Combinatorial optimization problem processing device, combinatorial optimization problem processing method and program

The device addresses the limitation of conventional quantum annealing machines by using an alternating optimization algorithm to transform and solve combinatorial optimization problems with terms of third order or higher, enabling efficient processing of such problems.

JP7579579B2Active Publication Date: 2024-11-08TOHOKU UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
JP2021553521
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2019-10-29
Filing Date
2020-10-22
Publication Date
2024-11-08
Estimated Expiration
2040-10-22

AI Technical Summary

Technical Problem

Conventional quantum annealing machines are limited to handling combinatorial optimization problems expressed in quadratic form, unable to process equations with terms of order three or higher.

Method used

A combinatorial optimization problem processing device that uses an annealing method to solve functions with terms of third order or higher by employing an alternating optimization algorithm, transforming the problem into a form with only quadratic or lower terms, and utilizing quantum annealing to find solution candidates.

Benefits of technology

Enables the processing of combinatorial optimization problems with terms of third order or higher, overcoming the limitations of conventional quantum annealing machines.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007579579000016
    Figure 0007579579000016
  • Figure 0007579579000017
    Figure 0007579579000017
  • Figure 0007579579000018
    Figure 0007579579000018
Patent Text Reader

Abstract

A combinatorial optimization problem processing device comprising: an annealing unit for acquiring, through an annealing procedure, a first solution candidate in an alternating optimization algorithm that repeats, until a termination condition is met, a first optimization process for acquiring a first solution that is a variable q satisfying a first condition related to a partition function, in which a function including third- or lower-order terms of a variable q that represents the combinatorial optimization problem to be analyzed is a Hamiltonian, the partition function being represented by the sum of a first expression in which the Hamiltonian is represented by integral representation based on a variable X and which includes only second- or lower-order terms of the variable q and a second expression that does not include the variable q, and a second optimization process for acquiring a second solution that is a variable X satisfying a second condition related to the partition function; a computation unit for acquiring the second solution; and a solution candidate acquisition unit for acquiring, when the termination condition has been met, the immediately preceding first solution candidate as a candidate for the solution of the combinatorial optimization problem.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a combinatorial optimization problem processing device, a combinatorial optimization problem processing method, and a program. This application claims priority based on Japanese Patent Application No. 2019-196324, filed on October 29, 2019, the contents of which are incorporated herein by reference. [Background technology]

[0002] In recent years, research on machine learning, deep learning, and artificial intelligence has progressed. In machine learning, deep learning, and artificial intelligence, one of the important issues is how to quickly process combinatorial optimization problems such as the traveling salesman problem. Therefore, expectations are rising for the implementation of a device that quickly processes combinatorial optimization problems. One of the devices that quickly processes such combinatorial optimization problems is a quantum annealing machine (Non-Patent Document 1) that processes optimization problems by quantum annealing. A quantum annealing machine is a computer that uses quantum bits as basic elements and is equipped with multiple quantum bits. A quantum annealing machine acquires information indicating a predetermined eigenstate (e.g., a ground state) in a system (hereinafter referred to as a "bit system") including multiple quantum bits equipped in the device by measuring the bit system (quantum annealing). The information indicating the eigenstate is, specifically, information indicating the distribution of values ​​indicated by each quantum bit in the eigenstate. Then, the quantum annealing machine outputs the acquired result as a candidate solution for the combinatorial optimization problem. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] "The D-Wave 2000Q System", [online], D Wave The Quantum Computing Company, [Retrieved October 8, 2019], Internet〈URL:https: / / www.dwavesys.com / d-wave-two-system> Summary of the Invention [Problem to be solved by the invention]

[0004] However, conventional quantum annealing machines have the problem that they can only handle combinatorial optimization problems expressed in quadratic forms because they use the interactions between adjacent quantum bits. Therefore, it was not possible to use annealing techniques to obtain candidate solutions for equations that contain cubic or higher terms in variables. Furthermore, this problem is not limited to quantum annealing machines, but is a common issue for devices that use annealing techniques to output candidate solutions to combinatorial optimization problems.

[0005] In view of the above circumstances, an object of the present invention is to provide a technique for acquiring candidate solutions to a combinatorial optimization problem expressed by a function including cubic or higher terms of variables, using an annealing technique. [Means for solving the problem]

[0006] In one aspect of the present invention, a partition function is defined as a partition function to be root-finded, the partition function being a Hamiltonian of a function to be analyzed that represents a combinatorial optimization problem to be analyzed and that includes third-order or higher terms of a variable q, the Hamiltonian being expressed by a predetermined integral representation by a variable X, the Hamiltonian being expressed by a sum of a first formula including only second-order or lower terms of the variable q and a second formula not including the variable q, a partition function to be root-finded, a process of acquiring a value of the variable q that satisfies a first predetermined condition regarding a maximum value of the partition function to be root-finded, the variable q being an independent variable only, is defined as a first optimization process, and the variable q that satisfies the first predetermined condition acquired by the first optimization process is defined as a first solution, The combinatorial optimization problem processing device includes: an annealing unit that acquires a candidate for the first solution by an annealing technique in an alternating optimization algorithm that repeats the first optimization process and the second optimization process until a predetermined termination condition is satisfied, the second optimization process being a process of acquiring a value of the variable X that satisfies a second predetermined condition related to a maximum value of the distribution function to be root-finded, with the variable X acquired by the second optimization process satisfying the second predetermined condition as a second solution; a calculation unit that acquires the second solution; and a solution candidate acquisition unit that acquires the first solution candidate that was acquired immediately before when the termination condition is satisfied, as a solution candidate for the combinatorial optimization problem.

[0007] One aspect of the present invention is the above-mentioned combinatorial optimization problem processing apparatus, wherein the first predetermined condition is a condition that a value of the distribution function is a maximum value.

[0008] One aspect of the present invention is the above-mentioned combinatorial optimization problem processing apparatus, wherein the second predetermined condition is a condition that a value of the distribution function is a maximum value.

[0009] One aspect of the present invention is the above-mentioned combinatorial optimization problem processing apparatus, wherein the predetermined integral representation is a Fourier integral representation.

[0010] One aspect of the present invention is the above-mentioned combinatorial optimization problem processing apparatus, wherein the annealing technique is quantum annealing.

[0011] One aspect of the present invention is the above-mentioned combinatorial optimization problem processing apparatus, wherein the calculation unit obtains the second solution by a root-finding algorithm.

[0012] In one aspect of the present invention, a partition function is defined as a partition function to be root-finded, the partition function being a Hamiltonian of a function to be analyzed that represents a combinatorial optimization problem to be analyzed and that includes third- or higher-order terms of a variable q, the Hamiltonian being a partition function expressed by a predetermined integral representation by a variable X, the Hamiltonian being expressed by a sum of a first formula including only second- or lower-order terms of the variable q and a second formula not including the variable q, a partition function to be root-finded, a process of acquiring a value of the variable q that satisfies a first predetermined condition regarding a maximum value of the partition function to be root-finded with only the variable q as an independent variable, a first optimization process is defined as a process of acquiring a value of the variable q that satisfies a first predetermined condition regarding a maximum value of the partition function to be root-finded, the variable q that satisfies the first predetermined condition acquired by the first optimization process is defined as a first solution, and a process of acquiring a value of the variable X only is defined as a first solution. a second optimization process for acquiring a value of the variable X that satisfies a second predetermined condition related to a maximum value of the distribution function to be root-finded, and the variable X that satisfies the second predetermined condition acquired by the second optimization process as a second solution, the first optimization process and the second optimization process are repeated until a predetermined termination condition is satisfied, an annealing step for acquiring a candidate for the first solution by an annealing technique, a calculation step for acquiring the second solution, and a solution candidate acquisition step for acquiring the first solution candidate that was acquired immediately before when the termination condition is satisfied, as a solution candidate for the combinatorial optimization problem.

[0013] One aspect of the present invention is a program for causing a computer to function as the above-mentioned combinatorial optimization problem processing device. Effect of the Invention

[0014] According to the present invention, it is possible to obtain candidate solutions to combinatorial optimization problems expressed by functions including cubic or higher terms of variables by utilizing an annealing technique. [Brief description of the drawings]

[0015] [Figure 1] FIG. 1 is a diagram showing an example of a functional configuration of a combinatorial optimization problem processing apparatus 1 according to an embodiment. [Diagram 2] FIG. 2 is a diagram showing an example of the functional configuration of a control unit 11 in the embodiment. [Diagram 3] FIG. 2 is an explanatory diagram for explaining an example of a state measuring unit 15 in the embodiment. [Figure 4] FIG. 2 is a diagram showing an example of the positional relationship of a plurality of physical quantum bits 101 according to the embodiment. [Diagram 5] 1 is a flowchart showing an example of a flow of processing executed by a combinatorial optimization problem processing apparatus 1 according to an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] (Embodiment) 1 is a diagram showing an example of a functional configuration of a combinatorial optimization problem processing apparatus 1 according to an embodiment. First, an overview of the combinatorial optimization problem processing apparatus 1 will be described.

[0017] <Overview of the Combinatorial Optimization Problem Processing Device 1> The combinatorial optimization problem processing device 1 is a device that uses an annealing technique to obtain solution candidates for a combinatorial optimization problem to be analyzed, such as optimization of traffic management. More specifically, the combinatorial optimization problem processing device 1 obtains solution candidates for a combinatorial optimization problem to be analyzed, such as optimization of traffic management, for a third or higher degree term g i The device utilizes an annealing technique to obtain a solution to a combinatorial optimization problem expressed by a function including (hereinafter referred to as the "analysis target function"), in which: Specifically, the solution to the combinatorial optimization problem expressed by the analysis target function is obtained by utilizing the annealing technique to obtain candidate values ​​of variables that satisfy predetermined conditions (hereinafter referred to as the "combination conditions") that represent the analysis target combinatorial optimization problem.

[0018] The combination condition may be any condition that represents the combinatorial optimization problem to be analyzed. The combination condition is, for example, a condition that the value of the function to be analyzed is a minimum value. Depending on the combinatorial optimization problem to be analyzed, the combination condition is, for example, a condition that the value of the function to be analyzed is a minimum value. Depending on the combinatorial optimization problem to be analyzed, the combination condition is, for example, a condition that the value of the function to be analyzed is a maximum value. For ease of explanation, the combinatorial optimization problem processing device 1 will be explained below taking as an example a case where the combination condition is a condition that the value of the function to be analyzed is a minimum value. Hereinafter, the value of the variable that satisfies the combination condition will be referred to as a conditional solution of the function to be analyzed. In other words, the conditional solution of the function to be analyzed is the solution of the combinatorial optimization problem to be analyzed. Note that i is an integer between 1 and I (I is an integer greater than or equal to 1). The different i's are the sum of the third and higher terms g i The analysis target function is, for example, an equation expressed by equation (1).

[0019]

number

[0020]

number

[0021] In formula (1), f(q) represents the function to be analyzed. In formula (1), q represents a variable in the function to be analyzed. The variable q is a scalar or a vector. For ease of explanation, the combinatorial optimization problem processing apparatus 1 will be explained below using an example in which q is a vector having N elements. Each element of the variable q is a binary variable.

[0022] In formula (1), f0(q) represents a term contained in the analysis target function f(q) that is a quadratic term of the variable q. In formula (1), the second term represents the sum of the cubic or higher terms of the variable q among the terms of the analysis target function f(q). In formula (2), the superscript T represents the transpose of the variable q. In formula (2), Q represents a coefficient matrix.

[0023] When the function f(q) to be analyzed is regarded as a Hamiltonian representing a certain system such as a quantum system, the partition function Z represented by the Hamiltonian f(q) is expressed by the following equation (3).

[0024]

number

[0025] In equation (3), β represents the inverse temperature of the system when the state of the system is measured by the annealing technique described below. In equation (3), the sigma symbol Σ means calculating the value of the term written to the right of the sigma symbol Σ for each element of variable q, and taking the sum of the calculated values. If the exponent of the exponential function with Napier's number as the base on the right side of equation (3) is expressed in Fourier integral notation using variable x, equation (3) can be expressed as the following equation (4). Hereinafter, equation (4) will be referred to as the first Fourier representation. The variable x is x i It is an I-dimensional vector with elements . If q is a vector, xi is also a vector.

[0026]

number

[0027] In equation (4), δ(···) represents the Dirac delta function. Thus, the first Fourier representation includes a delta function whose argument is the difference between the variable x and the variable q.

[0028] If the delta function of equation (4) is expressed by the Fourier integral representation of the variable X, the delta function of equation (4) is expressed by the following equation (5). The variable X is X i It is an I-dimensional vector with elements. If q is a vector, Xi is also a vector.

[0029]

number

[0030] From equations (3), (4), and (5), the distribution function Z is expressed by the following equation (6): Hereinafter, equation (6) will be referred to as the second Fourier representation.

[0031]

number

[0032] Comparing equation (3) and equation (6), it can be seen that the analysis target function f(q) is expressed by the following equation (7).

[0033]

number

[0034] Here, H1 in formula (7) is expressed by the following formula (8), and H2 in formula (7) is expressed by the following formula (9). Hereinafter, H1 will be referred to as the first analysis target term. Hereinafter, H2 will be referred to as the second analysis target term.

[0035]

number

[0036]

number

[0037] In this way, the analysis target function f(q), which is a function including cubic or higher terms of the variable q, is transformed into a function including only quadratic or lower terms of the variable q by using the variables X and x. More specifically, the analysis target function f(q) is transformed into an equation that is the sum of a first analysis target term including only quadratic or lower terms of the variable q and a second analysis target term that does not include the variable q by using the variables X and x. In the case of an equation indicating that the value of a function including only quadratic or lower terms of the variable q is equal to 0, it is possible to obtain the solution by an annealing technique such as quantum annealing (for example, see Non-Patent Document 1).

[0038] As is well known in quantum physics such as quantum statistical mechanics, when the partition function is at its maximum, the Hamiltonian is at its minimum. Therefore, the conditional solutions of the analyte function are the variable q, the variable q that gives the minimum value of the function with variables X and x, and the variable q in variables X and x. In particular, when the variable x is a function of the variable q or the variable X, the conditional solutions of the analyte function are the variable q and the variable q in variable X that give the minimum value of the function with variables q and X.

[0039] As described later, since the variable x is a function of the variable q, the combinatorial optimization problem processing apparatus 1 acquires a conditional solution to the analysis target function by executing an alternating optimization algorithm described below using an annealing method and a non-annealing method. The non-annealing method means a method other than annealing.

[0040] The alternating optimization algorithm executed by the combinatorial optimization problem processing device 1 is an algorithm that alternately repeats a first optimization process and a second optimization process until a predetermined termination condition is satisfied. The first optimization process is a process of acquiring a variable q that satisfies a first predetermined condition (hereinafter referred to as a "first distribution function condition") regarding the maximum value of the distribution function Z, using only the variable q as an independent variable. Hereinafter, the variable q that satisfies the first distribution function condition is referred to as a first solution. The first distribution function condition is, for example, a condition that the difference between the value of the distribution function Z and the maximum value of the distribution function Z is less than a predetermined value. The first distribution function condition may be, for example, a condition that the value of the distribution function Z is the maximum value of the distribution function Z. The first distribution function condition may be a condition that the value of the distribution function Z is a maximum value. Hereinafter, for simplicity of explanation, the combinatorial optimization problem processing device 1 will be explained taking as an example a case where the first distribution function condition is a condition that the value of the distribution function Z is a maximum value.

[0041] The second optimization process is a process of acquiring a variable X that satisfies a second predetermined condition (hereinafter referred to as the "second distribution function condition") regarding the maximum value of the distribution function Z, with the variable X being the only independent variable. Hereinafter, the variable X that satisfies the second distribution function condition is referred to as the second solution. The second distribution function condition is, for example, a condition that the difference between the value of the distribution function Z and the maximum value of the distribution function Z is less than a predetermined value. The second distribution function condition may be, for example, a condition that the value of the distribution function Z is the maximum value of the distribution function Z. The second distribution function condition may be a condition that the value of the distribution function Z is a maximum value. Hereinafter, for simplicity of explanation, the combinatorial optimization problem processing apparatus 1 will be explained taking as an example a case where the second distribution function condition is a condition that the value of the distribution function Z is a maximum value.

[0042] More specifically, in the alternating optimization algorithm executed by the combinatorial optimization problem processing device 1, first, a first solution is acquired by executing a first optimization process. Next, the first solution is substituted into the distribution function Z. Next, a variable X (i.e., a second solution) that gives a maximum value of the distribution function Z after substitution with the variable X as an independent variable is acquired. Next, a value of the variable q that gives a maximum value of the distribution function Z into which the variable q is substituted with the second solution as an independent variable is acquired. Next, a variable X that gives a maximum value of the distribution function Z after substitution of the acquired variable q with the variable X as an independent variable is acquired. In the alternating optimization algorithm executed by the combinatorial optimization problem processing device 1, the process of acquiring the first solution and the second solution is repeated until a predetermined termination condition (hereinafter referred to as "optimization termination condition") is satisfied. The optimization termination condition is, for example, a condition that the change in the value of the first solution is less than a predetermined value.

[0043] The combinatorial optimization problem processing device 1 executes the first optimization process by an annealing technique such as quantum annealing. Since the combinatorial optimization problem processing device 1 acquires the first optimization process by an annealing technique, the first solution acquired by the combinatorial optimization problem processing device 1 is a pair of the value of the variable q and the probability density of each value of the variable q.

[0044] Here, it will be explained that the first solution acquired by the combinatorial optimization problem processing device 1 is a set of the value of the variable q and the probability density of each value of the variable q. In the annealing method, the state of a system to be annealed (hereinafter referred to as the "annealing target system") is measured. The annealing target system is a system that includes a plurality of physical quantum bits and has a Hamiltonian expressed by equation (8). The annealing target system is, for example, a quantum system such as a spin system represented by the Ising model. Below, the combinatorial optimization problem processing device 1 will be explained taking as an example a case where the annealing method is quantum annealing and the annealing target system is a spin system represented by the Ising model. The value of the variable q indicates the state of physical quantum bits such as electron spin possessed by the annealing target system.

[0045] When the state of an annealing target system is measured, the value of a physical quantity and the probability density of that physical quantity are measured, as is well known in quantum physics such as quantum mechanics. Therefore, when the combinatorial optimization problem processing device 1 acquires the value of a variable q using a quantum annealing method, the combinatorial optimization problem processing device 1 measures the value of the physical quantity corresponding to the variable q and the probability density of the value of the physical quantity corresponding to the variable q. Note that the fact that not only the value of the physical quantity but also the probability density of the value of the physical quantity is measured is the same in any annealing method, not limited to quantum annealing.

[0046] In the annealing method, the values ​​of the variables X and x are pieces of information (hereinafter referred to as "annealing control information") for controlling the state of an annealing target system whose Hamiltonian is expressed by equation (8). For example, when the annealing method is quantum annealing and the annealing target system is a spin system expressed by the Ising model, the value of the variable X in the annealing method is a value related to the direction and strength of the magnetic field applied to the spin. The direction and strength of the magnetic field are specifically determined by the value of the partition function Z and the variable X. In such a case, the value of the variable x in the annealing method is the expected value of the variable q indicated by the first solution obtained in the immediately preceding first optimization process. The expected value of the variable q is specifically a value according to the partition function Z. The annealing control information includes format information, formula coefficient information, and temperature information, which will be described later, in addition to the values ​​of the variables X and x.

[0047] The values ​​of the variables x and X may be obtained by any device that obtains values ​​by a non-annealing technique. An example of a device that obtains values ​​by a non-annealing technique is a von Neumann type computer.

[0048] <Annealing method details> Here, the annealing technique for obtaining the value of the variable q will be described in more detail. As described above, the analysis object function f(q) can be regarded as a Hamiltonian representing a certain system such as a quantum system, so the first analysis object term H1 can also be regarded as a Hamiltonian of a subsystem of the Hamiltonian f(q). The partition function Z1 represented by the Hamiltonian H1 is expressed by the following formula (10).

[0049]

number

[0050] As described above, the combinatorial optimization problem processing device 1 uses an annealing technique to obtain the variable q that gives the maximum value of the distribution function Z expressed by formula (6). Since the second analysis target term does not explicitly include the variable q, the process of obtaining the variable q that gives the maximum value of the distribution function Z is equivalent to the process of obtaining the variable q that gives the maximum value of the distribution function Z1. Therefore, the combinatorial optimization problem processing device 1 uses an annealing technique to obtain a solution that sets the value of the first derivative of the distribution function with the first analysis target term as the Hamiltonian to 0. That is, the combinatorial optimization problem processing device 1 uses an annealing technique to obtain a solution that satisfies the following formulas (11) and (12).

[0051]

number

[0052]

number

[0053] The process of obtaining a solution that satisfies formulas (11) and (12) using the annealing technique is, for example, when the annealing technique is quantum annealing, a process of measuring a quantum state of a quantum system whose Hamiltonian is expressed by formula (8) and whose partition function Z1 is maximum. Hereinafter, the quantum state of a quantum system whose Hamiltonian is expressed by formula (8) and whose partition function Z1 is maximum is referred to as a target quantum state. The quantum state whose partition function Z1 is maximum is the ground state of a quantum system whose Hamiltonian is expressed by formula (8).

[0054] <About the second optimization process> Here, the second optimization process will be described in more detail. In the second optimization process, the expected value of the variable q obtained by the immediately preceding annealing is substituted for the variable x. That is, the variable x is updated by the second optimization process to a value expressed by the following formula (13).

[0055]

number

[0056] The second solution is obtained, for example, by a root-finding algorithm whose update equation is the following equation (14):

[0057]

number

[0058] The update equation of equation (14) is an update equation for a root-finding algorithm for finding a solution to equation (15) below when the value of variable x is the value expressed by equation (13).

[0059]

number

[0060] Equation (15) indicates that the value of the second analyzed term is equal to zero.

[0061] The root-finding algorithm may be any algorithm that can find a solution to the following equation (15) when the value of the variable x is the value expressed by equation (13). The root-finding algorithm may be, for example, a gradient method. The root-finding algorithm may be, for example, a bisection method. The root-finding algorithm may be, for example, an iterative substitution method.

[0062] <Equipment details> Hereinafter, the combinatorial optimization problem processing apparatus 1 will be described more specifically by explaining each functional unit of the combinatorial optimization problem processing apparatus 1.

[0063] The combinatorial optimization problem processing device 1 includes a control unit 11 including a processor 91, such as a CPU (Central Processing Unit) connected by a bus, and a memory 92, and executes a program. The combinatorial optimization problem processing device 1 functions as a device including the control unit 11, a storage unit 12, an input unit 13, an output unit 14, and a state measurement unit 15 by executing the program.

[0064] All or part of the functions of the combinatorial optimization problem processing apparatus 1 may be realized using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). The program may be recorded on a computer-readable recording medium. Examples of computer-readable recording media include portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. The program may be transmitted via an electric communication line.

[0065] The control unit 11 controls the operation of each functional unit of the combinatorial optimization problem processing apparatus 1. Fig. 2 is a diagram showing an example of a functional configuration of the control unit 11 in the embodiment. Fig. 2 shows functional units related to acquisition of solution candidates of the analysis target function. The control unit 11 includes a condition acquisition unit 111, a solution candidate acquisition unit 112, a non-annealing calculation unit 113, and a state measurement control unit 114.

[0066] The condition acquisition unit 111 acquires the analysis target function and information indicating the temperature of the system when the state of the system is measured by an annealing technique such as inverse temperature (hereinafter referred to as "temperature information"). Specifically, acquiring the analysis target function means acquiring information indicating the form of the analysis target function such as the variables and orders constituting each term of the analysis target function, information indicating the coefficient values ​​of each term of the analysis target function, and information indicating that the value of the analysis target function is 0. The information indicating the form of the analysis target function (hereinafter referred to as "form information"), information indicating the coefficient values ​​of each term of the analysis target function (hereinafter referred to as "formula coefficient information"), and information indicating that the value of the analysis target function is 0 (hereinafter referred to as "zero information") are each a type of annealing control information. The temperature information may indicate, for example, the inverse temperature.

[0067] The condition acquisition unit 111 may acquire the analysis target function and the temperature information in any manner. For example, the condition acquisition unit 111 may acquire the analysis target function and the temperature information by acquiring the analysis target function and the temperature information input to the input unit 13 via a bus. For example, the condition acquisition unit 111 may acquire the analysis target function and the value indicating the temperature of the system by reading out the analysis target function and the temperature information stored in the storage unit 12 from the storage unit 12.

[0068] The solution candidate acquisition unit 112 determines whether or not an optimization end condition is satisfied, and if the optimization end condition is satisfied, acquires the first solution acquired by the immediately preceding first optimization process as a solution candidate for the analysis target function. The solution candidate acquired by the solution candidate acquisition unit 112 is a solution candidate for the combinatorial optimization problem to be analyzed.

[0069] The non-annealing calculation unit 113 updates the value of the variable x stored in the memory unit 12, the details of which will be described later, to the value represented by equation (13) based on the measurement result of the state measurement unit 15, the details of which will be described later. The non-annealing calculation unit 113 also acquires a second solution based on the measurement result of the state measurement unit 15, the details of which will be described later. Specifically, the measurement result of the state measurement unit 15 is the value of the variable q and the probability density of each value of the variable q. The non-annealing calculation unit 113 acquires the second solution by a non-annealing method. The non-annealing calculation unit 113 updates the value of the variable X stored in the memory unit 12, the details of which will be described later, based on the acquired second solution.

[0070] The state measurement control unit 114 controls the operation of the state measurement unit 15 based on the annealing control information. The state measurement control unit 114 outputs an annealing start instruction to the state measurement unit 15 every time a second solution is acquired until an optimization end condition is satisfied. The annealing start instruction is a signal that causes the state measurement unit 15 to start annealing. The state measurement unit 15 also outputs an annealing start instruction to the state measurement control unit 114 when a function to be analyzed and temperature information are input to the input unit 13. Input of a function to be analyzed means that format information, formula coefficient information, and zero information are input to the input unit 13.

[0071] The control unit 11 may operate in any manner as long as the condition acquisition unit 111, the solution candidate acquisition unit 112, the non-annealing calculation unit 113, and the state measurement control unit 114 operate. The control unit 11 may operate in a manner similar to that of a von Neumann-type computer, for example.

[0072] Returning to the explanation of Figure 1. The storage unit 12 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device. The storage unit 12 stores various information related to the device itself. The storage unit 12 stores, for example, the measurement results of the state measurement unit 15. The storage unit 12 stores, for example, annealing control information. More specifically, the storage unit 12 stores, for example, initial values ​​of variables X and x. Furthermore, the storage unit 12 stores the updated variables X and x each time the variables X and x are updated by, for example, executing the second optimization process. Furthermore, the storage unit 12 stores, for example, an analysis target function. Storing an analysis target function specifically means storing format information, formula coefficient information, and zero information. Furthermore, the storage unit 12 stores, for example, temperature information.

[0073] The storage unit 12 stores, for example, a first analysis target term. Storing the first analysis target term specifically means storing information indicating the form of the first analysis target term, such as variables constituting the first analysis target term and an order, and the value of the coefficient of the first analysis target term. The storage unit 12 stores, for example, a second analysis target term. Storing the second analysis target term specifically means storing information indicating the form of the second analysis target term, such as variables constituting the second analysis target term and an order, and the value of the coefficient of the second analysis target term. The storage unit 12 stores, for example, a first solution. The storage unit 12 stores, for example, a second solution. The storage unit 12 stores, for example, combination conditions in advance.

[0074] The input unit 13 includes input devices such as a mouse, a keyboard, and a touch panel. The input unit 13 may be configured as an interface that connects these input devices to the device itself. The input unit 13 accepts input of various information for the device itself. The input unit 13 accepts input of a function to be analyzed. To be specific, accepting input of a function to be analyzed means accepting input of format information, equation coefficient information, and zero information. The input unit 13 outputs the input function to be analyzed to the control unit 11 and the memory unit 12.

[0075] The output unit 14 outputs the conditional solution candidates of the analysis target function acquired by the solution candidate acquisition unit 112. The output unit 14 may be of any type as long as it is capable of outputting the acquired conditional solution candidates of the analysis target function. The output unit 14 includes a display device such as a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro-Luminescence) display. The output unit 14 may be configured as an interface that connects these display devices to the output unit itself.

[0076] The state measurement unit 15 measures the target quantum state of the annealing target system by a quantum annealing technique. The state measurement unit 15 acquires a first solution by measuring the target quantum state. Each time the state measurement unit 15 acquires an annealing start instruction, the state measurement unit 15 controls the state of the annealing target system based on the annealing control information. The state measurement unit 15 measures the target quantum state, which is one of the states of the annealing target system and one of the states controlled by the state measurement unit 15.

[0077] Controlling the state of the annealing target system based on the annealing control information specifically means controlling the state of the annealing target system so that it is in a state that satisfies the following Hamiltonian condition. The Hamiltonian condition is a condition that the Hamiltonian of the annealing target system is the Hamiltonian expressed by equation (8), and the values ​​of each variable and constant other than the variable q in equation (8) are values ​​indicated by the annealing control information. The annealing target system is controlled, for example, by application of a magnetic field to the annealing target system by the state measurement unit 15. The temperature of the annealing target system when the state measurement unit 15 measures the state is the temperature indicated by the temperature information.

[0078] Note that, until the execution of the second optimization process starts, the values ​​of the variables X and x during the execution of the first optimization process are the initial values ​​of the variables X and x.

[0079] FIG. 3 is an explanatory diagram illustrating an example of the state measuring unit 15 in the embodiment. The state measurement unit 15 includes a quantum bit body 501 , a state control unit 502 , a measurement device 503 , and a first solution acquisition unit 504 .

[0080] The quantum bit body 501 has a plurality of physical quantum bits 101. The quantum bit body 501 is, for example, a semiconductor substrate having quantum dots. In such a case, the physical quantum bit 101 is, for example, the spin of an electron in a quantum dot. The quantum bit body 501 may be a diamond substrate having nitrogen defects. In such a case, the physical quantum bit 101 is, for example, the spin of an electron in a nitrogen defect. The quantum bit body 501 may be a superconducting circuit. In such a case, the physical quantum bit 101 is, for example, a superconducting quantum bit. The quantum bit body 501 is an example of an annealing target system.

[0081] The state control unit 502 controls the state of the quantum bit body 501. More specifically, the state control unit 502 controls the state of the quantum bit body 501 so that the Hamiltonian of the quantum bit body 501 is a Hamiltonian that satisfies a Hamiltonian condition.

[0082] For example, when the physical quantum bit 101 of the quantum bit body 501 is a superconducting quantum bit, the state control unit 502 controls the state of the quantum bit body 501 by controlling the current flowing through each superconducting circuit and controlling the magnetic field applied to the quantum bit body 501. In such a case, the state control unit 502 is a device including, for example, a circuit for controlling the current flowing through the superconducting circuit and a magnetic field generator. The state control unit 502 may further control the temperature. The state control unit 502 may also control the state of the quantum bit body 501 by temperature control. In such a case, the state control unit 502 includes, for example, a heater or a cooler. The state control unit 502 controls, for example, the temperature to a temperature indicated by the temperature information. It is desirable to measure the state of the quantum bit body 501 at a low temperature below room temperature. In particular, it is desirable to perform the measurement at 0 K. Hereinafter, for the sake of simplicity, the combinatorial optimization problem processing apparatus 1 will be described taking, as an example, a case in which the physical quantum bit 101 is the spin of an electron and the temperature at the time of measurement is 0 K.

[0083] The measurement device 503 measures the quantum state of the quantum bit body 501. The measurement device 503 may be any device capable of acquiring the quantum state of the quantum bit body 501. The quantum state of the quantum bit body 501 is specifically the quantum state of the multiple physical quantum bits 101. The measurement device 503 may be, for example, a device equipped with a flux meter, or may be a device equipped with an optical sensor. The measurement device 503 determines whether the state of the quantum bit body 501 is the target quantum state based on the measurement result. More specifically, the measurement device 503 determines that the state of the quantum bit body 501 is the target quantum state when a predetermined condition regarding the maximum value of the partition function Z is satisfied. The predetermined condition regarding the maximum value of the partition function Z (hereinafter referred to as the "maximum condition") is, for example, a condition that the value of the first derivative of the partition function Z is within a predetermined range including 0, and the sign of the first derivative of the partition function Z changes from positive to negative.

[0084] Here, the operations of the state control unit 502 and the measurement device 503 will be described in more detail. Upon receiving an instruction to start annealing, the state control unit 502 changes the state of the quantum bit body 501. Specifically, the state of the quantum bit body 501 is changed so as to change the quantum state of each physical quantum bit 101.

[0085] The measurement device 503 measures the state of the quantum bit body 501 changed by the state control unit 502. The measurement device 503 determines whether or not the state of the quantum bit body 501 is the target quantum state based on the measurement result. If the measurement device 503 determines that the state is not the target quantum state, the state control unit 502 changes the state of the quantum bit body 501 again. It is desirable that the change in the state of the quantum bit body 501 is performed adiabatically. Then, the measurement device 503 measures the state of the quantum bit body 501 after the change. The same process is executed thereafter until the maximum condition is satisfied. By repeating such a process, the measurement device 503 measures the state of each physical quantum bit 101 in the target quantum state.

[0086] The first solution acquisition unit 504 acquires a value indicating the state of each physical quantum bit 101 in the target quantum state of the quantum bit body 501 and a probability density of each state occurring based on the measurement result of the measurement device 503. The state of the physical quantum bit 101 is, for example, the direction and magnitude of the electron spin if the physical quantum bit 101 is an electron spin. The value indicating the state of each physical quantum bit 101 acquired by the first solution acquisition unit 504 is the value of the variable q in the first solution. The probability density of the state occurring is the probability density of each value of the variable q in the first solution. That is, the first solution acquisition unit 504 acquires the first solution by measurement. The first solution acquired by the first solution acquisition unit 504 is stored in the storage unit 12. The first solution acquired by the first solution acquisition unit 504 may be output to the control unit 11.

[0087] Fig. 4 is a diagram showing an example of the positional relationship of a plurality of physical quantum bits 101 in an embodiment. Fig. 4 is a diagram showing a plurality of physical quantum bits 101 located in a quantum bit body 501. Fig. 4 shows that the plurality of physical quantum bits 101 are located in a lattice pattern.

[0088] FIG. 5 is a flowchart showing an example of the flow of processing executed by the combinatorial optimization problem processing apparatus 1 of the embodiment. The condition acquisition unit 111 acquires the analysis target function and temperature information (step S101). Next, the state measurement control unit 114 outputs an annealing start instruction to the state measurement unit 15 (step S102). Next, the state measurement unit 15 measures the state of the quantum bit body 501 while changing the state of the quantum bit body 501 so as to satisfy the Hamiltonian condition until the maximal condition is satisfied (step S103). Next, the state measurement unit 15 measures the state of each physical quantum bit 101 in the target quantum state (step S104). Since the state of each physical quantum bit 101 measured in step S104 indicates the value of the variable q and the probability density of each value of the variable q, the processing from step S103 to step S104 is a first optimization processing, which is a processing for acquiring a first solution.

[0089] After step S104, the non-annealing calculation unit 113 acquires a second solution by a non-annealing method based on the result of step S104 (step S105). The non-annealing calculation unit 113 updates the value of the variable X stored in the storage unit 12 with the acquired second solution. In addition, when acquiring the second solution, the non-annealing calculation unit 113 updates the value of the variable x stored in the storage unit 12 to a value represented by equation (13) based on the result of step S104. More specifically, the value of the variable x is updated to the expected value of the variable q using the value of the variable q acquired in the immediately preceding step S104 and the probability density of each value of the variable q. The process of step S105 is a second optimization process.

[0090] After step S105, the solution candidate acquisition unit 112 determines whether or not an optimization end condition is satisfied (step S106). If the optimization end condition is satisfied (step S106: YES), the solution candidate acquisition unit 112 acquires the first solution acquired by the immediately preceding first optimization process as a candidate for the conditional solution of the analysis target function (step S107). After step S107, the output unit 14 outputs the candidate for the conditional solution acquired in step S107 (step S108).

[0091] On the other hand, if the optimization end condition is not satisfied in step S106, the process returns to step S102.

[0092] The combinatorial optimization problem processing device 1 of the embodiment configured in this way acquires candidates for a conditional solution of the analysis target function using a partition function Z with the Hamiltonian of the formula (1) rather than the formula (1) itself. The partition function Z is transformed into an equation including only quadratic or lower terms of the variable q by using a variable X that does not exist in the variables (1).

[0093] Therefore, the combinatorial optimization problem processing device 1 alternately repeats the first optimization process and the second optimization process. The first optimization process and the second optimization process are repeated to obtain a variable q that satisfies the first distribution function condition. Of the variables q obtained by repeating the first optimization process and the second optimization process, the variable q at the time when the optimization end condition is satisfied is a candidate for the solution of the combinatorial optimization problem represented by the analysis target function.

[0094] Incidentally, a solution to an equation that contains only quadratic or lower terms of the variable q can be obtained by an annealing technique. Therefore, the combinatorial optimization problem processing device 1 includes a state measurement unit 15 that executes annealing. Therefore, the combinatorial optimization problem processing device 1 configured in this manner can obtain, by using the annealing technique, candidate solutions to combinatorial optimization problems that are expressed as functions that contain cubic or higher terms of the variables.

[0095] (Modification) It is not necessary for the state measurement unit 15 to obtain the first solution by quantum annealing. The state measurement unit 15 may obtain the first solution by, for example, non-quantum annealing. An apparatus that obtains the first solution by non-quantum annealing is, for example, a CMOS (Complementary MOS) annealing machine.

[0096] Note that the delta function of equation (4) and the exponent of the distribution function Z expressed by equation (6) do not necessarily have to be expressed by a Fourier integral representation as long as they are a predetermined integral representation by the variable X. More specifically, the predetermined integral representation may be any integral representation as long as it is a representation in which an auxiliary variable can be introduced into the exponent of the exponential function expressing the distribution function Z.

[0097] The combination conditions do not necessarily have to be stored in advance in the storage unit 12, and may be input via the input unit 13, for example.

[0098] The combinatorial optimization problem processing device 1 may be implemented using a plurality of information processing devices communicably connected via a network. In this case, each functional unit of the combinatorial optimization problem processing device 1 may be distributed and implemented in a plurality of information processing devices. For example, the condition acquisition unit 111, the solution candidate acquisition unit 112, the non-annealing calculation unit 113, the state measurement control unit 114, and the state measurement unit 15 may each be implemented in different information processing devices. The first solution acquisition unit 504 may be provided in the control unit 11.

[0099] The distribution function Z expressed by the formula (6) is an example of a distribution function to be root-finded. The state measurement unit 15 is an example of an annealing unit. The non-annealing calculation unit 113 is an example of a calculation unit. The first analysis target term is an example of a first formula. The second analysis target term is an example of a second formula.

[0100] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and designs that do not deviate from the gist of the present invention are also included. [Explanation of symbols]

[0101] 1... Combinatorial optimization problem processing device, 11... Control unit, 12... Memory unit, 13... Input unit, 14... Output unit, 15... State measurement unit, 111... Condition acquisition unit, 112... Solution candidate acquisition unit, 113... Non-annealing calculation unit, 114... State measurement control unit, 501... Quantum bit body, 502... State control unit, 503... Measurement device, 504... First solution acquisition unit

Claims

1. a partition function having a Hamiltonian that is a function expressing a combinatorial optimization problem to be analyzed and that includes third or higher terms of a variable q, the partition function being expressed by a predetermined integral representation using a variable X, the partition function being expressed by the sum of a first formula including only second or lower terms of the variable q and a second formula not including the variable q, and a first optimization process being a process of acquiring a value of the variable q that satisfies a first predetermined condition regarding the maximum value of the partition function to be root-finded, with the variable q being the only independent variable; an annealing unit that acquires candidates for the first solution by an annealing technique in an alternating optimization algorithm that repeats the first optimization process and the second optimization process until a predetermined termination condition is satisfied, the variable q that is acquired by the first optimization process and satisfies the first predetermined condition is defined as a first solution, the variable X is an independent variable only, and a process of acquiring a value of the variable X that satisfies a second predetermined condition related to a maximum value of the distribution function to be root-finded is defined as a second optimization process, and the variable X that is acquired by the second optimization process and satisfies the second predetermined condition is defined as a second solution; A calculation unit for obtaining the second solution; a solution candidate acquisition unit that acquires, when the termination condition is satisfied, the first solution candidate that has been obtained immediately before as a solution candidate for the combinatorial optimization problem; A combinatorial optimization problem processing device comprising:

2. the first predetermined condition is that the value of the partition function is a maximum value; The combination optimization problem processing apparatus according to claim 1 .

3. the second predetermined condition is that the value of the partition function is a maximum value; The combination optimization problem processing apparatus according to claim 1 or 2.

4. The predetermined integral representation is a Fourier integral representation. The combination optimization problem processing apparatus according to claim 1 .

5. The annealing method is quantum annealing. The combination optimization problem processing apparatus according to claim 1 .

6. The calculation unit obtains the second solution by a root-finding algorithm. The combination optimization problem processing apparatus according to claim 1 .

7. A process in which a computer determines a partition function to be a root-finding target as a partition function having a Hamiltonian that is a function expressing a combinatorial optimization problem to be analyzed and that includes third- or higher-order terms of a variable q, the partition function being expressed by a predetermined integral representation using a variable X, and the Hamiltonian being expressed as a sum of a first equation including only second- or lower-order terms of the variable q and a second equation not including the variable q, and the variable q is used as the only independent variable, and the value of the variable q that satisfies a first predetermined condition regarding the maximum value of the partition function to be a root-finding target is determined as a first optimization process. an annealing step of acquiring candidates for the first solution by an annealing technique in an alternating optimization algorithm in which the variable q that satisfies the first predetermined condition acquired by the first optimization process is defined as a first solution, a process of acquiring the value of the variable X that satisfies a second predetermined condition related to the maximum value of the distribution function to be root-finded, with only the variable X as an independent variable, and the variable X that satisfies the second predetermined condition acquired by the second optimization process as a second solution, and the first optimization process and the second optimization process are repeated until a predetermined termination condition is satisfied; A calculation step in which the computer obtains the second solution; a solution candidate acquisition step in which the computer acquires, when the termination condition is satisfied, the first solution candidate that has been obtained immediately before as a solution candidate for the combinatorial optimization problem; A combinatorial optimization problem processing method comprising the steps of:

8. A program for causing a computer to function as the combinatorial optimization problem processing device according to any one of claims 1 to 6.

Citation Information

Patent Citations

  • Systems and methods for generating and using higher-order interactions between quantum devices

    JP2018533106A

  • Simulation device, computer program, and simulation method

    WO2018074006A1