|  | 
|  | 1 | +package com.thealgorithms.maths; | 
|  | 2 | + | 
|  | 3 | +/** | 
|  | 4 | + * In numerical analysis, Chebyshev iteration is an iterative method for solving | 
|  | 5 | + * systems of linear equations Ax = b. It is designed for systems where the | 
|  | 6 | + * matrix A is symmetric positive-definite (SPD). | 
|  | 7 | + * | 
|  | 8 | + * <p> | 
|  | 9 | + * This method is a "polynomial acceleration" method, meaning it finds the | 
|  | 10 | + * optimal polynomial to apply to the residual to accelerate convergence. | 
|  | 11 | + * | 
|  | 12 | + * <p> | 
|  | 13 | + * It requires knowledge of the bounds of the eigenvalues of the matrix A: | 
|  | 14 | + * m(A) (smallest eigenvalue) and M(A) (largest eigenvalue). | 
|  | 15 | + * | 
|  | 16 | + * <p> | 
|  | 17 | + * Wikipedia: https://en.wikipedia.org/wiki/Chebyshev_iteration | 
|  | 18 | + * | 
|  | 19 | + * @author Mitrajit Ghorui(KeyKyrios) | 
|  | 20 | + */ | 
|  | 21 | +public final class ChebyshevIteration { | 
|  | 22 | + | 
|  | 23 | +    private ChebyshevIteration() { | 
|  | 24 | +    } | 
|  | 25 | + | 
|  | 26 | +    /** | 
|  | 27 | +     * Solves the linear system Ax = b using the Chebyshev iteration method. | 
|  | 28 | +     * | 
|  | 29 | +     * <p> | 
|  | 30 | +     * NOTE: The matrix A *must* be symmetric positive-definite (SPD) for this | 
|  | 31 | +     * algorithm to converge. | 
|  | 32 | +     * | 
|  | 33 | +     * @param a The matrix A (must be square, SPD). | 
|  | 34 | +     * @param b The vector b. | 
|  | 35 | +     * @param x0 The initial guess vector. | 
|  | 36 | +     * @param minEigenvalue The smallest eigenvalue of A (m(A)). | 
|  | 37 | +     * @param maxEigenvalue The largest eigenvalue of A (M(A)). | 
|  | 38 | +     * @param maxIterations The maximum number of iterations to perform. | 
|  | 39 | +     * @param tolerance The desired tolerance for the residual norm. | 
|  | 40 | +     * @return The solution vector x. | 
|  | 41 | +     * @throws IllegalArgumentException if matrix/vector dimensions are | 
|  | 42 | +     * incompatible, | 
|  | 43 | +     * if maxIterations <= 0, or if eigenvalues are invalid (e.g., minEigenvalue | 
|  | 44 | +     * <= 0, maxEigenvalue <= minEigenvalue). | 
|  | 45 | +     */ | 
|  | 46 | +    public static double[] solve(double[][] a, double[] b, double[] x0, double minEigenvalue, double maxEigenvalue, int maxIterations, double tolerance) { | 
|  | 47 | +        validateInputs(a, b, x0, minEigenvalue, maxEigenvalue, maxIterations, tolerance); | 
|  | 48 | + | 
|  | 49 | +        int n = b.length; | 
|  | 50 | +        double[] x = x0.clone(); | 
|  | 51 | +        double[] r = vectorSubtract(b, matrixVectorMultiply(a, x)); | 
|  | 52 | +        double[] p = new double[n]; | 
|  | 53 | + | 
|  | 54 | +        double d = (maxEigenvalue + minEigenvalue) / 2.0; | 
|  | 55 | +        double c = (maxEigenvalue - minEigenvalue) / 2.0; | 
|  | 56 | + | 
|  | 57 | +        double alpha = 0.0; | 
|  | 58 | +        double alphaPrev = 0.0; | 
|  | 59 | + | 
|  | 60 | +        for (int k = 0; k < maxIterations; k++) { | 
|  | 61 | +            double residualNorm = vectorNorm(r); | 
|  | 62 | +            if (residualNorm < tolerance) { | 
|  | 63 | +                return x; // Solution converged | 
|  | 64 | +            } | 
|  | 65 | + | 
|  | 66 | +            if (k == 0) { | 
|  | 67 | +                alpha = 1.0 / d; | 
|  | 68 | +                System.arraycopy(r, 0, p, 0, n); // p = r | 
|  | 69 | +            } else { | 
|  | 70 | +                double beta = c * alphaPrev / 2.0 * (c * alphaPrev / 2.0); | 
|  | 71 | +                alpha = 1.0 / (d - beta / alphaPrev); | 
|  | 72 | +                double[] pUpdate = scalarMultiply(beta / alphaPrev, p); | 
|  | 73 | +                p = vectorAdd(r, pUpdate); // p = r + (beta / alphaPrev) * p | 
|  | 74 | +            } | 
|  | 75 | + | 
|  | 76 | +            double[] xUpdate = scalarMultiply(alpha, p); | 
|  | 77 | +            x = vectorAdd(x, xUpdate); // x = x + alpha * p | 
|  | 78 | + | 
|  | 79 | +            // Recompute residual for accuracy | 
|  | 80 | +            r = vectorSubtract(b, matrixVectorMultiply(a, x)); | 
|  | 81 | +            alphaPrev = alpha; | 
|  | 82 | +        } | 
|  | 83 | + | 
|  | 84 | +        return x; // Return best guess after maxIterations | 
|  | 85 | +    } | 
|  | 86 | + | 
|  | 87 | +    /** | 
|  | 88 | +     * Validates the inputs for the Chebyshev solver. | 
|  | 89 | +     */ | 
|  | 90 | +    private static void validateInputs(double[][] a, double[] b, double[] x0, double minEigenvalue, double maxEigenvalue, int maxIterations, double tolerance) { | 
|  | 91 | +        int n = a.length; | 
|  | 92 | +        if (n == 0) { | 
|  | 93 | +            throw new IllegalArgumentException("Matrix A cannot be empty."); | 
|  | 94 | +        } | 
|  | 95 | +        if (n != a[0].length) { | 
|  | 96 | +            throw new IllegalArgumentException("Matrix A must be square."); | 
|  | 97 | +        } | 
|  | 98 | +        if (n != b.length) { | 
|  | 99 | +            throw new IllegalArgumentException("Matrix A and vector b dimensions do not match."); | 
|  | 100 | +        } | 
|  | 101 | +        if (n != x0.length) { | 
|  | 102 | +            throw new IllegalArgumentException("Matrix A and vector x0 dimensions do not match."); | 
|  | 103 | +        } | 
|  | 104 | +        if (minEigenvalue <= 0) { | 
|  | 105 | +            throw new IllegalArgumentException("Smallest eigenvalue must be positive (matrix must be positive-definite)."); | 
|  | 106 | +        } | 
|  | 107 | +        if (maxEigenvalue <= minEigenvalue) { | 
|  | 108 | +            throw new IllegalArgumentException("Max eigenvalue must be strictly greater than min eigenvalue."); | 
|  | 109 | +        } | 
|  | 110 | +        if (maxIterations <= 0) { | 
|  | 111 | +            throw new IllegalArgumentException("Max iterations must be positive."); | 
|  | 112 | +        } | 
|  | 113 | +        if (tolerance <= 0) { | 
|  | 114 | +            throw new IllegalArgumentException("Tolerance must be positive."); | 
|  | 115 | +        } | 
|  | 116 | +    } | 
|  | 117 | + | 
|  | 118 | +    // --- Vector/Matrix Helper Methods --- | 
|  | 119 | +    /** | 
|  | 120 | +     * Computes the product of a matrix A and a vector v (Av). | 
|  | 121 | +     */ | 
|  | 122 | +    private static double[] matrixVectorMultiply(double[][] a, double[] v) { | 
|  | 123 | +        int n = a.length; | 
|  | 124 | +        double[] result = new double[n]; | 
|  | 125 | +        for (int i = 0; i < n; i++) { | 
|  | 126 | +            double sum = 0; | 
|  | 127 | +            for (int j = 0; j < n; j++) { | 
|  | 128 | +                sum += a[i][j] * v[j]; | 
|  | 129 | +            } | 
|  | 130 | +            result[i] = sum; | 
|  | 131 | +        } | 
|  | 132 | +        return result; | 
|  | 133 | +    } | 
|  | 134 | + | 
|  | 135 | +    /** | 
|  | 136 | +     * Computes the subtraction of two vectors (v1 - v2). | 
|  | 137 | +     */ | 
|  | 138 | +    private static double[] vectorSubtract(double[] v1, double[] v2) { | 
|  | 139 | +        int n = v1.length; | 
|  | 140 | +        double[] result = new double[n]; | 
|  | 141 | +        for (int i = 0; i < n; i++) { | 
|  | 142 | +            result[i] = v1[i] - v2[i]; | 
|  | 143 | +        } | 
|  | 144 | +        return result; | 
|  | 145 | +    } | 
|  | 146 | + | 
|  | 147 | +    /** | 
|  | 148 | +     * Computes the addition of two vectors (v1 + v2). | 
|  | 149 | +     */ | 
|  | 150 | +    private static double[] vectorAdd(double[] v1, double[] v2) { | 
|  | 151 | +        int n = v1.length; | 
|  | 152 | +        double[] result = new double[n]; | 
|  | 153 | +        for (int i = 0; i < n; i++) { | 
|  | 154 | +            result[i] = v1[i] + v2[i]; | 
|  | 155 | +        } | 
|  | 156 | +        return result; | 
|  | 157 | +    } | 
|  | 158 | + | 
|  | 159 | +    /** | 
|  | 160 | +     * Computes the product of a scalar and a vector (s * v). | 
|  | 161 | +     */ | 
|  | 162 | +    private static double[] scalarMultiply(double scalar, double[] v) { | 
|  | 163 | +        int n = v.length; | 
|  | 164 | +        double[] result = new double[n]; | 
|  | 165 | +        for (int i = 0; i < n; i++) { | 
|  | 166 | +            result[i] = scalar * v[i]; | 
|  | 167 | +        } | 
|  | 168 | +        return result; | 
|  | 169 | +    } | 
|  | 170 | + | 
|  | 171 | +    /** | 
|  | 172 | +     * Computes the L2 norm (Euclidean norm) of a vector. | 
|  | 173 | +     */ | 
|  | 174 | +    private static double vectorNorm(double[] v) { | 
|  | 175 | +        double sumOfSquares = 0; | 
|  | 176 | +        for (double val : v) { | 
|  | 177 | +            sumOfSquares += val * val; | 
|  | 178 | +        } | 
|  | 179 | +        return Math.sqrt(sumOfSquares); | 
|  | 180 | +    } | 
|  | 181 | +} | 
0 commit comments