WO2016011440A1 - Quantum processor schedule control - Google Patents
Quantum processor schedule control Download PDFInfo
- Publication number
- WO2016011440A1 WO2016011440A1 PCT/US2015/041090 US2015041090W WO2016011440A1 WO 2016011440 A1 WO2016011440 A1 WO 2016011440A1 US 2015041090 W US2015041090 W US 2015041090W WO 2016011440 A1 WO2016011440 A1 WO 2016011440A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- hamiltonian
- quantum
- computer
- states
- subspace
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N10/00—Quantum computing, i.e. information processing based on quantum-mechanical phenomena
Definitions
- This description relates to the solution of problems (e.g., binary optimization problems) by using quantum computers, and more particularly to controlling a quantum computer according to an evolution schedule.
- problems e.g., binary optimization problems
- PUBO Polynomial Unconstrained Binary Optimization
- a cost function that receives an N-bit configuration as input can be written according to the above expression.
- the coefficients c s completely specify the cost function / .
- Many problems that are not initially formulated as optimization problems can be rephrased in a form compatible with the minimization of a cost function. In this case as well, the solution of the optimization problem is the N-bit configuration that minimizes
- each problem variable x i corresponds to a physical element (usually a quantum two-level system) of the physical machine, typically referred to as a qubit.
- a single variable x i is associated with several physical qubits.
- H p (f) can always be expressed as
- Adiabatic quantum computation is based on the adiabatic quantum theorem: if a system is initially in the ground state of a certain Hamiltonian that is later changed continuously in time, the system will dynamically evolve to be, at any instant, in the ground state of the instantaneous Hamiltonian. This is correct only if the Hamiltonian is changed in time slowly enough to be considered adiabatic.
- AQC consists in slowly changing the following time-dependent Hamiltonian
- H D typically referred to as the "driver" Hamiltonian, for which the ground state is known and its preparation is practical on the physical machine.
- H D also has the property of not commuting with H p .
- the schedule function s(t) controls the quantum computer, and, together with H D and H p , determines the probability of success of the computation.
- the evolution time T (which is given by s -1 (1) ) determines how long the quantum computation takes to complete the adiabatic optimization.
- the evolution time T should be chosen as small as possible to reduce the adiabatic computational time.
- T should be selected to be sufficiently long so that the quantum computer evolution is adiabatic and it is therefore ensured that the system remains in its instantaneous ground state at any time.
- a method for controlling a quantum computer to solve a problem includes: accepting a first data representation of a problem Hamiltonian corresponding to a final quantum state of the quantum computer, where a measurement of the final quantum state at an end time provides at least one solution to the problem; accepting a second data representation of a driver Hamiltonian corresponding to an initial quantum state of the quantum computer at a start time; applying a computer- implemented procedure to determine a schedule function that determines an interpolation of a total Hamiltonian from the driver Hamiltonian to the problem Hamiltonian at multiple times between the start time and the end time, the procedure including: forming a third data representation of the total Hamiltonian from a combination of the first data representation and the second data representation, determining multiple states from the third data representation, forming an overlap matrix from the multiple states, the overlap matrix effectively defining a Hilbert space, and determining a spectrum of the total Hamiltonian restricted to a subspace of the Hilbert space, where the dimensionality of the subspace is smaller
- aspects can include one or more of the following features.
- the procedure further includes forming a block diagonal matrix from the overlap matrix.
- Forming a block diagonal matrix from the overlap matrix includes transforming a basis of the overlap matrix.
- the subspace corresponds to a selected block of the block diagonal matrix.
- the block diagonal matrix contains only a single block that is larger than 1 x 1 in size, which is the selected block.
- Determining the spectrum of the subspace includes determining eigenvalues of the selected block.
- the procedure further includes determining, based on the determined spectrum, a spectral gap at each of the multiple times between the start time and the end time, the spectral gap at a particular time corresponding to a difference between a ground state energy and a first exited state energy of an effective Hamiltonian corresponding to the selected block.
- Controlling the time evolution of the configuration of the quantum computer between the start time and the end time according to the determined schedule function includes varying the total Hamiltonian more slowly when the spectral gap is smaller and more quickly when the spectral gap is larger.
- the third data representation includes: (1) a first term representing a first Hamiltonian that is highly degenerate with M distinct energy levels, and (2) a second term representing a second Hamiltonian that is a sum of k projectors on k respective pure quantum states.
- At least one of the first Hamiltonian or the second Hamiltonian includes contributions from both the driver Hamiltonian and the problem Hamiltonian.
- Both of the first Hamiltonian and the second Hamiltonian include contributions from both the driver Hamiltonian and the problem Hamiltonian.
- Determining multiple states from the third data representation includes determining k M states corresponding to the k pure quantum states of the second Hamiltonian projected onto each of multiple eigensubspaces of the first Hamiltonian.
- the Hilbert space has a dimensionality of 2 N and the subspace has a dimensionality of IM , where k and / and M are integers and / ⁇ k .
- a quantum state representing the quantum computer and subject to the total Hamiltonian undergoes adiabatic dynamics that are confined to the subspace.
- a quantum computing system comprises a scheduling computer configured to control a quantum computer by performing all the steps of any of the methods.
- software comprises instructions stored on a non- transitory computer-readable medium for causing a scheduling computer to control a quantum computer by performing all the steps of any of the methods.
- the (sub-)optimal schedules are devised by analyzing the spectral gap of the AQC quantum Hamiltonian.
- the techniques described herein are not limited to these particular classes of problems. Instead, the effective dimensionality can be reduced without relying on any explicit symmetry.
- the techniques also go beyond the strict distinction of driver and problem contributions to the Hamiltonian, which is made possible by the recognition of a hidden block diagonal structure in the total Hamiltonian that leads to the identification of a small subspace in which the relevant part of the energy spectrum, and in particular the energy gap between ground state and first excited state, can be efficiently computed. The relevant eigenstates are effectively confined to this small subspace.
- a scheduling computer can be configured to efficiently and exactly compute, at any point during the adiabatic process and at zero temperature, the energy spectrum of an effective Hamiltonian.
- the computation can be performed using quantities that can be computed exactly or using efficient numerical approaches that require only a polynomial amount of classical resources.
- FIG. 1 is a block diagram of a quantum computing system.
- a strategy for determining a schedule that is closer to optimal than a simple linear schedule is based on additional characteristics of the specific problem, or class of problems, to be solved.
- the strategy includes choosing s(t) to be the solution of the following differential equation with the energy gap between the ground state and the first excited state of H AQC at fixed and a small parameter. The total evolution time T is chosen accordingly as
- the calculation of the spectrum of an arbitrary H AQC is accomplished by identifying a convenient splitting of this total Hamiltonian into two terms representing distinct contributions that simplify at least some of the subsequent operations.
- the two terms are defined using the two Hamiltonians H A (s) and H B (s)
- H is a highly degenerate Hamiltonian with only M distinct energy levels. is the sum of k projectors on pure quantum states (also called "rank-1"
- H A (s) corresponds to either H p or H D .
- the spectrum of the total Hamiltonian H AQC (t) is calculated by using information extracted from the Hamiltonians H A (s ) and H B (s ) , as described in more detail below.
- a problem specification 102 including data representing a problem Hamiltonian and a driver Hamiltonian are provided to a (Classical) scheduling computer 104.
- the scheduling computer 104 includes at least one processor 106 that executes a computer program 108 for generating the schedule data 1 10 that will be used to control a quantum computer 1 12.
- the scheduling computer 104 executes the program 108 to apply an analysis procedure to determine the minimum spectral gap for the input
- a controller 1 14 in the quantum computer 1 12 controls signals coupled to an arrangement of qubits 1 16 of the quantum computer 1 12 according to the schedule data 1 10, and ultimately provides a problem solution 1 18 according to a final configuration of the arrangement of qubits 1 16 of the quantum computer 1 12 after operation for the time T .
- operation of the quantum computing system 100 includes selecting an evolution time T , and then operating the quantum computer during an interval of time between 0 and T according to an evolution schedule function s ⁇ t).
- a schedule function represents the time-variation of the Hamiltonian that governs the dynamics of the quantum computer 1 12.
- the change of the Hamiltonian is given by a time-variation of the degree of combination of a first Hamiltonian representing a default initial choice (e.g., a "driver" Hamiltonian) and a second Hamiltonian encoding the problem to be solved.
- the evolution schedule may be determined based on data representations of the first and the second Hamiltonians
- the computations performed by the scheduling computer 104 include determining the evolution schedule based on data representations of modified Hamiltonians H A (s) and H B (s ) , which enable additional efficiency in the Classical computation, as described in more detail below.
- a first step is projecting each of the k states involved in H B (s) on the eigensubspaces of H A (s) . This results in kM states, labelled by ⁇
- This calculation involves the normalization of the projected states.
- a second step includes calculation of an overlap matrix, whose entries are expressed below.
- the overlap matrix has dimensions (k M) x (k M) .
- Preliminary input representation of H A (s) is available in at least the following three different ways: analytical analysis of the problem that gives its "a priori" characterization, Monte Carlo simulations in the form of entropic sampling, or possible future methods and data algorithms to evaluate the necessary properties.
- a third step includes extracting from the ⁇ E a ) states a smaller set of linearly independent, pairwise orthogonal states ⁇ ⁇ ⁇ ) ⁇ This is achieved by applying the
- a fourth step to obtain the desired basis is to complete the set ⁇ ⁇ ⁇ ⁇ ⁇ ⁇ into a basis of the complete space. This can be done by adding pairwise orthogonal states belonging to the eigensubspaces of H A (s) and orthogonal to all ⁇ ⁇ ⁇ ) .
- the Hamiltonian restricted to the subspace generated by the states ⁇ ⁇ ⁇ ) is called the "effective" Hamiltonian.
- the relevant part of the energy spectrum of the total Hamiltonian H AQC (t) is obtained by solving the eigenvalue problem associated with the effective Hamiltonian at any instant (in practice every small step As ) during the adiabatic evolution.
- Some AQC or quantum annealing machines operate according to a fixed, problem independent, schedule.
- the techniques described herein provide a schedule that is closer to an optimal schedule, tailored either on the specific problem of interest or on its specific class of problems.
- the driver Hamiltonian H D can assume a variety of forms, but only a few regularly appear in the literature:
- the "Grover style” driver Hamiltonian (or simply Grover driver Hamiltonian),
- H A and H B search for those in which H A is a highly degenerate Hamiltonian with only M distinct energy levels and H B consists of the sum of k rank-1 projectors, namely
- ⁇ £ is the subspace associated
- Hamiltonian H eS (s) results to be an effective (K x M) -level Hamiltonian
- H AQC (i) presents a peculiar block diagonal structure when represented in such basis: the only block with dimension larger than l x l is a (KM) x (KM) block exactly reproduced by
- the procedure selects a highly degenerate H A with only M distinct energies, and a Hamiltonian H B formed by a small number k of rank-1 projectors.
- the relevant part of the energy spectrum of H AQC (t) can be obtained assuming an effective system.
- the set ⁇ ⁇ is by construction contained in the kernel of H B , such that
- Such matrix is a square matrix with linear dimension k and can be shown to be Hermitian and positive- semidefinite. Therefore it admits a Cholesky decomposition:
Landscapes
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Mathematical Analysis (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Condensed Matter Physics & Semiconductors (AREA)
- Computational Mathematics (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
An approach to controlling a quantum computer provides a technical advantage by reducing the computation time. Such shorter computation time is obtained by controlling the schedule that determines the instantaneous Hamiltonian for the quantum computer. The approach uses characterization of a problem and driver Hamiltonian that allows computationally efficient calculation of the schedule, which is based on reducing the effective dimensionality of the quantum system.
Description
QUANTUM P RO C E S S O R S CHEDULE C ONTRO L
Cross-reference to Related Applications
[001] This application claims the benefit of U.S. Provisional Application No.
62/026,090 filed on July 18, 2014, which is incorporated herein by reference.
Statement as to Federally Sponsored Research
[002] This invention was made with government support under contract number FA9550-12-1-0046 awarded by the United States Air Force Office of Scientific Research and contract number DE-SC0001088 awarded by the Department of Energy. The government has certain rights in the invention.
Background
[003] This description relates to the solution of problems (e.g., binary optimization problems) by using quantum computers, and more particularly to controlling a quantum computer according to an evolution schedule.
[004] Approaches described in this document primarily focus on analogic adiabatic quantum computation. An example of such an approach, and a physical realization of a quantum computer to host such an approach, is described in US Pat. Pub. 2010/0306142, titled "Methods for Adiabatic Quantum Computation." This publication, is incorporated herein by reference.
[005] The class of binary optimization problems referred to as Polynomial Unconstrained Binary Optimization (PUBO) consists in finding an optimal assignment y of N bits, where such that y minimizes a certain N-bit cost function
unique representation as:
[006] A cost function that receives an N-bit configuration as input can be written according to the above expression. The coefficients cs completely specify the cost function / .
[007] Many problems that are not initially formulated as optimization problems can be rephrased in a form compatible with the minimization of a cost function. In this case as well, the solution of the optimization problem is the N-bit configuration that minimizes
[008] The coefficients cs also specify a "problem" Hamiltonian Hp (f) , used to configure the quantum computer (namely the physical machine). In many common implementations and configurations of a quantum computer, each problem variable xi corresponds to a physical element (usually a quantum two-level system) of the physical machine, typically referred to as a qubit. In some implementations, a single variable xi is associated with several physical qubits.
[010] where | x) represents the quantum state associated with the N-bit configuration x .There is a one-to-one correspondence between the cost function / and the problem Hamiltonian Hp(f) . Accordingly, the problem is solved when the state corresponding to the ground state (namely the lowest energy state) of Hp{f) is produced and measured.
[011] Adiabatic quantum computation (AQC) is based on the adiabatic quantum theorem: if a system is initially in the ground state of a certain Hamiltonian that is later changed continuously in time, the system will dynamically evolve to be, at any instant, in the ground state of the instantaneous Hamiltonian. This is correct only if the Hamiltonian is changed in time slowly enough to be considered adiabatic.
[012] Several criteria have been proposed to ensure that the evolution is adiabatic, i.e. to verify that the Hamiltonian is changed slowly enough. The relevant conditions for fulfilling these criteria are described below.
with HD typically referred to as the "driver" Hamiltonian, for which the ground state is known and its preparation is practical on the physical machine. HD also has the property of not commuting with Hp .
[014] It should be understood that the techniques described in this document are not limited to a particular choice of HD .
[015] The adiabatic schedule function s(t) is an arbitrary function such that s(0) = 0 and s(T) = 1 , where T is the total evolution time. The schedule function s(t) controls the quantum computer, and, together with HD and Hp , determines the probability of success of the computation.
[016] The evolution time T (which is given by s-1(1) ) determines how long the quantum computation takes to complete the adiabatic optimization.
[017] On one hand, the evolution time T should be chosen as small as possible to reduce the adiabatic computational time. On the other hand, T should be selected to be sufficiently long so that the quantum computer evolution is adiabatic and it is therefore ensured that the system remains in its instantaneous ground state at any time.
dependent and it is defined as the minimum energy gap between the ground state and the first excited state of HAQC during the whole adiabatic computation.
[019] Another choice, for example adopted by the D-wave machines (as described in Johnson et at. Nature, 473(7346): 194-198, May 201 1 , incorporated herein by reference), is to follow a non-linear schedule that is problem independent, but specific to the hardware.
[020] Both procedures described in the two previous paragraphs have proven to be not optimal in general.
Summary
[021] In one aspect, in general, a method for controlling a quantum computer to solve a problem includes: accepting a first data representation of a problem Hamiltonian corresponding to a final quantum state of the quantum computer, where a measurement of the final quantum state at an end time provides at least one solution to the problem; accepting a second data representation of a driver Hamiltonian corresponding to an initial quantum state of the quantum computer at a start time; applying a computer- implemented procedure to determine a schedule function that determines an
interpolation of a total Hamiltonian from the driver Hamiltonian to the problem Hamiltonian at multiple times between the start time and the end time, the procedure including: forming a third data representation of the total Hamiltonian from a combination of the first data representation and the second data representation, determining multiple states from the third data representation, forming an overlap matrix from the multiple states, the overlap matrix effectively defining a Hilbert space, and determining a spectrum of the total Hamiltonian restricted to a subspace of the Hilbert space, where the dimensionality of the subspace is smaller than the dimensionality of the Hilbert space; and controlling a time evolution of a configuration of the quantum computer between the start time and the end time according to the determined schedule function.
[022] Aspects can include one or more of the following features.
[023] The procedure further includes forming a block diagonal matrix from the overlap matrix.
[024] Forming a block diagonal matrix from the overlap matrix includes transforming a basis of the overlap matrix.
[025] The subspace corresponds to a selected block of the block diagonal matrix.
[026] The block diagonal matrix contains only a single block that is larger than 1 x 1 in size, which is the selected block.
[027] Determining the spectrum of the subspace includes determining eigenvalues of the selected block.
[028] The procedure further includes determining, based on the determined spectrum, a spectral gap at each of the multiple times between the start time and the end time, the spectral gap at a particular time corresponding to a difference between a ground state energy and a first exited state energy of an effective Hamiltonian corresponding to the selected block.
[029] Controlling the time evolution of the configuration of the quantum computer between the start time and the end time according to the determined schedule function includes varying the total Hamiltonian more slowly when the spectral gap is smaller and more quickly when the spectral gap is larger.
[030] The third data representation includes: (1) a first term representing a first Hamiltonian that is highly degenerate with M distinct energy levels, and (2) a second term representing a second Hamiltonian that is a sum of k projectors on k respective pure quantum states.
[031] At least one of the first Hamiltonian or the second Hamiltonian includes contributions from both the driver Hamiltonian and the problem Hamiltonian.
[032] Both of the first Hamiltonian and the second Hamiltonian include contributions from both the driver Hamiltonian and the problem Hamiltonian.
[033] Determining multiple states from the third data representation includes determining k M states corresponding to the k pure quantum states of the second Hamiltonian projected onto each of multiple eigensubspaces of the first Hamiltonian.
[034] The Hilbert space has a dimensionality of 2N and the subspace has a dimensionality of IM , where k and / and M are integers and / < k .
[035] A quantum state representing the quantum computer and subject to the total Hamiltonian undergoes adiabatic dynamics that are confined to the subspace.
[036] In another aspect, in general, a quantum computing system comprises a scheduling computer configured to control a quantum computer by performing all the steps of any of the methods.
[037] In another aspect, in general, software comprises instructions stored on a non- transitory computer-readable medium for causing a scheduling computer to control a quantum computer by performing all the steps of any of the methods.
[038] Aspects can have one or more of the following advantages.
[039] The techniques described herein are able to identify either optimal or sub- optimal schedules that are better than both the linear schedule and the problem- independent schedule, providing superior performance for AQC in many applications.
[040] The (sub-)optimal schedules are devised by analyzing the spectral gap of the AQC quantum Hamiltonian.
[041] The calculation of the spectral gap is impossible without reducing the dimensionality of the AQC Hamiltonian for the desired problem. Therefore, we provide
techniques to effectively and efficiently reduce the dimensionality of AQC Hamiltonians.
[042] Unlike previous techniques which are based on the existence of explicit symmetries of the AQC Hamiltonian, the techniques described herein are not limited to these particular classes of problems. Instead, the effective dimensionality can be reduced without relying on any explicit symmetry. The techniques also go beyond the strict distinction of driver and problem contributions to the Hamiltonian, which is made possible by the recognition of a hidden block diagonal structure in the total Hamiltonian that leads to the identification of a small subspace in which the relevant part of the energy spectrum, and in particular the energy gap between ground state and first excited state, can be efficiently computed. The relevant eigenstates are effectively confined to this small subspace. Using this more efficient representation, a scheduling computer can be configured to efficiently and exactly compute, at any point during the adiabatic process and at zero temperature, the energy spectrum of an effective Hamiltonian. The computation can be performed using quantities that can be computed exactly or using efficient numerical approaches that require only a polynomial amount of classical resources.
[043] Other features and advantages of the invention are apparent from the following description, and from the claims.
Description of Drawings
[044] FIG. 1 is a block diagram of a quantum computing system.
Description
[045] A strategy for determining a schedule that is closer to optimal than a simple linear schedule is based on additional characteristics of the specific problem, or class of problems, to be solved. In some implementations, the strategy includes choosing s(t) to be the solution of the following differential equation
with the energy gap between the ground state and the first excited state of H AQC at fixed and
a small parameter. The total evolution time T is chosen accordingly as
[046] In both the linear case and the nonlinear case described above, the spectrum of HAQC is needed to compute the schedule for controlling the quantum computer.
[047] In one aspect, the calculation of the spectrum of an arbitrary H AQC is accomplished by identifying a convenient splitting of this total Hamiltonian into two terms representing distinct contributions that simplify at least some of the subsequent operations. The two terms are defined using the two Hamiltonians HA (s) and HB (s)
(with time dependence of s being implicit), where their sum with appropriate coefficients is able to reproduce the total Hamiltonian H AQC (t) , corresponding to each value of s{t) , as follows
[048] The choice of HA (s ) and is made according to the following criteria.
[049]
H is a highly degenerate Hamiltonian with only M distinct energy levels. is the sum of k projectors on pure quantum states (also called "rank-1"
projectors, or "single-state" projectors, or "unidimensional" projectors).
[050] It should be understood that the techniques described in this document are not limited to the particular case in which HA (s) corresponds to either Hp or HD . The spectrum of the total Hamiltonian H AQC (t) is calculated by using information extracted from the Hamiltonians HA (s ) and HB (s ) , as described in more detail below.
[051] Under some quite general conditions, the quantity of classical computational resources used by these techniques to compute the spectrum of HAQC grows only polynomially with N as opposed to direct methods involving the solution of eigenvalue problems for matrices with dimensions
With the techniques described herein, the required computational resources always scale polynomially in cases where both k and M are polynomial functions of N.
[052] Referring to FIG. 1 , in one implementation of a quantum computing system
100, a problem specification 102, including data representing a problem Hamiltonian and a driver Hamiltonian are provided to a (Classical) scheduling computer 104. The scheduling computer 104 includes at least one processor 106 that executes a computer program 108 for generating the schedule data 1 10 that will be used to control a quantum computer 1 12. The scheduling computer 104 executes the program 108 to apply an analysis procedure to determine the minimum spectral gap for the input
Hamiltonians, and use this minimum spectral gap to determine schedule data (e.g., including a representation of the schedule function s(t) at multiple times between a start time of t = 0 and an end time equal to the a total evolution time T ) that is passed to the quantum computer 1 12. A controller 1 14 in the quantum computer 1 12 then controls signals coupled to an arrangement of qubits 1 16 of the quantum computer 1 12 according to the schedule data 1 10, and ultimately provides a problem solution 1 18 according to a final configuration of the arrangement of qubits 1 16 of the quantum computer 1 12 after operation for the time T .
[053] In embodiments in which the quantum computer 1 12 is an adiabatic quantum computer, operation of the quantum computing system 100 includes selecting an evolution time T , and then operating the quantum computer during an interval of time between 0 and T according to an evolution schedule function s{t). Such a schedule function represents the time-variation of the Hamiltonian that governs the dynamics of the quantum computer 1 12. The change of the Hamiltonian is given by a time-variation of the degree of combination of a first Hamiltonian representing a default initial choice (e.g., a "driver" Hamiltonian) and a second Hamiltonian encoding the problem to be solved. While the evolution schedule may be determined based on data representations of the first and the second Hamiltonians, in some embodiments, the computations performed by the scheduling computer 104 include determining the evolution schedule based on data representations of modified Hamiltonians HA (s) and HB (s ) , which enable additional efficiency in the Classical computation, as described in more detail below.
[054] When the scheduling computer 104 has been configured with the total Hamiltonian H AQC (t) divided into the distinct contributions HA (s ) and HB (s ) , a sequence of computations are performed to obtain a basis in which the total Hamiltonian results in a block diagonal matrix for every s. The data representation of
the only relevant block is obtained by selecting the only block whose size is larger than l x l .
[055] While the specific equations used to perform these computations are described in greater detail below, the following is a high-level summary of the steps involved in the schedule generation program 108. A first step is projecting each of the k states involved in HB (s) on the eigensubspaces of HA (s) . This results in kM states, labelled by { | Ea) } , so that those states belonging to different eigensubspaces of HA (s) are orthogonal, while the k states belonging to the same eigensubspace are not necessarily orthogonal. This calculation involves the normalization of the projected states.
[056] A second step includes calculation of an overlap matrix, whose entries are expressed below. The overlap matrix has dimensions (k M) x (k M) .
[057] The normalization of the \ Ea } states and the construction of their overlap matrix make use of a preliminary representation of the Hamiltonian H A (s) , in particular of the structure of its eigensubspaces. In several cases of interest (for example when HB (s) corresponds to a Graver style driver Hamiltonian and HA (s) is a generic problem matrix, a representation of the density of states of HA (s) is sufficient. For more general situations, multi-point correlation functions at fixed energy have to be provided in addition to the density of states.
[058] Preliminary input representation of H A (s) is available in at least the following three different ways: analytical analysis of the problem that gives its "a priori" characterization, Monte Carlo simulations in the form of entropic sampling, or possible future methods and data algorithms to evaluate the necessary properties.
[059] A third step includes extracting from the \ Ea) states a smaller set of linearly independent, pairwise orthogonal states \ εα)■ This is achieved by applying the
Cholesky decomposition to the overlap matrix. Since the overlap matrix is a Hermitian positive-semidefmite matrix, its Cholesky decomposition may be not unique, but it always exists. This reflects a certain freedom in choosing the states \ εα) , but their number is always the same and never larger than k M .
[060] A fourth step to obtain the desired basis is to complete the set { \ εα} } into a basis of the complete space. This can be done by adding pairwise orthogonal states belonging to the eigensubspaces of HA (s) and orthogonal to all \ εα) .
[061] The only subspace relevant to AQC is the one spanned by the states \ εα) . When the total Hamiltonian H AQC (t) is represented in the basis constructed in the four steps above, this relevant subspace appears as a separate block along the diagonal of the resulting block diagonal matrix. In fact, this block is the only block with dimension larger than 1 x 1 .
[062] The relevant diagonal block of the total Hamiltonian H AQC (t) is now represented by a matrix of dimensions smaller than (k M) x (k M) , where k is the number of states on which the Hamiltonian HB (s) acts non-trivially and M is the number of distinct energies composing H A (s) . While in the most generic case (kM^ can be exponential in N, if we consider HA (s) to be the problem Hamiltonian corresponding to many concrete cases (3 SAT, QUBO, Max-2SAT, Exact Cover, . ..), then M is polynomial (quadratic or cubic) in N. In this case, the value of k depends on the explicit choice of the driver Hamiltonian. The consideration above is not intended to limit the applicability of the techniques described herein to the particular case HA (s) = Hp , but to provide a concrete example where M is polynomial, and typically only quadratic or cubic, in N.
[063] The Hamiltonian restricted to the subspace generated by the states \ εα) is called the "effective" Hamiltonian. The relevant part of the energy spectrum of the total Hamiltonian H AQC (t) is obtained by solving the eigenvalue problem associated with the effective Hamiltonian at any instant (in practice every small step As ) during the adiabatic evolution.
[064] The reduction method introduced herein allows the calculation of the energies of HAQC (s) to be performed on an ordinary classical computer with order of (kM)3 operations. Any algorithm devised to diagonalize matrices can be used to such aim.
[065] The difference between the two lowest eigenenergies determines the spectral g Oap I . Its minimum value between s = 0 and s = 1 is denoted g o mm. . From g o mm. it is possible to estimate the evolution time expected for a successful computation as
[066] The knowledge of the spectral gap g(s) , obtained thanks to the reduction method, leads to the optimal schedule as described above. Such optimal schedule can improve the scaling of T on the problem size N .
[067] Some AQC or quantum annealing machines operate according to a fixed, problem independent, schedule. The techniques described herein provide a schedule that is closer to an optimal schedule, tailored either on the specific problem of interest or on its specific class of problems.
[068] These techniques concern not only the calculation of the spectral gap g(s) , but of the full spectrum (eigenvalues and eigenstates) of the relevant part of the total Hamiltonian.
[069] The following description includes examples of equations that can be used to program the scheduling computer 104.
[070] The driver Hamiltonian HD can assume a variety of forms, but only a few regularly appear in the literature: The "Grover style" driver Hamiltonian (or simply Grover driver Hamiltonian),
where is the X Pauli matrix acting on the i -th qubit, which physically corresponds to a quantum transverse field. Despite their diversity, both HD are invariant under the exchange of any pair of qubits, have the same ground state and do not commute
with any Hp apart from the trivial Hp∞ 1 case.
[071] To determine whether the system evolution is adiabatic, several conditions have been proposed that all relate the evolution time T to the inverse of the spectral gap
and the first excited state energy of the adiabatic quantum Hamiltonian HAQC (t^ . In
particular, a widely adopted condition implies with the
minimum spectral gap achieved during the evolution from t = 0 to t = T .
initial driver or problem Hamiltonian and, in general, depend on s . In the following, we will omit any obvious dependence on s to maintain the notation as readable as possible.
[073] A main idea is that, among the many possible choices of HA and HB , the procedure performed by the scheduling computer 104 searches for those in which HA is a highly degenerate Hamiltonian with only M distinct energy levels and HB consists of the sum of k rank-1 projectors, namely
with the eigenvalue E of HA and ΡΩ the corresponding projector. Therefore, the
E
procedure will lead to an exponential reduction whenever both k and M depend polynomially on n .
[074] The non-trivial action of H AQC (t) is limited to the space spanned by the set of states | Ea) oc ΡΩ^ \ ψα)■ Restricting the action of the Hamiltonian to this subspace, we obtain
Hamiltonian HeS (s) results to be an effective (K x M) -level Hamiltonian, where
approximated version of the original HAQC (t) , but an exact description of its relevant
part. In fact, if we extend the set to a complete basis by adding orthonormal
vectors belonging to eigensubspaces of HA , then HAQC (i) presents a peculiar block diagonal structure when represented in such basis: the only block with dimension larger than l x l is a (KM) x (KM) block exactly reproduced by
[075] It is important to appreciate a subtlety: in most cases, we do not know the exact form of the states | Ea) , for example because they are related to the eigenstates of H A .
computer 104 to perform the numerical analysis? The answer, as described in more detail below, is that all the entries of ff can be computed as far as the overlaps
be computed either analytically or numerically by using algorithms that require only a polynomial amount of classical resources. Problems with a variety of forms of problem and driver Hamiltonians can be soved. In some cases, the procedure can successfully represent problems in which neither the problem nor the driver Hamiltonian are of Grover-style form.
[076] As described above, the procedure starts with the decomposition of the total Hamiltonian given as the sum of two contributions, HA and HB , having the form: )
In particular, the procedure selects a highly degenerate HA with only M distinct energies, and a Hamiltonian HB formed by a small number k of rank-1 projectors. The relevant part of the energy spectrum of HAQC (t) can be obtained assuming an effective system. Here, we present the derivation in the case of k = 1 , i.e., for a Grover- style Hamiltonian > where the procedure is more intuitive.
[077] In the definition of HA , E represents one of the M distinct eigenvalues and ΡΩ
E
the associated eigensubspace whose degeneracy is denoted by λ(Ε) . From the completeness of HA we have and For each energy E , we
introduce [ orthonormal states to obtain a basis of
and then
orthogonal because
since the terms (5) are factorized from the remaining terms (3) and (4). Thus, the relevant part of the AQC Hamiltonian results in
which is an effective M-level Hamiltonian, where M is the number of distinct energy levels of the contribution HA .
[080] Next, we extend the derivation above to the general form of HB . We will show that it is then possible to reduce a generic AQC Hamiltonian to a (M x K) -level effective Hamiltonian, where M is the number of energy levels of HA and K is an integer number smaller than or equal to the number k of states over which the term HB acts non-trivially. Let us consider the Hamiltonian
{| Ea )}a=l k and the other representing its orthogonal complement ωΕ . As for the 1 - state case, the set ωΕ is by construction contained in the kernel of HB , such that
neglected in the effective Hamiltonian. Moreover, since it is not said that
smaller set of rc(E) < mm{k ,λ(Ε)} pair-wise orthonormal states
[082] In Eq. (16), we already removed all terms in ωΕ because they are factorized with respect to the relevant part of the AQC Hamiltonian. As one can see, Eq. (16) describes an effective (M x K) -level Hamiltonian, where ■ Correctly, if k = \ we
obtain the AQC Hamiltonian reported in Eq. (10).
[083] It is important to observe that we reduced the original AQC Hamiltonian in Eq. (1) to an effective (M x K) - level Hamiltonian, and then we reduced the Hilbert space from 2" states to (M x K) states. Therefore, if both K and M are polynomial in the number of spins n , the reduced effective Hamiltonian in Eq. (16) can be expressed using
only a polynomial number of states. That is to say that we obtained an exponential reduction of the Hilbert space. We observe that the calculation of Za (E) and
might be non trivial for arbitrary states | ψα ) and Hamiltonian HA .
and interpret the above values as the entries of a certain matrix V . Such matrix is a square matrix with linear dimension k and can be shown to be Hermitian and positive- semidefinite. Therefore it admits a Cholesky decomposition:
where U is an upper triangular matrix with real and positive diagonal entries. While every Hermitian positive-definite matrix has a unique Cholesky decomposition, this does not need to be the case for Hermitian positive-semidefinite matrices and this reflects a certain freedom in choosing the states · It appears clear that the expansion
coefficients Τμα are the entries of a particular choice of such matrix U = T .
[085] Expressin the Hamiltonian HB in the basis of the effective subspace, we have
In this way, we have expressed all the necessary operators in the reduced basis.
[086] These techniques concern not only the calculation of spectral gap for a total Hamiltonian composed of contributions from only the driver and problem Hamiltonians, but include cases in which a third contribution is considered. This additional term can be associated with noise contributions (see section 4 of Mandra et al., arXiv: 1407.8183v2 [quant-ph], incorporated herein by reference), with a heuristic modification of the total Hamiltonian (see section "Energy gap for multi-parameter family of Hamiltonians" of Mandra et al., arXiv: 1407.8183vl [quant-ph], incorporated herein by reference), with control terms (for example to counteract the anti-adiabatic part of the dynamics), or with other future improvements of the AQC approach. It is to be understood that the foregoing description is intended to illustrate and not to limit the scope of the invention, which is defined by the scope of the appended claims. Other embodiments are within the scope of the following claims.
Claims
1. A method for controlling a quantum computer to solve a problem, the method comprising:
accepting a first data representation of a problem Hamiltonian corresponding to a final quantum state of the quantum computer, where a measurement of the final quantum state at an end time provides at least one solution to the problem;
accepting a second data representation of a driver Hamiltonian corresponding to an initial quantum state of the quantum computer at a start time;
applying a computer-implemented procedure to determine a schedule function that determines an interpolation of a total Hamiltonian from the driver Hamiltonian to the problem Hamiltonian at multiple times between the start time and the end time, the procedure including:
forming a third data representation of the total Hamiltonian from a combination of the first data representation and the second data representation,
determining multiple states from the third data representation, forming an overlap matrix from the multiple states, the overlap matrix effectively defining a Hilbert space, and
determining a spectrum of the total Hamiltonian restricted to a subspace of the Hilbert space, where the dimensionality of the subspace is smaller than the dimensionality of the Hilbert space; and controlling a time evolution of a configuration of the quantum computer between the start time and the end time according to the determined schedule function.
2. The method of claim 1 , wherein the procedure further includes forming a block diagonal matrix from the overlap matrix.
3. The method of claim 2, wherein forming a block diagonal matrix from the overlap matrix includes transforming a basis of the overlap matrix.
4. The method of claim 2, wherein the subspace corresponds to a selected block of the block diagonal matrix.
5. The method of claim 4, wherein the block diagonal matrix contains only a single block that is larger than 1 x 1 in size, which is the selected block.
6. The method of claim 4, wherein determining the spectrum of the subspace includes determining eigenvalues of the selected block.
7. The method of claim 6, wherein the procedure further includes determining, based on the determined spectrum, a spectral gap at each of the multiple times between the start time and the end time, the spectral gap at a particular time corresponding to a difference between a ground state energy and a first exited state energy of an effective Hamiltonian corresponding to the selected block.
8. The method of claim 7, wherein controlling the time evolution of the configuration of the quantum computer between the start time and the end time according to the determined schedule function includes varying the total Hamiltonian more slowly when the spectral gap is smaller and more quickly when the spectral gap is larger.
9. The method of claim 1 , wherein the third data representation includes: (1) a first term representing a first Hamiltonian that is highly degenerate with M distinct energy levels, and (2) a second term representing a second Hamiltonian that is a sum of k projectors on k respective pure quantum states.
10. The method of claim 9, wherein at least one of the first Hamiltonian or the second Hamiltonian includes contributions from both the driver Hamiltonian and the problem Hamiltonian.
1 1. The method of claim 10, wherein both of the first Hamiltonian and the second Hamiltonian include contributions from both the driver Hamiltonian and the problem Hamiltonian.
12. The method of claim 9, wherein determining multiple states from the third data representation includes determining k M states corresponding to the k pure quantum states of the second Hamiltonian projected onto each of multiple eigensubspaces of the first Hamiltonian.
13. The method of claim 9, wherein the Hilbert space has a dimensionality of 2N and the subspace has a dimensionality of / M , where k and / and M are integers and /≤ k .
14. The method of claim 1 , wherein a quantum state representing the quantum computer and subject to the total Hamiltonian undergoes adiabatic dynamics that are confined to the subspace.
15. A quantum computing system comprising a scheduling computer configured to control a quantum computer by performing all the steps of any of claims 1 to 14.
16. Software comprising instructions stored on a non-transitory computer-readable medium for causing a scheduling computer to control a quantum computer by performing all the steps of any of claims 1 to 14.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US201462026090P | 2014-07-18 | 2014-07-18 | |
US62/026,090 | 2014-07-18 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2016011440A1 true WO2016011440A1 (en) | 2016-01-21 |
Family
ID=55079122
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2015/041090 WO2016011440A1 (en) | 2014-07-18 | 2015-07-20 | Quantum processor schedule control |
Country Status (1)
Country | Link |
---|---|
WO (1) | WO2016011440A1 (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018106556A1 (en) * | 2016-12-08 | 2018-06-14 | Microsoft Technology Licensing, Llc | Tomography and generative data modeling via quantum boltzmann training |
CN111527503A (en) * | 2017-12-29 | 2020-08-11 | 谷歌有限责任公司 | Non-uniform quantum annealing scheduling |
US10977570B2 (en) | 2017-06-19 | 2021-04-13 | Rigetti & Co, Inc. | Distributed quantum computing system |
US11010145B1 (en) | 2018-02-21 | 2021-05-18 | Rigetti & Co, Inc. | Retargetable compilation for quantum computing systems |
CN113377441A (en) * | 2021-06-16 | 2021-09-10 | 南方科技大学 | Universal time sequence controller framework for quantum physical experiment platform |
US11132617B2 (en) | 2017-12-22 | 2021-09-28 | Microsoft Technology Licensing, Llc | Quantum simulation of real time evolution of lattice Hamiltonians |
CN113609704A (en) * | 2021-08-20 | 2021-11-05 | 四川元匠科技有限公司 | Quantum open system simulation method based on different measurement modes, storage medium and terminal |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030023651A1 (en) * | 2001-04-11 | 2003-01-30 | Whaley K. Birgit | Quantum computation |
US20110047201A1 (en) * | 2005-07-11 | 2011-02-24 | Macready William G | Systems, methods and apparatus for factoring numbers |
US20110313741A1 (en) * | 2010-06-21 | 2011-12-22 | Spectral Associates, Llc | Methodology and its computational implementation for quantitative first-principles quantum-mechanical predictions of the structures and properties of matter |
-
2015
- 2015-07-20 WO PCT/US2015/041090 patent/WO2016011440A1/en active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20030023651A1 (en) * | 2001-04-11 | 2003-01-30 | Whaley K. Birgit | Quantum computation |
US20110047201A1 (en) * | 2005-07-11 | 2011-02-24 | Macready William G | Systems, methods and apparatus for factoring numbers |
US20110313741A1 (en) * | 2010-06-21 | 2011-12-22 | Spectral Associates, Llc | Methodology and its computational implementation for quantitative first-principles quantum-mechanical predictions of the structures and properties of matter |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018106556A1 (en) * | 2016-12-08 | 2018-06-14 | Microsoft Technology Licensing, Llc | Tomography and generative data modeling via quantum boltzmann training |
US11157828B2 (en) | 2016-12-08 | 2021-10-26 | Microsoft Technology Licensing, Llc | Tomography and generative data modeling via quantum boltzmann training |
US10977570B2 (en) | 2017-06-19 | 2021-04-13 | Rigetti & Co, Inc. | Distributed quantum computing system |
US11727299B2 (en) | 2017-06-19 | 2023-08-15 | Rigetti & Co, Llc | Distributed quantum computing system |
EP3711003B1 (en) * | 2017-12-22 | 2023-08-02 | Microsoft Technology Licensing, LLC | Quantum simulation of real time evolution of lattice hamiltonians |
US11132617B2 (en) | 2017-12-22 | 2021-09-28 | Microsoft Technology Licensing, Llc | Quantum simulation of real time evolution of lattice Hamiltonians |
CN111527503A (en) * | 2017-12-29 | 2020-08-11 | 谷歌有限责任公司 | Non-uniform quantum annealing scheduling |
US11809961B2 (en) | 2017-12-29 | 2023-11-07 | Google Llc | Inhomogeneous quantum annealing schedules |
US11010145B1 (en) | 2018-02-21 | 2021-05-18 | Rigetti & Co, Inc. | Retargetable compilation for quantum computing systems |
CN113377441B (en) * | 2021-06-16 | 2023-07-11 | 南方科技大学 | Universal time sequence controller architecture for quantum physical experiment platform |
CN113377441A (en) * | 2021-06-16 | 2021-09-10 | 南方科技大学 | Universal time sequence controller framework for quantum physical experiment platform |
CN113609704B (en) * | 2021-08-20 | 2023-08-01 | 四川元匠科技有限公司 | Quantum open system simulation method based on different measurement modes, storage medium and terminal |
CN113609704A (en) * | 2021-08-20 | 2021-11-05 | 四川元匠科技有限公司 | Quantum open system simulation method based on different measurement modes, storage medium and terminal |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2016011440A1 (en) | Quantum processor schedule control | |
Luo | Understanding diffusion models: A unified perspective | |
Elenberg et al. | Streaming weak submodularity: Interpreting neural networks on the fly | |
US20210350056A1 (en) | Method and system for quantum computing | |
Benatti et al. | Entanglement of identical particles | |
Finster et al. | Causal fermion systems as a candidate for a unified physical theory | |
Nedic et al. | On stochastic subgradient mirror-descent algorithm with weighted averaging | |
Yuan et al. | Truncated Power Method for Sparse Eigenvalue Problems. | |
Zhu et al. | General framework for verifying pure quantum states in the adversarial scenario | |
Hein et al. | Entanglement properties of multipartite entangled states under the influence of decoherence | |
Heinosaari et al. | Noise robustness of the incompatibility of quantum measurements | |
Arad et al. | Connecting global and local energy distributions in quantum spin models on a lattice | |
Acharya et al. | A comparative study of estimation methods in quantum tomography | |
Bollt et al. | On matching, and even rectifying, dynamical systems through Koopman operator eigenfunctions | |
Macek et al. | Excited-state quantum phase transitions in systems with two degrees of freedom. III. Interacting boson systems | |
Eltschka et al. | Distribution of entanglement and correlations in all finite dimensions | |
Katsura et al. | Entanglement in valence-bond-solid states on symmetric graphs | |
JP2023514596A (en) | Method and apparatus for locally simulating resource-optimized fermions on a quantum computer for quantum chemistry | |
Bauerschmidt et al. | Spectral gap critical exponent for Glauber dynamics of hierarchical spin models | |
Sun et al. | An iterative approach to rank minimization problems | |
Ostilli et al. | Thermalization of noninteracting quantum systems coupled to blackbody radiation: A Lindblad-based analysis | |
Chiribella et al. | Optimal asymptotic cloning machines | |
Wagner et al. | Computational renormalization scheme for quantum field theories | |
Boettcher et al. | Renormalization of the unitary evolution equation for coined quantum walks | |
US20230169383A1 (en) | Methods and systems for quantum computational chemistry modeling |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 15822856 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 15822856 Country of ref document: EP Kind code of ref document: A1 |