WO2023162004A1 - Synthesis condition generation method, synthesis condition generation device, and program - Google Patents

Abstract

A synthesis condition generation method according to one embodiment of the present invention generates a synthesis condition in a trial run in which generation of a substance that conforms to the synthesis condition, measurement of an evaluation value for evaluating the substance, and generation of the synthesis condition are executed in the stated order. In the synthesis condition generation method, a computer executes: an evaluation value conversion procedure in which, when there exists a trial run n' in which the evaluation value yn' (n'∈{1, …, N}) of a substance could not be measured among the past N rounds of trial runs, an evaluation value yn' in the trial run n' is replaced with the worst evaluation value; a prediction procedure in which the evaluation value y of an untried synthesis condition x is predicted using a synthesis condition xn and an evaluation value yn in a trial run (n=1, …, N); and a generation procedure in which a synthesis condition that yields a better evaluation value than the best value of an evaluation value yn (n=1, …, N) is generated, as a synthesis condition xN+1 in a trial run N+1, using the result of having predicted the evaluation value y of the synthesis condition x of the untried trial run.

Description

Synthesis condition generation method, synthesis condition generation device and program

The present invention relates to a synthesis condition generation method, a synthesis condition generation device, and a program.

Conventionally, it has been performed to search for synthesis conditions that give good evaluation values while repeating trials to measure evaluation values of substances produced according to given synthesis conditions (recipes). For example, as shown in FIG. 1, where n is the number of trials, (1) synthesis conditions _xn are generated, (2) samples (substances) are generated according to the synthesis conditions _xn , and (3) the samples are evaluated. and measure the evaluation value y _n , while repeating trials to search for a synthesis condition that gives a good evaluation value.

Methods for producing thin films (substances, samples) according to given synthesis conditions include, for example, the molecular beam epitaxy method (Non-Patent Document 1). In addition to this, for example, sputtering method, pulse laser ablation method, physical vapor deposition method, chemical vapor deposition method, spin coating method, atomic layer deposition method, floating zone method, flux method, Czochralski method, etc. Bulk synthesis method and the like can be mentioned. Also, Bayesian optimization and the like are known as techniques for generating synthesis conditions to be tried next under problem settings in which evaluation values for synthesis conditions are always obtained (Patent Document 1, Non-Patent Documents 2 and 3). In addition, since the process of generating a substance according to the synthesis conditions and measuring the evaluation value of the substance with an evaluation index designed in advance according to the use of the substance is complicated, the relationship between the synthesis conditions and the evaluation value is explicit. There may be situations that cannot be described, but Bayesian optimization is known as a method that can search for synthesis conditions even in such situations.

JP 2018-147075 A

However, the conventional method of searching for a synthesis condition that gives a good evaluation value has the problem that there may be synthesis conditions for which no evaluation value is obtained. This is because a stable substance is not always produced under all synthesis conditions within the search range of synthesis conditions. In addition, under certain synthesis conditions, it is possible to obtain a substance having a composition or crystal structure different from that of the target substance.

One embodiment of the present invention has been made in view of the above points, and aims to generate synthesis conditions under which an optimum substance can be obtained under a given evaluation index.

In order to achieve the above object, a method for generating synthesis conditions according to one embodiment provides the above-described A synthesis condition generation method for generating a synthesis condition, wherein an evaluation value y _n′ (n′ε{1, . If there is a trial n ', an evaluation value conversion procedure for replacing the evaluation value y _{n '} in the trial n ' with the worst evaluation value, and a synthesis condition x _n and A prediction procedure for predicting an evaluation value y of an untried synthetic condition x using an evaluation value y _n and an evaluation value y _n (n =1, . . . , N) to generate a synthesis condition that takes a better evaluation value than the best value of x _N+ 1 in trial N+1.

It is possible to generate synthesis conditions that yield the optimal substance under the given evaluation index.

FIG. 10 is a diagram showing an example of trials for obtaining synthesis conditions with good evaluation values; It is a figure showing an example of the whole composition condition generation system concerning this embodiment. It is a figure which shows an example of the hardware constitutions of the synthesis condition production|generation apparatus which concerns on this embodiment. It is a figure showing an example of functional composition of a synthetic condition generating device concerning this embodiment. 6 is a flowchart showing an example of synthesis condition generation processing according to the embodiment; It is a figure which shows the evaluation value in a simulation experiment. It is a figure which shows the result of a simulation experiment. FIG. 3 shows the results of SrRuO ₃ experiments.

An embodiment of the present invention will be described below. In the following embodiments, a synthesis condition generation system 1 capable of generating synthesis conditions for obtaining an optimum substance under given evaluation indices will be described.

<Overall Configuration of Synthesis Condition Generation System 1>
FIG. 2 shows an example of the overall configuration of the synthesis condition generation system 1 according to this embodiment. As shown in FIG. 2 , the synthesis condition generation system 1 according to this embodiment includes a synthesis condition generation device 10 , a substance generation device 20 and a substance evaluation device 30 .

The synthesis condition generation device 10 sets the index representing the number of trials to n, and generates the synthesis conditions to be used in the current trial, using the synthesis conditions (recipes) obtained up to the previous trials and their evaluation values. Synthesis conditions are conditions for obtaining a certain substance by proceeding with the reaction of one or more raw materials by a certain method (for example, the amount of constituent elements, reaction temperature, timing of introducing raw materials, etc.), and are generally expressed in a vector expressed. As an example, in the task of producing a _SrRuO3 thin film, the synthesis conditions are (a) the supply amount of Ru, (b) the temperature of the substrate on which the thin film is deposited, and (c) the temperature of the ozone nozzle supplying oxygen atoms O and the substrate. It is represented by a vector having three elements: the distance between Hereinafter, the n-th trial will also be referred to as "trial n", and the synthetic conditions used to generate the substance in trial n will be "xn _" .

The substance generation device 20 advances the reaction of raw materials according to the synthesis conditions generated by the synthesis condition generation device 10 to generate substances (samples). Examples of the substance generating device 20 include molecular beam epitaxy, sputtering, pulsed laser ablation, physical vapor deposition, chemical vapor deposition, spin coating, atomic layer deposition, floating zone, flux, and czochral. It is a device that generates substances by thin film synthesis and bulk synthesis methods such as the ski method. As an example, in the task of producing SrRuO ₃ thin films, a molecular beam epitaxy thin film production device can be used as the material production device 20 .

The substance evaluation device 30 measures an evaluation value obtained by evaluating the substance (sample) generated by the substance generation device 20 using a predetermined evaluation index. As the evaluation index, an existing evaluation index corresponding to the desired substance and its shape (thin film, bulk, etc.) may be used. As an example, in the task of producing a SrRuO ₃ thin film, it is conceivable to use a residual resistance ratio (RRR) as an evaluation index for measuring the goodness of arrangement of elements in the thin film.

Hereinafter, the evaluation value is composed of binary information indicating whether a substance (sample) is a stable substance or a desired substance and a scalar value indicating the quality of the substance, Take the scalar value if the material is a stable or desired material and NaN if the material is not a stable or desired material. For example, when obtaining an insulator with high electrical resistance, the evaluation value is the electrical resistance value if a desired sample is obtained, and NaN if not. Note that NaN represents Not a Number.

Also, hereinafter, the evaluation value for the substance generated by the substance generation device 20 according to the synthesis condition _xn is defined as “y _n ”, and the larger the evaluation value, the better the evaluation. Note that if the smaller the evaluation value, the better the evaluation, for example, by inverting the sign of the evaluation value, it is possible to apply the present embodiment in the same manner.

Here, the synthesis condition generation system 1 according to the present embodiment includes synthesis conditions obtained in previous trials and their evaluation values {(x _n , y _n )|n=1, . . . , N} (where , N is the number of _trials so far _. The device 30 evaluates and measures the evaluation value y _N+1 , and repeats trials to search for a synthesis condition that gives a good evaluation value. In the following, the substance generation device 20 and the substance evaluation device 30 are treated as black boxes, and the synthesis conditions and their evaluation values {(x _n , y _n )|n=1, . . . , N }, the synthesis condition generation device 10 generates a synthesis condition x _N+1 that gives a better evaluation value in the N+1th trial. Thus, by repeating this trial, it becomes possible to generate synthesis conditions for obtaining an optimum substance.

However, the overall configuration of the synthesis condition generation system 1 shown in FIG. 2 is an example, and is not limited to this. For example, the synthesis condition generation device 10 and two or more of the substance generation device 20 and the substance evaluation device 30 may be realized by the same device.

<Hardware Configuration of Synthesis Condition Generation Device 10>
FIG. 3 shows a hardware configuration example of the synthesis condition generation device 10 according to this embodiment. As shown in FIG. 3, the synthesis condition generation device 10 according to the present embodiment includes an input device 101, a display device 102, an external I/F 103, a communication I/F 104, a RAM (random access memory) 105, It has a ROM (Read Only Memory) 106 , an auxiliary storage device 107 and a processor 108 . Each of these pieces of hardware is communicably connected via a bus 109 .

The input device 101 is, for example, a keyboard, mouse, touch panel, or the like. The display device 102 is, for example, a display, a display panel, or the like. Note that the synthesis condition generation device 10 may not have at least one of the input device 101 and the display device 102, for example.

The external I/F 103 is an interface with an external device such as the recording medium 103a. The synthesizing condition generating device 10 can perform reading and writing of the recording medium 103a via the external I/F 103. FIG. Examples of the recording medium 103a include flexible disks, CDs (Compact Discs), DVDs (Digital Versatile Disks), SD memory cards (Secure Digital memory cards), USB (Universal Serial Bus) memory cards, and the like.

The communication I/F 104 is an interface for connecting the synthesis condition generation device 10 to a communication network. A RAM 105 is a volatile semiconductor memory (storage device) that temporarily holds programs and data. The ROM 106 is a non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. The auxiliary storage device 107 is, for example, a storage device (storage device) such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The processor 108 is, for example, a computing device such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit).

By having the hardware configuration shown in FIG. 3, the synthesis condition generation device 10 according to the present embodiment can implement the synthesis condition generation processing described later. Note that the hardware configuration shown in FIG. 3 is an example, and the hardware configuration of the synthesis condition generation device 10 is not limited to this. For example, the synthesis condition generation device 10 may have multiple auxiliary storage devices 107 and multiple processors 108, and may have various hardware other than the illustrated hardware.

<Functional Configuration of Synthesis Condition Generation Device 10>
FIG. 4 shows a functional configuration example of the synthesis condition generation device 10 according to this embodiment. As shown in FIG. 4 , the synthesis condition generation device 10 according to this embodiment has an evaluation value input unit 201 , an evaluation value prediction unit 202 , a synthesis condition search unit 203 and a synthesis condition output unit 204 . Each of these units is implemented by, for example, one or more programs installed in the synthesis condition generation device 10 causing the processor 108 or the like to execute processing. Further, the synthesis condition generation device 10 according to this embodiment has a storage unit 205 . The storage unit 205 is realized by, for example, the auxiliary storage device 107 or the like. Note that the storage unit 205 may be realized by a storage device such as a database server connected to the synthesis condition generation device 10 via a communication network, for example.

The evaluation value input unit 201 inputs the evaluation value _yN evaluated by the substance evaluation device 30 in trial N (previous trial). Note that this evaluation value _yN is stored in the storage unit 205 .

Here, since each evaluation value y _n (n=1, . . . , N) can take NaN, conversion processing when y _n =NaN is necessary. Moreover, there is a possibility that the prediction accuracy when the evaluation value prediction unit 202 predicts the evaluation value deteriorates due to the difference in the scale of each element constituting the synthesis condition _xn and the scale of the evaluation value _yn itself. For this reason, in order to secure the prediction accuracy and efficiently search for the synthesis condition, it is necessary to normalize the synthesis condition _xn and its evaluation value _yn . Therefore, when the evaluation value y _N is input, the evaluation value input unit 201 converts the evaluation value y _n to y _{n ′} for each n=1, . Synthesis conditions x _n and y _n ' are each normalized by the following normalization processing.

- Conversion process The evaluation value input unit 201 converts each evaluation value y _n (n = 1, ..., N) using the following formula (1).

here,

is. That is, if y _n =NaN, it is the worst evaluation value in the trials so far. However, if y _n =NaN in all trials so far, then

and In this way, the worst evaluation value obtained so far is used as a substitute for a synthesis condition for which no evaluation value has been obtained. This solves the problem of the conventional method that "there may be synthetic conditions under which an evaluation value cannot be obtained".

-Normalization Processing Hereinafter, it is assumed that the synthesis condition _xn is a vector composed of D elements, and _xn = [xn _,1 ,...,xn _,D ] ^T . Here, T represents the transpose of a vector or matrix.

D elements x _n,d (d=1, . . . , D) represent conditions for obtaining a certain substance by a certain method. As an example, in the task of producing a _SrRuO3 thin film, xn _,1 is the Ru supply rate, xn _,2 is the temperature of the substrate on which the thin film is deposited, and _xn,3 is the ozone nozzle supplying oxygen atoms O and the substrate. represents the distance between each

The lower and upper limits of the search range of the d-th element x _{n, d} are respectively

and however,

is.

At ^this time, the evaluation value input unit 201 normalizes the synthesis condition x _n =[x _n,1 , _.

is determined by the following formula (2).

Similarly, the approximate value range for the evaluation value

is given, the evaluation value input unit 201 obtains a normalized evaluation value obtained by normalizing y _n ′ using the following formula (3) using the range.

On the other hand, if the approximate value range for the evaluation value is not given, the evaluation value input unit 201

, the normalized evaluation value obtained by normalizing y _n ' is obtained by the equation (3).

Hereinafter, in the text of the specification, the normalized synthesis condition obtained by normalizing the synthesis condition x _n will be referred to as " ^~ x _n ", and the normalized evaluation value obtained by normalizing the evaluation value y _n ' after conversion processing will be referred to as " ^~ y _n '". In addition, when there is no risk of misunderstanding, the normalized synthesis condition is also simply referred to as "synthesis condition", and the normalized evaluation value is simply referred to as "evaluation value". Further, let the variable representing the normalized synthesis condition be “ ^~ x” and the variable representing the normalized evaluation value be “ ^~ y”.

The evaluation value prediction unit 202 calculates the normalized synthesis condition obtained by the evaluation value input unit 201 and the normalized evaluation value D _N :={( ^~ x _n , ^~ y _n ')|n=1, . . . , N } to predict the synthesis condition ^~ evaluation value taken by x ^~ y that has never been tried before. where ^~ y is a random ^variable following a normal distribution with mean μ ( ^~ x) and ^variance σ

, μ ( ^~ x) and σ ² ( ^~ x) depending on the synthesis condition ^~ x are calculated by a Gaussian process using D _N as .

In a Gaussian process, we model the joint _{distributions} of ^∼y _{1 ′ ,} ^.

where K _mn is the m-by-n element of the covariance matrix K _N . k ( ^~ x _m , ^~ x _n ) constituting each element of the covariance matrix K _N is called a kernel function and represents the similarity between ^~ x _m and ^~ x _n . That is, when ^~ x _m and ^~ x _n have similar values, K _mn takes a large value, and as a result, the correlation between ^~ y _m ' and ^~ y _n ' increases (similar values tend to be taken). is a model. In this embodiment, as an example, an RBF (radial basis function) kernel is adopted as the kernel function k.

For RBF kernels with kernel function k, where ^~ _xm and ^~ _xn are D-dimensional vectors, the kernel function k is given by:

where A>0 is a parameter given to the kernel function.

A kernel function other than the RBF kernel may be employed as the kernel function k. For example, the Matern kernel described in Non-Patent Document 2 may be employed when the evaluation value changes sharply with respect to the synthesis condition ^~ x. Other than this, various kernel functions can be adopted as needed.

At this time, the evaluation value prediction unit 202 uses the above kernel function k(·,·) and D _N to calculate μ( ^˜x ) and σ ² ( ^˜x ) in Equation (4) as follows.

Therefore, the parameters of this normal distribution can be used to predict the evaluation value for the untried synthetic condition ^~ x by equation (4). Note that this explicitly describes the relationship between the synthesis conditions and the evaluation values.

Synthesis condition search section 203 searches for a synthesis condition ^~ x that gives a better evaluation value ^~ y using the distribution of evaluation values predicted by equation (4). Here, when a certain synthesis condition ^~ x is given, μ ( ^~ x) and σ ² ( ^~ x) corresponding to it are obtained from equation (9). Therefore, as a measure of the goodness of a certain synthesis condition ^~ x, the highest evaluation value in trials so far

Expected value of improvement from

Use However, p ( ^~ y) is the probability density function of normal distribution with mean μ ( ^~ x) and variance σ ² ( ^~ x). Also, Φ(·) and φ(·) are the cumulative density function and probability density function of the standard normal distribution with mean 0 and variance 1, respectively.

In the example shown in formula (10), the expected value of the improvement of the evaluation value (Reference 1) is used, but other criteria can be used to improve search efficiency (References 2 and 3). .

Thus, the problem of searching for a good synthesis condition ^˜x becomes the problem of searching for a synthesis condition ^˜x that makes equation (10) large.

return to

Here, if the dimensionality of the search space of the synthesis condition ^~ x is large, the number of points for evaluating equation (10) increases as an exponential function of the dimensionality. Therefore, as a device to reduce the number of evaluations of formula (10), after first evaluating formula (10) with a coarse grid, a region having several points with high expected improvement values as vertices is selected. Equation (11) is approximately maximized by increasing the resolution of the grid in the evaluation. In other words, the search space is first searched and evaluated with a coarse-resolution grid, and then searched and evaluated with a finer-resolution grid in a region having vertices in which the top several points with the highest evaluation results are located.
This yields a synthesis condition ^~ x _{N+1 that} approximately maximizes the expected value of equation (10).

Synthesis condition output unit 204 performs inverse normalization processing (processing for undoing normalization) shown in the following equation (12) for each element of synthesis condition ^˜x _N+1 searched by synthesis condition search unit 203. and output x _N+1 =[x _N+1,1 , . . . , x _N+1,D ] ^T .

As a result, in the N+1-th trial, the substance (sample) is produced and evaluated under the synthesis conditions xN ₊₁ . Note that this synthesizing condition x _N+1 is stored in the storage unit 205 .

The storage unit 205 stores the synthesis conditions obtained in previous trials and their evaluation values {(x _n , y _n )|n=1, . . . , N}. The storage unit 205 also stores the normalized synthesis conditions and normalized evaluation values {( ^~ x _n , ^~ y _n ′)|n=1, . . . , N}.

<Synthesis condition generation processing>
Next, synthesis condition generation processing according to this embodiment will be described with reference to FIG. In the following, a case will be described in which, after N trials have been performed, a synthesis condition x _N+1 that gives a better evaluation value in the N+1th trial is generated.

First, the evaluation value input unit 201 inputs the evaluation value _yN of the previous trial N (step S101). This evaluation value y _N is associated with the synthesis condition x _N in the previous trial N and stored in the storage unit 205 as (x _N , y _N ). Note that the synthesis condition x _N for the previous trial N has already been stored in the storage unit 205 .

Next _, the evaluation value input unit 201 converts the evaluation values y _n (n=1, .

Next, the evaluation value input unit 201 inputs the synthesis condition x _n (n=1, . . . , N) and the evaluation value y _n ' (n=1, . Each is normalized by normalization processing (step S103).

That is, the evaluation value input unit 201 normalizes the synthesis condition x _n according to equation (2) to generate the normalized synthesis condition ^˜x _n , and normalizes the evaluation value y _n ′ after conversion processing according to equation (3). ^y _n '. These normalized synthesizing conditions and normalized evaluation values ( ^~ x _n , ^~ y _n ') are stored in the storage unit 205 .

The following steps S104 to S107 are repeatedly performed for untried synthesis conditions ^~ x in the search space. As described above, first, the following steps S104 to S107 are repeated with a rough grid in the search space as an untried synthesis condition. It is preferred that the following steps S104 to S107 be repeated by increasing the resolution of the grids within the region and using those grids as untried synthesis conditions. Steps S104 to S107 in a certain repetition will be described below.

The evaluation value prediction unit 202 selects an untried synthesis condition ^~ x from the search space (step S104).

Next, the evaluation value prediction unit 202 predicts the evaluation value of the synthesis condition ^~ x selected in step S104 (step S105). Specifically, the evaluation value prediction unit 202 calculates the average μ ( ^∼ x) and the variance σ ² ( ^∼ x) that determine the distribution of the evaluation values of the synthesis condition ^∼ x according to Equation (9).

Next, the synthesis condition searching unit 203 uses the distribution of evaluation values predicted by expression (4) to calculate the expected value of expression (10) (improvement from the highest evaluation value obtained in the N-th trial). expected value) is calculated (step S106).

Then, the synthesis condition search unit 203 evaluates the expected values obtained in step S106 (for example, sorts the expected values obtained so far in ascending or descending order, specifies the maximum value, etc.). (step S107).

By repeating the above steps S104 to S107, the synthesizing condition ^˜x _N+1 shown in Equation (11) is approximately obtained.

The synthesis condition output unit 204 performs the inverse normalization process of Equation (12) on the synthesis condition ^~ x _N+1 finally obtained by repeating steps S104 to S107 to generate synthesis condition x _N+1. (step S108). Note that this synthesizing condition x _N+1 is stored in the storage unit 205 .

Then, the synthesis condition output unit 204 outputs the synthesis condition x _N+1 obtained in step S108 to the substance generation device 20 (step S109).

As described above, the synthesis condition _N+1 for the N+1th trial is obtained. Therefore, in the N+1th trial, a sample (substance) is generated by the substance generation device 20 according to the synthesis condition _N+1 , and this sample is evaluated by the substance evaluation device 30 to measure the evaluation value y _N+1 . By repeating such a trial a predetermined number of times, the synthesis conditions used to generate the sample with the best evaluation value (for example, the maximum evaluation value) are obtained as the synthesis conditions for obtaining the optimum substance. be done.

<Experimental results>
Two types of experimental results by the synthesis condition generation system 1 according to this embodiment will be described below. The first is an experiment using simulation data (hereinafter referred to as a simulation experiment), in which a two-dimensional synthesis condition search space is given as simulation data. The second is an experiment for producing SrRuO ₃ by molecular beam epitaxy (hereinafter referred to as SrRuO ₃ experiment).

- Simulation experiment In this experiment, the evaluation value represented by the following formula (13) was used.

However, the values of each constant are as follows.

||v|| ₁ =Σ _d |v _d | represents the L ₁ norm of vector v.

The above evaluation value becomes maximum at y≈1.24 near x≈[0.48, 0.52] ^T .

The search range of the synthesis condition is x∈[-1,1]×[-1,1]. however,

The evaluation value follows the formula (13) only within the range of , and y=NaN in other ranges. FIG. 6 shows the evaluation values used in this experiment.

The results of this experiment are shown in Figure 7. In FIG. 7, Our method represents the synthesis condition generation system 1 according to this embodiment. Pad is a conventional method, which treats NaN as an evaluation value of zero. In FIG. 7, the horizontal axis indicates the number of trials to obtain the evaluation value, and the vertical axis indicates the highest evaluation value obtained by the trial excluding NaN (that is, the upper left graph (higher evaluation value with fewer trials) the better.).

As shown in FIG. 7, with the Our method, the evaluation value exceeded 1.20 after about 15 trials, while with the pad, the evaluation value was less than 1.20 even after 30 trials. Therefore, it can be seen that with Our method, a better evaluation value is obtained with a smaller number of trials than with the conventional method.

• SrRuO ₃ experiment Table 1 below shows each element of synthesis conditions and its search range.

Residual resistance ratio y:=ρ(300K)/ρ(4K) was used as an evaluation value. However, y=NaN depending on the value of the synthesis condition x used for thin film formation.

The results of this experiment are shown in Figure 8. In FIG. 8, the horizontal axis indicates the number of trials to obtain the evaluation value, and the vertical axis indicates the maximum evaluation value obtained up to the trial. As shown in FIG. 8, it can be confirmed that the evaluation value continues to improve as the number of trials increases.

<Summary>
As described above, in the synthesis condition generation system 1 according to the present embodiment, it is possible to automatically search in the synthesis condition space where the substance generation is successful or not, and to discover synthesis conditions with good evaluation values. It can be used for creation, etc. In addition, in the synthesis condition generation system 1 according to the present embodiment, it is possible to optimize (maximize or minimize) the evaluation value with fewer trials than the conventional method, reducing the development period / development time of new materials and reducing the cost You can expect the effect of reduction.

The present invention is not limited to the specifically disclosed embodiments described above, and various modifications, alterations, combinations with known techniques, etc. are possible without departing from the scope of the claims. .

[References]
Reference 1: J. Mockus, V. Tiesis, and A. Zilinskas: The application of Bayesian methods for seeking the extremum. Towards Global Optimization 2, 1978.
Reference 2: N. Srinivas, A. Krause, S. Kakade, and M. Seeger: Gaussian Process Optimization in the Bandit Setting: No Regret and Experimental Design, International Conference on Machine Learning, 2010.
Reference 3: E. Contal, V. Perchet, and N. Vayatis: Gaussian Process Optimization with Mutual Information, International Conference on Machine Learning, 2014.

1 synthesis condition generation system 10 synthesis condition generation device 20 substance generation device 30 substance evaluation device 101 input device 102 display device 103 external I/F
103a recording medium 104 communication I/F
105 RAMs
106 ROMs
107 auxiliary storage device 108 processor 109 bus 201 evaluation value input unit 202 evaluation value prediction unit 203 synthesis condition search unit 204 synthesis condition output unit 205 storage unit

Claims (7)

1. A synthesis condition generation method for generating the synthesis conditions in a trial for sequentially executing the generation of a substance according to synthesis conditions, the measurement of an evaluation value for evaluating the substance, and the generation of the synthesis conditions,
   If there is a trial n' in which the substance evaluation value y _n' (n'∈{1, ..., N}) could not be measured among the N trials so far, the evaluation value in the trial n' an evaluation value conversion procedure for replacing y _n′ with the worst evaluation value;
   A prediction procedure for predicting an evaluation value y of an untried synthetic condition x using the synthetic condition x _n and the evaluation value y _n in the trial n (n=1, . . . , N);
   Using the result of predicting the evaluation value y of the untried synthesis condition x, a synthesis condition that takes a better evaluation value than the best evaluation value y _n (n=1, . a generation procedure that generates as synthesis condition x _N+1 in
   A computer-executed synthesis condition generation method.
2. The prediction procedure includes:
   Using synthetic condition x _n and evaluation value y _n in ^trials n (n=1, . Calculated as a result of said prediction,
   The generating procedure includes:
   Using the average μ(x) and the variance σ ² (x), calculate the expected value that the evaluation value y of the untried synthetic condition x takes a better evaluation value than the best value, and the expected value is the maximum 2. The method of generating a synthesis condition according to claim 1, wherein an untried synthesis condition such that is generated as the synthesis condition xN ₊₁ .
3. The prediction procedure includes:
   selecting an untried synthesis condition x from a predetermined first resolution grid point in a search space of synthesis conditions, predicting an evaluation value y of the selected untried synthesis condition;
   After all the untried synthesis conditions x have been selected from the grid points of the first resolution, a predetermined number of the untried synthesis conditions x are selected in descending order of the expected value, and the selected untried synthesis conditions x In the region having the synthesis condition x at the vertex, an untried synthesis condition x is selected from grid points of a second resolution higher than the first resolution, and the evaluation value y of the selected untried synthesis condition 3. The synthesis condition generating method according to claim 2, wherein
4. The synthesis condition generation method includes:
   further comprising a normalization procedure for normalizing the synthesis condition x _N and the evaluation value y _N in trial N;
   The prediction procedure includes:
   4. The synthesis condition generation method according to any one of claims 1 to 3, wherein the synthesis condition _xn and the evaluation value _yn after normalization are used to predict the evaluation value y of the untried synthesis condition x.
5. The evaluation value conversion procedure includes:
   Claims 1 to 4, wherein y _N =0 when all evaluation values y _n (n=1, . . . , N) in trials n (n=1, . . . , N) cannot be measured. The synthesis condition generation method according to any one of .
6. A synthesis condition generation device for generating the synthesis conditions in a trial for sequentially executing the generation of a substance according to synthesis conditions, the measurement of an evaluation value for evaluating the substance, and the generation of the synthesis conditions,
   If there is a trial n' in which the substance evaluation value y _n' (n'∈{1, ..., N}) could not be measured among the N trials so far, the evaluation value in the trial n' an evaluation value conversion unit that replaces y _n′ with the worst evaluation value;
   a prediction unit configured to predict an evaluation value y of an untried synthesis condition x using synthesis conditions x _n and evaluation values y _n in trials n (n=1, . . . , N);
   Using the result of predicting the evaluation value y of the untried synthesis condition x, a synthesis condition that takes a better evaluation value than the best evaluation value y _n (n=1, . a generating unit configured to generate as a synthesis condition x _N+1 in
   Synthesis condition generation device having
7. A program for causing a computer to execute the synthesis condition generation method according to any one of claims 1 to 5.

Images