Methods, apparatuses, and media for controlling training quantum evolution using sub-logic

Abstract

This disclosure relates to methods, apparatus, and media for training quantum evolution using sub-logic control. In one aspect, the method includes the following actions: accessing quantum hardware, wherein the quantum hardware comprises: a quantum system including one or more multi-level quantum subsystems; one or more control devices operating the one or more multi-level quantum subsystems according to one or more corresponding control parameters, the one or more corresponding control parameters being related to parameters of the physical environment in which the multi-level quantum subsystems reside; initializing the quantum system in an initial quantum state, wherein an initial set of control parameters forms a parameterization defining the initial quantum state; obtaining one or more quantum system observables and one or more target quantum states; and iteratively training until a completion event occurs.

Description

[0001] Case Analysis

[0002] This application is a divisional application of Chinese invention patent application 201680087746.8, filed on December 19, 2016. Background Technology

[0003] This manual relates to quantum computing.

[0004] Quantum variational eigenvalue solvers have been proposed as a method for preparing and studying the states of physically interesting systems. Digital implementations of quantum variational eigenvalue solvers utilize quantum logic gates that perform precisely known operations on qubits. Summary of the Invention

[0005] This specification describes a technique for training the quantum evolution of an initial quantum state to achieve a target quantum state with defined target properties. The quantum evolution of the initial quantum state is trained using tunable analog evolution defined by tuning basic hardware components (such as control knobs typically used to calibrate individual quantum gates).

[0006] Typically, an innovative aspect of the subject matter described in this specification can be implemented in a method comprising: accessing quantum hardware, wherein the quantum hardware comprises: a quantum system comprising one or more multilevel quantum subsystems; one or more control devices that operate the one or more multilevel quantum subsystems according to one or more corresponding control parameters relating to parameters of the physical environment in which the multilevel quantum subsystems reside; initializing the quantum system in an initial quantum state, wherein an initial set of control parameters forms a parameterization defining the initial quantum state; obtaining one or more quantum system observables and one or more target quantum states; and iteratively training until a completion event occurs.

[0007] Other embodiments of this aspect include corresponding computer systems, devices, and computer programs recorded on one or more computer storage devices, each computer system, device, and computer program being configured to perform actions of the method. A system of one or more computers may be configured to perform a specific operation or action by software, firmware, hardware, or a combination thereof installed on the system, which causes the system to perform actions during operation. One or more computer programs may be configured to perform a specific operation or action by including instructions that cause a device to perform actions when executed by a data processing device.

[0008] The foregoing and other embodiments may optionally include one or more of the following features, individually or in combination. In some embodiments, iterative training includes iteratively training changes from an initial quantum state to achieve one or more target quantum states.

[0009] In another implementation, iterative training includes iteratively training the evolution of the initial quantum state and subsequent quantum states to achieve one or more target quantum states until a completion event occurs.

[0010] In some cases, evolution is simulated evolution.

[0011] In some implementations, for each iteration, iterative training includes: determining the value of a cost function based on the current quantum state for the iteration and one or more observables of the quantum system; minimizing the value of the cost function to determine updated values ​​of control parameters defining the current quantum state; and determining whether a completion event has occurred.

[0012] In other implementations, minimizing the value of the cost function to determine the updated value of the control parameters includes adjusting the control parameters.

[0013] In some cases, the method further includes: in response to determining that a completion event has occurred, providing one or more target quantum states for experimental probing.

[0014] In some implementations, (i) at least one of the observable quantities of the quantum system includes the Hamiltonian of the quantum system, (ii) one or more target quantum states include one or more eigenstates of the Hamiltonian, and (iii) experimental probing includes measuring the energy of one or more eigenstates to determine the corresponding energy eigenvalue of the eigenstate.

[0015] In other embodiments, (i) the system observable is the molecular electronic structure Hamiltonian, (ii) one or more target quantum states include the ground state of the molecular electronic structure Hamiltonian, and (iii) experimental probing includes measuring the target quantum state to determine the ground state energy.

[0016] In some cases, the value of the cost function based on one or more of the quantum states and system observables is the expected value of one or more of the quantum states and system observables.

[0017] In other cases, determining the expected value of one or more of the quantum states and system observables involves: repeatedly initializing the quantum system in an initial quantum state; measuring one or more system observables for each initialized quantum state to determine a set of measurement results; and determining the expected value of one or more of the quantum states and system observables based on the set of measurement results.

[0018] In some cases, each initialized quantum state is different from every other initialized quantum state.

[0019] In some implementations, determining the expected value of one or more of the quantum state and system observables includes determining the expected value of the density operator and one or more system observables.

[0020] In other embodiments, obtaining one or more target quantum states includes encoding the solution to the optimization problem into the ground state of the quantum system.

[0021] In some cases, the method further includes obtaining a solution to the optimization problem from experimental probes.

[0022] In some implementations, (i) an initial quantum state encodes the training data, (ii) one or more system observables act as prediction functions, and (iii) iterative training from the initial quantum state includes solving a machine learning problem.

[0023] In some cases, quantum hardware includes quantum circuits.

[0024] In some cases, the control device includes one or more quantum gates that operate the quantum system through one or more corresponding control parameters.

[0025] In some embodiments, the method further includes calibrating one or more quantum gates, including: for each quantum gate to be calibrated: defining the correct action of the quantum gate on the quantum system; performing a measurement to determine the action of the quantum gate on the quantum system; and adjusting the corresponding control parameters for the quantum gate in response to determining that the action of the quantum gate on the quantum system is incorrect.

[0026] In some cases, iterative training involves combining iterative training with one or more combinations of calibrating quantum gates.

[0027] In other cases, minimizing the value of the cost function to determine the updated values ​​of the control parameters includes performing a gradient-free greedy minimization method.

[0028] In some implementations, the completion event is the convergence of a cost function based on the determination of one or more of the quantum states and system observables.

[0029] In some cases, the initial quantum state is the state of a resonator coupled to a superconducting quantum bit.

[0030] The subject matter described in this specification may be implemented in a particular manner to achieve one or more of the following advantages.

[0031] Training quantum evolution using sub-logic control parameters the quantum evolution using the natural control parameters of the quantum system by performing a variational minimization procedure by directly adjusting the control parameters of the control devices included in the system (e.g., by adjusting the voltage on the digital-to-analog converter). By utilizing low-level control to parameterize the fitting, training quantum evolution using sub-logic control, in contrast to training quantum evolution using tunable quantum gates, avoids the need for precise knowledge of the effective circuitry by exchanging a global mapping that is agnostic to system errors and robust to many control and calibration problems. Furthermore, using low-level control allows for a fundamentally more accurate representation of the desired state because of the increased control over the evolution of the state. Hardware-level control provides the greatest possible control over the evolution and the state to be prepared, thus creating a more accurate fitting for the model being simulated, even in the presence of errors.

[0032] Training quantum evolution using sub-logic control adaptively trains quantum evolution to achieve target quantum states with desired properties without using parameterized digital quantum circuits, thus reducing the experimental complexity of the system, as such quantum circuits can be very complex to implement. Instead, unlike other quantum evolution training systems, training quantum evolution using sub-logic control abandons the concept of digital quantum gates, supporting tunable analog evolution defined using control parameters typically reserved for calibrating individual gates. By operating at the level of basic hardware components, training quantum evolution using sub-logic control does not require precise knowledge of the effective circuitry, but instead performs a global mapping that is agnostic to system errors and robust to many control and calibration problems.

[0033] Digital quantum circuits can include quantum logic gates that perform precise operations on qubits, or in some examples, multidimensional qudits. In many settings, a quantum logic gate model algorithm specifies the quantum circuit and requires the experimenter to execute the specified quantum circuit with minimal error. This can be a challenging task because implementing a quantum logic gate model algorithm may require many quantum logic gates. Therefore, scalable computation based on quantum logic gate models can require expensive processes, such as quantum error correction. Furthermore, each gate in the circuit must be calibrated before execution. Systems that do not train quantum evolution using sublogic control can be calibrated by carefully tuning hardware control parameters and performing classical minimization loops to perfect each individual gate. Therefore, applying quantum circuits may require considerable effort, which is never error-free. In fact, the system may exhibit a large number of errors, such as control errors (which are often caused by pulses that do not fully form the gates) or errors caused by noise coupled to the system.

[0034] Training quantum evolution using sub-logic control systems can be applied to various settings and used to manipulate various types of controllable states, such as those existing in resonators attached to superconducting qubits. These resonator states are controllable and have appropriately defined variational assumptions.

[0035] The system can be trained to use quantum evolution with sub-logic control instead of or in combination with standard methods for gate calibration to improve system scalability and performance. Quantum gates can be generated by the exact evolution of a quantum system under a specific Hamiltonian. For example, in the case of a superconducting Gmon qubit, the system can be described by a Hamiltonian given by the following equation (1).

[0036]

[0037] In equation (1), X i Y i Z i ; is the Pauli operator, and a i (t), b i (t), c i (t) and g ij (t) is the time-dependent curve generated by a microwave pulse transmitted through wires in the hardware. To implement a quantum circuit, it may be necessary for the experimenter to first calibrate all the quantum gates included in the quantum circuit by transmitting a specified pulse through wires in the hardware. In the case of superconducting electronics, the pulse can be considered as formed by a Fourier series; for example, a pulse inducing a local Y-field on a qubit q can be generated by… Given, where ω0 is the fundamental frequency, and A K B K Determine the pulse shape. At the hardware level, pulses Y1 and Y2 can be designed using a digital-to-analog converter (DAC), which sets the voltage and time for each line.

[0038] It is possible to determine precisely or approximately which pulses are needed to execute the desired quantum gate, then perform a Fourier transform on the determined pulses and program them into an appropriate DAC. However, the precise mapping between the DAC setup and the actual pulses seen by the system through the pulses is imprecise for actual experimental systems in the laboratory. For example, a DAC can be programmed in a specific way to perform a specific task; however, in reality, the physical behavior of the DAC may deviate from the expected behavior, for example, due to unknown persistent systematic errors. For instance, a DAC in the laboratory may be at room temperature, while the quantum chip may be at millikelvin, leading to unpredictable or undesirable transitions as the signal travels down the waveguide to the chip due to temperature variations.

[0039] Unlike other systems, training quantum evolution in systems using sublogic control may not require protocols from quantum control theory, such as chopped random basis schemes, to form clean quantum gates. Similarly, unlike other systems that implement quantum control procedures to tune individual gates, as mentioned above, training quantum evolution in systems using sublogic control can be used to calibrate or train the entire circuit. In some implementations, for example, for systems with advanced architectures, combining training quantum evolution using sublogic control with the standard methods described above for gate calibration may be further advantageous.

[0040] Furthermore, training quantum evolution using sublogic control can be more robust than training it without, and avoids introducing certain control or system errors into the quantum system. For example, experimenters can often use control theory feedback to form gates from sublogic control because the control parameters may not be fully known or predictable regarding the system's actions. However, such control or system errors are not a problem for training quantum evolution using sublogic control because the classical minimization process used to determine control parameter updates is agnostic to what is happening in the hardware. This is not a problem if the parameterization is assumed to be performed in a certain way, but the actual degree of parameterization is slightly different if the parameterization remains unchanged.

[0041] Furthermore, training quantum evolution using sub-logic control allows for leakage outside the qubit state, a significant problem for most quantum computing setups. Defining variational fittings on a qubit manifold while allowing states or parts of states to leave the manifold is permissible, equivalent to performing non-single operations and reorganizing the states.

[0042] Unlike other quantum evolution training systems, training quantum evolution using sub-logic control can be applied to multi-level quantum systems, such as multi-level systems beyond qubits, because any parameterized quantum evolution of any quantum system can be used to define the hypothesis. Therefore, the practicality and applicability of training quantum evolution using sub-logic control are greatly improved, since quantum hardware often suffers from qubit leakage; that is, qubits may only be approximate qubits and sometimes occupy higher energy levels. Furthermore, training quantum evolution using sub-logic control can be applied to training quantum evolution and can achieve quantum states with the desired properties even in the presence of noise; that is, training quantum evolution using sub-logic control does not require closed-loop evolution.

[0043] Training quantum evolution using sub-logic control can be applied to a wide range of settings, specifically those with industrial value. Generally, any physical system that is difficult to study due to quantum mechanics can benefit from implementing systems that utilize training quantum evolution using sub-logic control. For example, training quantum evolution using sub-logic control can be used to prepare the ground state of the Hamiltonian of the molecular electronic structure, which describes the energetics of the molecule—a compelling industrial application for small quantum computers. Solving such problems on a quantum computer would provide an energy surface describing chemical reactions and could be used to predict chemical rates, thus significantly accelerating drug discovery, solar cell design, and the development of industrial catalysts. As another example, training quantum evolution using sub-logic control can be used to study the properties of high-temperature superconductors, for example, by studying the Fermi-Hubbard model. As yet another example, training quantum evolution using sub-logic control can be used to simulate models through condensed matter physics, for example, for investigating and designing material properties.

[0044] Details of one or more embodiments of the subject matter of this description are set forth in the following drawings and description. Other features, aspects, and advantages of the subject matter will become apparent from the description, drawings, and claims. Attached Figure Description

[0045] Figure 1 An example quantum evolution training system is described.

[0046] Figure 2 This is a flowchart of an example process for training quantum evolution using sub-logic control.

[0047] Figure 3 This is a flowchart of an example iteration of quantum evolution used to train quantum states.

[0048] In the various figures, the same reference numerals and names indicate the same elements. Detailed Implementation

[0049] This specification describes an apparatus and method for adaptively training quantum evolution to achieve quantum states with target properties using hardware-level control. The apparatus and method have applications in various settings, including machine learning tasks and quantum simulations, such as preparing and studying physically interesting states.

[0050] For example, it might be necessary to prepare or solve for the quantum state |ψ> of a quantum system, which is the lowest energy eigenvalue of the Hamiltonian function H such that H|ψ>=E0|ψ). One approximate method for preparing |ψ> is based on the vector... The parameterization of the polynomial number of parameters is called the hypothesized (ansatz) conjectured wavefunction. Then, the quantum variational principle states that when hour, They are equal. Therefore, by solving... |ψ> can be approximated as This makes the above inequalities as tight as possible within the parameterization. This specification describes a method for controlling parameterization using low-level (i.e., sub-logic) control. The apparatus and method, wherein the low-level control is parameterized based on the natural control knob of the corresponding quantum system.

[0051] Example operating environment

[0052] Figure 1 An example quantum evolution training system 100 is depicted. Example system 100 is an example of a system implemented as a classical or quantum computer program on one or more classical computers or quantum computing devices at one or more locations, wherein the systems, components, and techniques described below can be implemented.

[0053] The quantum evolution training system 100 may include quantum hardware 102 that communicates data with a classical processor 104. The quantum evolution training system 100 may receive data as input, including data specifying one or more quantum system observables and one or more target quantum states (e.g., quantum states with corresponding target properties), such as observables and target properties 106. The evolution training system 100 may generate data specifying one or more target quantum states (e.g., target quantum state 110) as output.

[0054] One or more observable quantities of a quantum system may include measurable operators, such as Hamiltonian operators, momentum operators, or position operators. The target quantum state may include one or more eigenstates of the Hamiltonian operator, such as the ground state of the Hamiltonian operator. In some implementations, the solution to the optimization problem may be encoded into the ground state of the Hamiltonian operator. Data specifying the target quantum state 110 may be further provided for experimental probing or post-processing; for example, the energy of the target quantum state may be measured to determine the corresponding energy eigenvalues.

[0055] Quantum hardware 102 may include quantum system 112, control device 114, and data specifying configuration 116. Quantum system 112 may include one or more multilevel quantum subsystems, such as qubits or multidimensional qubits. In some embodiments, the multilevel quantum subsystem may be a superconducting qubit, such as a Gmon qubit. The type of multilevel quantum subsystem used by system 100 depends on the application to which system 100 is applied. For example, in some cases, it may be convenient to include one or more resonators attached to one or more superconducting qubits (e.g., Gmon or Xmon qubits). In other cases, ion traps, photonic devices, or superconducting cavities (which can prepare states without requiring qubits) may be used. Other examples of implementing multilevel quantum subsystems include fluxmon qubits, silicon quantum dots, or phosphorus-impurity qubits.

[0056] One or more control devices 114 may be configured to operate the multi-level quantum subsystem 112 via one or more corresponding control parameters 118 (e.g., one or more physical control parameters). For example, in some embodiments, the multi-level quantum subsystem may be a superconducting quantum bit, and the control device 114 may include one or more digital-to-analog converters (DACs) with corresponding voltage physical control parameters. In other embodiments, the quantum system 112 may include quantum circuits, and the control device 114 may include one or more quantum logic gates that operate the quantum system 112 via microwave pulse physical control parameters transmitted through wires included in the quantum hardware 102. Further examples of the control device include an arbitrary waveform generator capable of generating signals controlled by the DAC. The control parameters may include the quantum bit frequency.

[0057] The data specifying scheme 116 includes a set of parameters 118 and is selected based on knowledge of quantum system 112 and control device 114 acting on quantum system 112. For example, scheme 116 may be a variational scheme that uses information about quantum hardware 102 (such as control device 114 and its corresponding control parameters 118) to determine the parameterization of the state of quantum system 112. In some embodiments, quantum hardware 102 is used directly to parameterize scheme 116, which is a set of variational parameters forming the variational scheme 116, and may include sub-logic-physical control parameters of control device 114.

[0058] Quantum hardware 102 can be configured to perform quantum measurements on quantum system 112 and send the measurement results to classical processor 104. Additionally, quantum hardware 102 can be configured to receive data from classical processor 104 specifying updated physical control parameter values ​​120. Quantum hardware 102 can use the received updated physical control parameter values ​​120 to update the actions of control device 114 on quantum system 112, thereby training the evolution of quantum state 112. For example, quantum hardware can receive data specifying new values ​​representing the voltage strength of one or more DACs included in control device 114 and can update the actions of the DACs on quantum system 112 accordingly.

[0059] The classical processor 104 can be configured, for example, to initialize the quantum system 112 in an initial quantum state by sending data to quantum hardware 102 with an initial set of specified parameters, and to iteratively train the simulated evolution of the initial quantum state and subsequent quantum states to achieve a target quantum state 100 with target property 106. The classical processor can be configured to iteratively train the simulated evolution of the initial quantum state until a completion event occurs, for example, until the received measurement result 108 converges. The classical processor 104 can determine that the completion event has occurred and provide the target quantum state 110 with the defined target property 106 for experimental detection, as described above.

[0060] The classical processor 104 can be further configured to: determine the value of a cost function based on the quantum state of the quantum system 112 and one or more system observables 106, and minimize the value of the cost function to determine updated values ​​of the physical control parameters 120. For example, the classical processor can be configured to perform a minimization method, such as a gradient-free minimization method including Powell's method or Nelder-Mead.

[0061] In some implementations, the value of the cost function based on the quantum states of quantum system 112 and one or more system observables 106 is the expected value of the quantum states and system observables. For example, classical processor 104 may be configured to repeatedly initialize the quantum system in an initial quantum state, and for each initialized quantum state, initiate measurements of one or more system observables to determine a set of measurement results, e.g., measurement result set 108. Based on the measurement result set, classical processor 104 may be configured to determine the corresponding expected value of the quantum states and one or more system observables. In other examples, classical processor 104 may be configured to determine the expected value of the density operator and one or more system observables.

[0062] The classical processor 104 can be further configured to calibrate one or more quantum gates that may be included in the quantum hardware 102. For example, the classical processor can be configured to: define the correct action of the quantum gates on the quantum system 112, perform or initiate a measurement of the quantum system 112 to determine the action of the quantum gates on the quantum system 112, and, in response to determining that the action of the quantum gates on the quantum system is incorrect, adjust the corresponding physical control parameters for the quantum gates, and provide the quantum hardware 102 with the adjusted updated physical control parameters 120. In some embodiments, the classical processor can be configured to iteratively train the simulated evolution of the quantum state to achieve a target quantum state with target properties by combining iterative training of the simulated evolution of the initial quantum state and subsequent quantum states with gate calibration techniques.

[0063] Programming the hardware

[0064] Figure 2 This is a flowchart of an example process 200 for training quantum evolution using sub-logic control. For convenience, process 200 will be described as being executed by a system of one or more classical or quantum computing devices located in one or more locations. For example, a quantum evolution training system appropriately programmed according to this specification (e.g., Figure 1 The quantum evolution training system 100 can execute process 200.

[0065] The system accesses quantum hardware, for example, Figure 1 The quantum hardware 102 (step 202). The quantum hardware may include: a quantum system comprising one or more multi-level quantum subsystems, for example, Figure 1 A quantum system 112; and one or more control devices for operating one or more multi-level quantum subsystems according to one or more corresponding control parameters, for example, Figure 1 The control device 114 and the corresponding control parameters 118. The corresponding control parameters may include physical control parameters. (Refer to the above.) Figure 1 An example multi-level quantum system and a control device for operating on it are described.

[0066] Quantum hardware may include quantum circuits, which in turn may include one or more quantum logic gates that operate on a quantum system via one or more corresponding control parameters. In some embodiments, the system can calibrate one or more of the quantum logic gates included in the quantum hardware using a control device. For example, for each quantum gate to be calibrated, the system may define the correct action of the quantum gate on the quantum system and perform measurements to determine the action of the quantum gate on the quantum system. In response to determining that the quantum gate is not acting correctly on the quantum system, the system may adjust the corresponding control parameters for the quantum gate. In some embodiments, the system may combine process 200 with quantum gate calibration techniques, for example, when the quantum hardware includes an advanced architecture.

[0067] The system initializes the quantum system in an initial quantum state (step 204). An initial set of control parameters can form a parameterization defining the initial quantum state. The initialized quantum state can be a parameterized quantum state, for example, a simulated or "guessed" wavefunction, where the control parameters form the parameterization of the quantum state. In some embodiments, the initial quantum state is the state of a resonator coupled to a superconducting quantum bit. This resonator state is controllable and can form a variational simulation.

[0068] The system obtains one or more quantum system observables and one or more target quantum states (step 206).

[0069] In some implementations, the process 200 for training quantum evolution using sub-logic control can be applied to quantum simulation tasks, such as preparing and studying physically interesting states. In this case, the system can receive one or more quantum system observables that include the Hamiltonian of the quantum system. The system can further receive one or more target quantum states that include one or more eigenstates of the defined Hamiltonian. For example, the received quantum system observable may be a molecular electronic structure Hamiltonian, and the corresponding target quantum state may include the ground state of the molecular electronic structure Hamiltonian. In another example, the received quantum system observable may be a Fermi-Habbard model Hamiltonian, and the corresponding target quantum state may include one or more eigenstates of the Hamiltonian. Typically, process 200 can be applied to quantum simulation tasks involving the study of any physical system that is difficult to study due to quantum mechanics.

[0070] In other embodiments, the process 200 for training quantum evolution using sub-logic control can be applied to machine learning tasks, such as solving optimization tasks. In this case, the system can receive one or more quantum system observables that may include the Hamiltonian of the quantum system. The system can further receive a target quantum state that encodes the solution to the optimization task into the ground state of the quantum system. (See below for reference.) Figure 3 The application of training quantum evolution to machine learning tasks by using sub-logic control is described in more detail.

[0071] The system trains iteratively until the completion event occurs (step 208). The system can iteratively train changes from the initial quantum state to achieve one or more target states. In some implementations, the system can iteratively train the evolution of the initial quantum state and subsequent quantum states to achieve one or more target quantum states. The evolution can be simulated evolution. Iterative training can be performed until the completion event occurs, for example, until the simulated evolution of the initial quantum state and subsequent quantum states converges. See below. Figure 3A more detailed description of the simulated evolution of the quantum state during iterative training.

[0072] The system provides a target quantum state for experimental probing (step 210). In response to determining that a completion event has occurred, the system can provide a target quantum state for experimental probing. For example, as described above with reference to step 206, in some embodiments, process 200 can be applied to quantum simulation tasks, such as preparing and studying physically interesting states. In an example where at least one of the system observables includes a Hamiltonian of the quantum system and one or more target quantum states include one or more eigenstates of the Hamiltonian, experimental probing may include measuring the energy of one or more eigenstates to determine the corresponding energy eigenvalue of the eigenstate. For example, in an embodiment where the system observable is a molecular electronic structure Hamiltonian and the target quantum state includes the ground state of the molecular electronic structure Hamiltonian, experimental probing may include measuring the target quantum state to determine the ground state energy.

[0073] Similarly, as described above with reference to step 206, in other embodiments, process 200 can be applied to machine learning tasks, such as solving an optimization task. In examples where the target characteristics of the target quantum state include encoding the solution to the optimization task into the ground state of the quantum system, experimental probing can include solving a machine learning problem, such as obtaining a solution to an optimization problem.

[0074] Figure 3 This is a flowchart of an example process 300 for simulating the evolution of a quantum state for training. For example, process 300 might describe iteratively training an initial quantum state and the evolution of subsequent quantum states to achieve a target quantum state, as shown above. Figure 2 The process described in step 208. For convenience, process 300 will be described as being performed by one or more computing devices located in one or more locations. For example, a quantum evolution training system appropriately programmed according to this specification (e.g., Figure 1 The quantum evolution training system 100 can execute process 300.

[0075] The quantum evolution training system determines the value of the cost function (step 302). The cost function can be based on the quantum states for the iteration and one or more system observables.

[0076] In some implementations, the value of the cost function based on the quantum state and one or more system observables can be the expected value of one or more system observables among the quantum state and system observables. A quantum evolution training system can determine the expected value of one or more system observables among the quantum state and system observables by repeatedly initializing the quantum system in an initial quantum state. In some implementations, each initialized quantum state is different from every other initialized quantum state. For each initialized quantum state, the system can measure one or more system observables to determine a set of measurement results. Based on the set of measurement results, the quantum evolution training system can determine the expected value of one or more system observables among the quantum state and system observables. In some implementations, the quantum evolution training system can determine the expected value of one or more system observables among the quantum state and system observables by determining the expected value of the density operator and one or more system observables.

[0077] In some implementations, the value of the cost function based on the quantum state and one or more system observables may not be the expected value of the quantum state and the system Hamiltonian, but rather the minimum value of an objective function defined by another observable of the quantum system. For example, as referenced above... Figure 2 The processes 200 and 300 described herein can be applied to machine learning tasks, such as solving optimization tasks, which can always be represented as tasks that require minimizing the corresponding objective function.

[0078] An example of this machine learning task is training a binary classifier using noisy data. This can be viewed as a machine learning problem where the training data is provided as x i The vector represents N features. There can be M training examples, each of which can be associated with a vector denoted as y. i The binary labels (e.g., 0 or 1) are associated with each other. The training problem can be formulated to determine the best classifier for predicting data by classifying example i. Therefore, the goal might be to find a function F such that for all (x) in the training set... i y i ), F(x) i )=y i The problem can become formally difficult when the presence of label noise in the data might prevent it from satisfying all examples. To train a classifier using noisy data, a non-convex loss function can be used as a 0-to-1 loss, which is robust to label noise. For example, if based on the variable... Parameterizing F allows the training problem to be formulated by minimizing the amount of Method selection A classifier can be viewed as a hyperplane in the feature space that divides data points into negative and positive categories. The distance that example i falls from the classification hyperplane can be called the margin. In this example, the sign function is the loss function, and y i F(x i ) is the margin. Negative margins can represent the classification relative to the training label, while positive margins can represent the classification consistent with the training label.

[0079] This optimization can be performed using quantum evolution trained with sub-logic control (e.g., by using processes 200 and 300). For example, in step 204 of process 200, for instance, it can be performed on the control of |φ i > represents one of the x i The corresponding quantum hardware is initialized in the encoding state, and F can be defined as the output state. The observable quantities, of which, It is possible to use each x i The training objective for quantum circuit computation in classification. Therefore, it can be achieved through... The value of the cost function to be minimized is given. The margin, loss, and empirical risk can then be calculated using standard techniques. Once the cost function is defined, the process of training quantum evolution using sublogic control can be used to find the quantum state that minimizes the defined cost function.

[0080] An additional example of machine learning tasks to which processes 200 and 300 can be applied is training neural networks. For instance, the training of a data network can be viewed as a process of training quantum evolution by using sub-logic control, the cost function being determined by the state. The Kullback-Leibler divergence representation between the associated probability distribution and the model implied by the training data.

[0081] The quantum evolution training system minimizes the value of the cost function to determine updated values ​​of the control parameters (step 304). In some implementations, the quantum evolution training system minimizes the value of the cost function to determine updated values ​​of the control parameters by adjusting the control parameters (e.g., adjusting physical control parameters). For example, parameters that determine the pulse shape that can induce a local field on a particular qubit can be adjusted by changing the corresponding pixel on the voltage DAC. As another example, when the quantum hardware includes an ion trap, the qubit can be controlled by using a laser, and parameters that determine the shape or intensity of the laser pulse can be adjusted. Other control parameters that can be adjusted include the qubit frequency or chip temperature. The quantum hardware training system can minimize the value of the cost function to determine updated values ​​of the control parameters by performing gradient-free greedy minimization methods (e.g., Powell's method or the simplex method). For example, if the cost function is the expected value of the quantum state and one or more system observables, the quantum evolution training system can perform greedy minimization of the energy picture to suggest new settings for system parameters (e.g., hardware control parameters).

[0082] The quantum evolution training system iteratively executes steps 302 and 304 until a completion event is determined to have occurred (step 306). In some implementations, the completion event is the convergence of a determined value of a cost function based on one or more of the state and system observables as described above with reference to step 302.

[0083] The digital and / or quantum themes and implementations of digital functional operations and quantum operations described in this specification may be implemented in digital electronic circuit systems, suitable quantum circuit systems or more generally quantum computing systems, tangible digital and / or quantum computer software or firmware, digital and / or quantum computer hardware (including the structures disclosed in this specification and their structural equivalents), or combinations thereof. The term "quantum computing system" may include, but is not limited to, quantum computers, quantum information processing systems, quantum encryption systems, or quantum simulators.

[0084] The embodiments of the digital and / or quantum themes described in this specification can be implemented as one or more digital and / or quantum computer programs, i.e., one or more modules of digital and / or quantum computer program instructions encoded on a tangible, non-transitory storage medium for execution by a data processing device or for controlling the operation of such data processing device. The digital and / or quantum computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, one or more qubits, or a combination thereof. Alternatively or additionally, the program instructions can be encoded on an artificially generated propagated signal capable of encoding digital and / or quantum information, such as a machine-generated electrical, optical, or electromagnetic signal generated to encode digital and / or quantum information for transmission to a suitable receiver device for execution by a data processing device.

[0085] The terms quantum information and quantum data refer to information or data carried, held, or stored by a quantum system, wherein the smallest nontrivial system is a qubit, i.e., a system that defines the unit of quantum information. It should be understood that the term "qubit" encompasses all quantum systems that can be appropriately approximated as a two-level system in the corresponding context. Such quantum systems can include multi-level systems, for example, having two or more levels. For example, such systems can include atomic, electron, photon, ionic, or superconducting qubits. In many implementations, the fundamental state of computation is identified using a ground state and a first excited state; however, it should be understood that other arrangements using higher-level excited states to identify computational states are possible. The term "data processing device" refers to digital and / or quantum data processing hardware and encompasses various devices, apparatuses, and machines for processing digital and / or quantum data, including, for example, programmable digital processors, programmable quantum processors, digital computers, quantum computers, multiple digital and quantum processors or computers, and combinations thereof. The device may also or further include dedicated logic circuit systems, such as FPGAs (Field-Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), or quantum simulators, i.e., quantum data processing devices designed to simulate or generate information about a particular quantum system. Specifically, a quantum simulator is a dedicated quantum computer that does not have the capability to perform general-purpose quantum computing. In addition to the hardware, the device may optionally include code that creates an execution environment for digital and / or quantum computer programs, such as code constituting processor firmware, protocol stack, database management system, operating system, or a combination of one or more of these.

[0086] Digital computer programs (which may also be referred to or described as programs, software, software applications, modules, software modules, scripts, or code) can be written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and can be deployed in any form (including as stand-alone programs or as modules, components, subroutines, or other units suitable for digital computing environments). Quantum computer programs (which may also be referred to or described as programs, software, software applications, modules, software modules, scripts, or code) can be written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and can be translated into a suitable quantum programming language, or can be written in a quantum programming language, such as QCL or Quipper.

[0087] Digital and / or quantum computer programs may, but are not necessarily, correspond to files in a file system. Programs can be stored as a part of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), or as a single file dedicated to the program under development, or as multiple collaborative files (e.g., a file storing one or more modules, subroutines, or portions of code). Digital and / or quantum computer programs can be deployed to execute on a single digital or quantum computer, or on multiple digital and / or quantum computers located at one site or distributed across multiple sites and interconnected via digital and / or quantum data communication networks. A quantum data communication network is understood as a network that can transmit quantum data using quantum systems (e.g., qubits). Typically, digital data communication networks cannot transmit quantum data; however, quantum data communication networks can transmit both quantum data and digital data.

[0088] The processes and logic flows described in this specification can be executed by one or more programmable digital and / or quantum computers, appropriately operating in conjunction with one or more digital and / or quantum processors, to execute one or more digital and / or quantum computer programs to perform functions by manipulating input digital and quantum data and generating outputs. The processes and logic flows can also be executed by a dedicated logic circuit system, and the device can also be implemented as a dedicated logic circuit system, such as an FPGA or ASIC or a quantum simulator, or by a combination of a dedicated logic circuit system or a quantum simulator and one or more programmable digital and / or quantum computers.

[0089] A system of one or more digital and / or quantum computers "configured" to perform a specific operation or action means that the system has software, firmware, hardware, or a combination thereof installed on it, which in operation causes the system to perform the operation or action. One or more digital and / or quantum computer programs configured to perform a specific operation or action means that one or more programs include instructions that, when executed by a digital and / or quantum data processing device, cause the device to perform an operation or action. A quantum computer can receive instructions from a digital computer that, when executed by a quantum computing device, cause the device to perform an operation or action.

[0090] Digital and / or quantum computers suitable for executing digital and / or quantum computer programs can be based on general-purpose or special-purpose digital and / or quantum processors or both, or any other type of central digital and / or quantum processing unit. Typically, the central digital and / or quantum processing unit receives instructions and digital and / or quantum data from read-only memory, random access memory, or a quantum system suitable for transmitting quantum data (e.g., photons), or a combination thereof.

[0091] The fundamental components of a digital and / or quantum computer are a central processing unit (CPU) for executing or implementing instructions and one or more memory devices for storing instructions and digital and / or quantum data. The CPU and memory may be supplemented by or incorporated into a dedicated logic circuit system or quantum simulator. Typically, a digital and / or quantum computer will also include, or be operatively coupled to, one or more mass storage devices for storing digital and / or quantum data, to receive digital and / or quantum data from or to, or both, such as magnetic disks, magneto-optical disks, optical disks, or quantum systems suitable for storing quantum information. However, a digital and / or quantum computer does not necessarily require such devices.

[0092] Digital and / or quantum computer-readable media suitable for storing digital and / or quantum computer program instructions and digital and / or quantum data include all forms of non-volatile digital and / or quantum memories, media, and memory devices, such as: semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices; disks, such as internal hard disks or removable disks; magneto-optical disks; CD-ROMs and DVD-ROMs; and quantum systems, such as trapped atoms or electrons. To understand this, quantum memory is a device capable of storing quantum data with high fidelity and efficiency for extended periods, such as optical material interfaces for light transmission and materials for storing and preserving quantum characteristics of quantum data, such as superposition or quantum coherence.

[0093] Control of the various systems or portions thereof described in this specification may be implemented in a digital and / or quantum computer program product including instructions stored on one or more non-transitory machine-readable storage media and executable on one or more digital and / or quantum processing devices. The systems or portions thereof described in this specification may be implemented as apparatus, methods, or systems that may include one or more digital and / or quantum processing devices and memories to store executable instructions to perform the operations described in this specification.

[0094] While this specification contains numerous specific implementation details, these details should not be construed as limiting the scope of what may be claimed, but rather as descriptions of features that may be implemented for a particular embodiment. Certain features described in this specification within the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments. Furthermore, although features may be described above as functioning in certain combinations and initially or even equally claimed, in some cases one or more features from the claimed combination may be removed from the combination, and the claimed combination may point to a sub-combination or a variation of the sub-combination.

[0095] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring the operations to be performed in the specific order shown or in a sequential order, or to perform all the operations shown to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Moreover, the separation of the various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0096] Specific embodiments of this subject matter have been described. Other embodiments are within the scope of the following claims. For example, the actions described in the claims can be performed in a different order and still achieve the desired result. As an example, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous.

Claims

1. A method executed by one or more classical processors communicating with quantum hardware, said quantum hardware comprising: i) a quantum system comprising one or more qubits, and ii) one or more control devices operating on the one or more qubits, wherein one or more control parameters of the one or more control devices: i) include physical control parameters, and ii) form a variational fit on the qubit manifold, the method comprising: Adjust one or more control parameters of the one or more control devices that form the variational hypothesis to define the initial hypothesis wave function of the quantum system; Receive data specifying one or more observables of a quantum system and one or more target quantum states; and The one or more control parameters of the one or more control devices are adjusted to iteratively train the evolution of the initial simulated wavefunction and the subsequent quantum state of the quantum system within the variational simulation until a completion event occurs to achieve the one or more target quantum states, wherein, during one or more periods of the evolution, the quantum state of the quantum system leaves the qubit manifold.

2. The method according to claim 1, wherein, Each qubit includes a qubit energy level and higher energy levels.

3. The method according to claim 1, wherein, The evolution includes open evolution.

4. The method according to claim 1, wherein, The iterative training for each iteration includes: Determine the value of the cost function, which is based on the initial hypothesized wave function for the iteration or the subsequent quantum state of the quantum system and one or more quantum system observables. Minimize the value of the cost function to determine the updated value of the physical control parameters; and Determine whether the completion event has occurred.

5. The method according to claim 4, wherein, Minimizing the value of the cost function to determine the updated value of the physical control parameters includes adjusting the physical control parameters.

6. The method according to claim 4, wherein, The value of the cost function is the expected value of one or more of the initial hypothesized wave function or subsequent quantum state and the observables of the quantum system.

7. The method according to claim 6, wherein, Determining the expected value of the initial hypothesized wavefunction or subsequent quantum state and one or more of the system observables includes: Repeat the preparation of the initial proposed wavefunction; For each prepared hypothetical wavefunction, measure one or more observables of the quantum system to determine the set of measurement results; Based on the set of measurement results, the subsequent quantum state of the quantum system and the expected value of one or more system observables are determined.

8. The method according to claim 7, wherein, Determining the expected value of the initial hypothesized wavefunction or subsequent quantum state and one or more quantum system observables includes determining the density operator and the expected value of the one or more system observables.

9. The method according to claim 4, wherein, The completion event is the convergence of the determined value of the cost function.

10. The method of claim 1, further comprising providing one or more realized target quantum states for experimental detection.

11. The method according to claim 10, wherein, (i) at least one of the system observables includes the Hamiltonian of the quantum system, (ii) the one or more target quantum states include one or more eigenstates of the Hamiltonian, and (iii) experimental probing includes measuring the energy of one or more of the eigenstates to determine the corresponding energy eigenvalue of the eigenstate.

12. The method according to claim 11, wherein, The solution to the optimization problem is encoded into the ground state of the Hamiltonian, and the experimental probing includes obtaining the solution to the optimization problem.

13. The method according to claim 10, wherein, (i) the observable quantity of the system is the molecular electronic structure Hamiltonian, (ii) the one or more target quantum states include the ground state of the molecular electronic structure Hamiltonian, and (iii) experimental probing includes measuring the target quantum state to determine the ground state energy.

14. The method according to claim 1, wherein, (i) the proposed wave function encodes the training data, (ii) the one or more quantum system observables are used as prediction functions, and (iii) iteratively training the initial proposed wave function and the simulated evolution of the subsequent quantum states includes solving a machine learning problem.

15. The method according to claim 1, wherein, The quantum hardware includes quantum circuits, and the control device includes one or more quantum gates that operate the quantum system through one or more corresponding physical control parameters.

16. The method of claim 15, further comprising calibrating one or more of the quantum gates, wherein for each quantum gate to be calibrated, the method includes: Define the correct action of the quantum gate on the quantum system; Perform measurements to determine the action of the quantum gate on the quantum system; as well as In response to determining that the quantum gate's action on the quantum system is incorrect, the corresponding physical control parameters for the quantum gate are adjusted.

17. The method according to claim 16, wherein, Iteratively training the simulated evolution of the initial hypothesized wavefunction and the subsequent quantum states of the quantum system to achieve the target quantum state includes combining iterative training of the evolution of the initial hypothesized wavefunction and the subsequent quantum states of the quantum system with calibration of one or more of the quantum gates.

18. An apparatus comprising: Quantum hardware, including: A quantum system comprising one or more qubits, and One or more control devices for operating on the one or more qubits, wherein one or more control parameters of the one or more control devices: i) include physical control parameters, and ii) form a variational scheme on the qubit manifold; and One or more classic processors, wherein the one or more classic processors are configured to: Adjust one or more control parameters of the one or more control devices that form the variational hypothesis to define the initial hypothesis wave function of the quantum system; Receive data specifying one or more observables of a quantum system and one or more target quantum states; and The one or more control parameters of the one or more control devices are adjusted to iteratively train the evolution of the initial simulated wavefunction and the subsequent quantum state of the quantum system within the variational simulation until a completion event occurs to achieve the one or more target quantum states, wherein, during one or more periods of the evolution, the quantum state of the quantum system leaves the qubit manifold.

19. The apparatus according to claim 18, wherein, The iterative training for each iteration includes: Determine the value of the cost function, which is based on the initial hypothetical wave function for the iteration or the subsequent quantum state of the quantum system and one or more system observables. Minimize the value of the cost function to determine the updated value of the physical control parameters; and Determine whether the completion event has occurred.

20. The apparatus according to claim 19, wherein, Minimizing the value of the cost function to determine the updated value of the physical control parameters includes adjusting the physical control parameters.

21. A method executed by one or more classical processors communicating with quantum hardware, said quantum hardware comprising: i) comprising one or more multi-level quantum subsystems, and ii) one or more control devices operating the one or more multi-level quantum subsystems, wherein one or more control parameters of the one or more control devices: i) include physical control parameters, and ii) form a variational hypothesis, the method comprising: Adjust one or more control parameters of the one or more control devices that form the variational hypothesis to define the initial hypothesis wave function of the quantum system; Receive data specifying one or more observables of a quantum system and one or more target quantum states; and The one or more control parameters of the one or more control devices are adjusted to iteratively train the evolution of the initial simulated wavefunction and the subsequent quantum state of the quantum system within the variational simulation until a completion event occurs to achieve the one or more target quantum states, wherein the evolution is a simulated evolution.

22. The method according to claim 21, wherein, The iterative training for each iteration includes: Determine the value of the cost function, which is based on the initial hypothesized wave function for the iteration or the subsequent quantum state of the quantum system and one or more quantum system observables. Minimize the value of the cost function to determine the updated value of the physical control parameters; and Determine whether the completion event has occurred.

23. The method according to claim 22, wherein, Minimizing the value of the cost function to determine the updated value of the physical control parameters includes adjusting the physical control parameters.

24. The method according to claim 22, wherein, The value of the cost function is the expected value of one or more of the initial hypothesized wave function or subsequent quantum state and the observables of the system.

25. The method according to claim 24, wherein, Determining the expected value of the initial hypothesized wavefunction or subsequent quantum state and one or more of the system observables includes: Repeat the preparation of the initial proposed wavefunction; For each prepared hypothetical wavefunction, measure one or more observables of the quantum system to determine the set of measurement results; Based on the set of measurement results, the subsequent quantum state of the quantum system and the expected value of one or more system observables are determined.

26. The method of claim 25, wherein, Determining the expected value of the initial hypothesized wavefunction or subsequent quantum state and one or more quantum system observables includes determining the density operator and the expected value of the one or more system observables.

27. The method according to claim 22, wherein, The completion event is the convergence of the determined value of the cost function.

28. The method of claim 21, further comprising providing one or more realized target quantum states for experimental detection.

29. The method according to claim 28, wherein, (i) at least one of the system observables includes the Hamiltonian of the quantum system, (ii) the one or more target quantum states include one or more eigenstates of the Hamiltonian, and (iii) experimental probing includes measuring the energy of one or more of the eigenstates to determine the corresponding energy eigenvalue of the eigenstate.

30. The method according to claim 29, wherein, The solution to the optimization problem is encoded into the ground state of the Hamiltonian, and the experimental probing includes obtaining the solution to the optimization problem.

31. The method according to claim 28, wherein, (i) the observable quantity of the system is the molecular electronic structure Hamiltonian, (ii) the one or more target quantum states include the ground state of the molecular electronic structure Hamiltonian, and (iii) experimental probing includes measuring the target quantum state to determine the ground state energy.

32. The method according to claim 21, wherein, (i) the proposed wave function encodes the training data, (ii) the one or more quantum system observables are used as prediction functions, and (iii) iteratively training the initial proposed wave function and the simulated evolution of the subsequent quantum states includes solving a machine learning problem.

33. The method according to claim 21, wherein, The quantum hardware includes quantum circuits, and the control device includes one or more quantum gates that operate the quantum system through one or more corresponding physical control parameters.

34. The method of claim 33, further comprising calibrating one or more of the quantum gates, wherein for each quantum gate to be calibrated, the method includes: Define the correct action of the quantum gate on the quantum system; Perform measurements to determine the action of the quantum gate on the quantum system; as well as In response to determining that the quantum gate's action on the quantum system is incorrect, the corresponding physical control parameters for the quantum gate are adjusted.

35. The method according to claim 34, wherein, Iteratively training the simulated evolution of the initial hypothesized wavefunction and the subsequent quantum states of the quantum system to achieve the target quantum state includes combining iterative training of the simulated evolution of the initial hypothesized wavefunction and the subsequent quantum states of the quantum system with calibration of one or more of the quantum gates.

36. An apparatus comprising: Quantum hardware, including: Quantum systems including one or more multi-level quantum subsystems, and One or more control devices for operating the one or more multi-level quantum subsystems, wherein one or more control parameters of the one or more control devices are: i) includes physical control parameters, and ii) forms a variational hypothesis; and One or more classic processors, wherein the one or more classic processors are configured to: Adjust one or more control parameters of the one or more control devices that form the variational hypothesis to define the initial hypothesis wave function of the quantum system; Receive data specifying one or more observables of a quantum system and one or more target quantum states; and The one or more control parameters of the one or more control devices are adjusted to iteratively train the evolution of the initial simulated wavefunction and the subsequent quantum state of the quantum system within the variational simulation until a completion event occurs to achieve the one or more target quantum states, wherein the evolution is a simulated evolution.

37. The apparatus according to claim 36, wherein, The iterative training for each iteration includes: Determine the value of the cost function, which is based on the initial hypothetical wave function for the iteration or the subsequent quantum state of the quantum system and one or more system observables. Minimize the value of the cost function to determine the updated value of the physical control parameters; and Determine whether the completion event has occurred.

38. The apparatus according to claim 37, wherein, Minimizing the value of the cost function to determine the updated value of the physical control parameters includes adjusting the physical control parameters.

39. The apparatus according to claim 37, wherein, The value of the cost function is the expected value of one or more of the initial hypothesized wave function or subsequent quantum state and the observables of the system.

40. The apparatus according to claim 39, wherein, Determining the expected value of the initial hypothesized wavefunction or subsequent quantum state and one or more of the system observables includes: Repeat the preparation of the initial proposed wavefunction; For each prepared hypothetical wavefunction, measure one or more observables of the quantum system to determine the set of measurement results; Based on the set of measurement results, the subsequent quantum state of the quantum system and the expected value of one or more system observables are determined.

Images