N rounds of operations may include a second round of operations performed after the first round of operations. Evolving the quantum system according to the sequence of unitary operators of the second round of operations may result in a second quantum state of the quantum system. Performing the measurement in the second round may include measuring an energy of the second quantum state. Measuring the energy of the second quantum state may include measuring a Hamiltonian, such as the total Hamiltonian.
The method described herein may include comparing the energy of the first quantum state with the energy of the second quantum state. The method may include determining the sequence of unitary operators to be applied in a third round of the N rounds of operations, wherein the third round is to be performed after the second round. The sequence of unitary operators to be applied in the third round may be determined based at least on the comparison of the energy of the first quantum state with the energy of the second quantum state.
For example, if the comparison of the energy of the first quantum state with the energy of the second quantum state reveals that the energy of the first quantum state is smaller than the energy of the second quantum state, the user may conclude that the first quantum state is closer to the ground state of the measured Hamiltonian (e.g. the total Hamiltonian) than the second quantum state. In light thereof, the user may reject the sequence of unitary operators of the second round and return to the sequence of unitary operations of the first round. Starting from the sequence of unitary operations of the first round, the user may make a small perturbation to said sequence, e.g. by replacing one or just a few operators from said sequence by different operators. The resulting sequence may be the sequence of unitary operators to be applied in the third round of operations.
Alternatively, if the comparison of the energy of the first quantum state with the energy of the second quantum state reveals that the energy of the first quantum state is larger than (or equal to) the energy of the second quantum state, the user may conclude that the second quantum state is closer to the ground state of the measured Hamiltonian than the first quantum state. In light thereof, the user may accept the sequence of unitary operators of the second round. Starting from the sequence of unitary operations of the second round, the user may make a small adjustment or perturbation to said sequence. The resulting adjusted sequence may be the sequence of unitary operators to be applied in the third round of operations.
The user can proceed in a similar manner throughout all rounds of operations, namely: (i) measure the energy of the quantum state obtained after applying the sequence of unitary operations of the current round of operations (e.g. by measuring the total Hamiltonian); (ii) compare the measured energy of the current round with a measured energy of a previous round; (iii) if the measured energy of the current round is larger than the measured energy of the previous round, reject the quantum state of the current round and accept the sequence of unitary operations of the previous round; alternatively, if the measured energy of the current round is smaller than the measured energy of the previous round, accept the sequence of unitary operations of the current round; (iv) starting from the accepted sequence of unitary operations, perturb said accepted sequence to obtain a sequence of operations for a next round of operations.
As the number N of rounds is increased, an increasingly larger set of quantum states is prepared, wherein the energy of subsequent quantum states gradually decreases (or at least does not increase). Accordingly, the energy gradually approaches the ground state energy of the measured Hamiltonian. In light thereof, embodiments described herein provide a gradually improving approximation to the ground state of the measured Hamiltonian.
By measuring a suitable Hamiltonian in the respective rounds of operations, a solution to the computational problem can be determined. According to embodiments, a plurality of rounds of the N rounds of operations (in particular, substantially all of the N rounds) may each include measuring the total Hamiltonian of the quantum system. As described herein, the total Hamiltonian has a ground state containing information about a solution to the computational problem. Accordingly, if the quantum system is in the ground state of the total Hamiltonian, or close to the ground state, the information in question may be revealed by measuring the quantum system. A solution to the computational problem can be determined.
Although above an example is given of a method wherein the measured Hamiltonian is the total Hamiltonian, other Hamiltonians may also be measured. In particular, the total Hamiltonian may be modified or transformed so that the form of the modified Hamiltonian may differ from the total Hamiltonian, but the modified Hamiltonian still has the property that the ground state of the modified Hamiltonian encodes a solution to the computational problem. For example, modifying the total Hamiltonian by a change of basis of each constituent, or of a plurality of small groups of constituents, may change the form of the Hamiltonian, but does not change the property that the ground state of the modified Hamiltonian encodes the solution to the computational problem.
Although above an example is given of a method wherein one or more energies are measured, further or alternative measurements may also be performed. For example, instead of measuring an energy of a quantum state, the quantum fidelity of the quantum state with respect to a target state may be measured. For example, the target state may be a ground state of a target Hamiltonian (such as the total Hamiltonians as described herein, or other Hamiltonians). The quantum fidelity of two quantum states |ψ1> and |ψ2> refers to the quantity |<ψ1|ψ2>|.
The method according to embodiments described herein includes outputting a result of the quantum computation. The result of the quantum computation may be based on one or more measurement outcomes of one or more measurements performed in the N rounds of operations. The method may include processing, e.g. by a classical computing system as described herein, one or more measurement outcomes of one or more measurements performed in the N rounds op operations. The method may include outputting the result of the quantum computation based on the one or more processed measurement outcomes.
The outputted result of the quantum computation may be a solution, particularly a trial solution, of the computational problem. A trial solution may, for example, be an approximate solution of the computational problem.
It may be beneficial to choose the first subset S1 of summand constraint Hamiltonian in a manner such that the summand constraint Hamiltonians in said first subset effectively define a partitioning of the quantum system into smaller groups of constituents, called subsystems. This approach is also referred to herein as “modularization”.
In
The spatial arrangement of the summand constraint Hamiltonians 320 may define a subdivision of the quantum system 300 into subsystems 450, wherein each subsystem 450 consists of a subgroup of constituents 302, as shown in
The quantum system may include subsystems. Each subsystem may include a subset of the constituents of the quantum system. The subsystems may be disjoint. Different subsystems may not have a constituent in common. Each subsystem may have boundary constituents forming part of, and in particular forming, a boundary between the subsystem and one or more adjacent subsystems.
Each boundary constituent may participate in a quantum interaction represented by a summand constraint Hamiltonian from the first subset S1 of the summand constraint Hamiltonians of the constraint Hamiltonian. That a boundary constituent participates in a quantum interaction represented by a summand constraint Hamiltonian means that said summand constraint Hamiltonian acts on the boundary constituent in question. For each subsystem, each boundary constituent of the subsystem may be coupled to a boundary constituent of an adjacent subsystem by a quantum interaction that is represented by a summand constraint Hamiltonian of the first subset of summand constraint Hamiltonians.
Each subsystem may have a total number of constituents that is much smaller than the total number of constituents of the quantum system. For example, the total number of constituents of each subsystem may be 30% or less, 20% or less, or 10% or less, or even 1% or less, of the total number of constituents of the quantum system. Each subsystem may have a total number of constituents that is independent of a size of the computational problem. The total number of constituents of each subsystem may be a constant. In other words, the subdivision of the quantum system into subsystem by virtue of the spatial arrangement of the summand constraint Hamiltonians from the first subset may be such that the number of constituents in each subsystem is bounded by a constant independent of the size of the computational problem, and, correspondingly, independent of the total number of constituents of the entire quantum system.
According to some embodiments, each unitary driver operator may act fully inside one of the subsystems of the quantum system. A unitary driver operator may act only on constituents belonging to a same subsystem. In cases where the size of each subsystem of the quantum system is relatively small as compared to the size of the entire quantum system, a unitary operator acting fully inside one of the subsystems is a manageable object of reduced complexity. In particular, the circuit depth required for implementing such a unitary driver operator may be relatively small. For example, if the size (total number of constituents) of the subsystem is constant, i.e. independent of the size of the computational problem, then a unitary driver operator acting fully inside said subsystem can be implemented by a quantum circuit of constant depth, i.e. a highly parallelized quantum circuit, which is advantageous because it requires only a small amount of computational (time) resources.
Each unitary driver operator may be realizable, or realized, by a quantum circuit of constant depth. The term “depth” as used in the present disclosure refers to the notion of circuit depth of a circuit of logic gates as known in the field of computer science. A circuit of quantum gates, i.e. unitary operators, acting on a set of constituents of the quantum system can be said to be parallelizable to a depth D if the quantum gates in the circuit can be grouped into D layers (slices) of gates, such that in each layer there are no two quantum gates acting on the same constituent. In other words, within each layer, each constituent is acted upon by at most one quantum gate. The depth is a measure of how much a circuit can be parallelized. Operations within a same layer of the circuit can be performed in the same time step (“in parallel”), since the gates within one layer act on different constituents. Therefore, a circuit which is parallelizable to a depth D can be carried out in D time steps. For a more detailed discussion of the notion of depth, reference is made to WO 2020/156680 A1. The document WO 2020/156680 A1 is incorporated herein.
A constant depth refers to a depth which is independent of the number of constituents of the quantum system. A constant depth may be a depth which is much smaller than the number of constituents in the quantum system. For example, a constant depth may be a depth which is 30% or less, in particular 20% or less, more particularly 10% or less, of the number of constituents of the quantum system. As the method described herein is used for solving computational problems of increasing sizes, quantum systems of increasing system sizes are needed. According to embodiments, irrespective of the size of the computational problem, each unitary driver operator may be realized by a quantum circuit of constant depth D. That is, unlike the number of constituents in the quantum system, the depth D does not grow as a function of the size of the computational problem but is bounded from above by a constant. For example, D may be at most 100.
The constituents of the quantum system may be arranged according to a first two-dimensional lattice (such as the constituents 302 shown in
The set formed by the boundary constituents of all subsystems of the quantum system (e.g. the boundary constituents 420 shown in
The quantum system and its constituents (such as qubits) are physical entities, as explained herein. Specific implementations of the quantum system/the constituents and of the interactions involved in the method described herein are briefly discussed below. Further details can be found in EP 3 113 084 B1 and WO 2020/156680 A1. However, the method described herein can be carried out on any other specific implementation of said physical entities and of their interactions, and the exemplary implementations shall not be considered as limiting.
The constituents may be superconducting qubits, e.g. transmon or flux qubits. Superconducting currents propagating clockwise and counter-clockwise, respectively, in the primary superconducting loop can form the quantum basis states |1> and |0> of the superconducting qubit. Further, a magnetic flux bias through the secondary superconducting loop can couple the quantum basis states |0> and |1>.
A single-body Hamiltonian of the form Σk ak {circumflex over (σ)}z(k) can be realized by magnetic fluxes interacting with the superconducting qubits. A constraint Hamiltonian, for example a plaquette Hamiltonian, can be realized using ancillary qubits, wherein an ancillary qubit may be arranged inside each plaquette. Interactions between qubits of the form Kkm{circumflex over (σ)}z(k){circumflex over (σ)}z(m) can be realized by an inductive coupling unit including a superconducting quantum interference device. Applying an adjustable magnetic flux bias to the superconducting quantum interference device allows tuning the coefficient Kkm. A summand constraint Hamiltonian can then be realized by Hsr,p=C(σz(1)+σz(2)+σz(3)+σz(4)−2σz(p)−1)2, which includes only pairwise interactions of the form σz(k)σz(m) and single-body σz(l) terms corresponding to imposed energy differences between the |0> and |1> quantum basis states. Here, σz(p) represents the ancilla qubit. Alternatively, a constraint Hamiltonian can be realized without ancillary qubits, e.g., using three-island superconducting devices as transmon qubits.
Further, a magnetic flux bias through the primary superconducting loop of the superconducting qubit may be set such that the basis states |0> and |1> have the same energy, i.e. the energy difference for these basis states is zero. Further, a magnetic flux bias through the secondary superconducting loop can couple the basis states |0> and |1>. Accordingly, a Hamiltonian of the form h{circumflex over (σ)}x(k) can be realized for the superconducting qubit.
For superconducting charge or flux qubits, CNOT operations can be realized with an additional capacitive element coupled to two qubits. The interaction strength is tuned by magnetic or electric flux applied to the additional element. Alternatively, the two qubits are coupled to two modes of a Josephson ring modulator. Single-body unitary operators exp(it{circumflex over (σ)}x(k)) or exp(it{circumflex over (σ)}z(k)) can be realized with controlled external magnetic or electric flux.
For superconducting qubits, the qubit states |0> and |1> can be measured with high fidelity using a measurement device including a plurality of superconducting quantum interference devices, in particular N hysteretic DC superconducting quantum interference devices and N RF superconducting quantum interference device latches controlled by bias lines, wherein the number of bias lines scales according to N.
Alternatively, the quantum system may be realized by systems of trapped ions, ultracold atoms, impurities in solid-state crystals (such as NV Centers), quantum dots, and the like. For background on how Hamiltonians, unitary operators and measurements can be implemented in such systems, reference is made to EP 3 113 084 B1 and WO 2020/156680 A1. As already mentioned above, these are exemplary implementations, which shall not be considered as limiting.
According to a further embodiment, and as illustrated in
Each round of the N rounds of operations may include preparing, for example by at least one of the unitary evolution device and the measurement device, an initial quantum state.
The apparatus may include a controller. The controller may include or be the classical computing system. The controller may be connected to the quantum processing system. The controller may be configured to instruct the unitary evolution device to evolve the quantum system according to the sequence of unitary operators of each round of the N rounds of operations. The controller may be configured to instruct the measurement device to perform a measurement of one or more constituents of the quantum system in each of the N rounds. The classical computing system may be configured to receive a set of measurement outcomes from the measurement device, the measurement outcomes resulting from measurements performed during one or more of the N rounds of operations. The classical computing system may be configured to determine a sequence of unitary operators to be performed in a future round of the N round of operations based on one or more received measurement outcomes. The classical computing system may be configured to determine a result of the quantum computation, such as a solution of the computational problem, based on one or more received measurement outcomes.
The unitary evolution device and the measurement device may be configured for performing any unitary operator and any measurement, respectively, as described in relation to the method described herein. For example, the unitary evolution device may be configured to perform a quantum circuit comprising quantum gates to implement a unitary operation of a sequence of unitary operations of any round of the N rounds of operations. For example, the measurement device may be configured to perform any energy measurement, e.g. measurement of the total Hamiltonian, as described herein, such as measuring the energy of the first quantum state and/or the second quantum state as described herein.
A classical computing system is distinguished from a quantum computing system. A classical computing system can be understood as a computing system that stores and processes information using only classical information carriers, such as classical bits. A classical computing system may not use quantum information carriers, such as qubits, for processing information. A classical computing system may include a central processing unit (CPU) for processing information with classical bits and/or a memory for storing information with classical bits. A classical computing system may include one or more conventional computers and/or a network of conventional computers, such as personal computers (PCs).
The classical computing system of the apparatus described herein may be configured to determine a driver Hamiltonian being a sum of summand driver Hamiltonians, wherein the driver Hamiltonian commutes with every summand constraint Hamiltonian of the second subset of the summand constraint Hamiltonians. Each unitary driver operator may be a unitary time evolution operator of a summand driver Hamiltonian of the driver Hamiltonian or may be a unitary time evolution operator of a sum of two or more summand driver Hamiltonians of the driver Hamiltonian.
The classical computing system may be configured to perform any classical computational operation of the method described herein, such as the feed forward of information in the adaptive rounds of operations, the comparison of measured energies to determine the sequence of unitary operations to be applied in a future round of operations, and the like.
The Quantum Approximate Optimization Algorithm (QAOA) is a gate-based algorithm designed for solving combinatorial optimization problems on contemporary noisy quantum devices. This algorithm can in principle be applied to generic combinatorial optimization problems. However, the limited inter-qubit connectivity of quantum devices complicates its implementation and can be detrimental for practical QAOA performance. The parity architecture described below resolves the mismatch between the connectivity of the problem- and hardware graph by mapping problem-defining interactions onto a single-body problem Hamiltonian, while restricting the enlarged Hilbert space by a short-range constraint Hamiltonian. In particular, starting with a classical compilation step to adjust the arrangement of qubits and constraints to the optimization problem, the parity architecture enables mapping of generic, i.e. long-range and higher-order connected, optimization problems onto a fixed qubit layout utilizing only short-range quantum interactions.
QAOA implementations for the parity architecture in which the constraint terms are explicitly enforced through an energy penalty can be considered. Alternatively, conserving the constraint terms can be achieved implicitly by making the involved operators commute with the constraint Hamiltonian. While the first approach enables a parallelizable implementation with low circuit depth in the QAOA, the latter can lead to significantly improved success probabilities.
According to embodiments described herein, a hybrid approach is provided, which keeps the required quantum circuit depth constant while reducing the number of constraint terms to be enforced explicitly and thus increasing performance. This is achieved by partitioning the constraint terms into a subset that is enforced explicitly (the first subset of summand constraint Hamiltonians, as described herein) and a subset where the constraints are conserved implicitly by adapting the driver Hamiltonians (the second subset of summand constraint Hamiltonians, as described herein).
In other words,
The task of finding solutions to computational problems such as combinatorial optimization problems can be formulated as energy minimization of general (classical) N-spin Hamiltonian functions of the form
where si=±1 denote classical spin variables (also called “problem spins” herein) and the coefficients {Ji(1), Jij(2), Jijl(3), . . . } describe the strengths of single-body terms and (potentially long-range) products of up to k spin variables. We denote by K the number of non-zero coefficients in the above expression for the Hamiltonian function H. Further, nd denotes the number of spin-flip symmetries in the Hamiltonian function H.
Due to the two-body and quasi-local nature of physical interactions between constituents, such as qubits, of a quantum system, it is difficult, if not unfeasible, to implement couplings as occurring in Equation 1 directly in quantum hardware. In the present disclosure, we utilize the parity mapping described in EP 3 113 084 B1 and WO 2022/008057 A1, that allows to encode arbitrary higher-order long-range k-body terms into a Hamiltonian (the “total Hamiltonian” as described herein) of a quantum system comprising qubits (or more generally constituents) arranged on a square lattice, wherein the total Hamiltonian involves only short-range interactions. This involves mapping the k-fold product of a subset of problem spins si onto a single qubit (called herein “parity qubit”, denoted by {circumflex over (σ)}), e.g. Jijl(3)sisjslJm{circumflex over (σ)}z(m), where we label each parity qubit with the corresponding k-tuple of problem spin indices. That is to say, in the example given, the index m is shorthand for the 3-tuple ijl. As a result, the K non-zero interaction terms of Equation 1, which define the computational problem, are represented by K≥N single-body quantum operations of strength Jm acting on K respective parity qubits. The corresponding single-body Hamiltonian, referred to herein as problem Hamiltonian, has the form ĤP=Σm Jm {circumflex over (σ)}z(m) (wherein the terms Jmσz(m) are summand problem Hamiltonians as described herein). The Hilbert space formed by the K parity qubits, possibly supplemented with ancillary qubits, is denoted by
parity. In order to ensure that the optimization problem is embedded in the low-energy subspace (ground space) of
parity, K−N+nd constraint terms Ĉl are provided in addition to the problem Hamiltonian. Accordingly, a constraint Hamiltonian of the form ĤC=Σl Ĉl is provided, where Ĉl are referred to herein as summand constraint Hamiltonians, or constraint terms for short. This gives rise to a total Hamiltonian of the form
The solution of the computational problem is encoded in the ground space of the total Hamiltonian Ĥtotal. The summand constraint Hamiltonians Ĉl represent short-range interactions between small groups of qubits. Specifically, the summand constraint Hamiltonians are short-range three-, or four-body Hamiltonians acting on, or within, 2×2 plaquettes of qubits (cf.
with the constraint strength cl>0. Here, the li denote labels of qubits, such that all the indices of the corresponding problem spins appear an even number of times within each constraint term. The square brackets around {circumflex over (σ)}z(l-state (or |1
-state) per constraint.
For further details regarding the parity mapping, reference is made to EP 3 113 084 B1 and WO 2022/008057 A1.
The Quantum Approximate Optimization Algorithm (QAOA) is designed to find low energy solutions of the quantum mechanical implementation Ĥ of the Hamiltonian function H of Equation 1 (wherein Ĥ is obtained by replacing each classical spin si of H by a {circumflex over (σ)}z Pauli operator), by evolving a quantum state alternately with a unitary time evolution of a driver Hamiltonian ĤBΣiN{circumflex over (σ)}x(i) and a unitary time evolution of the Hamiltonian Ĥ for variable durations. To implement the QAOA in the parity architecture, the single-qubit driver Hamiltonian ĤX=ΣiN{circumflex over (σ)}x(i) may be chosen in an analogous manner, while the Hamiltonian Ĥ may be replaced by the total Hamiltonian Ĥtotal=ĤP+ĤC described above; see WO 2020/156680 A1. A parity-QAOA sequence of depth p thus corresponds to variationally evolving the quantum system with Hamiltonians ĤX, ĤP and ĤC as
where the variational parameters βj, γj and Ωj are optimized in a quantum-classical feedback loop (adaptive rounds of operations, as described herein), in order to minimize ψ|Ĥtotal|ψ
. As a consequence, during optimization, the constraint terms (summand constraint Hamiltonians) of the constraint Hamiltonian are treated on the same footing as the problem-encoding single-body terms (summand problem Hamiltonians) of the problem Hamiltonian. That is to say, as regards the constraint terms, it is the case that unitary time evolution operators of the form e−iΩĤ
Instead of enforcing the constraint Hamiltonian explicitly in the QAOA protocol by means of time evolution unitary operators of the form e−iΩĤ
and restrict the subsequent dynamics to that subspace by providing a driver Hamiltonian ĤXimp which is configured such that the constraint terms are conserved implicitly, in other words the time evolution defined by the driver Hamiltonian is such that the quantum system never leaves the constraint-fulfilling subspace, thereby removing the need for time evolution operators of the form e−iΩĤ
may be provided.
Generally, we may define the elements of a constraint-preserving driver Hamiltonian ĤXimp as follows.
Let us consider a set Qμ of qubits that can be simultaneously flipped without changing which constraint terms Ĉl are fulfilled (that a constraint term is “fulfilled” means that the quantum system is in a ground state of the constraint term). These qubits are typically arranged on the layout along a line (see for example
The index μ enumerates the driver lines for a given computational problem. With each driver line, we associate a driver term
which has the property
We refer to the number of qubits in a driver line as the length of the driver line. A set D of driver lines is called independent if and only if no element Qμ∈D can be obtained via symmetric difference of (multiple) other elements in D. Furthermore, we call a set D of driver lines valid if and only if D is independent and |D|=N−nd holds. The set of driver terms associated with a valid set of driver lines allows for all operations that correspond to flipping problem spins. Two driver lines Qμ and Qv are said to overlap if and only if Qμ∩Qv≠Ø.
In contrast to the QAOA approach enforcing all constraints explicitly, we now consider the performance of parity-QAOA utilizing a constraint-preserving driver Hamiltonian of the form
consisting of the operators associated with a valid set of constraint-preserving driver lines. In the following, we will use D to refer to a set of driver lines as well as to the set of its associated driver terms. Provided that we start from a constraint-fulfilling quantum state, a unitary time evolution defined by such a driver Hamiltonian only introduces transitions to other constraint-fulfilling quantum states and therefore restricts the dynamics to CF.
For the example shown in
In this context, the superscript (μv) denotes the qubit involving the problem spin indices μ and v, in accordance with the labelling in
Note that the constraint-preserving driver Hamiltonian ĤXimp commutes with all constraint terms Ĉl, thus ensuring that the number of violated constraints stays constant during time evolution. In particular, as long as the initial quantum state is prepared in the constraint-fulfilling subspace, this guarantees that the variational quantum state that results from the unitary time evolution corresponds to a valid configuration of the problem spins and is therefore a potential solution of the computational problem throughout the time evolution.
Using the constraint-preserving driver Hamiltonian ĤXimp, the variational QAOA-state can be prepared using the protocol
with |ψ0 being an appropriately chosen initial quantum state fulfilling all parity constraints Ĉl. Usually, |ψ0
is chosen to be the equal superposition of all constraint-fulfilling computational basis states; for details on the preparation of this state, see Sec. 4. In the QAOA protocol defined by Equation 11, we have completely removed the time evolution operators of the form e−iΩĤ
Implementing the unitary time evolution e−iβĤ
Let us start in the fully implicit implementation. Switching a constraint term from an implicit to an explicit implementation doubles the dimension of the reachable subspace. Therefore, the driver Hamiltonian can contain one additional term, which does not commute with the constraint term in question. Consider a set of nC constraint terms (referred to herein as the first subset of summand constraint Hamiltonians) which are to be explicitly enforced, i.e. by way of a unitary time evolution operator. The remaining constraint terms (referred to herein as the second subset of summand constraint Hamiltonians) are to be enforced implicitly, i.e. by providing a suitable driver Hamiltonian that conserves the constraints in question. We define the hybrid Hilbert space hyb as the space spanned by the computational basis states which fulfill all constraints terms that are to be enforced implicitly (i.e. all summand constraint Hamiltonians in the second subset). Note that
We now generalize the concepts of the implicit approach introduced in Sec. 2.2. A hybrid driver line Qμ is a set of qubits that can be simultaneously flipped without leaving the hybrid subspace hyb, given that the current state of the qubits lies inside that space. The index μ enumerates the hybrid driver lines for a given computational problem. In the same manner as described above for constraint-preserving driver lines, we associate a (hybrid) driver term {circumflex over (X)}(μ) with each hybrid driver line, corresponding to the product of all {circumflex over (σ)}x-operators acting on the qubits involved in the hybrid driver line in question. The definitions of independence and overlap of hybrid driver lines are analog to the definitions given in Sec. 2.2 for fully constraint-preserving driver lines. Note that just like before, the term “driver line” here is used irrespective of the actual geometrical arrangement of qubits in the layout.
A set D of hybrid driver lines is valid if and only if it is independent and any computational basis state in the constraint-fulfilling Hilbert space CF can transformed to any other by applying operators associated with driver lines in D only. This definition is less strict than the definition of validity for fully constraint-preserving driver lines: The set D can contain N−nd≤|D|≤N+nC−nd driver terms. Containing exactly N+nC−ndindependent elements is a sufficient but not necessary condition for a set of hybrid driver lines to be valid. It can contain less lines, provided that all constraint-fulfilling quantum states, i.e. all quantum states in the constraint-fulfilling Hilbert space
CF, can still be reached (and these are the only quantum states that we need to reach). In the following we will focus on the case|D|=N+nC−nd. The cases with less driver terms can be mapped back to this case by re-evaluating the partitioning of explicitly- and implicitly enforced constraint terms, as some of the originally explicitly enforced constraints will be naturally preserved by such a driver.
The hybrid driver lines are now sets of qubits that can be flipped simultaneously without violating any constraint terms, apart from the explicitly enforced constraint terms. The associated driver terms are defined analogous to Equation 7 and do not necessarily preserve the constraint fulfilling space CF but the hybrid Hilbert space
hyb.
In a computational problem with N classical spin variables and a constraint Hamiltonian having a total number of nCtot constraint terms of which nC are enforced explicitly, we can consider the hybrid driver Hamiltonian as
with driver terms {circumflex over (X)}(μ) associated with a valid set of hybrid driver lines. The first partial constraint Hamiltonian is defined by
and contains only the nC explicitly enforced constraint terms (i.e., only the summand constraint Hamiltonians from the first subset of summand constraint Hamiltonians). The problem Hamiltonian Ĥp encoding the computational problem in single-body terms is invariant with respect to the use of the original, the fully implicit or the hybrid protocol. (However, in contrast to the fully implicit implementation, in the hybrid protocol we can no longer associate the different driver terms with operations on the original problem spins.)
Note that the fully implicit and the fully explicit approach correspond to the limiting cases of the hybrid approach with nC=0 and nC=nCtot, respectively.
The hybrid QAOA-protocol is analogous to Equation 4, with replacements ĤX ĤXhyb and ĤC
ĤChyb and now reads
The initial quantum state |ψ0> can be chosen to be the equal superposition of all computational basis states in hyb (cf. Sec. 4).
In the following, we illustrate the strength of this new flexibility on the example of the complete graph and then show how this can be applied to arbitrary graphs.
Let us again consider a layout as shown in
Alternatively, we can also understand this construction in the following way: in the hybrid setting, the original driver lines of {circumflex over (X)}(μ({circumflex over (X)}(v), as the symmetric difference of two driver lines corresponds to the product of their associated driver terms. We thus arrive at a hybrid driver Hamiltonian
with driver terms
This corresponds to splitting the product in Equation 10 in two separate terms. We can parallelize the implementation of the above hybrid driver Hamiltonian by classifying the terms into groups A and B as in Equation 16. All terms in group A correspond to the hybrid drives lines depicted as solid lines in
Compiling general graphs (or hypergraphs) to the parity architecture can lead to a variety of placements of three- and four-body constraints. In the case of only four-body constraint terms, we can construct a hybrid driver Hamiltonian which conserves all constraint terms from only horizontal and vertical straight lines, as can be seen in
With additional optimization, it is possible to further reduce the number of explicitly enforced constraint terms: Some of the three-body constraint terms are automatically conserved by the above mentioned horizontal and vertical driver lines. In other cases, a small adjustment to the hybrid driver lines (for example adding a small set of additional qubits) can suffice to conserve even more three-body constraint terms. An example of such an optimization can be seen in
A more detailed discussion on how to choose which constraint terms should be enforced explicitly or implicitly and to find a valid set of hybrid driver lines is given in Appendix B.
With the procedure described in the previous section and in Appendix B, the average length of hybrid driver lines (and therefore the depth of the QAOA-circuit) may grow linearly with the device dimensions, i.e. with the number of qubits in the quantum system. We now utilize the concept of implicitly and explicitly enforced constraints to impose an upper bound for the length of the hybrid driver lines and thereby restrict the driver circuit depth to an adjustable constant, while avoiding the fully explicit approach.
In order to restrict the length of the hybrid driver lines, we introduce additional (typically equidistant) rows and columns of explicitly enforced constraint terms with a maximum spacing lmax. In
We refer to a portion of the layout surrounded by explicitly enforced constraints as a module, also referred to herein as a “subsystem” of the quantum system. For example, in
Further, the length of the hybrid driver lines within a module is limited. In particular, if all three-body constraint terms in the module are enforced explicitly (i.e. there are only strictly vertical and horizontal lines), the length of a hybrid driver line can be at most lmax.
The quantum circuits implementing the unitary time evolutions of the respective driver terms for each module can be executed at the same time. Therefore, with this approach, the circuit depth of the driver Hamiltonian implementation scales linearly with lmax, which is a user-determined quantity and can be chosen according to current needs. Thus, the circuit depth of the driver Hamiltonian implementation is a constant, independent of the problem- and device size.
Even in the more general case of enforcing some of the three-body constraint terms within a module implicitly, the problem of finding appropriate hybrid driver lines now reduces to smaller, separate problems for each module and the lengths of the hybrid driver lines will still be approximately lmax.
Thus, in
The implementations of e−iγĤ
As an initial quantum state for the optimization procedure, we may use the ground state of the (negative of the) driver Hamiltonian, corresponding to the equal superposition of all computational states spanning the considered Hilbert space parity,
hyb or
CF (fulfilling all implicitly enforced constraints), depending on whether the fully explicit, the hybrid or the fully implicit approach is used. While this can be easily achieved in the purely explicit approach by preparing each physical qubit in the equal superposition of the computational basis states, the initial state preparation is more challenging for the implicit and especially the hybrid approach. Consider a general hybrid driver Hamiltonian ĤXhyb involving a valid set D of hybrid driver lines, also including the limiting cases of fully implicit and explicit parity QAOA. The quantum state we wish to create is then the simultaneous eigenstate (with eigenvalue+1) of all driver terms in ĤXhyb and all constraint terms in ĤChyb. Since the quantum state in question is a stabilizer state, known methods for preparing stabilizer states can be used to construct a quantum circuit generating the stabilizer state in question from a product quantum state. The resulting circuits might result in large circuit depths on architectures with limited connectivity.
In the following, we propose a procedure to prepare the initial quantum state with a low circuit depth scaling linearly with the module size. We start by considering |D|=log2 dim(hyb) driver qubits to represent the states of the considered Hilbert space (satisfying all implicitly enforced constraints). This can be done by defining one qubit for every driver line, such that the associated driver term {circumflex over (X)}(μ) acts as the bit-flip operator for that driver qubit. In this picture, our desired initial state corresponds to all driver qubits being in the |+
-state.
We want to define an operator {circumflex over (Z)}(μ) corresponding to the phase-flip operator on driver qubit μ. In order for this to be a valid construction, our newly defined operators must fulfill the (anti-)commutation relations
for μ≠v. For a single hybrid driver line Qμ, it is easy to show that any operator σz(k) acting on a qubit k∈Qμ fulfills the desired commutation relations with the {circumflex over (X)}-rotation on the same driver line. As long as this qubit is not involved in any other driver line (i.e. ∃≠μ: k∈Qv), this remains a valid choice. In the fully implicit implementation depicted in
The goal is to prepare a state where every driver qubit μ is in the {circumflex over (X)}(μ)-eigenstate |+We start from the easily preparable and constraint-fulfilling state |↑
⊗K (in the space
parity), which corresponds to all driver qubits being in the state |↑
as well. To prepare the desired state from this, we have to perform the operations on the quantum system
parity corresponding to a π/2-rotation around the y-axis on all driver qubits, which can be decomposed into consecutive rotations
and thus implemented with the previously defined operators.
However, problems arise when a qubit is involved in multiple driver lines. Implementing a physical {circumflex over (σ)}z-operation on such a qubit has an effect on all involved driver lines and thus can introduce unwanted cross-talk which needs to be avoided. Whenever possible, we must therefore choose a qubit which is not involved in any other driver lines, to perform the phase operation on. This is possible for the fully implicit case, as shown in
If this is not possible, the {circumflex over (Z)}(μ)-operation on a qubit k for a driver line Qμ∃k can still be performed as long as all driver qubits associated with other driver lines Qv involving qubit k are in an eigenstate of {circumflex over (Z)}(v) and thus not affected by the rotation. In the initially prepared state |↑⊗K, all driver qubits are in the {circumflex over (Z)}-eigenstate. That enables us to find a sequence of driver rotations such that for every {circumflex over (Z)}(μ)-rotation there is at least one qubit of the corresponding driver line which is either not included in any other driver lines, or only involved in driver lines whose state has not been rotated yet.
This procedure allows to prepare the desired superposition state even for more general hybrid driver Hamiltonians. The circuit depth for state preparation scales similar to the implementation of the unitary for the time-evolution under a single driver Hamiltonian. Exact instructions for arbitrary layouts are provided in Appendix C.
Starting with the relative amount of explicitly enforced constraint terms nC/nCtot=0, the circuit depth grows linearly with the system size. The large coefficient in the circuit depth scaling is due to the excessive overlapping of driver lines in the fully implicit setup (
In order to demonstrate the advantages of this new approach, we compare the performance of the fully implicit, the hybrid and the fully explicit QAOA-approach in the parity scheme. The QAOA-parameters ranging in [0,2π) for p=3 QAOA-cycles are randomly initialized nreps=100 times. Note that for the fully implicit approach there is one QAOA-parameter less per cycle as the constraint part has been removed. For each initialization we use the following classical procedure to find a local optimum: We perform an update of a random QAOA-parameter. If the energy of the system decreases after a parameter update, the new parameters are accepted with probability paccept=1 and otherwise with a probability exponentially decreasing with the energy increase caused by the new parameters. This procedure is repeated until the objective value converges. Out of the nreps initializations the lowest energy expectation value E=ψ|Ĥphys|ψ
[cf. Equation 4] for the respective instance is kept. We calculate the residual energy Eres of the system state, defined as
after the optimization, as a function of the number of explicitly enforced constraints. In Equation 19, Emax and Emin denote the highest and lowest energy in the configuration space, respectively. The described procedure is applied for complete graphs with N∈{4, 5, 6} problem spins. The results are shown in
Clearly, the residual energy increases with increasing number of explicitly enforced constraints, which is related to the fact that also the size of the feasible subspace increases with the number of explicitly enforced constraints. Note that the simulations discussed in this section do not consider effects of quantum noise such as bit-flip errors or decoherence.
In summary, we have shown how to improve the parity-QAOA performance by interpolating between the standard single-qubit driver Hamiltonian to a driver Hamiltonian tailored to the computational problem. This proposed hybrid approach keeps the parallelizability of the fully explicit parity-QAOA while gaining performance, by reducing the search space. Given a fixed hardware layout, the trade-off between circuit depth and the size of the search space can be dynamically changed by tuning the size of implicitly driven and explicitly inter-connected submodules.
The ideas presented here can be readily realized on any grid arrangement of the qubits. This is necessary to address questions about the practical QAOA performance using modularized layouts for large problem sizes inaccessible for classical simulations.
In the following, we outline a general algorithm to find valid hybrid driver lines (we just call them lines in the following). Certain steps can be improved (solution is found faster, solution requires less explicitly enforced constraints or solution leads to smaller circuit depth) depending on how constraints/qubits/driver lines are picked, but any choice works in principle. Furthermore, the limitations for lines can follow different criteria, but they do not change the approach of the algorithm. Start with all constraints being implicitly enforced, apart from a grid of explicitly enforced constraints for modularization, as described in Sec. 3.3. Go through every module (connected set of implicitly enforced constraints, connected meaning connected through adjacency, diagonal/left/right/up/down):
At the latest, one will find a valid and acceptable (not exceeding the preset limits) set of driver lines once (a) all three-body constraint terms are explicitly enforced and connected to the boundary via explicitly enforced constraint terms and (b) the length limit is larger or equal to the maximal dimension of the modules (e.g. if the module contains 3×4 qubits, the maximal line length is 4). This is true even when not allowing branching or bending. If branching or bending is allowed, then one will find valid acceptable lines earlier.
The following instructions are valid for fully implicit, fully explicit and the general hybrid driver lines, but only required in the truly hybrid case, as the other cases are trivial:
First, assign an implementation priority Pμ determining the implementation order to every driver line Qμ, such that:
for each driver qubit λ in descending order of their priorities Pλ. Equal priorities can be implemented in any order, their required gate sequences can be performed in parallel (or as parallel as possible, if there are qubit overlaps of the driver lines). The priorities of the lines can be found iteratively, we call every line “unassigned” until it has been assigned a priority:
The initial state |ψ0> can then be prepared as
where Pmax is the highest assigned priority and Dκ⊆D is the subset of driver lines with priority κ. Note that the order of the products must be such that the terms with higher priority are applied first.
If with this procedure, not all lines can be assigned a priority, update the explicit/implicit grouping of the constraints according to Appendix B and try again. At the latest, it will work when all three-body constraints are explicitly enforced.
While the foregoing is directed to embodiments, other and further embodiments may be devised without departing from the scope determined by the claims.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2022/054557 | 2/23/2022 | WO |
