Estimation program, estimation device, and estimation method

Abstract

The physical quantities are estimated according to the type of substance. [Solution] The computer receives proton position information, which shows the position of protons in multiple atoms contained in the substance to be estimated, and atomic orbital information, which shows the atomic orbitals of electrons in multiple atoms. The computer uses a machine learning model trained with training data to estimate the physical quantities of the substance to be estimated from the proton position information and atomic orbital information. The training data includes information showing the position of protons in multiple atoms contained in a substance different from the substance to be estimated, and information showing the atomic orbitals of electrons in multiple atoms contained in a substance different from the substance to be estimated. The computer outputs estimation results showing the physical quantities of the substance to be estimated.

Description

[Technical Field]

[0001] This invention relates to estimation techniques using machine learning models. [Background technology]

[0002] Quantum chemical calculations are techniques for analyzing the structure and properties of compounds through theoretical calculations based on quantum mechanics. For example, quantum chemical calculations are used to calculate the energy of a molecule's ground state or excited state. One technique for calculating ground state energy using a quantum computer is the Variational Quantum Eigensolver (VQE).

[0003] A quantum computer is a computer that performs calculations by utilizing the phenomena of quantum mechanics. A quantum computer changes the quantum state of qubits using quantum circuits, which are quantum computation models that describe quantum algorithms.

[0004] Regarding quantum chemical calculations, neural network architectures using self-attention are known (see, for example, Non-Patent Document 1). Metavariational quantum eigenvalue solvers are also known (see, for example, Non-Patent Document 2). Quantum self-attention networks (QSANs) are also known (see, for example, Non-Patent Document 3).

[0005] Explainable transducer-transformers are also known (see, for example, Patent Document 1). [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Special Publication No. 2024-500182 [Non-patent literature]

[0007] [Non-Patent Document 1] I. von Glehn et al., “A Self-Attention Ansatz for Ab-initio Quantum Chemistry”, arXiv:2211.13672v2, 2023 [Non-Patent Document 2] A. Cervera-Lierta et al., “Meta-Variational Quantum Eigensolver: Learning Energy Profiles of Parameterized Hamiltonians for Quantum Simulation”, PRX QUANTUM 2, 020329, 2021 [Non-Patent Document 3] J. Shi et al., “QSAN: A Near-term Achievable Quantum Self-Attention Network”, arXiv:2207.07563v4, 2023. [Overview of the project] [Problems that the invention aims to solve]

[0008] In the techniques described in Non-Patent Document 1 or Non-Patent Document 2, it is difficult to estimate the ground state energy of various molecules using a single model.

[0009] Furthermore, this problem arises not only when estimating the ground state energy of molecules, but also when estimating various physical quantities of various substances.

[0010] In one aspect, the present invention aims to estimate physical quantities according to the type of substance. [Means for solving the problem]

[0011] One approach involves having the estimation program perform the following steps on the computer:

[0012] The computer receives proton position information indicating the positions of protons of each of a plurality of atoms included in the substance to be estimated, and atomic orbital information indicating the atomic orbitals of electrons of each of the plurality of atoms.

[0013] The computer estimates a physical quantity of the substance to be estimated from the proton position information and the atomic orbital information using a machine learning model trained using training data. The training data includes information indicating the positions of protons of each of a plurality of atoms included in a substance different from the substance to be estimated, and information indicating the atomic orbitals of electrons of each of the plurality of atoms included in a substance different from the substance to be estimated. The computer outputs an estimation result indicating the physical quantity of the substance to be estimated.

Advantages of the Invention

[0014] According to one aspect, a physical quantity can be estimated according to the type of substance.

Brief Description of the Drawings

[0015] [Figure 1] It is a functional configuration diagram of the estimation device of the embodiment. [Figure 2] It is a flowchart of the first estimation process. [Figure 3] It is a configuration diagram of a quantum computer system. [Figure 4] It is a diagram showing the first estimation model. [Figure 5] It is a diagram showing a conversion module included in the first estimation model. [Figure 6] It is a diagram showing a feature conversion circuit. [Figure 7] It is a diagram showing a state conversion circuit. [Figure 8] It is a diagram showing a MPS circuit diagram. [Figure 9] It is a diagram showing a HEA circuit that converts the first integrated feature into the second integrated feature. [Figure 10] It is a diagram showing a HEA circuit that converts the second integrated feature into the third integrated feature. [Figure 11] It is a hardware configuration diagram of a quantum computer. [Figure 12] This is a functional configuration diagram of the server. [Figure 13] This is a flowchart of the training process. [Figure 14] This is a flowchart of the second estimation process. [Figure 15] This figure shows the second estimation model. [Figure 16] This figure shows the transformation modules included in the second estimation model. [Figure 17] This is a hardware configuration diagram of an information processing device. [Modes for carrying out the invention]

[0016] The embodiments will be described in detail below with reference to the drawings.

[0017] Figure 1 shows an example of the functional configuration of the estimation device according to the embodiment. The estimation device 101 in Figure 1 includes a receiving unit 111, an estimation unit 112, and an output unit 113.

[0018] Figure 2 is a flowchart showing an example of the first estimation process performed by the estimation device 101 in Figure 1. First, the reception unit 111 receives proton position information indicating the position of protons in each of the multiple atoms contained in the substance to be estimated, and atomic orbital information indicating the atomic orbitals of electrons in each of the multiple atoms (step 201).

[0019] Next, the estimation unit 112 estimates the physical quantities of the target substance from proton position information and atomic orbital information using a machine learning model trained with training data (step 202). The training data includes information showing the positions of protons in multiple atoms contained in a substance different from the target substance, and information showing the atomic orbitals of electrons in multiple atoms contained in a substance different from the target substance. Then, the output unit 113 outputs the estimation results showing the physical quantities of the target substance (step 203).

[0020] According to the estimation device 101 in Figure 1, physical quantities can be estimated according to the type of substance.

[0021] Figure 3 shows an example configuration of a quantum computer system including the estimation device 101 of Figure 1. The quantum computer system in Figure 3 includes a quantum computer 301, a server 302, and a terminal device 303. The quantum computer 301, server 302, and terminal device 303 are hardware.

[0022] The quantum computer system shown in Figure 3 is used, for example, in quantum chemical calculations for materials development or drug discovery to estimate the ground state energy of a substance. The ground state energy of a substance is an example of a physical quantity of the substance being estimated.

[0023] Server 302 is a classical computer, and terminal device 303 is a user information processing device (computer). Server 302 corresponds to the estimation device 101 in Figure 1.

[0024] Server 302 communicates with quantum computer 301 and terminal device 303 via communication network 304. Communication network 304 is, for example, a WAN (Wide Area Network) or LAN (Local Area Network).

[0025] Server 302 stores the estimated model to be trained. During the training process, Server 302 generates a trained estimated model by performing machine learning using the quantum computer 301 to train the estimated model. The estimated model is an example of a machine learning model.

[0026] Server 302 trains an estimation model using training data. For example, proton configuration information and atomic orbital information of various materials are used as training data. Server 302 uses the training model to determine the ground state of a material from the proton configuration information and atomic orbital information included in the training data. Proton configuration information is an example of proton position information.

[0027] In the estimation process for the substance to be estimated, the server 302 receives, for example, proton configuration information and atomic orbital information of the substance to be estimated from the terminal device 303. Then, the server 302 uses a trained estimation model to determine the ground state of the substance from the proton configuration information and atomic orbital information of the substance to be estimated.

[0028] Figure 4 shows an example of the first estimation model. The estimation model in Figure 4 includes a transformation module 411 and a quantum circuit 412, and determines the ground state of substance A from the proton configuration information and atomic orbital information of substance A. Substance A is the substance represented by the training data or the substance to be estimated. Substance A may be a molecule, a compound, or an ion. The ground state of substance A is an example of the state of the substance to be estimated.

[0029] Proton configuration information indicates the position of each proton in the nucleus of each of the multiple atoms contained in substance A. For example, the three-dimensional coordinates (x,y,z) of each proton can be used as proton configuration information. If substance A contains m protons (where m is an integer greater than or equal to 2), the proton configuration information will include m three-dimensional coordinates (x,y,z).

[0030] This type of proton configuration information allows for a uniform representation of the differences in the nuclides of atoms contained in matter represented by multiple training data sets, using the number of protons, m.

[0031] Atomic orbital information indicates the atomic orbitals of each electron contained in each of the multiple atoms in substance A. Atomic orbitals are represented by wave functions that describe the distribution of electrons within an atom, and are distinguished using symbols such as 1s, 2s, and 2p.

[0032] Atomic orbital information can be expressed using, for example, orbital IDs, which are numbers assigned to atomic orbitals in descending order of their energy levels. Orbital IDs are common numbers that are independent of the nuclide. If substance A contains n electrons (where n is an integer greater than or equal to 2), the atomic orbital information will include n orbital IDs. Each orbital ID can be represented, for example, by a one-hot vector.

[0033] For example, if substance A is a molecule of lithium hydride (LiH), then m=n=4. In this case, the proton configuration information includes the three-dimensional coordinates of each of the three protons contained in the nucleus of the lithium atom (Li) and the three-dimensional coordinates of the one proton contained in the nucleus of the hydrogen atom (H).

[0034] Furthermore, the atomic orbital information includes orbital IDs "1" to "3" for the three electrons in the lithium atom (Li), and orbital ID "1" for the one electron in the hydrogen atom (H). Orbital IDs "1" and "2" represent the 1s orbital, and orbital ID "3" represents the 2s orbital.

[0035] As another example, if substance A is a molecule of beryllium hydride (BeH2), then m=n=6. In this case, the proton configuration information includes the three-dimensional coordinates of each of the four protons contained in the nucleus of the beryllium atom (Be), and the three-dimensional coordinates of the one proton contained in the nucleus of each of the two hydrogen atoms (H).

[0036] Furthermore, the atomic orbital information includes the orbital IDs "1" to "4" of the four electrons contained in the beryllium atom Be, and the orbital ID "1" of the one electron contained in each of the two hydrogen atoms H. Orbital IDs "1" and "2" represent the 1s orbital, and orbital IDs "3" and "4" represent the 2s orbital.

[0037] According to this atomic orbital information, the initial position of each electron in material A is associated with an orbital ID corresponding to the energy level of the atomic orbital, so the initial position of electrons can be uniformly represented using the orbital ID.

[0038] The conversion module 411 converts proton configuration information and atomic orbital information into characteristic information for material A. The characteristic information for material A includes feature vectors that represent the characteristics of each of the m × n combinations of protons and electrons contained in material A.

[0039] The quantum circuit 412 represents an operation that generates state information representing the ground state of material A from the characteristic information of material A, and includes a feature transformation circuit 421, quantum transformers 422-1 to 422-L (where L is an integer greater than or equal to 1), and a state transformation circuit 423.

[0040] The feature conversion circuit 421 represents the operation of converting the feature information of material A into quantum feature information. Each quantum transformer 422-s (s=1~L) includes a self-attention circuit and represents the operation of converting quantum feature information.

[0041] As the quantum transformer 422-s, for example, the QSAN described in Non-Patent Document 3 can be used. QSAN includes quantum gates for unitary transformations that compute the Query, Key, and Value, respectively, and each quantum gate includes parameters. The parameters included in QSAN are an example of the second parameter.

[0042] The state conversion circuit 423 represents the operation of converting the quantum feature information output from the quantum transformer 422-L into state information that represents the ground state of the n-electron system.

[0043] Server 302 uses the conversion module 411 to generate characteristic information of material A from proton configuration information and atomic orbital information, and then has the quantum computer 301 execute the operations represented by the quantum circuit 412, thereby obtaining state information of material A from the quantum computer 301. Server 302 then uses the state information of material A to determine the energy of the ground state of material A.

[0044] Figure 5 shows an example of the transformation module 411 in Figure 4. The transformation module 411 in Figure 5 includes a transformation layer 511, addition 512, subtraction 513, norm calculation 514, concatenation operation 515, transformation layer 516, and a tanh function 517.

[0045] The transformation layer 511 converts the orbital ID of the j-th (j=1 to n) electron E(j) contained in the atomic orbital information into a 3D vector V(j) representing the relative position of the electron with respect to the atomic nucleus. The transformation layer 511 obtains n 3D vectors V(j). The transformation layer 511 may be, for example, a neural network including parameters. The transformation layer 511 is an example of a first module, and the parameters included in the transformation layer 511 are an example of first parameters.

[0046] By using a three-dimensional vector V(j) representing the relative position of electrons, it is possible to change the electron position according to the difference in proton positions, even if the positions of protons in the material represented by each of the multiple training data sets are different. Furthermore, by optimizing the parameters of the transformation layer 511 through training, it becomes possible to accurately determine the three-dimensional vector V(j) of n-electron systems contained in various materials.

[0047] Addition 512 adds the 3D vector V(j) output from transformation layer 511 to the 3D coordinates of the i-th proton P(i) (i=1~m) included in the proton configuration information. This allows us to obtain the 3D coordinates of the electron E(j) included in the combination of proton P(i) and electron E(j).

[0048] Subtraction 513 subtracts the 3D coordinates of proton P(i) from the 3D coordinates of electron E(j) obtained by addition 512. This gives the 3D vector RP(i,j) representing the relative position of electron E(j) to proton P(i). Subtraction 513 gives m × n 3D vectors RP(i,j). Addition 512 and subtraction 513 are examples of the second module.

[0049] Norm calculation 514 calculates the L2 norm R(i,j) of the 3D vector RP(i,j) obtained by subtraction 513. Norm calculation 514 calculates m × n L2 norms R(i,j). The L2 norm R(i,j) represents the Euclidean distance between the proton P(i) and the electron E(j).

[0050] The concatenation operation 515 generates a 4-dimensional vector V1(i,j) by concatenating a 3-dimensional vector RP(i,j) with an L2 norm R(i,j). The concatenation operation 515 generates m × n 4-dimensional vectors V1(i,j).

[0051] The transformation layer 516 transforms the 4-dimensional vector V1(i,j) generated by the concatenation operation 515 into a d-dimensional (d is an integer greater than or equal to 2) vector V2(i,j). The transformation layer 516 obtains m × n vectors V2(i,j). The transformation layer 516 may be, for example, a neural network including parameters.

[0052] The tanh function 517 is a hyperbolic tangent function that generates a feature vector FV(i,j) by transforming each of the d elements of the vector V2(i,j) to a value in the range of -1 to 1. The tanh function 517 generates m × n feature vectors FV(i,j). The tanh function 517 is sometimes called an activation function. Norm calculation 514, concatenation operation 515, transformation layer 516, and the tanh function 517 are examples of the third module.

[0053] Server 302 uses the conversion module 411 to generate m × n feature vectors FV(i,j) from atomic orbital information and proton configuration information. These m × n feature vectors FV(i,j) are used as the feature information of material A. Server 302 transmits the feature information of material A and the information of quantum circuit 412 to quantum computer 301.

[0054] The quantum computer 301 sets m × n × d qubits to an initial state |0>, and then performs the operations represented by the quantum circuit 412 to obtain a measurement value for each qubit. The measurement value for each qubit represents either a logical value of "0" or a logical value of "1".

[0055] As an example, let's assume that substance A is a molecule of beryllium hydride BeH2. In this case, m=n=6. Proton P(1) represents the proton contained in the first hydrogen atom H, protons P(2) to P(5) represent the four protons contained in the nucleus of the beryllium atom Be, and proton P(6) represents the proton contained in the second hydrogen atom H.

[0056] Electron E(1) represents the electron contained in the first hydrogen atom H, electrons E(2) to E(5) represent the four electrons contained in the beryllium atom Be, and electron E(6) represents the electron contained in the second hydrogen atom H.

[0057] In this example, the three-dimensional coordinates (x,y,z) of protons P(1) to P(6) included in the proton configuration information are as follows:

[0058] P(1) (0,0,-2.5) P(2) (0,0,0) P(3) (0,0,0) P(4) (0,0,0) P(5) (0,0,0) P(6) (0,0,2.5)

[0059] The orbital IDs for electrons E(1) to E(6) included in the atomic orbital information are as follows:

[0060] E(1) 1 E(2) 1 E(3) 2 E(4) 3 E(5) 4 E(6) 1

[0061] Orbital IDs "1" and "2" represent the 1s orbital, while orbital IDs "3" and "4" represent the 2s orbital. The one-hot vectors representing orbital IDs "1" through "4" are as follows:

[0062] 1 (1,0,0,0) 2 (0,1,0,0) 3 (0,0,1,0) 4 (0,0,0,1)

[0063] The 3D vectors V(1) to V(6) output from transformation layer 511 are as follows:

[0064] V(1)=(x(1),y(1),z(1)) V(2)=(x(1),y(1),z(1)) V(3)=(x(2),y(2),z(2)) V(4)=(x(3),y(3),z(3)) V(5)=(x(4),y(4),z(4)) V(6)=(x(1),y(1),z(1))

[0065] The three-dimensional coordinates of electrons E(1) to E(6), obtained by addition 512, are as follows:

[0066] E(1) (x(1),y(1),z(1)-2.5) E(2) (x(1),y(1),z(1)) E(3) (x(2),y(2),z(2)) E(4) (x(3),y(3),z(3)) E(5) (x(4),y(4),z(4)) E(6) (x(1),y(1),z(1)+2.5)

[0067] The 3D vectors RP(1,1) to RP(6,1) of electron E(1), obtained by subtraction 513, are as follows:

[0068] RP(1,1)=(x(1),y(1),z(1)) RP(2,1)=(x(1),y(1),z(1)-2.5) RP(3,1)=(x(1),y(1),z(1)-2.5) RP(4,1)=(x(1),y(1),z(1)-2.5) RP(5,1)=(x(1),y(1),z(1)-2.5) RP(6,1) = (x(1), y(1), z(1) - 5)

[0069] The 3D vectors RP(1,j) to RP(6,j) of the other electrons E(j) (j = 2 to 6) are also obtained in the same way. The L2 norms R(1,1) to R(6,1) of the electron E(1) obtained by the norm calculation 514 are as follows.

[0070] R(1,1) = (x(1) ,

[0072] , , + y(1) 2 + z(1) 2 ) 1 / 2 R(2,1) = (x(1) 2 + y(1) 2 + (z(1) - 2.5) 2 ) 1 / 2 R(3,1) = (x(1) 2 + y(1) 2 + (z(1) - 2.5) 2 ) 1 / 2 R(4,1) = (x(1) 2 + y(1) 2 + (z(1) - 2.5) 2 ) 1 / 2 R(5,1) = (x(1) 2 + y(1) 2 + (z(1) - 2.5) 2 ) 1 / 2 R(6,1) = (x(1) 2 + y((1) 2 + (z(1) - 5) 2 ) 1 / 2

[0071] The L2 norms R(1,j) to R(6,j) of the other electrons E(j) (j = 2 to 6) are also obtained in the same way. The 4D vectors V1(1,1) to V1(6,1) of the electron E(1) generated by the concatenation operation 515 are as follows. [[ID=,66]]

[0072] V1(1,1) = (x(1), y(1), z(1), R(1,1)) V1(2,1)=(x(1),y(1),z(1)-2.5,R(2,1)) V1(3,1)=(x(1),y(1),z(1)-2.5,R(3,1)) V1(4,1)=(x(1),y(1),z(1)-2.5,R(4,1)) V1(5,1)=(x(1),y(1),z(1)-2.5,R(5,1)) V1(6,1)=(x(1),y(1),z(1)-5,R(6,1))

[0073] The 4-dimensional vectors V1(1,j) to V1(6,j) of other electrons E(j) (j=2 to 6) can be obtained in the same way.

[0074] According to the conversion module 411 in Figure 5, m × n feature vectors FV(i,j) are generated from atomic orbital information and proton configuration information, determined according to the number of protons m and the number of electrons n. By using such feature vectors FV(i,j), the characteristic information of various materials input to the quantum circuit 412 can be represented uniformly.

[0075] Figure 6 shows an example of the feature transformation circuit 421 from Figure 4. The feature transformation circuit 421 in Figure 6 includes an arccos function 611 and an embedding circuit 612.

[0076] The arccos function 611 generates m × n × d angles by converting each of the d elements of the feature vector FV(i,j) contained in the feature information output from the transformation module 411 into an angle.

[0077] The embedded circuit 612 represents a rotation operation applied to each of the m × n × d qubits set to the initial state |0>. The rotation operation for each qubit is performed using each of the m × n × d angles generated by the arccos function 611. The quantum gate representing the rotation operation may also be a Y rotation gate.

[0078] By performing the operations represented by the feature transformation circuit 421, the information of m × n feature vectors FV(i,j) is embedded into m × n × d qubits. This generates quantum feature information containing m × n × d quantum features.

[0079] Figure 7 shows an example of the state conversion circuit 423 in Figure 4. The state conversion circuit 423 in Figure 7 includes n × d MPS (Matrix Product State) circuits 711-jp (j=1~n, p=1~d), d HEA (Hardware Efficient Ansatz) circuits 712-p (p=1~d), and a HEA circuit 713.

[0080] Each MPS circuit 711-jp represents a 2-bond MPS Ansatz operation that transforms the p-th element of the feature vectors FV(1,j) to FV(m,j) into the first unified feature of electron E(j). The first unified feature of electron E(j) represents the feature obtained by unifying the p-th elements of the feature vectors FV(1,j) to FV(m,j).

[0081] Each HEA circuit 712-p represents the operation of converting n first unified features output from MPS circuits 711-1-p to 711-np into n second unified features. HEA circuit 713 represents the operation of converting n × d second unified features output from HEA circuits 712-1 to 712-d into n × d third unified features.

[0082] The quantum computer 301 obtains q measurement values ​​by performing measurement 714 on the q qubits from the 1st to the qth qubits, which are among the n × d qubits representing the third unification feature. The quantum computer 301 then transmits the obtained q measurement values ​​to the server 302 as state information for material A.

[0083] Figure 8 shows an example of the MPS circuit 711-jp from Figure 7. The MPS circuit 711-jp in Figure 8 contains m-1 Ups. Up represents a unitary transform and contains 4×4 parameters. The parameters included in Up are examples of the second parameter. FV(i,j)(p)(i=1~m) represents the p-th element of the feature vector FV(i,j).

[0084] The first Up represents the unitary transformation of FV(1,j)(p) and FV(2,j)(p). The i-th (i=2~m-1) Up represents the result of the transformation by the (i-1)th Up and the unitary transformation of FV(i+1,j)(p). The result of the transformation by the (m-1)th Up is output as the unified feature MPS(j,p). The unified feature MPS(j,p) corresponds to the first unified feature of electron E(j).

[0085] Figure 9 shows an example of the HEA circuit 712-p from Figure 7. Ry represents a Y rotation gate, and Rz represents a Z rotation gate. Ry and Rz include parameters. The parameters included in Ry and Rz are examples of second parameters. A line segment connecting two adjacent qubits represents a CZ gate. () × g indicates that the circuit inside the parentheses is repeated g stages (where g is an integer greater than or equal to 1).

[0086] In the first stage of the circuit, Ry transforms the integrated feature MPS(j,p) of electron E(j), and Rz transforms the result of the transformation by Ry. The CZ gate transforms the quantum states of two adjacent qubits. The circuits from the second to g stages represent the same operation as the first stage.

[0087] The transformation result from the g-stage circuit is transformed by the final Ry, and the transformation result from the final Ry is transformed by the final Rz. The transformation result from the final Rz is output as the integrated feature HEA(j,p). The integrated feature HEA(j,p) corresponds to the second integrated feature of electron E(j).

[0088] Figure 10 shows an example of the HEA circuit 713 from Figure 7. Ry represents a Y rotation gate, and Rz represents a Z rotation gate. Ry and Rz include parameters. The parameters included in Ry and Rz are examples of second parameters. A line segment connecting two adjacent qubits represents a CZ gate. () × h indicates that the circuit inside the parentheses is repeated h stages (where h is an integer greater than or equal to 1).

[0089] In the first stage of the circuit, Ry transforms the integrated feature HEA(j,p), and Rz transforms the result of the transformation by Ry. The CZ gate transforms the quantum states of two adjacent qubits. The circuits from the second to the hth stage represent the same operation as the first stage.

[0090] The conversion result from the h-stage circuit is converted by the final Ry, and the conversion result from the final Ry is converted by the final Rz. The conversion result from the final Rz corresponds to the third integrated feature. Of the n × d final Rz conversion results, the q conversion results from the 1st to the qth are output as integrated feature HEA(1) to integrated feature HEA(q).

[0091] Quantum computer 301 obtains the measured values ​​for each of the integrated feature HEA(1) to integrated feature HEA(q) by performing measurement 714.

[0092] Figure 11 shows an example of the hardware configuration of the quantum computer 301 in Figure 3. The quantum computer 301 in Figure 11 includes a communication interface 1111, a control unit 1112, and a quantum system 1113. These components are hardware.

[0093] The communication interface 1111 is a communication circuit connected to the communication network 304 and performs data conversion associated with communication. The quantum system 1113 includes a quantum device that realizes multiple qubits.

[0094] The communication interface 1111 receives characteristic information of material A and information of the quantum circuit 412 from the server 302 and outputs it to the control device 1112. The control device 1112 generates a control signal using the characteristic information of material A and information of the quantum circuit 412 and outputs it to the quantum system 1113.

[0095] The quantum system 1113 obtains a measurement value for each qubit by manipulating the qubits according to the control signal. The control device 1112 then transmits the measurement value for each qubit obtained by the quantum system 1113 to the server 302 via the communication interface 1111.

[0096] The quantum computer 301 performs M measurements 714 by repeating the operation represented by the quantum circuit 412 M times (where M is an integer greater than or equal to 1). The quantum computer 301 then transmits the state information obtained from each of the M measurements 714 to the server 302.

[0097] Figure 12 shows an example of the functional configuration of the server 302 in Figure 3. The server 302 in Figure 12 includes a communication unit 1211, an acquisition unit 1212, a quantization unit 1213, a training unit 1214, an estimation unit 1215, and a storage unit 1216. The communication unit 1211, the acquisition unit 1212, and the estimation unit 1215 correspond to the output unit 113, the receiving unit 111, and the estimation unit 112 in Figure 1, respectively.

[0098] The communication unit 1211 communicates with the quantum computer 301 and the terminal device 303 via the communication network 304. The communication unit 1211 can also communicate with a database server (not shown) or the like via the communication network 304. The storage unit 1216 stores the estimated model 1221 to be trained.

[0099] The acquisition unit 1212 acquires a training data set 1222 containing multiple training data from a database server or the like via the communication unit 1211 and stores it in the storage unit 1216. The acquisition unit 1212 may also acquire the training data set 1222 input from the operator via a user interface or portable recording medium.

[0100] During the training process, the quantization unit 1213 generates a Hamiltonian HM of the n-electron system of the material represented by each training data in the training data set 1222, and maps the ground state of the n-electron system to q qubits.

[0101] The training unit 1214 communicates with the quantum computer 301 via the communication unit 1211 and trains the estimation model 1221 using the training data set 1222 to generate a trained estimation model 1223, which is then stored in the storage unit 1216.

[0102] The training unit 1214 uses the conversion module 411 included in the estimation model 1221 to generate material characteristic information from the proton configuration information and atomic orbital information included in each training data. The training unit 1214 then transmits the material characteristic information and the information of the quantum circuit 412 included in the estimation model 1221 to the quantum computer 301 via the communication unit 1211, and receives M state information of the material from the quantum computer 301.

[0103] A typical Hamiltonian HM is represented by a linear combination of the products of multiple Pauli operators. In this case, the training unit 1214 can calculate the expectation value of the Hamiltonian HM using the q logical values ​​contained in each of the M received state information. The calculated expectation value of the Hamiltonian HM represents the energy of the ground state of matter.

[0104] The training unit 1214 updates the parameters of the transformation module 411 and the quantum circuit 412 in a direction that decreases the expected value of the Hamiltonian HM, using backpropagation. The parameters of the transformation module 411 are those included in the transformation layers 511 and 516. The parameters of the quantum circuit 412 are those included in the quantum transformer 422-s, the MPS circuit 711-jp, the HEA circuit 712-p, and the HEA circuit 713.

[0105] The training unit 1214 can update each parameter by, for example, calculating the gradient using the parameter shift method. The training unit 1214 determines the value of each parameter by, for example, repeating the parameter update over N epochs (where N is an integer greater than or equal to 1). Then, the training unit 1214 generates a trained estimation model 1223 by setting the determined parameter values ​​in the transformation module 411 and the quantum circuit 412.

[0106] By optimizing the parameters of the transformation layer 511 through training, it becomes possible to accurately determine the characteristic information of various materials. Furthermore, by optimizing the parameters of the quantum circuit 412 through training, it becomes possible to accurately determine the state information of various materials.

[0107] In the estimation process, the acquisition unit 1212 acquires proton configuration information and atomic orbital information of the target substance from the terminal device 303 via the communication unit 1211, and accepts the acquired proton configuration information and atomic orbital information. The acquisition unit 1212 may also acquire proton configuration information and atomic orbital information input from the operator via a user interface or portable recording medium.

[0108] The quantization unit 1213 generates the Hamiltonian HM of the n-electron system of the estimated material and maps the ground state of the n-electron system to q qubits.

[0109] The estimation unit 1215 communicates with the quantum computer 301 via the communication unit 1211, and uses the estimation model 1223 to estimate the ground state energy of the target material, generating an estimation result showing the estimated energy. The estimation unit 1215 then transmits the estimation result to the terminal device 303 via the communication unit 1211. The terminal device 303 displays the received estimation result on its screen.

[0110] The estimation unit 1215 uses the conversion module 411 included in the estimation model 1223 to generate characteristic information of the material from the proton configuration information and atomic orbital information of the material to be estimated. Then, the estimation unit 1215 transmits the characteristic information of the material to be estimated and the information of the quantum circuit 412 included in the estimation model 1223 to the quantum computer 301 via the communication unit 1211, and receives M state information of the material to be estimated from the quantum computer 301.

[0111] The estimation unit 1215 then calculates the expected value of the Hamiltonian HM using the q logical values ​​contained in each of the M received state information. The calculated expected value of the Hamiltonian HM represents the ground state energy of the substance being estimated.

[0112] Of the conversion module 411 and quantum circuit 412 included in the estimation model 1223, the quantum computer 301 can perform the operation represented by the quantum circuit 412, thereby enabling the rapid acquisition of state information for the estimated substance.

[0113] Figure 13 is a flowchart showing an example of the training process performed by the server 302 in Figure 12. First, the training unit 1214 initializes the parameters of the transformation module 411 and the quantum circuit 412 included in the estimated model 1221 (step 1301).

[0114] Next, server 302 repeats the training loop process from steps 1302 to 1308 N times, where N is the number of epochs.

[0115] In the training loop, the quantization unit 1213 generates a Hamiltonian HM of the n-electron system of the material represented by each training data in the training data set 1222 (step 1302). Then, the quantization unit 1213 maps the ground state of the n-electron system to q qubits (step 1303).

[0116] Next, the training unit 1214 uses the conversion module 411 included in the estimated model 1221 to generate material characteristic information from the proton configuration information and atomic orbital information included in each training data (step 1304). Then, the training unit 1214 transmits the material characteristic information and the information of the quantum circuit 412 included in the estimated model 1221 to the quantum computer 301 via the communication unit 1211 (step 1305).

[0117] The quantum computer 301 uses the material's characteristic information to generate M state information of the material by repeating the operation represented by the quantum circuit 412 M times. The quantum computer 301 then transmits the generated M state information to the server 302.

[0118] The training unit 1214 receives M state information from the quantum computer 301 via the communication unit 1211 (step 1306). Then, the training unit 1214 calculates the expectation value of the Hamiltonian HM using the q logical values ​​contained in each of the M received state information (step 1307).

[0119] Next, the training unit 1214 updates the parameters of the transformation module 411 and the quantum circuit 412 in a direction that decreases the expected value of the Hamiltonian HM using the backpropagation method (step 1308).

[0120] After the training loop has been executed N times, the training unit 1214 stores the estimated model 1221, which includes the updated parameters, in the storage unit 1216 as the estimated model 1223 (step 1309).

[0121] In each training loop, the training unit 1214 may use all of the training data included in the training data set 1222, or it may use only some of the training data. In each training loop, the training unit 1214 may randomly select the training data.

[0122] For example, if the training data set 1222 contains training data representing molecules of hydrogen (H2), beryllium hydride (BeH2), and H4 (where hydrogen atoms are arranged at the vertices of a rectangle), then in each training loop, the training unit 1214 randomly selects one of these molecules. The training unit 1214 then determines the position of each proton in the selected molecule by setting the bond lengths between atoms to random values. This generates different training data for each training loop.

[0123] By training the estimation model 1221 using training data of molecules with various bond lengths (hydrogen H2, beryllium hydride BeH2, and hydrogen in a rectangular arrangement H4), an estimation model 1223 can be generated to estimate the ground state energy of various substances.

[0124] Figure 14 is a flowchart showing an example of the second estimation process performed by the server 302 in Figure 12. First, the acquisition unit 1212 receives material information of the substance to be estimated from the terminal device 303 via the communication unit 1211 (step 1401).

[0125] The material information includes multiple proton configurations and atomic orbital information. Each proton configuration represents the configuration of multiple protons contained in the estimated material, and each of the multiple proton configurations represents a different configuration. The configurations of multiple protons correspond to the configurations of multiple atoms contained in the estimated material.

[0126] Next, the quantization unit 1213 generates the Hamiltonian HM of the n-electron system of the material to be estimated (step 1402). Then, the quantization unit 1213 maps the ground state of the n-electron system to q qubits (step 1403).

[0127] Next, the estimation unit 1215 selects one proton configuration from among the multiple proton configurations received (step 1404). Then, the estimation unit 1215 uses the conversion module 411 included in the estimation model 1223 to generate characteristic information of the target material from the selected proton configuration and atomic orbital information (step 1405).

[0128] Next, the estimation unit 1215 transmits the characteristic information of the material to be estimated and the information of the quantum circuit 412 included in the estimation model 1223 to the quantum computer 301 via the communication unit 1211 (step 1406).

[0129] The quantum computer 301 generates M state information for a substance by repeating the operation represented by the quantum circuit 412 M times using the characteristic information of the substance to be estimated. Then, the quantum computer 301 transmits the generated M state information to the server 302.

[0130] The estimation unit 1215 receives M state information from the quantum computer 301 via the communication unit 1211 (step 1407). Then, the estimation unit 1215 calculates the expectation value of the Hamiltonian HM using the q logical values ​​contained in each of the M received state information (step 1408).

[0131] Next, the estimation unit 1215 checks whether all proton configuration information has been selected (step 1409). If there is still unselected proton configuration information remaining (step 1409, NO), the estimation unit 1215 repeats the process from step 1404 onwards for the next proton configuration information.

[0132] If all proton configuration information is selected (step 1409, YES), the estimation unit 1215 generates an estimation result that includes the expected value of the Hamiltonian HM obtained using each of the multiple proton configuration information (step 1410). Then, the estimation unit 1215 transmits the estimation result to the terminal device 303 via the communication unit 1211 (step 1411).

[0133] The terminal device 303 generates a space for plotting the ground state energy surface of the target material using variables that distinguish between the arrangements of multiple atoms represented by each proton arrangement information. Variables used to distinguish between the arrangements of multiple atoms include the bond length between two atoms and the bond angle formed by three atoms.

[0134] The terminal device 303 generates a ground state energy surface by plotting each expected value included in the estimation result at a position corresponding to the value of each variable in the generated space, and displays the generated energy surface on the screen.

[0135] According to the quantum computer system shown in Figure 3, various molecules, compounds, ions, and other materials are represented using common information such as the number of protons (m), the number of electrons (n), and the orbital ID of each electron, thus increasing the versatility of estimation model 1223. Therefore, by calculating the ground state energy using estimation model 1223, the ground state energy can be efficiently estimated according to the type of material.

[0136] As an example, consider a scenario where estimation model 1223 is generated using training data of molecules with various bond lengths for hydrogen (H2), beryllium hydride (BeH2), and hydrogen atoms arranged in a rectangular configuration (H4). In this case, the generated estimation model 1223 can be used to estimate the ground state energy of a lithium hydride molecule (LiH) that has a different molecular structure from the training data.

[0137] A hydrogen atom (H) contains one proton, a beryllium atom (Be) contains four protons, and a lithium atom (Li) contains three protons. The number of protons in a lithium atom (Li) is greater than that of a hydrogen atom (H), but less than that of a beryllium atom (Be).

[0138] In this case, using the generated estimation model 1223, an effect is obtained in which the state of the Li-H bond is interpolated using the state of the HH bond and the state of the Be-H bond. Therefore, even for lithium hydride LiH molecules with molecular structures different from the training data, an effective estimate of the ground state energy can be obtained.

[0139] Users can analyze reaction rates, molecular structure changes, and other factors in chemical reactions based on the shape of the energy surface displayed on the screen. Therefore, using the quantum computer system shown in Figure 3 can streamline simulations in the fields of materials development or drug discovery.

[0140] Estimation model 1223 can also be used to estimate physical quantities other than the ground state energy of a substance. These other physical quantities may be the position or momentum of each electron in the ground state, or the energy of an excited state.

[0141] Figure 15 shows an example of the second estimation model. The estimation model in Figure 15 has the same structure as the estimation model in Figure 4, but with the conversion module 411 replaced by the conversion module 1511.

[0142] Nuclear configuration information indicates the position of the nucleus of each of the multiple atoms contained in substance A. For example, the three-dimensional coordinates (x,y,z) of each nucleus are used as nuclear configuration information. If substance A contains k atoms (where k is an integer greater than or equal to 2), the nuclear configuration information includes k three-dimensional coordinates (x,y,z). Nuclear configuration information is an example of proton position information.

[0143] The proton number information indicates the number of protons in each of the multiple atoms contained in substance A. Using this nuclear configuration information and proton number information, the differences in the nuclides of atoms contained in the substance represented by each of the multiple training data can be uniformly expressed using the number of atoms k and the number of protons in each atom.

[0144] Figure 16 shows an example of the conversion module 1511 in Figure 15. The conversion module 1511 in Figure 16 has a structure in which the multiplication 1611 is added to the conversion module 411 in Figure 5. The proton number information includes the number of protons m(i) contained in the atomic nucleus AN(i) (i=1~k) of each atom.

[0145] The conversion layer 511 converts the orbital ID of the j-th electron E(j) (j=1 to n) included in the atomic orbital information into a three-dimensional vector V(j) that represents the relative position of the electron with respect to the atomic nucleus.

[0146] Addition 512 adds the 3D vector V(j) output from transformation layer 511 to the 3D coordinates of the i-th nucleus AN(i) included in the nucleus arrangement information. This allows us to obtain the 3D coordinates of the electron E(j) included in the combination of nucleus AN(i) and electron E(j).

[0147] Subtraction 513 subtracts the 3D coordinates of the atomic nucleus AN(i) from the 3D coordinates of the electron E(j) obtained by addition 512. This gives the 3D vector RP(i,j) that represents the relative position of the electron E(j) to the atomic nucleus AN(i). Subtraction 513 gives k × n 3D vectors RP(i,j).

[0148] Norm calculation 514 calculates the L2 norm R(i,j) of the 3D vector RP(i,j) obtained by subtraction 513. Norm calculation 514 calculates k × n L2 norms R(i,j). The L2 norm R(i,j) represents the Euclidean distance between the atomic nucleus AN(i) and the electron E(j).

[0149] The concatenation operation 515 generates a 4-dimensional vector V1(i,j) by concatenating a 3-dimensional vector RP(i,j) with an L2 norm R(i,j). The concatenation operation 515 generates k × n 4-dimensional vectors V1(i,j).

[0150] The transformation layer 516 transforms the 4-dimensional vector V1(i,j) generated by the concatenation operation 515 into a d-dimensional vector V2(i,j). The transformation layer 516 obtains k × n vectors V2(i,j).

[0151] The tanh function 517 generates a feature vector FV(i,j) by converting each of the d elements of the vector V2(i,j) to a value in the range of -1 to 1. The tanh function 517 generates k × n feature vectors FV(i,j).

[0152] Multiplication 1611 generates the amplified feature vector AFV(i,j) by multiplying the value of each element of the feature vector FV(i,j) by the number of protons m(i) in the atomic nucleus AN(i) included in the proton number information. Norm calculation 514, concatenation operation 515, transformation layer 516, tanh function 517, and multiplication 1611 are examples of the third module.

[0153] In the training or estimation process using the estimation model in Figure 15, nuclear configuration information and proton number information are used instead of proton configuration information. In this case, the server 302 uses the conversion module 1511 to generate k × n feature vectors AFV(i,j) from the atomic orbital information, nuclear configuration information, and proton number information. The k × n feature vectors AFV(i,j) are used as the feature information of material A. The server 302 transmits the feature information of material A and the information of the quantum circuit 412 to the quantum computer 301.

[0154] The structure of quantum circuit 412 is similar to the estimated model in Figure 4. However, the arccos function 611 in Figure 6 generates k × n × d angles by converting d elements of the feature vector AFV(i,j) into angles instead of the feature vector FV(i,j). The embedded circuit 612 represents the rotation operation applied to each of the k × n × d qubits set in the initial state |0>.

[0155] In this case, each MPS circuit 711-jp in Figure 7 represents a 2-bond MPS Ansatz operation that transforms the p-th element of the feature vector AFV(1,j) to feature vector AFV(k,j) into the first unified feature of electron E(j).

[0156] The quantum computer 301 generates M state information for material A by repeating the operation represented by the quantum circuit 412 M times using the characteristic information of material A. Then, the quantum computer 301 transmits the generated M state information to the server 302.

[0157] According to the conversion module 1511 in Figure 16, k × n feature vectors AFV(i,j) are generated from atomic orbital information, nuclear configuration information, and proton number information, with the number of nuclei k and the number of electrons n being determined accordingly. Each element of the feature vector AFV(i,j) has a value corresponding to the number of protons m(i) in the nucleus AN(i). By using such feature vectors AFV(i,j), the characteristic information of various materials input to the quantum circuit 412 can be uniformly represented.

[0158] By using k×n feature vectors AFV(i,j) instead of m×n feature vectors FV(i,j), the number of qubits used in the operation represented by quantum circuit 412 decreases from m×n×d to k×n×d. Therefore, using the estimation model in Figure 15 reduces the computational resources of quantum computer 301 compared to using the estimation model in Figure 4.

[0159] Server 302 does not necessarily need to use the quantum computer 301 to estimate the physical quantities of matter, and the estimation model does not necessarily need to include the quantum circuit 412. If the quantum computer 301 is not used, the quantum circuit 412 included in the estimation model is replaced with another machine learning model. This other machine learning model may be a neural network or a random forest.

[0160] The configuration of the estimation device 101 shown in Figure 1 is merely an example, and some components may be omitted or changed depending on the application or conditions of the estimation device 101.

[0161] The configuration of the quantum computer system shown in Figure 3 is merely an example, and some components may be omitted or modified depending on the application or conditions of the quantum computer system. For example, if the estimation model does not include quantum circuit 412, quantum computer 301 can be omitted.

[0162] The configuration of the quantum computer 301 shown in Figure 11 is merely an example, and some components may be omitted or modified depending on the application or conditions of the quantum computer system.

[0163] The configuration of server 302 in Figure 12 is merely an example, and some components may be omitted or modified depending on the application or conditions of the quantum computer system. For example, if the training process is performed by an external device, the training unit 1214 can be omitted.

[0164] The flowcharts in Figures 2, 13, and 14 are merely examples, and some processes may be omitted or modified depending on the configuration or conditions of the estimation device 101 or the quantum computer system. For example, if the training process is performed by an external device, the server 302 may omit the training process shown in Figure 13.

[0165] The structures of the estimation models shown in Figures 4-10, 15, and 16 are merely examples, and the structure of the estimation model will vary depending on the application or conditions of the quantum computer system.

[0166] Figure 17 shows an example of the hardware configuration of an information processing device used as the estimation device 101 in Figure 1 and the server 302 in Figure 12. The information processing device in Figure 17 includes a CPU (Central Processing Unit) 1701, memory 1702, input device 1703, output device 1704, auxiliary storage device 1705, media drive device 1706, and network connection device 1707. These components are hardware and are connected to each other by a bus 1708.

[0167] Memory 1702 is, for example, a semiconductor memory such as ROM (Read Only Memory) or RAM (Random Access Memory), and stores the program and data used for processing. Memory 1702 may also operate as the storage unit 1216 in Figure 12.

[0168] The CPU 1701 (processor) operates as the reception unit 111 and estimation unit 112 in Figure 1, for example, by executing a program using the memory 1702. The CPU 1701 also operates as the acquisition unit 1212, quantization unit 1213, training unit 1214, and estimation unit 1215 in Figure 12, by executing a program using the memory 1702.

[0169] The input device 1703 is, for example, a keyboard, a pointing device, etc., and is used for inputting instructions or information from the operator. The output device 1704 is, for example, a display device, a printer, etc., and is used for inquiries or instructions to the operator and outputting processing results. The processing results may be estimated results.

[0170] The auxiliary storage device 1705 is, for example, a magnetic disk drive, an optical disk drive, a magneto-optical disk drive, a tape drive, etc. The auxiliary storage device 1705 may also be a hard disk drive or an SSD (Solid State Drive). The information processing device can store programs and data in the auxiliary storage device 1705 and load them into the memory 1702 for use. The auxiliary storage device 1705 may also operate as the storage unit 1216 in Figure 12.

[0171] The media drive unit 1706 drives the portable recording medium 1709 and accesses its recorded contents. The portable recording medium 1709 can be a memory device, a flexible disk, an optical disk, a magneto-optical disk, etc. The portable recording medium 1709 may also be a CD-ROM (Compact Disk Read Only Memory), a DVD (Digital Versatile Disk), a USB (Universal Serial Bus) memory, etc. The operator can store programs and data on the portable recording medium 1709 and load them into the memory 1702 for use.

[0172] Thus, the computer-readable recording medium that stores the programs and data used in the processing is a physical (non-temporary) recording medium such as memory 1702, auxiliary storage device 1705, or portable recording medium 1709.

[0173] The network connection device 1707 is a communication circuit connected to the communication network 304 that performs data conversion associated with communication. The network connection device 1707 may also operate as the communication unit 1211 in Figure 12. The information processing device can receive programs and data from external devices via the network connection device 1707, load them into the memory 1702, and use them.

[0174] Note that the information processing device does not need to include all the components shown in Figure 17, and some components may be omitted depending on the intended use or conditions of the information processing device. For example, if an interface with an operator is not required, the input device 1703 and output device 1704 may be omitted. If the portable recording medium 1709 is not used, the media drive device 1706 may be omitted.

[0175] As the terminal device 303 in Figure 3, an information processing device with the same configuration as in Figure 17 can be used.

[0176] While embodiments of the disclosure and their advantages have been described in detail, those skilled in the art will be able to make various modifications, additions, and omissions without departing from the scope of the invention as expressly stated in the claims.

[0177] With reference to Figures 1 to 17, the following additional information is disclosed regarding the embodiments described. (Note 1) The system receives proton position information indicating the position of protons in each of several atoms contained in the substance to be estimated, and atomic orbital information indicating the atomic orbitals of electrons in each of the aforementioned atoms. Using a machine learning model trained with training data that includes information indicating the positions of protons in multiple atoms contained in a substance different from the substance to be estimated, and information indicating the atomic orbitals of electrons in multiple atoms contained in a substance different from the substance to be estimated, the physical quantities of the substance to be estimated are estimated from the proton position information and the atomic orbital information. Outputting estimation results showing the physical quantities of the substance to be estimated. An estimation program that causes a computer to perform a process. (Note 2) The aforementioned machine learning model, A conversion module that converts the proton position information and the atomic orbital information into characteristic information of the estimated target substance, A quantum circuit that generates state information indicating the state of the estimated target substance from characteristic information of the estimated target substance, Includes, The process for estimating the physical quantity of the substance to be estimated is as follows: Using the aforementioned conversion module, a process is performed to generate characteristic information of the estimated target substance from the proton position information and the atomic orbital information. The process involves having a quantum computer perform calculations on the quantum circuit, thereby obtaining the state information from the quantum computer, A process to determine the physical quantity of the estimated target substance based on the state information, The estimation program described in Appendix 1, characterized by including the following: (Note 3) The characteristic information of the estimated target substance includes the characteristics of each of the multiple combinations of protons and electrons contained in the estimated target substance. The aforementioned conversion module is A first module that determines relative position information based on the atomic orbital information of each electron of a plurality of atoms contained in the estimated target substance and a first parameter, A second module that determines the relative position of the electron to the proton based on the proton position information and the relative position information for each of the aforementioned multiple combinations of protons and electrons, A third module generates characteristic information of the estimated target material based on the relative positions of electrons with respect to the protons in each of the aforementioned multiple combinations, Includes, The estimation program according to Appendix 2, characterized in that the value of the first parameter is determined by training the machine learning model using the training data. (Note 4) The position of the protons of each of the multiple atoms contained in the aforementioned estimated substance represents the position of the atomic nucleus of each of the multiple atoms contained in the aforementioned estimated substance. The characteristic information of the estimated substance includes the characteristics of each of the multiple combinations of atomic nuclei and electrons contained in the estimated substance. The aforementioned conversion module is A first module that determines relative position information based on the atomic orbital information of each electron of a plurality of atoms contained in the estimated target substance and a first parameter, A second module that determines the relative position of the electron to the atomic nucleus based on the proton position information and the relative position information for each of the aforementioned multiple combinations of atomic nuclei and electrons, A third module that generates characteristic information of the estimated target material based on the relative position of electrons with respect to the atomic nucleus of each of the plurality of combinations and the number of protons contained in the atomic nucleus of each of the plurality of combinations, Includes, The estimation program according to Appendix 2, characterized in that the value of the first parameter is determined by training the machine learning model using the training data. (Note 5) The quantum circuit generates the state information based on the characteristic information of the estimated target material and a second parameter. The estimation program according to any one of Appendix 2 to 4, characterized in that the value of the second parameter is determined by training the machine learning model using the training data. (Note 6) A receiving unit that receives proton position information indicating the position of protons in each of several atoms contained in the substance to be estimated, and atomic orbital information indicating the atomic orbitals of electrons in each of the several atoms, An estimation unit that estimates the physical quantities of the target substance from the proton position information and the atomic orbital information using a machine learning model trained with training data that includes information indicating the positions of protons of multiple atoms contained in a substance different from the target substance and information indicating the atomic orbitals of electrons of multiple atoms contained in a substance different from the target substance, An output unit that outputs estimation results showing the physical quantities of the substance to be estimated, An estimation device characterized by comprising: (Note 7) The aforementioned machine learning model, A conversion module that converts the proton position information and the atomic orbital information into characteristic information of the estimated target substance, A quantum circuit that generates state information indicating the state of the estimated target substance from characteristic information of the estimated target substance, Includes, The estimation device according to Appendix 6, characterized in that the estimation unit generates characteristic information of the target substance from the proton position information and the atomic orbital information using the conversion module, obtains the state information from the quantum computer by having the quantum computer perform calculations of the quantum circuit, and determines the physical quantities of the target substance based on the state information. (Note 8) The characteristic information of the estimated target substance includes the characteristics of each of the multiple combinations of protons and electrons contained in the estimated target substance. The aforementioned conversion module is A first module that determines relative position information based on the atomic orbital information of each electron of a plurality of atoms contained in the estimated target substance and a first parameter, A second module that determines the relative position of the electron to the proton based on the proton position information and the relative position information for each of the aforementioned multiple combinations of protons and electrons, A third module generates characteristic information of the estimated target material based on the relative positions of electrons with respect to the protons in each of the aforementioned multiple combinations, Includes, The estimation device according to Appendix 7, characterized in that the value of the first parameter is determined by training the machine learning model using the training data. (Note 9) The position of the protons of each of the multiple atoms contained in the aforementioned estimated substance represents the position of the atomic nucleus of each of the multiple atoms contained in the aforementioned estimated substance. The characteristic information of the estimated substance includes the characteristics of each of the multiple combinations of atomic nuclei and electrons contained in the estimated substance. The aforementioned conversion module is A first module that determines relative position information based on the atomic orbital information of each electron of a plurality of atoms contained in the estimated target substance and a first parameter, A second module that determines the relative position of the electron to the atomic nucleus based on the proton position information and the relative position information for each of the aforementioned multiple combinations of atomic nuclei and electrons, A third module that generates characteristic information of the estimated target material based on the relative position of electrons with respect to the atomic nucleus of each of the plurality of combinations and the number of protons contained in the atomic nucleus of each of the plurality of combinations, Includes, The estimation device according to Appendix 7, characterized in that the value of the first parameter is determined by training the machine learning model using the training data. (Note 10) The quantum circuit generates the state information based on the characteristic information of the estimated target material and a second parameter. The estimation device according to any one of Appendix 7 to 9, characterized in that the value of the second parameter is determined by training the machine learning model using the training data. (Note 11) The system receives proton position information indicating the position of protons in each of several atoms contained in the substance to be estimated, and atomic orbital information indicating the atomic orbitals of electrons in each of the aforementioned atoms. Using a machine learning model trained with training data that includes information indicating the positions of protons in multiple atoms contained in a substance different from the substance to be estimated, and information indicating the atomic orbitals of electrons in multiple atoms contained in a substance different from the substance to be estimated, the physical quantities of the substance to be estimated are estimated from the proton position information and the atomic orbital information. Outputting estimation results showing the physical quantities of the substance to be estimated. An estimation method characterized by the processing being performed by a computer. (Note 12) The aforementioned machine learning model, A conversion module that converts the proton position information and the atomic orbital information into characteristic information of the estimated target substance, A quantum circuit that generates state information indicating the state of the estimated target substance from characteristic information of the estimated target substance, Includes, The process for estimating the physical quantity of the substance to be estimated is as follows: Using the aforementioned conversion module, a process is performed to generate characteristic information of the estimated target substance from the proton position information and the atomic orbital information. The process involves having a quantum computer perform calculations on the quantum circuit, thereby obtaining the state information from the quantum computer, A process to determine the physical quantity of the estimated target substance based on the state information, The estimation method according to Appendix 11, characterized by including the following: (Note 13) The characteristic information of the estimated target substance includes the characteristics of each of the multiple combinations of protons and electrons contained in the estimated target substance. The aforementioned conversion module is A first module that determines relative position information based on the atomic orbital information of each electron of a plurality of atoms contained in the estimated target substance and a first parameter, A second module that determines the relative position of the electron to the proton based on the proton position information and the relative position information for each of the aforementioned multiple combinations of protons and electrons, A third module generates characteristic information of the estimated target material based on the relative positions of electrons with respect to the protons in each of the aforementioned multiple combinations, Includes, The estimation method according to Appendix 12, characterized in that the value of the first parameter is determined by training the machine learning model using the training data. (Note 14) The position of the protons of each of the multiple atoms contained in the aforementioned estimated substance represents the position of the atomic nucleus of each of the multiple atoms contained in the aforementioned estimated substance. The characteristic information of the estimated substance includes the characteristics of each of the multiple combinations of atomic nuclei and electrons contained in the estimated substance. The aforementioned conversion module is A first module that determines relative position information based on the atomic orbital information of each electron of a plurality of atoms contained in the estimated target substance and a first parameter, A second module that determines the relative position of the electron to the atomic nucleus based on the proton position information and the relative position information for each of the aforementioned multiple combinations of atomic nuclei and electrons, A third module that generates characteristic information of the estimated target material based on the relative position of electrons with respect to the atomic nucleus of each of the plurality of combinations and the number of protons contained in the atomic nucleus of each of the plurality of combinations, Includes, The estimation method according to Appendix 12, characterized in that the value of the first parameter is determined by training the machine learning model using the training data. (Note 15) The quantum circuit generates the state information based on the characteristic information of the estimated target material and a second parameter. The estimation method according to any one of Appendix 12 to 14, characterized in that the value of the second parameter is determined by training the machine learning model using the training data. [Explanation of Symbols]

[0178] 101 Estimation device 111 Reception Department 112, 1215 Estimation part 113 Output section 301 Quantum Computer 302 Server 303 Terminal device 304 Communication Network 411, 1511 conversion module 412 Quantum circuit 421 Feature Conversion Circuit 422-1~422-L Quantum Transformer 423 State conversion circuit 511, 516 Conversion Layer Add 512 513 Subtraction 514 Norm calculation 515 Concatenation operation 517 tanh function 611 arccos function 612 Embedded Circuit 711-1-1~711-nd MPS circuit 712-1~712-d, 713 HEA circuit 714 measurements 1111 Communication Interface 1112 Control device 1113 Quantum Systems 1211 Communications Department 1212 Acquisition Department 1213 Quantization section 1214 Training Department 1216 Storage section 1221, 1223 Estimated Models 1222 Training Data Set 1611 Multiplication 1701 CPU 1702 memory 1703 Input device 1704 Output device 1705 Auxiliary storage device 1706 Media drive device 1707 Network Connection Device 1708 Bus 1709 Portable recording media

Claims

1. The system receives proton position information indicating the position of protons in each of the multiple atoms contained in the substance to be estimated, and atomic orbital information indicating the atomic orbitals of electrons in each of the multiple atoms. Using a machine learning model trained with training data that includes information indicating the positions of protons in multiple atoms contained in a substance different from the substance to be estimated, and information indicating the atomic orbitals of electrons in multiple atoms contained in a substance different from the substance to be estimated, the physical quantities of the substance to be estimated are estimated from the proton position information and the atomic orbital information. Outputting estimation results showing the physical quantities of the substance to be estimated. An estimation program that causes a computer to perform a process.

2. The aforementioned machine learning model, A conversion module that converts the proton position information and the atomic orbital information into characteristic information of the estimated target substance, A quantum circuit that generates state information indicating the state of the estimated target substance from characteristic information of the estimated target substance, Includes, The process for estimating the physical quantity of the substance to be estimated is as follows: Using the aforementioned conversion module, a process is performed to generate characteristic information of the estimated target substance from the proton position information and the atomic orbital information. The process involves having a quantum computer perform calculations on the quantum circuit, thereby obtaining the state information from the quantum computer, A process to determine the physical quantity of the estimated target substance based on the state information, The estimation program according to claim 1, characterized by including the following:

3. The characteristic information of the estimated target substance includes the characteristics of each of the multiple combinations of protons and electrons contained in the estimated target substance. The aforementioned conversion module is A first module that determines relative position information based on the atomic orbital information of each electron of a plurality of atoms contained in the estimated target substance and a first parameter, A second module that determines the relative position of the electron to the proton based on the proton position information and the relative position information for each of the aforementioned multiple combinations of protons and electrons, A third module that generates characteristic information of the estimated target material based on the relative positions of electrons with respect to the protons in each of the plurality of combinations, Includes, The estimation program according to claim 2, characterized in that the value of the first parameter is determined by training the machine learning model using the training data.

4. The position of the protons of each of the multiple atoms contained in the aforementioned estimated substance represents the position of the atomic nucleus of each of the multiple atoms contained in the aforementioned estimated substance. The characteristic information of the estimated substance includes the characteristics of each of the multiple combinations of atomic nuclei and electrons contained in the estimated substance. The aforementioned conversion module is A first module that determines relative position information based on the atomic orbital information of each electron of a plurality of atoms contained in the estimated target substance and a first parameter, A second module that determines the relative position of the electron to the atomic nucleus based on the proton position information and the relative position information for each of the aforementioned multiple combinations of atomic nuclei and electrons, A third module that generates characteristic information of the estimated target material based on the relative position of electrons with respect to the atomic nucleus of each of the plurality of combinations and the number of protons contained in the atomic nucleus of each of the plurality of combinations, Includes, The estimation program according to claim 2, characterized in that the value of the first parameter is determined by training the machine learning model using the training data.

5. The quantum circuit generates the state information based on the characteristic information of the estimated target material and a second parameter. The estimation program according to any one of claims 2 to 4, characterized in that the value of the second parameter is determined by training the machine learning model using the training data.

6. A receiving unit that receives proton position information indicating the position of protons in each of several atoms contained in the substance to be estimated, and atomic orbital information indicating the atomic orbitals of electrons in each of the several atoms, An estimation unit that estimates the physical quantities of the target substance from the proton position information and the atomic orbital information using a machine learning model trained with training data that includes information indicating the positions of protons of multiple atoms contained in a substance different from the target substance and information indicating the atomic orbitals of electrons of multiple atoms contained in a substance different from the target substance, An output unit that outputs estimation results showing the physical quantities of the substance to be estimated, An estimation device characterized by comprising:

7. The system receives proton position information indicating the position of protons in each of the multiple atoms contained in the substance to be estimated, and atomic orbital information indicating the atomic orbitals of electrons in each of the multiple atoms. Using a machine learning model trained with training data that includes information indicating the positions of protons in multiple atoms contained in a substance different from the substance to be estimated, and information indicating the atomic orbitals of electrons in multiple atoms contained in a substance different from the substance to be estimated, the physical quantities of the substance to be estimated are estimated from the proton position information and the atomic orbital information. Outputting estimation results showing the physical quantities of the substance to be estimated. An estimation method characterized by the processing being performed by a computer.

Images