Exemplo n.º 1
0
        /// <summary>
        /// Solve the generalized eigenvalue problem in user-defined shift-invert mode.
        /// </summary>
        public IEigenSolverResult SolveGeneralized(int k, double sigma, ShiftMode mode, Spectrum job = Spectrum.LargestMagnitude)
        {
            if (!symmetric && !(mode == ShiftMode.None || mode == ShiftMode.Regular))
            {
                throw new InvalidOperationException("This mode is only available for symmetric eigenvalue problems.");
            }

            if (!CheckSquare(A))
            {
                throw new InvalidOperationException("Cannot solve eigenvalue problem with non-square matrix.");
            }

            if (!Job.Validate(symmetric, job))
            {
                throw new ArgumentException("Invalid job for symmetric eigenvalue problem.", "job");
            }

            var result = new ArpackResult(k, size, ComputeEigenVectors, symmetric);

            var handles = new List <GCHandle>();

            var a = GetMatrix(A, handles);
            var b = GetMatrix(B, handles);
            var e = result.GetEigenvalueStorage(handles);

            int conv = 0;

            if (symmetric)
            {
                char m = 'S';

                if (mode == ShiftMode.Buckling)
                {
                    m = 'B';
                }
                else if (mode == ShiftMode.Cayley)
                {
                    m = 'C';
                }

                conv = NativeMethods.ar_di_sg_shift(GetJob(job), m, k, ArnoldiCount, Iterations,
                                                    Tolerance, sigma, ref a, ref b, ref e);
            }
            else
            {
                conv = NativeMethods.ar_di_ng_shift(GetJob(job), k, ArnoldiCount, Iterations,
                                                    Tolerance, sigma, ref a, ref b, ref e);
            }

            result.IterationsTaken      = e.iterations;
            result.ArnoldiCount         = e.ncv;
            result.ConvergedEigenValues = conv;
            result.ErrorCode            = e.info;

            InteropHelper.Free(handles);

            return(result);
        }
Exemplo n.º 2
0
        /// <summary>
        /// Special case solving the standard real generalized eigenvalue problem with complex shift.
        /// </summary>
        /// <param name="k">The number of eigenvalues to compute.</param>
        /// <param name="sigma_r">The real part of the complex shift.</param>
        /// <param name="sigma_i">The imaginary part of the complex shift.</param>
        /// <param name="part">Part to apply ('R' for real, 'I' for imaginary).</param>
        /// <param name="job">The part of the spectrum to compute.</param>
        /// <returns>The number of converged eigenvalues.</returns>
        public IEigenSolverResult SolveGeneralized(int k, double sigma_r, double sigma_i, char part, Spectrum job = Spectrum.LargestMagnitude)
        {
            if (symmetric)
            {
                throw new InvalidOperationException("Complex shift doesn't apply to real symmetric eigenvalue problems.");
            }

            if (!CheckSquare(A))
            {
                throw new InvalidOperationException("Cannot solve eigenvalue problem with non-square matrix.");
            }

            if (!Job.ValidateGeneral(job))
            {
                throw new ArgumentException("Invalid job for non-symmetric eigenvalue problem.", "job");
            }

            part = char.ToUpperInvariant(part);

            if (part != 'R' && part != 'I')
            {
                throw new ArgumentException("Invalid part specified for complex shift.", "part");
            }

            var result = new ArpackResult(k, size, ComputeEigenVectors, symmetric);

            var handles = new List <GCHandle>();

            var a = GetMatrix(A, handles);
            var b = GetMatrix(B, handles);
            var e = result.GetEigenvalueStorage(handles);

            int conv = 0;

            conv = NativeMethods.ar_di_ng_shift_cx(GetJob(job),
                                                   k, ArnoldiCount, Iterations, Tolerance,
                                                   part, sigma_r, sigma_i, ref a, ref b, ref e);


            result.IterationsTaken      = e.iterations;
            result.ArnoldiCount         = e.ncv;
            result.ConvergedEigenValues = conv;
            result.ErrorCode            = e.info;

            InteropHelper.Free(handles);

            return(result);
        }
Exemplo n.º 3
0
        /// <summary>
        /// Compute singular values and the partial singular value decomposition.
        /// </summary>
        /// <param name="k">The number of singular values to compute.</param>
        /// <param name="normal">Use normal equation to compute (squared) singular values.</param>
        /// <param name="job">The part of the spectrum to compute.</param>
        /// <returns>The number of converged singular values.</returns>
        /// <remarks>
        /// If <paramref name="normal"/> is true, the normal equation <c>(A'*A)*v = sigma*v</c>
        /// is considered, where A is an m-by-n real matrix. This formulation is appropriate
        /// when m >= n. The roles of A and A' must be reversed in the case that m &lt; n.
        ///
        /// The eigenvalues returned are the squared singular values of A. If requested, the
        /// returned eigenvectors correspond to the right singular vectors, if <c>A = U*S*V'</c>.
        /// The left singular vectors can be computed from the equation <c>A*v - sigma*u = 0</c>.
        ///
        /// If <paramref name="normal"/> is false, the symmetric system <c>[0  A; A' 0]</c> is
        /// considered (size m + n), where A is an m-by-n real matrix.
        ///
        /// This problem can be used to obtain the decomposition <c>A = U*S*V'</c>. The positive
        /// eigenvalues of this problem are the singular values of A (the eigenvalues come in
        /// pairs, the negative eigenvalues have the same magnitude of the positive ones and
        /// can be discarded). The columns of U can be extracted from the first m components
        /// of the eigenvectors y, while the columns of V can be extracted from the remaining
        /// n components.
        /// </remarks>
        public IEigenSolverResult SingularValues(int k, bool normal, Spectrum job = Spectrum.LargestMagnitude)
        {
            if (!Job.ValidateSymmetric(job))
            {
                throw new ArgumentException("Invalid job for singular value decomposition.", "job");
            }

            int m = A.RowCount;
            int n = A.ColumnCount;

            if (normal && m < n)
            {
                throw new InvalidOperationException("Number of columns must not be smaller than number of rows (use transposed matrix).");
            }

            int size = normal ? n : m + n;

            var result = new ArpackResult(k, size, ComputeEigenVectors, true);

            var handles = new List <GCHandle>();

            var a = GetMatrix(A, handles);
            var e = result.GetEigenvalueStorage(handles);

            int conv = 0;

            if (normal)
            {
                conv = NativeMethods.ar_di_svd_nrm(GetJob(job),
                                                   k, ArnoldiCount, Iterations, Tolerance, ref a, ref e);
            }
            else
            {
                conv = NativeMethods.ar_di_svd(GetJob(job),
                                               k, ArnoldiCount, Iterations, Tolerance, ref a, ref e);
            }

            result.IterationsTaken      = e.iterations;
            result.ArnoldiCount         = e.ncv;
            result.ConvergedEigenValues = conv;
            result.ErrorCode            = e.info;

            InteropHelper.Free(handles);

            return(result);
        }
Exemplo n.º 4
0
        /// <summary>
        /// Solve the generalized eigenvalue problem.
        /// </summary>
        public override IEigenSolverResult SolveGeneralized(int k, Spectrum job)
        {
            if (!CheckSquare(A))
            {
                throw new InvalidOperationException("Cannot solve eigenvalue problem with non-square matrix.");
            }

            if (!Job.Validate(symmetric, job))
            {
                throw new ArgumentException("Invalid job for given eigenvalue problem.", "job");
            }

            var result = new ArpackResult(k, size, ComputeEigenVectors, symmetric);

            var handles = new List <GCHandle>();

            var a = GetMatrix(A, handles);
            var b = GetMatrix(B, handles);
            var e = result.GetEigenvalueStorage(handles);

            int conv = 0;

            if (symmetric)
            {
                conv = NativeMethods.ar_di_sg(GetJob(job), k, ArnoldiCount,
                                              Iterations, Tolerance, ref a, ref b, ref e);
            }
            else
            {
                conv = NativeMethods.ar_di_ng(GetJob(job), k, ArnoldiCount,
                                              Iterations, Tolerance, ref a, ref b, ref e);
            }

            result.IterationsTaken      = e.iterations;
            result.ArnoldiCount         = e.ncv;
            result.ConvergedEigenValues = conv;
            result.ErrorCode            = e.info;

            InteropHelper.Free(handles);

            return(result);
        }