Method for producing multilayered optical filters using a genetic algorithm and a method for operating the genetic algorithm

1. Field of the Invention

The present invention relates to a method for producing optical multilayered optical filters, for appropriately splitting/merging signal lights having plural wavelength convoluted, used in an optical communication system sending/receiving wavelengths multiplexed light communications, and to a method for ICE operating the genetic algorithm for extracting the most suitable combination from candidates of complex combination.

2. Description of the Prior Art

As wavelength-multiplexed light communications have been developed, there are needs for multilayered splitting/merging filters having such characteristics as a narrow passband or small para-reflex characteristics.

For methods of designing multilayered optical filters each layer of which has refractive index Ni and thickness Di based on a desired optic characteristics, in general, it has been common to use a method of thickness optimization such as described in Japanese Published Unexamined Patent Application No. 2650048, by determining in advance refractive index Ni of each layer composing the filter, to preset the thickness Di, to determine optical characteristics for each wavelength, to change the thickness Di, to optimize the thickness Di so as to have maximum in the optical characteristics.

As another method of optimization, there is only a trial and error method in which refractive index of a layer is altered, and the thickness for every layer is optimized again.

In accordance with the method as described above, since the refractive index Ni is predetermined, the degree of freedom of design is not insufficient, and furthermore the results may often fall into a local solution.

In addition, there are a huge number of combinations of Ni and Di in a multilayered optical filter so that the optimization requires very long time for selection; therefore the selection of the most optimized combination has been practically impossible.

In the Prior Art local resolution obtained by using the thickness optimization method as have been described above was needed and the only practical result.

The present invention been made in view of the above circumstances and has an object to overcome the above problems and to provide a method of production of multilayered optical filters, used in the designing of multilayered optical filters having such complex combinations as described above, for selecting the most optimized combination of refractive index Ni and thickness Di for each of layers without falling into a local solution.

Also, the inventors of the present invention have been realized a method for selecting the most optimized solution, in order to determine the most optimized or almost optimized solution, in which a genetic algorithm (abbreviated as GA hereinafter) are used for converting the problems of optimization into the genetic sequences, such as the problems of optimization of the combinations of refractive index Ni and thickness Di in each layer in the designing of multilayered optical filters, comprising the steps of:

generating pattern groups, for generating pattern groups each comprised of a plurality of patterns;

extracting at least two patterns from within the generated group of patterns;

mutating and crossing over for generating new patterns by mutating or crossing over the extracted patterns;

evaluating for calculating the fitness of the optimization problem of the groups comprising the extracted patterns and newly generated patterns, for each of mutation and cross-over steps;

selecting to decrease the number of groups including extracted patterns and newly generated patterns to the number of extracted patterns;

substituting the selected patterns in place of the patterns extracted from the pattern groups, and

altering the contents of pattern groups by repeating between the extracting step and the selecting step.

The step of selecting in the genetic algorithm comprises either elite method, for selecting patterns of the predetermined numbers in the order of fitness among the object pattern group and removing others, or elite-roulette method, for selecting some patterns by using the elite method, and others by using the roulette method and removing the rest.

In both selection methods, there is no restriction of pattern selection, so that any duplicated patterns having the identical pattern elements were not rejected. If the selected patterns are filled with patterns having the identical elements, in some optimization problems, patterns having the same elements are inserted into the pattern group, and after repeating the genetic algorithm steps as described above, pattern group are totally filled with the patterns having the identical elements. This may result in an unexpected error in the retrieval search of the most optimized solution.

Often the search results obtained by the GA as above may be a local solution, which may or may not satisfy the goal.

Another object of the present invention is to avoid, in GA the predominance of selected pattern and of pattern group with the patterns having the same elements by preventing the patterns having the identical elements from coexisting in the selection patterns in order to retrieval search not only a local solution but also the most suitable one.

A method for producing multilayered optical filters comprises:

a generating step for generating a initial pattern comprising a matrix given by

P =( X 1 , X 2 , X 3 . . . , XS )  (1)

which is comprised of elemental matrices Xi, each of which comprises as element refractive index and thickness of i layers (i is an integer equal to or more than 1) of a multilayered optical filter having S layers (S is an integer equal to or more than 1);

a reproducting/mutating step for either increasing or decreasing, in an arbitrary element Xi of the initial pattern, either the refractive index or thickness of the initial pattern by a predetermined number, in terms of the initial pattern, to generate a predetermined number of mutation patterns which are mutually different one from other;

a cross over step for selecting at least one pair of patterns from the mutated patterns generated in the reproducting/mutating step and the initial patterns to cross over, in the pair of patterns selected, the matrix Xi in the pattern and/or the matrix obtained by the mutation of the matrix by the predetermined number to generate a predetermined number of crossed over patterns;

a selecting step for selecting the desired number of patterns having the most appropriate optical characteristics from the mutated pattern group generated in the reproducting/mutating step, the crossed over pattern group, and the pattern group comprised of the initial patterns; and

a repeating step for repeating a series of algorithmic process steps comprised of the reproducting/mutating step, the cross over step, and the selecting step,in terms of the predetermined number of patterns selected in the selecting step instead of the initial patterns, until the optical characteristics of the selected patterns obtained in the immediately preceding algorithmic steps may conform to the desired error range for the desired optical characteristics.

In accordance with the method of producing multilayered optical filters in accordance with the present invention, the repetition of a series of algorithmic process steps allows each layer in a multilayered optical filter to be set in such a manner as the optic characteristics may conform to a specific desired range. This may result in a better design and production of multilayered optical filters when compared with the conventional designing method.

When a series of the algorithmic process steps comprised of the reproducting and mutating step, the cross over step, and the selecting step, if the optical characteristics of the selected pattern obtained from the algorithmic process steps match with the optical characteristics of the selected pattern obtained from one of the repetitions preceding to the former, a second initial pattern, which is different from the initial pattern, is preferably set to repeat the algorithmic process steps using this second initial pattern.

In accordance with the method disclosed herein, even when the optical characteristics of the design obtained by the repetition of a series of algorithmic process steps show no progress in improvement, then the algorithm may recover from this stacked condition by restarting from the second initial pattern.

Preferably, the second initial pattern may be chosen which is generated by either increasing or decreasing the refractive index or the thickness in the selection patterns having the most appropriate optical characteristics obtained in the preceding process steps.

In accordance with the method disclosed herein, the information about optical elements, effective to the improvement of optical characteristics, in the patterns generated until the optical characteristics have been matched with, may be effectively used for the second initial pattern, allowing better and faster design of multilayered optical filters with better optical characteristics.

Another method for producing multilayered optical filters comprises:

a generating step for generating a initial pattern comprising a matrix given by

P =( X 1 , X 2 , X 3 . . . , XS )  (1)

which is comprised of elemental matrices Xi, each of which comprises as element refractive index and thickness of i layers (i is an integer equal to or more than 1) of a multilayered optical filter having S layers (S is an integer equal to or more than 1);

a first reproducting/mutating step for duplicating a predetermined number of patterns from the initial pattern for either increasing or decreasing, in an arbitrary element Xi of the duplicated pattern, either the refractive index or thickness of the initial pattern by a predetermined number;

a first selecting step for selecting the desired number of patterns having the most appropriate optical characteristics from the groups consisted of the mutation patterns generated in the immediately preceding step and the initial pattern;

a cross over step for selecting at least one set of a pair of patterns from the selected patterns generated in the selection step to cross over, the matrix Xi in the pattern and/or the matrix obtained by the mutation of the matrix by the predetermined number to generate a predetermined number of crossed over patterns;

a second reproducting/mutating step for selecting and replicating at least one arbitrary pattern from the pattern groups consisted of the crossed over pattern group and the selected pattern group, to generate mutated pattern by either increasing or decreasing, in an arbitrary element Xi of the duplicated pattern and/or in an element of the matrix obtained by the mutation of the matrix by the predetermined number;

a second selection step for selecting the desired number of patterns having the most appropriate optical characteristics from the pattern groups consisted of the mutated pattern group, the crossed over pattern group, and the selected pattern group;

a repeating step for repeating a series of algorithmic process steps comprised of the first reproducting/mutating step, the first selecting step, the cross over step, the second reproducting/mutating step, and the second selecting step, in terms of the predetermined number of patterns selected in the second selecting step instead of the initial pattern, until the optical characteristics of the second selected patterns obtained in the immediately preceding algorithmic steps may conform to the desired error range for the desired optical characteristics.

In accordance with another method of producing multilayered optical filters in accordance with the present invention, the repetition of a series of algorithmic process steps allows each layer in a multilayered optical filter to be set in such a manner as the optic characteristics may conform to a specific desired range. This may result in a better design and production of multilayered optical filters when compared with the conventional designing method.

When a series of the algorithmic process steps comprised of the first reproducting/mutating step, the first selecting step, the cross over step, the second reproducting/mutating step, and the second selecting step, if the optical characteristics of the second selected pattern obtained from the algorithmic process steps match with the optical characteristics of the second selected pattern obtained from one of the repetitions preceding to the former, a second initial pattern, which is different from the initial pattern, is preferably set to repeat the algorithmic process steps once again using this second initial pattern.

In accordance with the method disclosed herein, even when the optical characteristics of the design obtained by the repetition of a series of algorithmic process steps show no progress in improvement, then the algorithm may recover from this stacked condition by restarting from the second initial pattern.

Preferably, the second initial pattern may be chosen which is generated by either increasing or decreasing the refractive index or the thickness in the second selected patterns having the most appropriate optical characteristics obtained in the preceding process steps.

In accordance with the method disclosed herein, the information about optical elements, effective to the improvement of optical characteristics, in the patterns generated until the optical characteristics have been matched with, may be effectively used for the second initial pattern, allowing better and faster design of multilayered optical filters with better optical characteristics.

It is preferable to perform selection in the selecting step or the first and second selecting steps based on the order of Qj, largest-first, the Qj being given by: Q = 1 { ∑ λ     ( Rj  ( λ ) - Rr  ( λ ) ) 2 } ( 2 )

where Rj(λ) is the reflectance characteristics at each wavelength λ obtained from the pattern combination of elements in the matrix Xi, Rr(λ) is the desired reflectance characteristics.

If the disjunction X of the reflectance of a pattern combination from the desired reflectance characteristics, i.e., the denominator item of Q shrinks by the convergence, and the difference of X among patterns becomes small, larger difference among patterns can be taken because the Q is used as the fitness. Therefore finer extraction becomes possible even when the difference of X among patterns is small.

When the predetermined value of increasing or decreasing the refractive index and the predetermined value of increasing or decreasing the thickness in the second reproducting/mutating step is set to be 2 through 50 times, respectively, of the predetermined value of increasing or decreasing the refractive index and the predetermined value of increasing or decreasing the thickness in the first reproducting/mutating step, the convergence, around the solution may increase and the most appropriate solution may be extracted faster.

When the predetermined value of increasing or decreasing the refractive index and the predetermined value of increasing or decreasing the thickness in the second reproducting/mutating step is set to be 2 through 25 times of the predetermined value of increasing or decreasing the refractive index and the predetermined value of increasing or decreasing the thickness in the first reproducting/mutating step, the convergence around the solution increases more and the most appropriate solution may be extracted more faster.

The method of operating genetic algorithm in accordance with the present invention comprises:

pattern group generating step for generating pattern group consisted of a plurality of mutually different patterns, each of which patterns comprises elemental matrices Xi, and given by

P =( X 1 , X 2 , X 3 . . . , XS )  (1)

a manipulating step for extracting a predetermined number of patterns from the pattern group and operate on the elements of these patterns to generate operated patterns;

a selecting step for selecting the same number of patterns having mutually different characteristics from the extracted patterns and operated patterns based on the characteristics obtained from these patterns;

a substituting step for adding a predetermined patterns selected in the selecting step into the pattern group in place of the extracted patterns; and

a repeating step for repeating a series of algorithmic process steps comprised of the operating step, the selecting step, and the substituting step, until the best characteristics in the preceding pattern group obtained in the algorithmic process steps may conform to the desired error range for the desired characteristics.

In accordance with the method of operating on a genetic algorithm, in the GA, patterns having the identical elements in the selected patterns may not coexist, thus preventing the occupation of selected patterns by the pattern having the identical elements as well as the occupation of pattern groups, in order to allowing retrieval search to proceed until the best solution is found.

Preferably, the manipulating step may comprise a cross over step for extracting at least one set of a pair of patterns consisted of mutually different elements, swapping a part of the matrix in the patterns between thus extracted pattern pair to generate crossed over patterns.

In accordance with the method disclosed herein, since the cross-over is performed between patterns having mutually different elements, the crossed over patterns may be prevented from being identical to the extracted patterns. In addition, in the same process step, a new pattern having partial combination of elements effective to find the best solution in the pattern may be generated.

Preferably, the manipulating step may comprise a mutating step for extracting a predetermined number of patterns, mutating a part of the matrix of the pattern in the extracted patterns to generate mutated patterns.

In accordance with the method disclosed herein, a new pattern having partial combination of elements effective to find the best solution in the pattern and completely new elements in other parts may be generated.

Preferably, mutation method may be either a method of increasing or decreasing any element constituting the pattern by the predetermined amount of mutation, or substituting with one of predetermined candidate elements.

Former method may be effective when the genetic algorithm is applied to such a problem of finding the best solution that a solution of contiguously transforming functions should be determined, while on the other hand the latter may be effective when the genetic algorithmic is applied to such a problem of finding the best solution of combination that uses dissociative candidates.

Preferably, the operating step may comprise, in addition to the cross-over step, a mutating step for extracting a predetermined number of patterns and mutating a part of matrix of the patterns in the extracted patterns to generate mutated patterns.

The relationships between the cross-over step and the mutation step in the operating step is preferably such that the cross-over may be performed at first and then the mutation step may be performed thereafter. However, the mutation may be performed at first and then the cross-over, or either one of cross-over and mutation may be selectively performed for each repetition of the series of algorithmic process steps constituted of operating step, selecting step, and substituting step, or the cross-over and mutation may be alternatively performed for each repetition of the series of algorithmic process steps constituted of operating step, selecting step, and substituting step.

In accordance with the method as described above, diverse patterns may be newly generated by making use of both cross-over and mutation.

Preferably, the patterns selected in the selecting step may be comprised of the pattern having the best characteristics and the patterns selected by the roulette method.

In accordance with the method as described above, if there are the same patterns in the pattern groups comprised of manipulated patterns and extracted patterns, the best pattern and another pattern that is comprised of elements different from the best pattern and of elements mutually different each from other may be selected.

Preferably, the patterns selected in the selecting step may be comprised of the pattern having the best characteristics and the patterns selected by the random number method.

In accordance with the method as described above, if there are the same patterns in the pattern groups comprised of manipulated patterns and extracted patterns, a pattern may be selected which is comprised of elements different from the best pattern and of elements mutually different each from other.

Preferably, when the substitution step comprises a comparing step for comparing the characteristics of each pattern in the pattern group to be substituted in the substitution step with the characteristics of selected pattern, if there is a pattern having the identical characteristics to that of the selected pattern in the pattern group to be substituted in the substitution step, the substitution step may be omitted to proceed immediately to the operating step.

In accordance with the method disclosed herein, if there is already the selected pattern in the pattern group, the operating step may be repeated to prevent the identical patterns from occupying to allowing retrieval search to proceed until the best solution is found.

Although the steps of operating, substituting, and determining whether the algorithm has been completed, use the characteristics specific of patterns, the fitness of characteristics specific to patterns for each patterns with respect to the target characteristics may be determined and used instead. The decision of end of algorithm in this case is preferably such that the difference between the best fitness in the pattern group and the target fitness may be fitted into the error range of the target.

In accordance with the method disclosed herein, if the pattern characteristics are expressed by a plurality of values, the object to be compared with may be one fitness indication by determining the fitness with respect to the target characteristics, allowing the computational cost of comparison to be reduced.

The best values most suitable to the thickness of each layer and the refractive index may be:readily determined when designing multilayered optical filters by applying the method of operating on the genetic algorithm in accordance with the present invention. In such a case the elements of matrix Xi in the patterns should preferably be the thickness di and refractive index ni of the layer i.

In accordance with the method disclosed herein, the thickness di and refractive index ni of the layer i, the principal elements constituting the multilayered optical filters, may be simultaneously optimized, enabling larger freedom of design.

Additional objects and advantages of the invention will be according to part in the description which follows and in part will be obvious from the description, or may be learned by practice of the invention. The objects and advantages of the invention may be realized and attained by means of the instrumentalities and combinations particularly pointed out in the appended claims.

The accompanying drawings, which are incorporated in and constitute a part of this specification illustrate an embodiment of the invention and, together with the description, serve to explain the objects, advantages and principles of the invention in the drawings,

FIG. 1 shows a cross-sectional view of a multilayered optical filter with 5 layers, produced in accordance with first preferred embodiment of the present invention;

FIG. 2 shows a flowchart illustrating first preferred embodiment of the present invention;

FIG. 3 shows a schematic diagram illustrating the reproduction and small-scaled mutation process in the flowchart shown in FIG. 2;

FIG. 4 shows a flowchart illustrating the small-scaled mutation process in the flowchart shown in FIG. 2;

FIG. 5 shows a schematic diagram illustrating first selection process in the flowchart shown in FIG. 2;

FIG. 6 shows a schematic diagram illustrating cross-over process in the flowchart shown in FIG. 2;

FIG. 7 shows a flowchart illustrating cross-over process in the flowchart shown in FIG. 2;

FIG. 8 shows a schematic diagram illustrating selection, reproduction, and middle/large-scaled mutation process in the flowchart shown in FIG. 2;

FIG. 9 shows a flowchart illustrating middle-scaled mutation in the flowchart shown in FIG. 2;

FIG. 10 shows a flowchart illustrating large-scaled mutation in the flowchart shown in FIG. 2;

FIG. 11 shows a schematic diagram illustrating second selection process in the flowchart shown in FIG. 2;

FIG. 12 shows a graph illustrating reflectance of the best result of optimization of a multilayered optic filter having five layers produced in accordance with first example of the present invention;

FIG. 13 shows a graph illustrating reflectance of the best result of optimization of a multilayered optic filter having fifteen layers produced in accordance with second example of the present invention;

FIG. 14 shows a flowchart illustrating second preferred embodiment in accordance with the present invention;

FIG. 15 shows a schematic diagram illustrating reproduction and mutation process in the flowchart shown in FIG. 14;

FIG. 16 shows a flowchart illustrating mutation process in the flowchart shown in FIG. 14;

FIG. 17 shows a schematic diagram illustrating cross-over process in the flowchart shown in FIG. 14;

FIG. 18 shows a flowchart illustrating cross-over process in the flowchart shown in FIG. 14;

FIG. 19 shows a schematic diagram illustrating selection process in the flowchart shown in FIG. 14;

FIG. 20 shows a flowchart illustrating second initial pattern generation process in the flowchart shown in FIG. 14;

FIG. 21 shows a graph illustrating the shift of fitness with respect to the number of generations in the course of calculation for optimizing a multilayered optic filter having three layers obtained in third example of the present invention;

FIG. 22 shows a flowchart illustrating third preferred embodiment of the present invention;

FIG. 23 shows a flowchart illustrating cross-over process shown in the flowchart of FIG. 22;

FIG. 24 shows a flowchart illustrating selection process shown in the flowchart of FIG. 22;

FIG. 25 shows a flowchart illustrating fourth preferred embodiment in accordance with the present invention;

FIG. 26 shows a flowchart illustrating mutation process in the flowchart shown in FIG. 25;

FIG. 27 shows a flowchart illustrating fifth preferred embodiment in accordance with the present invention;

FIG. 28 shows a flowchart illustrating sixth preferred embodiment in accordance with the present invention;

FIG. 29 shows a cross-sectional view of a multilayered optical filter with three layers, produced in accordance with fourth example of the method in accordance with the present invention;

FIG. 30 shows a graph illustrating the shift of the best fitness in a pattern group with respect to the number of generations in the course of calculation for optimizing a multilayered optic filter having three layers obtained in fourth example of the method in accordance with the present invention;

FIG. 31 shows a graph illustrating the shift of dispersion of fitness in a pattern group with respect to the number of generations in the course of calculations for optimizing a multilayered optic filter having three layers obtained in fourth example of the method in accordance with the present invention; and

FIG. 32 shows a graph illustrating the refractive characteristics of the result of optimization of a multilayered optical filter having three layers produced in accordance with fourth example of the method of the present invention.

First Embodiment

A preferred embodiment in accordance with the present invention will be described below in greater details with reference to the accompanying drawings. In first preferred embodiment, the calculation for optimizing a multilayered optical filter comprised of five layers from the first layer 2 through fifth layer 6 formed on a substrate 1 as shown in FIG. 1 . In first preferred embodiment an algorithm is used which comprises the steps of generating initial pattern, first reproduction and mutation, first selection, cross-over, second reproduction and mutation, and second selection.

FIG. 2 shows a flowchart of designing a multilayered optical filter comprising five layers in accordance with first preferred embodiment of the present invention.

The first reproduction and mutation step is assumed to be a reproduction and small mutation process step, and the second reproduction and mutation step to be a selection, reproduction and middle/large-scaled mutation process step.

In the present preferred embodiment, random number α(=A, B, C, D, E, F, G, H) should be used in the operations of selection, extraction, deselection, cross-over, and mutation designated in steps S 2 through S 8 . The way how to generate this random number α is the congruence method given by the following equations (3) and (4):

α= Mn +2/10000  (3)

Mn +2 =MOD ( Mn +1 +Mn , 10000)  (4)

where MOD is a function designating to a modulus.

In step S, the initial values required for calculation (desired optical characteristics; total number of film layers; refractive index of the substrate; incident angle of the incident light; total number of patterns after the reproduction and small mutation process step; total number of patterns after the first selection step; total number of patterns after cross over step; total number of patterns after the selection, reproduction and middle/large-scaled mutation process step; total number of patterns after the second selection step; amount of small mutation; amount of middle/large-scaled mutation; number of repetition of the middle/large-scaled mutation step; target value of the fitness) or the range of numbers are specified.

The matrix Xi comprising the refractive index Ni and thickness Di as elements in the layer i of the multilayered film j may be expressed as following equation (5):

Xi =( Ni, Di )  (5)

By inputting initial values for element of each layer Xi=(Ni, Di) a matrix pattern Pj as expressed in (6) may be generated, which is constituted of the combination of Xi, indicating the multilayered film j (step of generating initial pattern group):

Pj =( X 1 , X 2 , . . . , Xi , . . . )  (6)

In the present embodiment, the initial values required for the calculation in step S 1 are as shown in Table 1. A pattern P 1 of combination of elements as expressed in the following expression (7) may be generated by the operation as described above:

P 1 =( X 1 , X 2 , X 3 , X 4 , X 5 )  (7)

Then, in step S 2 , the initial pattern is replicated to increase the total number of pattern group up to the number of patterns after the reproduction step (shown in FIG. 3 ).

In the present embodiment, total number of pattern groups after this operation will be P 1 through P 20 , patterns being expressed as the following equation (8):

P 1 = P 2 = P 3 . . . = P 20   (8)

Then, small mutation is performed on each of the replicated patterns P 2 through P 20 (as shown in FIG. 4 ). A random number A is generated in the range of 1≦A≦5 (=number of layers), and one element Xi={(Ni, Di)} in the patterns is selected by using the random number A (step S 2 - 2 ) Then, another random number B is generated in the range 0≦B≦1 (step S 2 - 3 ), and Di will be selected if the random number B is equal to or more than 0.5 (step S 2 - 4 a ), Ni will be selected if B is less than 0.5 (step S 2 - 4 b ). Then, still another random number C is generated in the range of 0≦C≦1 (steps S 2 - 5 a , S 2 - 5 b ), and + is selected if the random number C is less than 0.5 (steps S 2 - 6 a , S 2 - 6 d ), will be selected if C is equal to or more than 0.5 (steps S 2 - 6 b , S 2 - 6 c ). Small mutation of either positive or negative (+ or −Δnz, or + or −Δdz) is performed on Ni or Di by the above operation (reproduction/small mutation step S 2 ).

In the present embodiment, assuming that random numbers are A=3, B=0.2, and C=0.6 with respect to the replicated pattern P 2 , then the P 2 by the operation above may be expressed as the following equation (9) below:

P 2 =( X 1 , X 2 , X 3 , X 4 , X 5 )={ X 1 , X 2 , (N 3 +Δnz, D 3 ) X 4 , X 5 }  (9)

As shown in FIG. 2, in step S 3 , respective optical characteristics for each of patterns of combination of elements P 1 through P 20 generated through S 2 will be calculated in order to calculate the difference from the desired optical characteristics. The optical characteristics are referenced to as the reflection characteristics of the multilayered optical filter with respect to the incident light incident to the filter at the incident angle θ. It should be noted that the transmission characteristics may be used as the optical characteristics, instead of the reflection characteristics. Assuming that the reflection characteristics with respect to each wavelength λ of incident light for respective pattern Pj of combination of elements is Rj (λ), the desired reflection characteristics is Rr (λ), the bandwidth of wavelength of the desired reflection characteristics is λ1 through λ2, and the square of difference between Rj (λ) and Rr (λ) in λ1 through λ2 is X, then X may be expressed by the following equation (10): X = ∑ λ     ( Rj  ( λ ) - Rr  ( λ ) ) 2 ( 10 )

and the fitness Qj may be expressed by the following equation (2): Q = 1 { ∑ λ     ( Rj  ( λ ) - Rr  ( λ ) ) 2 } ( 2 )

That is, the larger the value Qj is, then the optical characteristics of the multilayered optical filter comprising that pattern of combination of elements are closer to the desired value.

Next, based on the fitness Qj with respect to the desired optical characteristics of the pattern Pj of combination of elements, calculated in step S 3 , as shown in FIG. 2, patterns of combination of elements are selected (shown in FIG. 5 ). To do this, Pj are sorted in the order of the fitness Qj from highest to lowest, then the sorted patterns are picked up by the number corresponding to the total number of patterns after selection from the highest. Patterns other than those picked up are deselected to reject (primary selection step S 4 ).

In the case of the present embodiment, if the size of fitness Q 1 through Q 20 are in the order as expressed by the equation (11) below:

Q 20 > Q 19 > Q 18 > . . . > Q 3 > Q 2 > 1   (11)

since the total number of individuals after selection is 10, the patterns of combination of elements extracted by the foregoing operation will be P 11 through P 20 .

Then, as shown by step, S 5 in FIG. 2, patterns are crossed over (shown in FIG. 6 and FIG. 7 ). FIG. 6 shows the generation of patterns in one cross-over step, while FIG. 7 shows a flowchart of cross over step including repetition. Here two random numbers D and E are generated in the range 1≦D, E≦10 (=total number of patterns after selection), then based on these random numbers D and E two patterns Pj′ and Pj″ (where 1≦j′, j″≦j) are picked up from among pattern groups P 1 through P 10 of combination of elements selected in step S 4 (step S 5 - 1 )

In the present embodiment, assuming that two patterns Pj′ and Pj″ may be expressed by the following equation 12 and 13, respectively (step S 5 - 2 ):

Pj ′=( X 1 ′, X 2 ′, X 3 ′, X 4 ′, X 5 ′)  (12)

Pj″=( X 1 ″, X 2 ″, X 3 ″ , X 4 ″, X 5 ″)  (13)

For each of patterns Pj′ and Pj″, another random number F may be generated in the range of 1≦F≦4 (={number of layers}−1), then based on this random number F the locations between the elements at which two patterns are to be crossed over are determined (step S 5 - 3 ).

In the present embodiment, assuming F=3, then the locations between the elements at which two patterns Pj′ and Pj″ are to be crossed over may be expressed as shown in equation 14 below: Pj ′ = ( X1 ′ , X2 ′ , X3 ′ ,  X4 ′ , X5 ′ )   ×   Pj ″ = ( X1 ″ , X2 ″ , X3 ″ ,  X4 ″ , X5 ″ ) 

    Crossing     over     point ( 14 )

Two patterns are crossed over at that locations between elements so as to generate newly combined two patterns Ck and Ck+1 as expressed in equations 16 and 17 below (steps 35 - 4 and S 5 - 5 ):

Ck =( X 1 ′, X 2 ′, X 3 ′, X 4 ″, X 5 ″)  (16)

Ck +1=( X 1 ″, X 2 ″, X 3 ″, X 4 ′, X 5 ′)  (17)

Then, the operation in the steps S 5 - 1 through S 5 - 5 will be repeated until the number of whole pattern groups will become equal to the predetermined total number of patterns after cross-over process step (cross-over step S 5 ).

In the present embodiment, as the total number of patterns after cross-over step is 20, and as the total number of patterns before cross-over step is 10, the operation of the steps S 5 - 1 through S 5 - 5 will be repeated until C 10 will have been generated (step S 5 - 6 ).

In step S 6 as shown in FIG. 2, a middle/large-scaled mutation is applied to the patterns Pj and Ck after cross-over to generate a new pattern CRp (shown in FIG. 8 ).

However, for the selection between middle- and large-scaled mutation in step S 6 , a random number G in the range of 0≦G≦1 at the completion of cross-over process step is generated, and middle-scaled mutation will be applied if the random number G is less than 0.8, or large mutation will be applied if G is 0.8 or more.

In step S 6 , a random number H is generated in the range of 0≦H≦(j+k), then based on this random number H one pattern is picked up from within the patterns Pj and Ck after cross-over and thus selected pattern is duplicated to generate a new pattern CRp-a. The operation of selecting and duplicating a pattern will be repeated until the total number of pattern groups will have become equal to 40, the total number of patterns after middle/large-scaled mutation process step. The selection and replication process will generate CR 1 - a through CR 20 - a . Thereafter, by applying a middle/large-scaled mutation to the CRp-a, patterns CRp having middle/large-scaled mutation applied will be generated (selection, reproduction, middle/large-scaled mutation process step S 6 ).

Now the process of middle-scaled mutation will be described below in greater details. As shown in FIG. 9, in the middle-scaled mutation step, a random number A is generated in the range of 1≦A≦5 (=number of layers) and thereby one element Xi={(Ni, Di)} in the patterns CRp-a will be selected (step S 6 - 1 - 2 ) Then another random number B will be generated in the range of 0≦B≦1 (step S 6 - 1 - 3 ), and Di will be selected if the random number B is 0.5 or more (step S 6 - 1 - 4 a ), or Ni will be selected if the random number B is less than 0.5 (step S 6 - 1 - 4 b ). Thereafter still another random number C will be generated in the range of 0≦C≦1 (steps S 6 - 1 - 5 a , S 6 - 1 - 5 b ), then + is selected if the random number C is less than 0.5 (steps S 6 - 1 - 6 a , S 6 - 1 - 6 d ), will be selected if C is equal to or more than 0.5 (steps S 6 - 1 - 6 b , S 6 - 1 - 6 c ). By the above operation, middle-scaled mutation of either positive or negative (+ or −Δnc, or + or −Δdc) will be performed on Ni or Di. The operation from S 6 - 1 - 2 to S 6 - 1 - 6 a , 6 b , 6 c , and 6 d will be repeated a predetermined number of times Nc (=5) (step S 6 - 1 - 7 ) to generate patterns CRp having middle-scaled mutation applied (step S 6 - 8 ).

Now the process of large-scaled mutation will be described below in greater details (shown in FIG. 10 ). In the large-scaled mutation process, a random number A is generated in the range of 1≦A≦5 (=number of layers) and thereby one element Xi={(Ni, Di)} in the patterns CRp-a will be selected (step S 6 - 2 - 2 ). Then another random number B will be generated in the range of 0≦B≦1 (step S 6 - 2 - 3 ), and Di will be selected if the random number B is 0.5 or more (step S 6 - 2 - 4 a ), or Ni will be selected if the random number B is less than 0.5 (step S 6 - 2 - 4 b ). Thereafter still another random number C will be generated in the range of 0≦C≦1 (steps S 6 - 2 - 5 a , S 6 - 2 - 5 b ), then + is selected if the random number C is less than 0.5 (steps S 6 - 2 - 6 a , S 6 - 2 - 6 d ), will be selected if C is equal to or more than 0.5 (steps S 6 - 2 - 6 b , S 6 - 2 - 6 c ). By the above operation, large-scaled mutation of either positive or negative (+ or −Δnd, or + or −Δdd) will be performed on Ni or Di. The operation from S 6 - 2 - 2 to S 6 - 2 - 6 a , 6 b , 6 c , 6 d will be repeated a predetermined number of times Nc (=10) (step S 6 - 2 - 7 ) to generate patterns CRp having large-scaled mutation applied (step S 6 - 8 ).

By applying the middle/large-scaled mutation process steps as above, pattern group of combination of elements having completely different element patterns may be incorporated into the pattern groups.

After the selection, reproduction, and middle/large-scaled mutation process step in step S 6 has been completed, then in step S 7 the optical characteristics of the patterns Pj, Ck, and CRp having passed through step S 6 are calculated so as to determine the difference in optical characteristics from the desired value. The optical characteristics refers to as the reflection characteristics of the multilayered optical filter with respect to the incident light incident to the filter at the incident angle θ, in a manner similar to step S 3 .

In accordance with the method as described above, the fitness Qm for each pattern (where 1≦m≦j+k+p) may be determined.

Next, in step S 8 , the patterns of combination of elements are selected based on the fitness Qm for the patterns Pj, Ck, CRp of combination of elements, determined in the step S 7 with respect to the desired optical characteristics (shown in FIG. 11 ). To do this, patterns Pj, Ck, CRp are sorted in the order of Qm from the highest to the lowest, then the sorted patterns are picked up from the highest by the number corresponding to the total number of patterns after selection. Patterns other than those picked up are deselected to reject (secondary selection step S 8 ).

In the present embodiment, if the patterns are in the order as shown in the following equation (18), by sorting on the fitness Qm:

C 1 , . . . C 10 , P 11 , . . . , P 20 , CR 1 , . . . , CR 20   (18)

since the number of individuals after selection is 10, the patterns of combination of elements extracted by the foregoing operation will be C 1 through C 10 .

Above steps S 1 through S 8 may be the first generation. When step S 8 has been completed, the pattern of combination of elements having the highest Qm is referred to as the best pattern of combination in that generation. In the present embodiment, C 1 may be the best pattern of combination in the first generation.

Once the first generation has been completed, the operation in steps S 2 through S 8 are repeated for the pattern group of combination of 10 elements obtained in the preceding step S 8 , until the best result of fitness Qm of the pattern of combination in respective generation becomes equal to or more than the predetermined value Q 0 (step S 9 ).

The pattern of combination of elements having the best fitness Qm obtained from the repetition of the foregoing steps may be used for the designing parameter of the multilayered optical filter.

Second Embodiment

Now another preferred embodiment of the present invention will be described below in greater details. In the second preferred embodiment, the calculation of optimization for a multilayered optical filter of 5 layers will be described, by way of example, as similar to that shown in FIG. 1 .

In the second preferred embodiment in accordance with the present invention, the optical characteristics of the selected pattern obtained in the algorithmic process steps of a given generation may be compared with the optical characteristics of the selected pattern obtained in the algorithmic process steps of an ancestor generation. If those two patterns are matched, a second initial pattern will be set, which may be different from the initial-pattern used in the preceding algorithmic process steps. By using this second initial pattern the algorithmic process steps will be repeated once again. In the following description the process steps using the second initial pattern will be primarily described and the description of the steps identical to the first preferred embodiment will be omitted.

FIG. 14 shows a flowchart of designing a multilayered optical filter having the structure of five layers in accordance with the second preferred embodiment of the present invention.

In the second preferred embodiment, random number α(=A, B, C, D, E, F) may be used in the operations of selection, extraction, deselection, cross-over, and mutation designated in steps S 21 through S 29 . The random number α may be generated by the congruence, as similar to the preceding first preferred embodiment.

In step S 21 , the initial values required for calculation (desired optical characteristics; total number of film layers; refractive index of the substrate; incident angle of the incident light; total number of patterns after the reproduction and mutation process step; amount of mutation; number of repetition of mutation step; total number of patterns after cross-over step; total number of patterns after the selection step; number of generations to be compared, amount of mutation in the second initial pattern generating step; total number of repetition of mutation step in the second initial pattern generating step; target value of the fitness) or the range of numbers are specified.

Then as similar to first preferred embodiment, the matrix Xi={Ni, Di} comprising the refractive index Ni and thickness Di as elements in the layer i of the multilayered film j may be used for generating a initial pattern Pj as shown in the following equation (6) (step of generating initial pattern group):

Pj =( X 1 , X 2 , . . . , Xi , . . . )  (6)

In the second preferred embodiment, initial values required for the calculation in step S 21 maybe those shown in Table 2 below. A pattern P 1 of combination of elements as expressed in the following expression (7) may be generated by the operation as described above:

P 1 =( X 1 , X 2 , X 3 , X 4 , X 5 )  (7)

Next, in step S 22 , the initial pattern is replicated to increase the total number of pattern group up to the number of patterns after the reproduction step (shown in FIG. 15 ).

In the present embodiment, total number of pattern groups after this operation will be P 1 through P 20 , patterns being expressed as the following equation (8):

P 1 = P 2 = P 3 = . . . = P 20   (8)

Then, mutation is performed on each of the replicated patterns P 2 through P 20 (as shown in FIG. 16 ). A random number A is generated in the range of 1≦A≦5 (=number of layers), and by using this random number one element Xi={(Ni, Di)} in the patterns is selected (step S 22 - 2 ). Then, another random number B is generated in the range 0≦B≦1 (step S 22 - 3 ), and Di will be selected if the random number B is equal to or more than 0.5 (step S 22 - 4 a ), Ni will be selected if B is less than 0.5 (step S 22 - 4 b ). Then, still another random number C is generated in the range of 0≦C≦1 (steps S 22 - 5 a , S 22 - 5 b ), and + is selected if the random number C is less than 0.5 (steps S 22 - 6 a , S 22 - 6 d ),—will be selected if C is equal to or more than 0.5 (steps S 22 - 6 b , S 22 - 6 c ). A mutation of either positive or negative (+ or −Δn, or + or −Δd) is performed on Ni or Di by the above operation. The operation from S 22 - 2 to S 22 - 6 a , 6 b , 6 c , and 6 d will be repeated a predetermined number of times Nh (step S 22 - 2 - 7 ) to generate patterns Pj having mutation applied (step S 22 - 8 ) (reproduction and mutation step: S 22 ).

In the present embodiment, assuming that random numbers are A=3, B=0.2, C=0.6 and Nh=1 with respect to the replicated pattern P 2 , then the P 2 by the operation above maybe expressed as the following equation (9) below:

P 2 =( X 1 , X 2 , X 3 ′, X 4 , X 5 )={ X 1 , X 2 , ( N 3 +Δ n, D 3 ), X 4 , X 5 }  (9)

Then, as shown by step S 23 in FIG. 14, patterns are crossed over (shown in FIG. 17 and FIG. 18 ). FIG. 17 shows the generation of patterns in one cross over step, while FIG. 18 shows a flowchart of cross over step including repetition. Here two random numbers D and E are generated in the range 1≦D, E≦20 (=total number of patterns after reproduction/selection), then based on these random numbers D and E two patterns Pj′ and Pj″ (where 1≦j′, j″≦j) are picked up from among pattern groups P 1 through P 20 generated in step S 22 or earlier (step S 23 - 1 ).

In the present embodiment, two patterns Pj′ and Pj″ may be expressed by the following equations 12 and 13, respectively (step S 23 - 2 ):

Pj ′=( X 1 ′, X 2 ′, X 3 ′, X 4 ′, X 5 ′)  (12)

Pj ″=( X 1 ″, X 2 ″, X 3 ″, X 4 ″, X 5 ″)  (13)

For each of patterns Pj′ and Pj″, another random number F may be generated in the range of 1≦F≦4 (={number of layers}−1), then based on this random number F the locations between the elements at which two patterns are to be crossed over are.determined (step S 23 - 3 ).

In the present embodiment, assuming F=3, then the locations between the elements at which two patterns Pj′ and Pj″ are to be crossed over may be expressed as shown in equation 14 below: Pj ′ = ( X1 ′ , X2 ′ , X3 ′ ,  X4 ′ , X5 ′ )   ×   Pj ″ = ( X1 ″ , X2 ″ , X3 ″ ,  X4 ″ , X5 ″ ) 

    Crossing     over     point ( 14 )

Two patterns are crossed over at that locations between elements so as to generate newly combined two patterns Ck and Ck+1 as expressed in equations 16 and 17 below (steps S 23 - 4 and S 23 - 5 ):

Ck =( X 1 ′, X 2 ′, X 3 ′, X 4 ″, X 5 ″)  (16)

Ck +1=( X 1 ″, X 2 ″, X 3 ″, X 4 ′, X 5 ′)  (17)

Then, the operation in the steps S 23 - 1 through S 23 - 5 will be repeated until the number of whole pattern groups will become equal to the predetermined total number of patterns after cross-over process step (cross-over step S 23 ).

In the present embodiment, the total number of patterns after cross-over step is 30 , and the total number of patterns before cross-over step is 20 , therefore the operation of the steps S 23 - 1 through S 23 - 5 will be repeated until C 10 will have been generated (step S 23 - 6 ).

After the cross-over process in step S 23 has been completed, in step S 24 , the optical characteristics of the patterns Pj, and Ck having passed through step S 23 are calculated so as to determine the difference in optical characteristics from the desired value. The optical characteristics refers to as the reflection characteristics of the multilayered optical filter with respect to the incident light incident to the filter at the incident angle θ, in a manner similar to first preferred embodiment.

In accordance with the method as described above, the fitness Qm for each pattern (where 1≦m≦j+k) may be determined.

Next, in step S 25 , the patterns of combination of elements are selected based on the fitness Qm for the patterns Pj, and Ck of combination of elements, determined in step S 24 with respect to the desired optical characteristics (shown in FIG. 19 ). To do this, patterns Pj, and Ck are sorted in the order of Qm from the highest to the lowest, then the sorted patterns are picked up from the highest by the number corresponding to the total number of patterns after selection. Patterns other than those picked up are deselected to reject (selection step S 25 ) In the present embodiment, if the patterns are in the order as shown in the following equation ( 19 ), by sorting on the fitness Qm:

C 1 , . . . , P 1 , . . . , P 20   (19)

since the number of individuals after selection is 10 the patterns of combination of elements extracted by the foregoing operation will be C 1 through C 10 .

Above steps S 21 through S 25 may be the first generation. When step S 25 has been completed, the pattern of combination of elements having the highest Qm is referred to as the best pattern of combination in that generation. In the present embodiment, C 1 may be the best pattern of combination in the first generation.

After the first generation has been completed, it may be determined whether or not the best fitness Qm will be equal to or more than the predetermined value, Q 0 , with respect to the pattern group of combination of 10 elements obtained in step S 25 (step S 26 ). If the best fitness value Qm of the combination pattern becomes the predetermined value Q 0 or more, then the pattern of combination of elements having the best fitness Qm may be used for the designing parameter of the multilayered optical filter.

If, on the other hand, the best fitness Qm of the combination pattern is not equal to or more than the predetermined Q 0 , then the local convergence of thus obtained combination pattern should be determined (step S 27 ). In step S 27 , the best fitness Qm of the combination pattern is determined whether or not to match with the best fitness in the combination patterns obtained from the current or preceding NN generations. Or the best fitness Qm of the combination pattern is determined whether to be the same in NN generations to the current generation.

If the best fitness Qm of the combination pattern is not matched with the best fitness in the combination pattern obtained from the NN generations preceding to the current generation, it will be determined that there is a local convergence, and the pattern group of combination of 10 elements obtained in step S 25 will be the initial pattern of step S 22 in the next generation (step S 28 ).

If matched, it will be determined that there is a local convergence, and second initial pattern will be generated (step S 29 ).

Second initial pattern may be generated by mutating the combination pattern of elements having the best fitness in the combination pattern of elements obtained between the first generation and the generation that matched (shown in FIG. 20 ). A random number A is generated in the range of 1≦A≦5 (=number of layers), and by using this random number A one element Xi={(Ni, Di)} in the patterns with the best fitness is selected (step S 29 - 2 ). Then , another random number B is generated in the range 0≦B≦1 (step S 29 - 3 ), and Di will be selected if the random number B is equal to or more than 0.5 (step S 29 - 4 a ), Ni will be selected if B is less than 0.5 (step S 29 - 4 b ). Then, still another random number C is generated in the range of 0≦C≦1 (steps S 29 - 5 a , S 29 - 5 b ), and + is selected if the random number C is less than 0.5 (steps S 29 - 6 a , S 29 - 6 d ),—will be selected if the random number C is equal to or more than 0.5 (steps S 29 - 6 b , S 29 - 6 c ). A middle-scaled mutation of either positive or negative (+ or −Δnss, or + or −Δdss) is performed on Ni or Di by the above operation. The operation from S 29 - 2 to S 29 - 6 a , 6 b , 6 c , and 6 d will be repeated a predetermined number of times Nss (step S 29 - 7 ) to generate second initial patterns PP 1 having middle-scaled mutation applied (step S 29 - 8 ). This second initial pattern PP 1 may be used as the initial pattern for step S 22 in the next generation to perform the operation in the step S 22 and later.

Now in the present embodiment, the optical characteristics of the best combination pattern of elements in a given generation are assumed to be matched with the optical characteristics of the pattern obtained in the selection step of the algorithm in the 10th generation earlier. If the combination pattern of elements having the best optical characteristics in generations earlier to the matched generation is C 1 {=(X 1 ′, X 2 ′, X 3 ′, X 4 ′, X 5 ′)}, and if in step S 29 the random number A=4, B=0.2, C=0.6, and Nh=1, then second initial pattern PP 1 may be expressed, based on the foregoing operation, as following equation (20):

PP 1 ={ X 1 ′, X 2 ′, X 3 ′, ( N 4 +Δ n, D 4 ), X 5 ′}  (20)

The operation in step S 22 through S 29 will be repeated until the best fitness Qm of the combination pattern of each of generations will become equal to or more than the predetermined value Q 0 .

The pattern of combination of elements having the best fitness Qm obtained from the repetition of the foregoing process steps may be used for the designing parameter of the multilayered optical filter.

In first preferred embodiment, steps S 27 through S 29 in the second preferred embodiment of the present invention, i.e., steps for determining the local convergence of the patterns having passed through step S 26 was not featured but these steps may be incorporated into the first preferred embodiment of the present invention.

Also in first preferred embodiment, the reproduction and mutation step and selection step are performed twice, however, the first selection step and second reproduction and mutation step may be omitted as similar to second preferred embodiment of the present invention.

Although the conventional optimization method of thickness of multilayered optical filters has small degree of freedom of design, and may obtain only multilayered structure having optical characteristics different from the desired optical characteristics, the method disclosed as first and second preferred embodiment of the present invention, in contrast, may obtain a multilayered structure more optimal to the desired optical characteristics.

Some examples carrying out the present invention will be described below in greater details with reference to the accompanying tables and drawings.

Now examples using the algorithm consisted of the steps of initial pattern generation, first reproduction and mutation, first selection, cross-over, second reproduction and mutation, and second selection (examples 1 and 2) will be described. Thereafter another example will be described, in which the optical characteristics obtained in a given algorithmic process is compared with the optical characteristics of selection pattern obtained in the earlier turns preceding to the given algorithmic process, and if matched, a second initial pattern different from the initial pattern used in the algorithmic process may be established and used to repeat the algorithmic process (example 3).

For the initial values of the multilayered structure used in the examples hereinbelow, multiple layers are alternated by higher refractive layers with lower refractive layers; thickness of each layer is set to λ/4 optical multilayered film having the mean reflection wavelength at 550 nanometers (nm). The target optical characteristics comprise the reflectance of 0.1 from 200 to 450 nm and from 650 to 1000 nm, and of 0.5 from 450 to 650 nm.

Some examples carrying out the method of operating a genetic algorithm in accordance with the preferred embodiment of the present invention will be described below in greater details with reference to the accompanying tables and drawings. In the examples the optimization will be described with respect to a multilayered film of 3 layers, comprising the 1st layer 102 , 2nd layer 103 , and 3rd layer 104 , for media substrate 101 , as shown in FIG. 29 .

The target optical characteristics in the example comprise the reflectance of 0.1 from 200 to 450 nm and from 650 to 1000 nm, and of 0.5 from 450 to 650 nm. Other characteristics will be the refractive index of substrate ns=1.5, refractive index of incident medium n 0 =1.0, and incident angle of incident lay θ=0 degree.

In the example 4, when comparing patterns, the fitness of the pattern characteristics with respect to the target characteristics will be used instead of the characteristics specific to a pattern. The parameters of the genetic algorithms when optimizing in the method of present embodiment will be given in Table 15 below:

In this example, each patterns in the pattern group may be comprised of elements randomized. More specifically, in order to form a pattern group, a random number is used for selecting the refractive index ni of each layer from the group { 1 . 5 , 1 . 6 , 1 . 7 , 1 . 8 , 1 . 9 , 2 . 0 , 2 . 1 , 2 . 2 , 2 . 3 }, while a random number is used for selecting the thickness di (nm.) of each layer from the group { 0 , 5 , 10 , . . . , 195 , 200 }.

The shift A 1 of maximum fitness in each generation with respect to the number of generations in the process of calculation for optimization is shown in FIG. 30 . In FIG. 30, the result of optimization calculation in case of genetic algorithm using the elite-roulette method in the selection step is also shown as B 1 . In the optimization in case of genetic algorithm using the delite-roulette method, the best fitness in each generation saturates at 0.28, stalls in a local resolution without achieving to the target fitness Q 0 =0.32. On the other hand, the optimization in accordance with the present invention achieved to the target fitness Q 0 =0.32 at the 5984th generation, having resolved the best solution.

FIG. 31 shows the shift of dispersion A 2 of the fitness of pattern group in each generation with respect to the number of generations the result of optimization calculation in case of genetic algorithm using the elite-roulette method in the selection step is also shown as B 2 . In the optimization in case of genetic algorithm using the elite-roulette method, the pattern group at younger generations possesses higher dispersion, however the dispersion abruptly drops when advancing the generations. Although not shown in this example, the occupation rate by the identical pattern in the pattern group was higher where the dispersion was lower.

This indicates that the pattern group was occupied by a single pattern or a pattern and its variant. Thus in such a pattern group which is occupied of a specific pattern, a pattern that has.a fitness in the proximity of the best result or that includes some effective elements may be generated but are not maintained and will be vanished.

In contrast, in accordance with the present invention, the dispersion is maintained at high level, thus the pattern group is not occupied by a specific pattern. A pattern group therefore can maintain a pattern that has its fitness in the proximity of the best solution or that includes some effective elements, and the cross-over or mutation thereafter allows the best pattern to be found without stalling on a local solution.

FIG. 32 shows the optical characteristics A 3 of a multilayered optical filter having its maximum fitness Q. For the purpose of comparison, the result B 3 of optimization when applying a genetic algorithms with the elite-roulette method, and the target optical characteristics C 3 are also shown.

As can be seen from FIG. 32, the present method in accordance with the present invention may obtain a multilayered structure having optical characteristics much closer to the target characteristics. Table 16 shows a multilayered structure of 3 layers, optimized by the present method.

Also, by applying the method disclosed herein to a structure having more than 4 layers, a multilayered optical filter having optical characteristics much closer to the desired optical characteristics may be designed and produced.

Effect of the Invention

As have been described in the foregoing discussion, the method of producing a multilayered optical filter in accordance with the present invention may provide an optical filter comprising a multilayered structure having the best Ni and Di, without falling into a local solution, in case of designing a multilayered optical filter with a desired optical characteristics.

The search of best result by using the genetic algorithms in accordance with the present invention, prevents patterns having the identical elements from coexisting in the selected patterns, so as to avoid the occupation of selected patterns thus the pattern groups, so that the search can find the best result without stalling on a local solution.

The foregoing description of the preferred embodiment of the invention has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed, and modifications and variations are possible in light of the above teachings or may be acquired from practice of the invention. The embodiment chosen and described in order to explain the principles of the invention and its practical application to enable one skilled in the art to utilize the invention in various embodiments and with various modifications as are suited to the particular use contemplated it is intended that the scope of the invention be defined by the claims appended hereto, and their equivalents.