# include # include # include # include # include # include # include "qr_solve.h" # include "r8lib.h" /******************************************************************************/ void daxpy ( int n, double da, double dx[], int incx, double dy[], int incy ) /******************************************************************************/ /* Purpose: DAXPY computes constant times a vector plus a vector. Discussion: This routine uses unrolled loops for increments equal to one. Licensing: This code is distributed under the MIT license. Modified: 30 March 2007 Author: C version by John Burkardt Reference: Jack Dongarra, Cleve Moler, Jim Bunch, Pete Stewart, LINPACK User's Guide, SIAM, 1979. Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh, Basic Linear Algebra Subprograms for Fortran Usage, Algorithm 539, ACM Transactions on Mathematical Software, Volume 5, Number 3, September 1979, pages 308-323. Parameters: Input, int N, the number of elements in DX and DY. Input, double DA, the multiplier of DX. Input, double DX[*], the first vector. Input, int INCX, the increment between successive entries of DX. Input/output, double DY[*], the second vector. On output, DY[*] has been replaced by DY[*] + DA * DX[*]. Input, int INCY, the increment between successive entries of DY. */ { int i; int ix; int iy; int m; if ( n <= 0 ) { return; } if ( da == 0.0 ) { return; } /* Code for unequal increments or equal increments not equal to 1. */ if ( incx != 1 || incy != 1 ) { if ( 0 <= incx ) { ix = 0; } else { ix = ( - n + 1 ) * incx; } if ( 0 <= incy ) { iy = 0; } else { iy = ( - n + 1 ) * incy; } for ( i = 0; i < n; i++ ) { dy[iy] = dy[iy] + da * dx[ix]; ix = ix + incx; iy = iy + incy; } } /* Code for both increments equal to 1. */ else { m = n % 4; for ( i = 0; i < m; i++ ) { dy[i] = dy[i] + da * dx[i]; } for ( i = m; i < n; i = i + 4 ) { dy[i ] = dy[i ] + da * dx[i ]; dy[i+1] = dy[i+1] + da * dx[i+1]; dy[i+2] = dy[i+2] + da * dx[i+2]; dy[i+3] = dy[i+3] + da * dx[i+3]; } } return; } /******************************************************************************/ double ddot ( int n, double dx[], int incx, double dy[], int incy ) /******************************************************************************/ /* Purpose: DDOT forms the dot product of two vectors. Discussion: This routine uses unrolled loops for increments equal to one. Licensing: This code is distributed under the MIT license. Modified: 30 March 2007 Author: C version by John Burkardt Reference: Jack Dongarra, Cleve Moler, Jim Bunch, Pete Stewart, LINPACK User's Guide, SIAM, 1979. Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh, Basic Linear Algebra Subprograms for Fortran Usage, Algorithm 539, ACM Transactions on Mathematical Software, Volume 5, Number 3, September 1979, pages 308-323. Parameters: Input, int N, the number of entries in the vectors. Input, double DX[*], the first vector. Input, int INCX, the increment between successive entries in DX. Input, double DY[*], the second vector. Input, int INCY, the increment between successive entries in DY. Output, double DDOT, the sum of the product of the corresponding entries of DX and DY. */ { double dtemp; int i; int ix; int iy; int m; dtemp = 0.0; if ( n <= 0 ) { return dtemp; } /* Code for unequal increments or equal increments not equal to 1. */ if ( incx != 1 || incy != 1 ) { if ( 0 <= incx ) { ix = 0; } else { ix = ( - n + 1 ) * incx; } if ( 0 <= incy ) { iy = 0; } else { iy = ( - n + 1 ) * incy; } for ( i = 0; i < n; i++ ) { dtemp = dtemp + dx[ix] * dy[iy]; ix = ix + incx; iy = iy + incy; } } /* Code for both increments equal to 1. */ else { m = n % 5; for ( i = 0; i < m; i++ ) { dtemp = dtemp + dx[i] * dy[i]; } for ( i = m; i < n; i = i + 5 ) { dtemp = dtemp + dx[i ] * dy[i ] + dx[i+1] * dy[i+1] + dx[i+2] * dy[i+2] + dx[i+3] * dy[i+3] + dx[i+4] * dy[i+4]; } } return dtemp; } /******************************************************************************/ double dnrm2 ( int n, double x[], int incx ) /******************************************************************************/ /* Purpose: DNRM2 returns the euclidean norm of a vector. Discussion: DNRM2 ( X ) = sqrt ( X' * X ) Licensing: This code is distributed under the MIT license. Modified: 30 March 2007 Author: C version by John Burkardt Reference: Jack Dongarra, Cleve Moler, Jim Bunch, Pete Stewart, LINPACK User's Guide, SIAM, 1979. Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh, Basic Linear Algebra Subprograms for Fortran Usage, Algorithm 539, ACM Transactions on Mathematical Software, Volume 5, Number 3, September 1979, pages 308-323. Parameters: Input, int N, the number of entries in the vector. Input, double X[*], the vector whose norm is to be computed. Input, int INCX, the increment between successive entries of X. Output, double DNRM2, the Euclidean norm of X. */ { double absxi; int i; int ix; double norm; double scale; double ssq; if ( n < 1 || incx < 1 ) { norm = 0.0; } else if ( n == 1 ) { norm = fabs ( x[0] ); } else { scale = 0.0; ssq = 1.0; ix = 0; for ( i = 0; i < n; i++ ) { if ( x[ix] != 0.0 ) { absxi = fabs ( x[ix] ); if ( scale < absxi ) { ssq = 1.0 + ssq * ( scale / absxi ) * ( scale / absxi ); scale = absxi; } else { ssq = ssq + ( absxi / scale ) * ( absxi / scale ); } } ix = ix + incx; } norm = scale * sqrt ( ssq ); } return norm; } /******************************************************************************/ void dqrank ( double a[], int lda, int m, int n, double tol, int *kr, int jpvt[], double qraux[] ) /******************************************************************************/ /* Purpose: DQRANK computes the QR factorization of a rectangular matrix. Discussion: This routine is used in conjunction with DQRLSS to solve overdetermined, underdetermined and singular linear systems in a least squares sense. DQRANK uses the LINPACK subroutine DQRDC to compute the QR factorization, with column pivoting, of an M by N matrix A. The numerical rank is determined using the tolerance TOL. Note that on output, ABS ( A(1,1) ) / ABS ( A(KR,KR) ) is an estimate of the condition number of the matrix of independent columns, and of R. This estimate will be <= 1/TOL. Licensing: This code is distributed under the MIT license. Modified: 21 April 2012 Author: C version by John Burkardt. Reference: Jack Dongarra, Cleve Moler, Jim Bunch, Pete Stewart, LINPACK User's Guide, SIAM, 1979, ISBN13: 978-0-898711-72-1, LC: QA214.L56. Parameters: Input/output, double A[LDA*N]. On input, the matrix whose decomposition is to be computed. On output, the information from DQRDC. The triangular matrix R of the QR factorization is contained in the upper triangle and information needed to recover the orthogonal matrix Q is stored below the diagonal in A and in the vector QRAUX. Input, int LDA, the leading dimension of A, which must be at least M. Input, int M, the number of rows of A. Input, int N, the number of columns of A. Input, double TOL, a relative tolerance used to determine the numerical rank. The problem should be scaled so that all the elements of A have roughly the same absolute accuracy, EPS. Then a reasonable value for TOL is roughly EPS divided by the magnitude of the largest element. Output, int *KR, the numerical rank. Output, int JPVT[N], the pivot information from DQRDC. Columns JPVT(1), ..., JPVT(KR) of the original matrix are linearly independent to within the tolerance TOL and the remaining columns are linearly dependent. Output, double QRAUX[N], will contain extra information defining the QR factorization. */ { int i; int j; int job; int k; double *work; for ( i = 0; i < n; i++ ) { jpvt[i] = 0; } work = ( double * ) malloc ( n * sizeof ( double ) ); job = 1; dqrdc ( a, lda, m, n, qraux, jpvt, work, job ); *kr = 0; k = i4_min ( m, n ); for ( j = 0; j < k; j++ ) { if ( fabs ( a[j+j*lda] ) <= tol * fabs ( a[0+0*lda] ) ) { return; } *kr = j + 1; } free ( work ); return; } /******************************************************************************/ void dqrdc ( double a[], int lda, int n, int p, double qraux[], int jpvt[], double work[], int job ) /******************************************************************************/ /* Purpose: DQRDC computes the QR factorization of a real rectangular matrix. Discussion: DQRDC uses Householder transformations. Column pivoting based on the 2-norms of the reduced columns may be performed at the user's option. Licensing: This code is distributed under the MIT license. Modified: 07 June 2005 Author: C version by John Burkardt. Reference: Jack Dongarra, Cleve Moler, Jim Bunch and Pete Stewart, LINPACK User's Guide, SIAM, (Society for Industrial and Applied Mathematics), 3600 University City Science Center, Philadelphia, PA, 19104-2688. ISBN 0-89871-172-X Parameters: Input/output, double A(LDA,P). On input, the N by P matrix whose decomposition is to be computed. On output, A contains in its upper triangle the upper triangular matrix R of the QR factorization. Below its diagonal A contains information from which the orthogonal part of the decomposition can be recovered. Note that if pivoting has been requested, the decomposition is not that of the original matrix A but that of A with its columns permuted as described by JPVT. Input, int LDA, the leading dimension of the array A. LDA must be at least N. Input, int N, the number of rows of the matrix A. Input, int P, the number of columns of the matrix A. Output, double QRAUX[P], contains further information required to recover the orthogonal part of the decomposition. Input/output, integer JPVT[P]. On input, JPVT contains integers that control the selection of the pivot columns. The K-th column A(*,K) of A is placed in one of three classes according to the value of JPVT(K). > 0, then A(K) is an initial column. = 0, then A(K) is a free column. < 0, then A(K) is a final column. Before the decomposition is computed, initial columns are moved to the beginning of the array A and final columns to the end. Both initial and final columns are frozen in place during the computation and only free columns are moved. At the K-th stage of the reduction, if A(*,K) is occupied by a free column it is interchanged with the free column of largest reduced norm. JPVT is not referenced if JOB == 0. On output, JPVT(K) contains the index of the column of the original matrix that has been interchanged into the K-th column, if pivoting was requested. Workspace, double WORK[P]. WORK is not referenced if JOB == 0. Input, int JOB, initiates column pivoting. 0, no pivoting is done. nonzero, pivoting is done. */ { int j; int jp; int l; int lup; int maxj; double maxnrm; double nrmxl; int pl; int pu; int swapj; double t; double tt; pl = 1; pu = 0; /* If pivoting is requested, rearrange the columns. */ if ( job != 0 ) { for ( j = 1; j <= p; j++ ) { swapj = ( 0 < jpvt[j-1] ); if ( jpvt[j-1] < 0 ) { jpvt[j-1] = -j; } else { jpvt[j-1] = j; } if ( swapj ) { if ( j != pl ) { dswap ( n, a+0+(pl-1)*lda, 1, a+0+(j-1), 1 ); } jpvt[j-1] = jpvt[pl-1]; jpvt[pl-1] = j; pl = pl + 1; } } pu = p; for ( j = p; 1 <= j; j-- ) { if ( jpvt[j-1] < 0 ) { jpvt[j-1] = -jpvt[j-1]; if ( j != pu ) { dswap ( n, a+0+(pu-1)*lda, 1, a+0+(j-1)*lda, 1 ); jp = jpvt[pu-1]; jpvt[pu-1] = jpvt[j-1]; jpvt[j-1] = jp; } pu = pu - 1; } } } /* Compute the norms of the free columns. */ for ( j = pl; j <= pu; j++ ) { qraux[j-1] = dnrm2 ( n, a+0+(j-1)*lda, 1 ); } for ( j = pl; j <= pu; j++ ) { work[j-1] = qraux[j-1]; } /* Perform the Householder reduction of A. */ lup = i4_min ( n, p ); for ( l = 1; l <= lup; l++ ) { /* Bring the column of largest norm into the pivot position. */ if ( pl <= l && l < pu ) { maxnrm = 0.0; maxj = l; for ( j = l; j <= pu; j++ ) { if ( maxnrm < qraux[j-1] ) { maxnrm = qraux[j-1]; maxj = j; } } if ( maxj != l ) { dswap ( n, a+0+(l-1)*lda, 1, a+0+(maxj-1)*lda, 1 ); qraux[maxj-1] = qraux[l-1]; work[maxj-1] = work[l-1]; jp = jpvt[maxj-1]; jpvt[maxj-1] = jpvt[l-1]; jpvt[l-1] = jp; } } /* Compute the Householder transformation for column L. */ qraux[l-1] = 0.0; if ( l != n ) { nrmxl = dnrm2 ( n-l+1, a+l-1+(l-1)*lda, 1 ); if ( nrmxl != 0.0 ) { if ( a[l-1+(l-1)*lda] != 0.0 ) { nrmxl = nrmxl * r8_sign ( a[l-1+(l-1)*lda] ); } dscal ( n-l+1, 1.0 / nrmxl, a+l-1+(l-1)*lda, 1 ); a[l-1+(l-1)*lda] = 1.0 + a[l-1+(l-1)*lda]; /* Apply the transformation to the remaining columns, updating the norms. */ for ( j = l + 1; j <= p; j++ ) { t = -ddot ( n-l+1, a+l-1+(l-1)*lda, 1, a+l-1+(j-1)*lda, 1 ) / a[l-1+(l-1)*lda]; daxpy ( n-l+1, t, a+l-1+(l-1)*lda, 1, a+l-1+(j-1)*lda, 1 ); if ( pl <= j && j <= pu ) { if ( qraux[j-1] != 0.0 ) { tt = 1.0 - pow ( fabs ( a[l-1+(j-1)*lda] ) / qraux[j-1], 2 ); tt = fmax ( tt, 0.0 ); t = tt; tt = 1.0 + 0.05 * tt * pow ( qraux[j-1] / work[j-1], 2 ); if ( tt != 1.0 ) { qraux[j-1] = qraux[j-1] * sqrt ( t ); } else { qraux[j-1] = dnrm2 ( n-l, a+l+(j-1)*lda, 1 ); work[j-1] = qraux[j-1]; } } } } /* Save the transformation. */ qraux[l-1] = a[l-1+(l-1)*lda]; a[l-1+(l-1)*lda] = -nrmxl; } } } return; } /******************************************************************************/ int dqrls ( double a[], int lda, int m, int n, double tol, int *kr, double b[], double x[], double rsd[], int jpvt[], double qraux[], int itask ) /******************************************************************************/ /* Purpose: DQRLS factors and solves a linear system in the least squares sense. Discussion: The linear system may be overdetermined, underdetermined or singular. The solution is obtained using a QR factorization of the coefficient matrix. DQRLS can be efficiently used to solve several least squares problems with the same matrix A. The first system is solved with ITASK = 1. The subsequent systems are solved with ITASK = 2, to avoid the recomputation of the matrix factors. The parameters KR, JPVT, and QRAUX must not be modified between calls to DQRLS. DQRLS is used to solve in a least squares sense overdetermined, underdetermined and singular linear systems. The system is A*X approximates B where A is M by N. B is a given M-vector, and X is the N-vector to be computed. A solution X is found which minimimzes the sum of squares (2-norm) of the residual, A*X - B. The numerical rank of A is determined using the tolerance TOL. DQRLS uses the LINPACK subroutine DQRDC to compute the QR factorization, with column pivoting, of an M by N matrix A. Licensing: This code is distributed under the MIT license. Modified: 10 September 2012 Author: C version by John Burkardt. Reference: David Kahaner, Cleve Moler, Steven Nash, Numerical Methods and Software, Prentice Hall, 1989, ISBN: 0-13-627258-4, LC: TA345.K34. Parameters: Input/output, double A[LDA*N], an M by N matrix. On input, the matrix whose decomposition is to be computed. In a least squares data fitting problem, A(I,J) is the value of the J-th basis (model) function at the I-th data point. On output, A contains the output from DQRDC. The triangular matrix R of the QR factorization is contained in the upper triangle and information needed to recover the orthogonal matrix Q is stored below the diagonal in A and in the vector QRAUX. Input, int LDA, the leading dimension of A. Input, int M, the number of rows of A. Input, int N, the number of columns of A. Input, double TOL, a relative tolerance used to determine the numerical rank. The problem should be scaled so that all the elements of A have roughly the same absolute accuracy EPS. Then a reasonable value for TOL is roughly EPS divided by the magnitude of the largest element. Output, int *KR, the numerical rank. Input, double B[M], the right hand side of the linear system. Output, double X[N], a least squares solution to the linear system. Output, double RSD[M], the residual, B - A*X. RSD may overwrite B. Workspace, int JPVT[N], required if ITASK = 1. Columns JPVT(1), ..., JPVT(KR) of the original matrix are linearly independent to within the tolerance TOL and the remaining columns are linearly dependent. ABS ( A(1,1) ) / ABS ( A(KR,KR) ) is an estimate of the condition number of the matrix of independent columns, and of R. This estimate will be <= 1/TOL. Workspace, double QRAUX[N], required if ITASK = 1. Input, int ITASK. 1, DQRLS factors the matrix A and solves the least squares problem. 2, DQRLS assumes that the matrix A was factored with an earlier call to DQRLS, and only solves the least squares problem. Output, int DQRLS, error code. 0: no error -1: LDA < M (fatal error) -2: N < 1 (fatal error) -3: ITASK < 1 (fatal error) */ { int ind; if ( lda < m ) { fprintf ( stderr, "\n" ); fprintf ( stderr, "DQRLS - Fatal error!\n" ); fprintf ( stderr, " LDA < M.\n" ); ind = -1; return ind; } if ( n <= 0 ) { fprintf ( stderr, "\n" ); fprintf ( stderr, "DQRLS - Fatal error!\n" ); fprintf ( stderr, " N <= 0.\n" ); ind = -2; return ind; } if ( itask < 1 ) { fprintf ( stderr, "\n" ); fprintf ( stderr, "DQRLS - Fatal error!\n" ); fprintf ( stderr, " ITASK < 1.\n" ); ind = -3; return ind; } ind = 0; /* Factor the matrix. */ if ( itask == 1 ) { dqrank ( a, lda, m, n, tol, kr, jpvt, qraux ); } /* Solve the least-squares problem. */ dqrlss ( a, lda, m, n, *kr, b, x, rsd, jpvt, qraux ); return ind; } /******************************************************************************/ void dqrlss ( double a[], int lda, int m, int n, int kr, double b[], double x[], double rsd[], int jpvt[], double qraux[] ) /******************************************************************************/ /* Purpose: DQRLSS solves a linear system in a least squares sense. Discussion: DQRLSS must be preceeded by a call to DQRANK. The system is to be solved is A * X = B where A is an M by N matrix with rank KR, as determined by DQRANK, B is a given M-vector, X is the N-vector to be computed. A solution X, with at most KR nonzero components, is found which minimizes the 2-norm of the residual (A*X-B). Once the matrix A has been formed, DQRANK should be called once to decompose it. Then, for each right hand side B, DQRLSS should be called once to obtain the solution and residual. Licensing: This code is distributed under the MIT license. Modified: 10 September 2012 Author: C version by John Burkardt Parameters: Input, double A[LDA*N], the QR factorization information from DQRANK. The triangular matrix R of the QR factorization is contained in the upper triangle and information needed to recover the orthogonal matrix Q is stored below the diagonal in A and in the vector QRAUX. Input, int LDA, the leading dimension of A, which must be at least M. Input, int M, the number of rows of A. Input, int N, the number of columns of A. Input, int KR, the rank of the matrix, as estimated by DQRANK. Input, double B[M], the right hand side of the linear system. Output, double X[N], a least squares solution to the linear system. Output, double RSD[M], the residual, B - A*X. RSD may overwite B. Input, int JPVT[N], the pivot information from DQRANK. Columns JPVT[0], ..., JPVT[KR-1] of the original matrix are linearly independent to within the tolerance TOL and the remaining columns are linearly dependent. Input, double QRAUX[N], auxiliary information from DQRANK defining the QR factorization. */ { int i; int j; int job; int k; double t; if ( kr != 0 ) { job = 110; dqrsl ( a, lda, m, kr, qraux, b, rsd, rsd, x, rsd, rsd, job ); } for ( i = 0; i < n; i++ ) { jpvt[i] = - jpvt[i]; } for ( i = kr; i < n; i++ ) { x[i] = 0.0; } for ( j = 1; j <= n; j++ ) { if ( jpvt[j-1] <= 0 ) { k = - jpvt[j-1]; jpvt[j-1] = k; while ( k != j ) { t = x[j-1]; x[j-1] = x[k-1]; x[k-1] = t; jpvt[k-1] = -jpvt[k-1]; k = jpvt[k-1]; } } } return; } /******************************************************************************/ int dqrsl ( double a[], int lda, int n, int k, double qraux[], double y[], double qy[], double qty[], double b[], double rsd[], double ab[], int job ) /******************************************************************************/ /* Purpose: DQRSL computes transformations, projections, and least squares solutions. Discussion: DQRSL requires the output of DQRDC. For K <= min(N,P), let AK be the matrix AK = ( A(JPVT[0]), A(JPVT(2)), ..., A(JPVT(K)) ) formed from columns JPVT[0], ..., JPVT(K) of the original N by P matrix A that was input to DQRDC. If no pivoting was done, AK consists of the first K columns of A in their original order. DQRDC produces a factored orthogonal matrix Q and an upper triangular matrix R such that AK = Q * (R) (0) This information is contained in coded form in the arrays A and QRAUX. The parameters QY, QTY, B, RSD, and AB are not referenced if their computation is not requested and in this case can be replaced by dummy variables in the calling program. To save storage, the user may in some cases use the same array for different parameters in the calling sequence. A frequently occuring example is when one wishes to compute any of B, RSD, or AB and does not need Y or QTY. In this case one may identify Y, QTY, and one of B, RSD, or AB, while providing separate arrays for anything else that is to be computed. Thus the calling sequence dqrsl ( a, lda, n, k, qraux, y, dum, y, b, y, dum, 110, info ) will result in the computation of B and RSD, with RSD overwriting Y. More generally, each item in the following list contains groups of permissible identifications for a single calling sequence. 1. (Y,QTY,B) (RSD) (AB) (QY) 2. (Y,QTY,RSD) (B) (AB) (QY) 3. (Y,QTY,AB) (B) (RSD) (QY) 4. (Y,QY) (QTY,B) (RSD) (AB) 5. (Y,QY) (QTY,RSD) (B) (AB) 6. (Y,QY) (QTY,AB) (B) (RSD) In any group the value returned in the array allocated to the group corresponds to the last member of the group. Licensing: This code is distributed under the MIT license. Modified: 07 June 2005 Author: C version by John Burkardt. Reference: Jack Dongarra, Cleve Moler, Jim Bunch and Pete Stewart, LINPACK User's Guide, SIAM, (Society for Industrial and Applied Mathematics), 3600 University City Science Center, Philadelphia, PA, 19104-2688. ISBN 0-89871-172-X Parameters: Input, double A[LDA*P], contains the output of DQRDC. Input, int LDA, the leading dimension of the array A. Input, int N, the number of rows of the matrix AK. It must have the same value as N in DQRDC. Input, int K, the number of columns of the matrix AK. K must not be greater than min(N,P), where P is the same as in the calling sequence to DQRDC. Input, double QRAUX[P], the auxiliary output from DQRDC. Input, double Y[N], a vector to be manipulated by DQRSL. Output, double QY[N], contains Q * Y, if requested. Output, double QTY[N], contains Q' * Y, if requested. Output, double B[K], the solution of the least squares problem minimize norm2 ( Y - AK * B), if its computation has been requested. Note that if pivoting was requested in DQRDC, the J-th component of B will be associated with column JPVT(J) of the original matrix A that was input into DQRDC. Output, double RSD[N], the least squares residual Y - AK * B, if its computation has been requested. RSD is also the orthogonal projection of Y onto the orthogonal complement of the column space of AK. Output, double AB[N], the least squares approximation Ak * B, if its computation has been requested. AB is also the orthogonal projection of Y onto the column space of A. Input, integer JOB, specifies what is to be computed. JOB has the decimal expansion ABCDE, with the following meaning: if A != 0, compute QY. if B != 0, compute QTY. if C != 0, compute QTY and B. if D != 0, compute QTY and RSD. if E != 0, compute QTY and AB. Note that a request to compute B, RSD, or AB automatically triggers the computation of QTY, for which an array must be provided in the calling sequence. Output, int DQRSL, is zero unless the computation of B has been requested and R is exactly singular. In this case, INFO is the index of the first zero diagonal element of R, and B is left unaltered. */ { int cab; int cb; int cqty; int cqy; int cr; int i; int info; int j; int jj; int ju; double t; double temp; /* Set INFO flag. */ info = 0; /* Determine what is to be computed. */ cqy = ( job / 10000 != 0 ); cqty = ( ( job % 10000 ) != 0 ); cb = ( ( job % 1000 ) / 100 != 0 ); cr = ( ( job % 100 ) / 10 != 0 ); cab = ( ( job % 10 ) != 0 ); ju = i4_min ( k, n-1 ); /* Special action when N = 1. */ if ( ju == 0 ) { if ( cqy ) { qy[0] = y[0]; } if ( cqty ) { qty[0] = y[0]; } if ( cab ) { ab[0] = y[0]; } if ( cb ) { if ( a[0+0*lda] == 0.0 ) { info = 1; } else { b[0] = y[0] / a[0+0*lda]; } } if ( cr ) { rsd[0] = 0.0; } return info; } /* Set up to compute QY or QTY. */ if ( cqy ) { for ( i = 1; i <= n; i++ ) { qy[i-1] = y[i-1]; } } if ( cqty ) { for ( i = 1; i <= n; i++ ) { qty[i-1] = y[i-1]; } } /* Compute QY. */ if ( cqy ) { for ( jj = 1; jj <= ju; jj++ ) { j = ju - jj + 1; if ( qraux[j-1] != 0.0 ) { temp = a[j-1+(j-1)*lda]; a[j-1+(j-1)*lda] = qraux[j-1]; t = -ddot ( n-j+1, a+j-1+(j-1)*lda, 1, qy+j-1, 1 ) / a[j-1+(j-1)*lda]; daxpy ( n-j+1, t, a+j-1+(j-1)*lda, 1, qy+j-1, 1 ); a[j-1+(j-1)*lda] = temp; } } } /* Compute Q'*Y. */ if ( cqty ) { for ( j = 1; j <= ju; j++ ) { if ( qraux[j-1] != 0.0 ) { temp = a[j-1+(j-1)*lda]; a[j-1+(j-1)*lda] = qraux[j-1]; t = -ddot ( n-j+1, a+j-1+(j-1)*lda, 1, qty+j-1, 1 ) / a[j-1+(j-1)*lda]; daxpy ( n-j+1, t, a+j-1+(j-1)*lda, 1, qty+j-1, 1 ); a[j-1+(j-1)*lda] = temp; } } } /* Set up to compute B, RSD, or AB. */ if ( cb ) { for ( i = 1; i <= k; i++ ) { b[i-1] = qty[i-1]; } } if ( cab ) { for ( i = 1; i <= k; i++ ) { ab[i-1] = qty[i-1]; } } if ( cr && k < n ) { for ( i = k+1; i <= n; i++ ) { rsd[i-1] = qty[i-1]; } } if ( cab && k+1 <= n ) { for ( i = k+1; i <= n; i++ ) { ab[i-1] = 0.0; } } if ( cr ) { for ( i = 1; i <= k; i++ ) { rsd[i-1] = 0.0; } } /* Compute B. */ if ( cb ) { for ( jj = 1; jj <= k; jj++ ) { j = k - jj + 1; if ( a[j-1+(j-1)*lda] == 0.0 ) { info = j; break; } b[j-1] = b[j-1] / a[j-1+(j-1)*lda]; if ( j != 1 ) { t = -b[j-1]; daxpy ( j-1, t, a+0+(j-1)*lda, 1, b, 1 ); } } } /* Compute RSD or AB as required. */ if ( cr || cab ) { for ( jj = 1; jj <= ju; jj++ ) { j = ju - jj + 1; if ( qraux[j-1] != 0.0 ) { temp = a[j-1+(j-1)*lda]; a[j-1+(j-1)*lda] = qraux[j-1]; if ( cr ) { t = -ddot ( n-j+1, a+j-1+(j-1)*lda, 1, rsd+j-1, 1 ) / a[j-1+(j-1)*lda]; daxpy ( n-j+1, t, a+j-1+(j-1)*lda, 1, rsd+j-1, 1 ); } if ( cab ) { t = -ddot ( n-j+1, a+j-1+(j-1)*lda, 1, ab+j-1, 1 ) / a[j-1+(j-1)*lda]; daxpy ( n-j+1, t, a+j-1+(j-1)*lda, 1, ab+j-1, 1 ); } a[j-1+(j-1)*lda] = temp; } } } return info; } /******************************************************************************/ void drot ( int n, double x[], int incx, double y[], int incy, double c, double s ) /******************************************************************************/ /* Purpose: DROT applies a plane rotation. Licensing: This code is distributed under the MIT license. Modified: 30 March 2007 Author: C version by John Burkardt Reference: Jack Dongarra, Cleve Moler, Jim Bunch, Pete Stewart, LINPACK User's Guide, SIAM, 1979. Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh, Basic Linear Algebra Subprograms for Fortran Usage, Algorithm 539, ACM Transactions on Mathematical Software, Volume 5, Number 3, September 1979, pages 308-323. Parameters: Input, int N, the number of entries in the vectors. Input/output, double X[*], one of the vectors to be rotated. Input, int INCX, the increment between successive entries of X. Input/output, double Y[*], one of the vectors to be rotated. Input, int INCY, the increment between successive elements of Y. Input, double C, S, parameters (presumably the cosine and sine of some angle) that define a plane rotation. */ { int i; int ix; int iy; double stemp; if ( n <= 0 ) { } else if ( incx == 1 && incy == 1 ) { for ( i = 0; i < n; i++ ) { stemp = c * x[i] + s * y[i]; y[i] = c * y[i] - s * x[i]; x[i] = stemp; } } else { if ( 0 <= incx ) { ix = 0; } else { ix = ( - n + 1 ) * incx; } if ( 0 <= incy ) { iy = 0; } else { iy = ( - n + 1 ) * incy; } for ( i = 0; i < n; i++ ) { stemp = c * x[ix] + s * y[iy]; y[iy] = c * y[iy] - s * x[ix]; x[ix] = stemp; ix = ix + incx; iy = iy + incy; } } return; } /******************************************************************************/ void drotg ( double *sa, double *sb, double *c, double *s ) /******************************************************************************/ /* Purpose: DROTG constructs a Givens plane rotation. Discussion: Given values A and B, this routine computes SIGMA = sign ( A ) if abs ( A ) > abs ( B ) = sign ( B ) if abs ( A ) <= abs ( B ); R = SIGMA * ( A * A + B * B ); C = A / R if R is not 0 = 1 if R is 0; S = B / R if R is not 0, 0 if R is 0. The computed numbers then satisfy the equation ( C S ) ( A ) = ( R ) ( -S C ) ( B ) = ( 0 ) The routine also computes Z = S if abs ( A ) > abs ( B ), = 1 / C if abs ( A ) <= abs ( B ) and C is not 0, = 1 if C is 0. The single value Z encodes C and S, and hence the rotation: If Z = 1, set C = 0 and S = 1; If abs ( Z ) < 1, set C = sqrt ( 1 - Z * Z ) and S = Z; if abs ( Z ) > 1, set C = 1/ Z and S = sqrt ( 1 - C * C ); Licensing: This code is distributed under the MIT license. Modified: 30 March 2007 Author: C version by John Burkardt Reference: Jack Dongarra, Cleve Moler, Jim Bunch, Pete Stewart, LINPACK User's Guide, SIAM, 1979. Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh, Basic Linear Algebra Subprograms for Fortran Usage, Algorithm 539, ACM Transactions on Mathematical Software, Volume 5, Number 3, September 1979, pages 308-323. Parameters: Input/output, double *SA, *SB, On input, SA and SB are the values A and B. On output, SA is overwritten with R, and SB is overwritten with Z. Output, double *C, *S, the cosine and sine of the Givens rotation. */ { double r; double roe; double scale; double z; if ( fabs ( *sb ) < fabs ( *sa ) ) { roe = *sa; } else { roe = *sb; } scale = fabs ( *sa ) + fabs ( *sb ); if ( scale == 0.0 ) { *c = 1.0; *s = 0.0; r = 0.0; } else { r = scale * sqrt ( ( *sa / scale ) * ( *sa / scale ) + ( *sb / scale ) * ( *sb / scale ) ); r = r8_sign ( roe ) * r; *c = *sa / r; *s = *sb / r; } if ( 0.0 < fabs ( *c ) && fabs ( *c ) <= *s ) { z = 1.0 / *c; } else { z = *s; } *sa = r; *sb = z; return; } /******************************************************************************/ void dscal ( int n, double sa, double x[], int incx ) /******************************************************************************/ /* Purpose: DSCAL scales a vector by a constant. Licensing: This code is distributed under the MIT license. Modified: 30 March 2007 Author: C version by John Burkardt Reference: Jack Dongarra, Cleve Moler, Jim Bunch, Pete Stewart, LINPACK User's Guide, SIAM, 1979. Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh, Basic Linear Algebra Subprograms for Fortran Usage, Algorithm 539, ACM Transactions on Mathematical Software, Volume 5, Number 3, September 1979, pages 308-323. Parameters: Input, int N, the number of entries in the vector. Input, double SA, the multiplier. Input/output, double X[*], the vector to be scaled. Input, int INCX, the increment between successive entries of X. */ { int i; int ix; int m; if ( n <= 0 ) { } else if ( incx == 1 ) { m = n % 5; for ( i = 0; i < m; i++ ) { x[i] = sa * x[i]; } for ( i = m; i < n; i = i + 5 ) { x[i] = sa * x[i]; x[i+1] = sa * x[i+1]; x[i+2] = sa * x[i+2]; x[i+3] = sa * x[i+3]; x[i+4] = sa * x[i+4]; } } else { if ( 0 <= incx ) { ix = 0; } else { ix = ( - n + 1 ) * incx; } for ( i = 0; i < n; i++ ) { x[ix] = sa * x[ix]; ix = ix + incx; } } return; } /******************************************************************************/ int dsvdc ( double a[], int lda, int m, int n, double s[], double e[], double u[], int ldu, double v[], int ldv, double work[], int job ) /******************************************************************************/ /* Purpose: DSVDC computes the singular value decomposition of a real rectangular matrix. Discussion: This routine reduces an M by N matrix A to diagonal form by orthogonal transformations U and V. The diagonal elements S(I) are the singular values of A. The columns of U are the corresponding left singular vectors, and the columns of V the right singular vectors. The form of the singular value decomposition is then A(MxN) = U(MxM) * S(MxN) * V(NxN)' Licensing: This code is distributed under the MIT license. Modified: 03 May 2007 Author: C version by John Burkardt. Reference: Jack Dongarra, Cleve Moler, Jim Bunch and Pete Stewart, LINPACK User's Guide, SIAM, (Society for Industrial and Applied Mathematics), 3600 University City Science Center, Philadelphia, PA, 19104-2688. ISBN 0-89871-172-X Parameters: Input/output, double A[LDA*N]. On input, the M by N matrix whose singular value decomposition is to be computed. On output, the matrix has been destroyed. Depending on the user's requests, the matrix may contain other useful information. Input, int LDA, the leading dimension of the array A. LDA must be at least M. Input, int M, the number of rows of the matrix. Input, int N, the number of columns of the matrix A. Output, double S[MM], where MM = min(M+1,N). The first min(M,N) entries of S contain the singular values of A arranged in descending order of magnitude. Output, double E[MM], where MM = min(M+1,N), ordinarily contains zeros. However see the discussion of INFO for exceptions. Output, double U[LDU*K]. If JOBA = 1 then K = M; if 2 <= JOBA, then K = min(M,N). U contains the M by M matrix of left singular vectors. U is not referenced if JOBA = 0. If M <= N or if JOBA = 2, then U may be identified with A in the subroutine call. Input, int LDU, the leading dimension of the array U. LDU must be at least M. Output, double V[LDV*N], the N by N matrix of right singular vectors. V is not referenced if JOB is 0. If N <= M, then V may be identified with A in the subroutine call. Input, int LDV, the leading dimension of the array V. LDV must be at least N. Workspace, double WORK[M]. Input, int JOB, controls the computation of the singular vectors. It has the decimal expansion AB with the following meaning: A = 0, do not compute the left singular vectors. A = 1, return the M left singular vectors in U. A >= 2, return the first min(M,N) singular vectors in U. B = 0, do not compute the right singular vectors. B = 1, return the right singular vectors in V. Output, int *DSVDC, status indicator INFO. The singular values (and their corresponding singular vectors) S(*INFO+1), S(*INFO+2),...,S(MN) are correct. Here MN = min ( M, N ). Thus if *INFO is 0, all the singular values and their vectors are correct. In any event, the matrix B = U' * A * V is the bidiagonal matrix with the elements of S on its diagonal and the elements of E on its superdiagonal. Thus the singular values of A and B are the same. */ { double b; double c; double cs; double el; double emm1; double f; double g; int i; int info; int iter; int j; int jobu; int k; int kase; int kk; int l; int ll; int lls; int ls; int lu; int maxit = 30; int mm; int mm1; int mn; int nct; int nctp1; int ncu; int nrt; int nrtp1; double scale; double shift; double sl; double sm; double smm1; double sn; double t; double t1; double test; int wantu; int wantv; double ztest; /* Determine what is to be computed. */ info = 0; wantu = 0; wantv = 0; jobu = ( job % 100 ) / 10; if ( 1 < jobu ) { ncu = i4_min ( m, n ); } else { ncu = m; } if ( jobu != 0 ) { wantu = 1; } if ( ( job % 10 ) != 0 ) { wantv = 1; } /* Reduce A to bidiagonal form, storing the diagonal elements in S and the super-diagonal elements in E. */ nct = i4_min ( m-1, n ); nrt = i4_max ( 0, i4_min ( m, n-2 ) ); lu = i4_max ( nct, nrt ); for ( l = 1; l <= lu; l++ ) { /* Compute the transformation for the L-th column and place the L-th diagonal in S(L). */ if ( l <= nct ) { s[l-1] = dnrm2 ( m-l+1, a+l-1+(l-1)*lda, 1 ); if ( s[l-1] != 0.0 ) { if ( a[l-1+(l-1)*lda] != 0.0 ) { s[l-1] = r8_sign ( a[l-1+(l-1)*lda] ) * fabs ( s[l-1] ); } dscal ( m-l+1, 1.0 / s[l-1], a+l-1+(l-1)*lda, 1 ); a[l-1+(l-1)*lda] = 1.0 + a[l-1+(l-1)*lda]; } s[l-1] = -s[l-1]; } for ( j = l+1; j <= n; j++ ) { /* Apply the transformation. */ if ( l <= nct && s[l-1] != 0.0 ) { t = - ddot ( m-l+1, a+l-1+(l-1)*lda, 1, a+l-1+(j-1)*lda, 1 ) / a[l-1+(l-1)*lda]; daxpy ( m-l+1, t, a+l-1+(l-1)*lda, 1, a+l-1+(j-1)*lda, 1 ); } /* Place the L-th row of A into E for the subsequent calculation of the row transformation. */ e[j-1] = a[l-1+(j-1)*lda]; } /* Place the transformation in U for subsequent back multiplication. */ if ( wantu && l <= nct ) { for ( i = l; i <= m; i++ ) { u[i-1+(l-1)*ldu] = a[i-1+(l-1)*lda]; } } if ( l <= nrt ) { /* Compute the L-th row transformation and place the L-th superdiagonal in E(L). */ e[l-1] = dnrm2 ( n-l, e+l, 1 ); if ( e[l-1] != 0.0 ) { if ( e[l] != 0.0 ) { e[l-1] = r8_sign ( e[l] ) * fabs ( e[l-1] ); } dscal ( n-l, 1.0 / e[l-1], e+l, 1 ); e[l] = 1.0 + e[l]; } e[l-1] = -e[l-1]; /* Apply the transformation. */ if ( l+1 <= m && e[l-1] != 0.0 ) { for ( j = l+1; j <= m; j++ ) { work[j-1] = 0.0; } for ( j = l+1; j <= n; j++ ) { daxpy ( m-l, e[j-1], a+l+(j-1)*lda, 1, work+l, 1 ); } for ( j = l+1; j <= n; j++ ) { daxpy ( m-l, -e[j-1]/e[l], work+l, 1, a+l+(j-1)*lda, 1 ); } } /* Place the transformation in V for subsequent back multiplication. */ if ( wantv ) { for ( j = l+1; j <= n; j++ ) { v[j-1+(l-1)*ldv] = e[j-1]; } } } } /* Set up the final bidiagonal matrix of order MN. */ mn = i4_min ( m + 1, n ); nctp1 = nct + 1; nrtp1 = nrt + 1; if ( nct < n ) { s[nctp1-1] = a[nctp1-1+(nctp1-1)*lda]; } if ( m < mn ) { s[mn-1] = 0.0; } if ( nrtp1 < mn ) { e[nrtp1-1] = a[nrtp1-1+(mn-1)*lda]; } e[mn-1] = 0.0; /* If required, generate U. */ if ( wantu ) { for ( i = 1; i <= m; i++ ) { for ( j = nctp1; j <= ncu; j++ ) { u[(i-1)+(j-1)*ldu] = 0.0; } } for ( j = nctp1; j <= ncu; j++ ) { u[j-1+(j-1)*ldu] = 1.0; } for ( ll = 1; ll <= nct; ll++ ) { l = nct - ll + 1; if ( s[l-1] != 0.0 ) { for ( j = l+1; j <= ncu; j++ ) { t = - ddot ( m-l+1, u+(l-1)+(l-1)*ldu, 1, u+(l-1)+(j-1)*ldu, 1 ) / u[l-1+(l-1)*ldu]; daxpy ( m-l+1, t, u+(l-1)+(l-1)*ldu, 1, u+(l-1)+(j-1)*ldu, 1 ); } dscal ( m-l+1, -1.0, u+(l-1)+(l-1)*ldu, 1 ); u[l-1+(l-1)*ldu] = 1.0 + u[l-1+(l-1)*ldu]; for ( i = 1; i <= l-1; i++ ) { u[i-1+(l-1)*ldu] = 0.0; } } else { for ( i = 1; i <= m; i++ ) { u[i-1+(l-1)*ldu] = 0.0; } u[l-1+(l-1)*ldu] = 1.0; } } } /* If it is required, generate V. */ if ( wantv ) { for ( ll = 1; ll <= n; ll++ ) { l = n - ll + 1; if ( l <= nrt && e[l-1] != 0.0 ) { for ( j = l+1; j <= n; j++ ) { t = - ddot ( n-l, v+l+(l-1)*ldv, 1, v+l+(j-1)*ldv, 1 ) / v[l+(l-1)*ldv]; daxpy ( n-l, t, v+l+(l-1)*ldv, 1, v+l+(j-1)*ldv, 1 ); } } for ( i = 1; i <= n; i++ ) { v[i-1+(l-1)*ldv] = 0.0; } v[l-1+(l-1)*ldv] = 1.0; } } /* Main iteration loop for the singular values. */ mm = mn; iter = 0; while ( 0 < mn ) { /* If too many iterations have been performed, set flag and return. */ if ( maxit <= iter ) { info = mn; return info; } /* This section of the program inspects for negligible elements in the S and E arrays. On completion the variables KASE and L are set as follows: KASE = 1 if S(MN) and E(L-1) are negligible and L < MN KASE = 2 if S(L) is negligible and L < MN KASE = 3 if E(L-1) is negligible, L < MN, and S(L), ..., S(MN) are not negligible (QR step). KASE = 4 if E(MN-1) is negligible (convergence). */ for ( ll = 1; ll <= mn; ll++ ) { l = mn - ll; if ( l == 0 ) { break; } test = fabs ( s[l-1] ) + fabs ( s[l] ); ztest = test + fabs ( e[l-1] ); if ( ztest == test ) { e[l-1] = 0.0; break; } } if ( l == mn - 1 ) { kase = 4; } else { for ( lls = l + 1; lls <= mn + 1; lls++ ) { ls = mn - lls + l + 1; if ( ls == l ) { break; } test = 0.0; if ( ls != mn ) { test = test + fabs ( e[ls-1] ); } if ( ls != l + 1 ) { test = test + fabs ( e[ls-2] ); } ztest = test + fabs ( s[ls-1] ); if ( ztest == test ) { s[ls-1] = 0.0; break; } } if ( ls == l ) { kase = 3; } else if ( ls == mn ) { kase = 1; } else { kase = 2; l = ls; } } l = l + 1; /* Deflate negligible S(MN). */ if ( kase == 1 ) { mm1 = mn - 1; f = e[mn-2]; e[mn-2] = 0.0; for ( kk = 1; kk <= mm1; kk++ ) { k = mm1 - kk + l; t1 = s[k-1]; drotg ( &t1, &f, &cs, &sn ); s[k-1] = t1; if ( k != l ) { f = -sn * e[k-2]; e[k-2] = cs * e[k-2]; } if ( wantv ) { drot ( n, v+0+(k-1)*ldv, 1, v+0+(mn-1)*ldv, 1, cs, sn ); } } } /* Split at negligible S(L). */ else if ( kase == 2 ) { f = e[l-2]; e[l-2] = 0.0; for ( k = l; k <= mn; k++ ) { t1 = s[k-1]; drotg ( &t1, &f, &cs, &sn ); s[k-1] = t1; f = - sn * e[k-1]; e[k-1] = cs * e[k-1]; if ( wantu ) { drot ( m, u+0+(k-1)*ldu, 1, u+0+(l-2)*ldu, 1, cs, sn ); } } } /* Perform one QR step. */ else if ( kase == 3 ) { /* Calculate the shift. */ scale = fmax ( fabs ( s[mn-1] ), fmax ( fabs ( s[mn-2] ), fmax ( fabs ( e[mn-2] ), fmax ( fabs ( s[l-1] ), fabs ( e[l-1] ) ) ) ) ); sm = s[mn-1] / scale; smm1 = s[mn-2] / scale; emm1 = e[mn-2] / scale; sl = s[l-1] / scale; el = e[l-1] / scale; b = ( ( smm1 + sm ) * ( smm1 - sm ) + emm1 * emm1 ) / 2.0; c = ( sm * emm1 ) * ( sm * emm1 ); shift = 0.0; if ( b != 0.0 || c != 0.0 ) { shift = sqrt ( b * b + c ); if ( b < 0.0 ) { shift = -shift; } shift = c / ( b + shift ); } f = ( sl + sm ) * ( sl - sm ) - shift; g = sl * el; /* Chase zeros. */ mm1 = mn - 1; for ( k = l; k <= mm1; k++ ) { drotg ( &f, &g, &cs, &sn ); if ( k != l ) { e[k-2] = f; } f = cs * s[k-1] + sn * e[k-1]; e[k-1] = cs * e[k-1] - sn * s[k-1]; g = sn * s[k]; s[k] = cs * s[k]; if ( wantv ) { drot ( n, v+0+(k-1)*ldv, 1, v+0+k*ldv, 1, cs, sn ); } drotg ( &f, &g, &cs, &sn ); s[k-1] = f; f = cs * e[k-1] + sn * s[k]; s[k] = -sn * e[k-1] + cs * s[k]; g = sn * e[k]; e[k] = cs * e[k]; if ( wantu && k < m ) { drot ( m, u+0+(k-1)*ldu, 1, u+0+k*ldu, 1, cs, sn ); } } e[mn-2] = f; iter = iter + 1; } /* Convergence. */ else if ( kase == 4 ) { /* Make the singular value nonnegative. */ if ( s[l-1] < 0.0 ) { s[l-1] = -s[l-1]; if ( wantv ) { dscal ( n, -1.0, v+0+(l-1)*ldv, 1 ); } } /* Order the singular value. */ for ( ; ; ) { if ( l == mm ) { break; } if ( s[l] <= s[l-1] ) { break; } t = s[l-1]; s[l-1] = s[l]; s[l] = t; if ( wantv && l < n ) { dswap ( n, v+0+(l-1)*ldv, 1, v+0+l*ldv, 1 ); } if ( wantu && l < m ) { dswap ( m, u+0+(l-1)*ldu, 1, u+0+l*ldu, 1 ); } l = l + 1; } iter = 0; mn = mn - 1; } } return info; } /******************************************************************************/ void dswap ( int n, double x[], int incx, double y[], int incy ) /******************************************************************************/ /* Purpose: DSWAP interchanges two vectors. Licensing: This code is distributed under the MIT license. Modified: 30 March 2007 Author: C version by John Burkardt Reference: Jack Dongarra, Cleve Moler, Jim Bunch, Pete Stewart, LINPACK User's Guide, SIAM, 1979. Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh, Basic Linear Algebra Subprograms for Fortran Usage, Algorithm 539, ACM Transactions on Mathematical Software, Volume 5, Number 3, September 1979, pages 308-323. Parameters: Input, int N, the number of entries in the vectors. Input/output, double X[*], one of the vectors to swap. Input, int INCX, the increment between successive entries of X. Input/output, double Y[*], one of the vectors to swap. Input, int INCY, the increment between successive elements of Y. */ { int i; int ix; int iy; int m; double temp; if ( n <= 0 ) { } else if ( incx == 1 && incy == 1 ) { m = n % 3; for ( i = 0; i < m; i++ ) { temp = x[i]; x[i] = y[i]; y[i] = temp; } for ( i = m; i < n; i = i + 3 ) { temp = x[i]; x[i] = y[i]; y[i] = temp; temp = x[i+1]; x[i+1] = y[i+1]; y[i+1] = temp; temp = x[i+2]; x[i+2] = y[i+2]; y[i+2] = temp; } } else { if ( 0 <= incx ) { ix = 0; } else { ix = ( - n + 1 ) * incx; } if ( 0 <= incy ) { iy = 0; } else { iy = ( - n + 1 ) * incy; } for ( i = 0; i < n; i++ ) { temp = x[ix]; x[ix] = y[iy]; y[iy] = temp; ix = ix + incx; iy = iy + incy; } } return; } /******************************************************************************/ double *normal_solve ( int m, int n, double a[], double b[], int *flag ) /******************************************************************************/ /* Purpose: NORMAL_SOLVE solves a linear system using the normal equations. Discussion: Given a presumably rectangular MxN system of equations A*x=b, this routine sets up the NxN system A'*A*x=A'b. Assuming N <= M, and that A has full column rank, the system will be solvable, and the vector x that is returned will minimize the Euclidean norm of the residual. One drawback to this approach is that the condition number of the linear system A'*A is effectively the square of the condition number of A, meaning that there is a substantial loss of accuracy. Thanks to David Doria for pointing out that this procedure was missing "free()" statements at the end, 08 April 2013. Licensing: This code is distributed under the MIT license. Modified: 08 April 2013 Author: John Burkardt Reference: David Kahaner, Cleve Moler, Steven Nash, Numerical Methods and Software, Prentice Hall, 1989, ISBN: 0-13-627258-4, LC: TA345.K34. Parameters: Input, int M, the number of rows of A. Input, int N, the number of columns of A. It must be the case that N <= M. Input, double A[M*N], the matrix. The matrix must have full column rank. Input, double B[M], the right hand side. Output, int *FLAG, 0, no error was detected. 1, an error occurred. Output, double NORMAL_SOLVE[N], the least squares solution. */ { double *at; double *ata; double *ata_c; double *atb; double *x; *flag = 0; if ( m < n ) { *flag = 1; return NULL; } at = r8mat_transpose_new ( m, n, a ); ata = r8mat_mm_new ( n, m, n, at, a ); ata_c = r8mat_cholesky_factor ( n, ata, flag ); if ( *flag != 0 ) { return NULL; } atb = r8mat_mv_new ( n, m, at, b ); x = r8mat_cholesky_solve ( n, ata_c, atb ); free ( at ); free ( ata ); free ( ata_c ); free ( atb ); return x; } /******************************************************************************/ double *qr_solve ( int m, int n, double a[], double b[] ) /******************************************************************************/ /* Purpose: QR_SOLVE solves a linear system in the least squares sense. Discussion: If the matrix A has full column rank, then the solution X should be the unique vector that minimizes the Euclidean norm of the residual. If the matrix A does not have full column rank, then the solution is not unique; the vector X will minimize the residual norm, but so will various other vectors. Licensing: This code is distributed under the MIT license. Modified: 11 September 2012 Author: John Burkardt Reference: David Kahaner, Cleve Moler, Steven Nash, Numerical Methods and Software, Prentice Hall, 1989, ISBN: 0-13-627258-4, LC: TA345.K34. Parameters: Input, int M, the number of rows of A. Input, int N, the number of columns of A. Input, double A[M*N], the matrix. Input, double B[M], the right hand side. Output, double QR_SOLVE[N], the least squares solution. */ { double *a_qr; int itask; int *jpvt; int kr; int lda; double *qraux; double *r; double tol; double *x; a_qr = r8mat_copy_new ( m, n, a ); lda = m; tol = DBL_EPSILON / r8mat_amax ( m, n, a_qr ); x = ( double * ) malloc ( n * sizeof ( double ) ); jpvt = ( int * ) malloc ( n * sizeof ( int ) ); qraux = ( double * ) malloc ( n * sizeof ( double ) ); r = ( double * ) malloc ( m * sizeof ( double ) ); itask = 1; dqrls ( a_qr, lda, m, n, tol, &kr, b, x, r, jpvt, qraux, itask ); free ( a_qr ); free ( jpvt ); free ( qraux ); free ( r ); return x; } /******************************************************************************/ double *svd_solve ( int m, int n, double a[], double b[] ) /******************************************************************************/ /* Purpose: SVD_SOLVE solves a linear system in the least squares sense. Discussion: The vector X returned by this routine should always minimize the Euclidean norm of the residual ||A*x-b||. If the matrix A does not have full column rank, then there are multiple vectors that attain the minimum residual. In that case, the vector X returned by this routine is the unique such minimizer that has the the minimum possible Euclidean norm, that is, ||A*x-b|| and ||x|| are both minimized. Licensing: This code is distributed under the MIT license. Modified: 21 April 2012 Author: John Burkardt Reference: David Kahaner, Cleve Moler, Steven Nash, Numerical Methods and Software, Prentice Hall, 1989, ISBN: 0-13-627258-4, LC: TA345.K34. Parameters: Input, int M, the number of rows of A. Input, int N, the number of columns of A. Input, double A[M*N], the matrix. Input, double B[M], the right hand side. Output, double SVD_SOLVE[N], the least squares solution. */ { double *a_copy; double *e; int i; int info; int lda; int ldu; int ldv; int job; double *sdiag; double smax; double stol; double *sub; double *u; double *ub; double *v; double *work; double *x; /* Get the SVD. */ a_copy = r8mat_copy_new ( m, n, a ); lda = m; sdiag = ( double * ) malloc ( i4_max ( m + 1, n ) * sizeof ( double ) ); e = ( double * ) malloc ( i4_max ( m + 1, n )* sizeof ( double ) ); u = ( double * ) malloc ( m * m * sizeof ( double ) ); ldu = m; v = ( double * ) malloc ( n * n * sizeof ( double ) ); ldv = n; work = ( double * ) malloc ( m * sizeof ( double ) ); job = 11; info = dsvdc ( a_copy, lda, m, n, sdiag, e, u, ldu, v, ldv, work, job ); free ( work ); if ( info != 0 ) { fprintf ( stderr, "\n" ); fprintf ( stderr, "SVD_SOLVE - Failure!\n" ); fprintf ( stderr, " The SVD could not be calculated.\n" ); fprintf ( stderr, " LINPACK routine DSVDC returned a nonzero\n" ); fprintf ( stderr, " value of the error flag, INFO = %d\n", info ); exit ( 1 ); } ub = r8mat_mtv_new ( m, m, u, b ); /* For singular problems, there may be tiny but nonzero singular values that should be ignored. This is a reasonable attempt to avoid such problems, although in general, the user might wish to control the tolerance. */ smax = r8vec_max ( n, sdiag ); if ( smax <= DBL_EPSILON ) { smax = 1.0; } stol = DBL_EPSILON * smax; sub = ( double * ) malloc ( n * sizeof ( double ) ); for ( i = 0; i < n; i++ ) { sub[i] = 0.0; if ( i < m ) { if ( stol <= sdiag[i] ) { sub[i] = ub[i] / sdiag[i]; } } } x = r8mat_mv_new ( n, n, v, sub ); free ( a_copy ); free ( e ); free ( sdiag ); free ( sub ); free ( u ); free ( ub ); free ( v ); return x; }