コード例 #1
0
ファイル: TestArpack.cs プロジェクト: wo80/csparse-interop
            /// <summary>
            /// Prints singular values and vectors.
            /// </summary>
            public static void SVD(SparseMatrix A, ArpackResult result)
            {
                if (!EnsureSuccess(result))
                {
                    return;
                }

                int m     = A.RowCount;
                int n     = A.ColumnCount;
                int nconv = result.ConvergedEigenValues;

                Console.WriteLine();
                Console.WriteLine("Testing ARPACK++ SVD");
                Console.WriteLine("Compute partial SVD: A = U*S*V'");
                Console.WriteLine();

                Console.WriteLine("Dimension of the system              : " + n);
                Console.WriteLine("Number of 'requested' singular values: " + result.Count);
                Console.WriteLine("Number of 'converged' singular values: " + nconv);
                Console.WriteLine("Number of Arnoldi vectors generated  : " + result.ArnoldiCount);
                Console.WriteLine("Number of iterations taken           : " + result.IterationsTaken);
                Console.WriteLine();

                var svals = result.EigenValuesReal();
                var svecs = result.EigenVectorsReal();

                int nAx = (n > m) ? n : m;

                // Printing singular values.

                Console.WriteLine("Singular values:");

                for (int i = 0; i < nconv; i++)
                {
                    Console.WriteLine("  sigma[" + (i + 1) + "]: " + svals[i]);
                }

                Console.WriteLine();

                if (svecs != null)
                {
                    var temp = new double[m + n];

                    var v = new double[n];
                    var u = new double[m];
                    var r = new double[nconv]; // residuals
                    var s = new double[nconv]; // residuals (transposed system)

                    for (int i = 0; i < nconv; i++)
                    {
                        var sigma = svals[i];

                        svecs.Column(i, temp);

                        // Compute A*v - sigma*u
                        Array.Copy(temp, m, v, 0, n);
                        Array.Copy(temp, 0, u, 0, m);

                        A.Multiply(1.0, v, -sigma, u);

                        r[i] = Vector.Norm(u) / Math.Abs(sigma);

                        // Compute A'*u - sigma*v
                        Array.Copy(temp, m, v, 0, n);
                        Array.Copy(temp, 0, u, 0, m);

                        A.TransposeMultiply(1.0, u, -sigma, v);

                        s[i] = Vector.Norm(v) / Math.Abs(sigma);
                    }

                    // Printing the residual norm || A*v - sigma*u ||
                    // for the nconv accurately computed vectors u and v.

                    for (int i = 0; i < nconv; i++)
                    {
                        Console.WriteLine("||A*v(" + (i + 1) + ") - sigma(" + (i + 1) + ")*u(" + (i + 1) + ")||: " + r[i]);
                    }

                    Console.WriteLine();

                    // Printing the residual norm || A'*u - sigma*v ||
                    // for the nconv accurately computed vectors u and v.

                    for (int i = 0; i < nconv; i++)
                    {
                        Console.WriteLine("||A'*u(" + (i + 1) + ") - sigma(" + (i + 1) + ")*v(" + (i + 1) + ")||: " + s[i]);
                    }

                    Console.WriteLine();
                }
            }
コード例 #2
0
ファイル: TestArpack.cs プロジェクト: wo80/csparse-interop
            /// <summary>
            /// Prints eigenvalues and eigenvectors of symmetric generalized eigen-problems.
            /// </summary>
            public static void Symmetric(SparseMatrix A, SparseMatrix B, ArpackResult result, ShiftMode mode)
            {
                if (!EnsureSuccess(result))
                {
                    return;
                }

                int n     = A.RowCount;
                int nconv = result.ConvergedEigenValues;

                Console.WriteLine();
                Console.WriteLine("Testing ARPACK++ class ARluSymGenEig");
                Console.WriteLine("Real symmetric generalized eigenvalue problem: A*x - lambda*B*x");
                Console.WriteLine();

                switch (mode)
                {
                case ShiftMode.None:
                    Console.WriteLine("Regular mode");
                    break;

                case ShiftMode.Regular:
                    Console.WriteLine("Shift and invert mode");
                    break;

                case ShiftMode.Buckling:
                    Console.WriteLine("Buckling mode");
                    break;

                case ShiftMode.Cayley:
                    Console.WriteLine("Cayley mode");
                    break;
                }

                Console.WriteLine();
                Console.WriteLine("Dimension of the system            : " + n);
                Console.WriteLine("Number of 'requested' eigenvalues  : " + result.Count);
                Console.WriteLine("Number of 'converged' eigenvalues  : " + nconv);
                Console.WriteLine("Number of Arnoldi vectors generated: " + result.ArnoldiCount);
                Console.WriteLine("Number of iterations taken         : " + result.IterationsTaken);
                Console.WriteLine();

                var evals = result.EigenValuesReal();
                var evecs = result.EigenVectorsReal();

                // Printing eigenvalues.

                Console.WriteLine("Eigenvalues:");

                for (int i = 0; i < nconv; i++)
                {
                    Console.WriteLine("  lambda[" + (i + 1) + "]: " + evals[i]);
                }

                Console.WriteLine();

                if (evecs != null)
                {
                    Symmetrize(ref A);
                    Symmetrize(ref B);

                    // Printing the residual norm || A*x - lambda*B*x ||
                    // for the nconv accurately computed eigenvectors.

                    var x = new double[n];
                    var y = new double[n];
                    var r = new double[nconv]; // residuals

                    for (int i = 0; i < nconv; i++)
                    {
                        var lambda = evals[i];

                        evecs.Column(i, x);

                        Vector.Copy(x, y);

                        // y = B*x
                        B.Multiply(x, y);

                        // y = A*x - lambda*B*x
                        A.Multiply(1.0, x, -lambda, y);

                        r[i] = Vector.Norm(y) / Math.Abs(lambda);
                    }

                    for (int i = 0; i < nconv; i++)
                    {
                        Console.WriteLine("||A*x(" + i + ") - lambda(" + i + ")*B*x(" + i + ")||: " + r[i]);
                    }

                    Console.WriteLine();
                }
            }