c_lanczos_tridiagonalization.cpp

Go to the documentation of this file.

/*
 *  SPDX-FileCopyrightText: Copyright 2021, Siavash Ameli <sameli@berkeley.edu>
 *  SPDX-License-Identifier: BSD-3-Clause
 *  SPDX-FileType: SOURCE
 *
 *  This program is free software: you can redistribute it and/or modify it
 *  under the terms of the license found in the LICENSE.txt file in the root
 *  directory of this source tree.
 */
 
 
// =======
// Headers
// =======
 
#include "./c_lanczos_tridiagonalization.h"
#include <cmath>  // sqrt
#include "./c_orthogonalization.h"  // cOrthogonalization
#include "../_c_basic_algebra/c_vector_operations.h"  // cVectorOperations
 
 
// ============================
// c lanczos tridiagonalization
// ============================
 
 
template <typename DataType>
IndexType c_lanczos_tridiagonalization(
        cLinearOperator<DataType>* A,
        const DataType* v,
        const LongIndexType n,
        const IndexType m,
        const DataType lanczos_tol,
        const FlagType orthogonalize,
        DataType* alpha,
        DataType* beta)
{
    // buffer_size is number of last orthogonal vectors to keep in the buffer V
    IndexType buffer_size;
    if (orthogonalize == 0 || orthogonalize == 1)
    {
        // At least two vectors must be stored in buffer for Lanczos recursion
        buffer_size = 2;
    }
    else if ((orthogonalize < 0) ||
             (orthogonalize > static_cast<FlagType>(m)))
    {
        // Using full reorthogonalization, keep all of the m vectors in buffer
        buffer_size = m;
    }
    else
    {
        // Orthogonalize with less than m vectors (0 < orthogonalize < m)
        buffer_size = orthogonalize;
    }
 
    // Allocate 2D array (as 1D array, and coalesced row-wise) to store
    // the last buffer_size of orthogonalized vectors of length n. New vectors
    // are stored by cycling through the buffer to replace with old ones.
    DataType* V = new DataType[n * buffer_size];
 
    // Allocate vector r
    DataType* r = new DataType[n];
 
    // Copy v into r
    cVectorOperations<DataType>::copy_vector(v, n, r);
 
    // Initial beta
    DataType initial_beta = cVectorOperations<DataType>::euclidean_norm(r, n);
 
    // Declare iterators
    IndexType j;
    IndexType lanczos_size = 0;
    IndexType num_ortho;
 
    // In the following, beta[j] means beta[j-1] in the Demmel text
    for (j=0; j < m; ++j)
    {
        // Update the size of Lanczos tridiagonal matrix
        ++lanczos_size;
 
        // Normalize r and copy to the j-th column of V
        if (j == 0)
        {
            cVectorOperations<DataType>::copy_scaled_vector(
                    r, n, 1.0/initial_beta, &V[(j % buffer_size)*n]);
        }
        else
        {
            cVectorOperations<DataType>::copy_scaled_vector(
                    r, n, 1.0/beta[j-1], &V[(j % buffer_size)*n]);
        }
 
        // Multiply A to the j-th column of V, write into r
        A->dot(&V[(j % buffer_size)*n], r);
 
        // alpha[j] is V[:, j] dot r
        alpha[j] = cVectorOperations<DataType>::inner_product(
                &V[(j % buffer_size)*n], r, n);
 
        // Subtract V[:,j] * alpha[j] from r
        cVectorOperations<DataType>::subtract_scaled_vector(
                &V[(j % buffer_size)*n], n, alpha[j], r);
 
        // Subtract V[:,j-1] * beta[j] from r
        if (j > 0)
        {
            cVectorOperations<DataType>::subtract_scaled_vector(
                    &V[((j-1) % buffer_size)*n], n, beta[j-1], r);
        }
 
        // Gram-Schmidt process (full re-orthogonalization)
        if (orthogonalize != 0)
        {
            // Find how many column vectors are filled so far in the buffer V
            if (j < buffer_size)
            {
                num_ortho = j+1;
            }
            else
            {
                num_ortho = buffer_size;
            }
 
            // Gram-Schmidt process
            cOrthogonalization<DataType>::gram_schmidt_process(
                    &V[0], n, buffer_size, j%buffer_size, num_ortho, r);
        }
 
        // beta is norm of r
        beta[j] = cVectorOperations<DataType>::euclidean_norm(r, n);
 
        // Exit criterion when the vector r is zero. If each component of a
        // zero vector has the tolerance epsilon, (which is called lanczos_tol
        // here), the tolerance of norm of r is epsilon times sqrt of n.
        if (beta[j] < lanczos_tol * sqrt(n))
        {
            break;
        }
    }
 
    // Free dynamic memory
    delete[] V;
    delete[] r;
 
    return lanczos_size;
}
 
 
// ===============================
// Explicit template instantiation
// ===============================
 
// lanczos tridiagonalization
template IndexType c_lanczos_tridiagonalization<float>(
        cLinearOperator<float>* A,
        const float* v,
        const LongIndexType n,
        const IndexType m,
        const float lanczos_tol,
        const FlagType orthogonalize,
        float* alpha,
        float* beta);
 
template IndexType c_lanczos_tridiagonalization<double>(
        cLinearOperator<double>* A,
        const double* v,
        const LongIndexType n,
        const IndexType m,
        const double lanczos_tol,
        const FlagType orthogonalize,
        double* alpha,
        double* beta);
 
template IndexType c_lanczos_tridiagonalization<long double>(
        cLinearOperator<long double>* A,
        const long double* v,
        const LongIndexType n,
        const IndexType m,
        const long double lanczos_tol,
        const FlagType orthogonalize,
        long double* alpha,
        long double* beta);