Functions
int64_t	lapack::heev (lapack::Job jobz, lapack::Uplo uplo, int64_t n, std::complex< double > A, int64_t lda, double W)
	Computes all eigenvalues and, optionally, eigenvectors of a Hermitian matrix A. More...

int64_t	lapack::heev (lapack::Job jobz, lapack::Uplo uplo, int64_t n, std::complex< float > A, int64_t lda, float W)

int64_t	lapack::heev_2stage (lapack::Job jobz, lapack::Uplo uplo, int64_t n, std::complex< double > A, int64_t lda, double W)
	Computes all eigenvalues and, optionally, eigenvectors of a Hermitian matrix A using the 2-stage technique for the reduction to tridiagonal. More...

int64_t	lapack::heev_2stage (lapack::Job jobz, lapack::Uplo uplo, int64_t n, std::complex< float > A, int64_t lda, float W)

int64_t	lapack::heevd (lapack::Job jobz, lapack::Uplo uplo, int64_t n, std::complex< double > A, int64_t lda, double W)
	Computes all eigenvalues and, optionally, eigenvectors of a. More...

int64_t	lapack::heevd (lapack::Job jobz, lapack::Uplo uplo, int64_t n, std::complex< float > A, int64_t lda, float W)

int64_t	lapack::heevd_2stage (lapack::Job jobz, lapack::Uplo uplo, int64_t n, std::complex< double > A, int64_t lda, double W)
	Computes all eigenvalues and, optionally, eigenvectors of a Hermitian matrix A using the 2-stage technique for the reduction to tridiagonal. More...

int64_t	lapack::heevd_2stage (lapack::Job jobz, lapack::Uplo uplo, int64_t n, std::complex< float > A, int64_t lda, float W)

int64_t	lapack::heevr (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, std::complex< double > A, int64_t lda, double vl, double vu, int64_t il, int64_t iu, double abstol, int64_t nfound, double W, std::complex< double > Z, int64_t ldz, int64_t *isuppz)
	Computes selected eigenvalues and, optionally, eigenvectors of a Hermitian matrix A. More...

int64_t	lapack::heevr (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, std::complex< float > A, int64_t lda, float vl, float vu, int64_t il, int64_t iu, float abstol, int64_t m, float W, std::complex< float > Z, int64_t ldz, int64_t *isuppz)

int64_t	lapack::heevr_2stage (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, std::complex< double > A, int64_t lda, double vl, double vu, int64_t il, int64_t iu, double abstol, int64_t nfound, double W, std::complex< double > Z, int64_t ldz, int64_t *isuppz)
	Computes selected eigenvalues and, optionally, eigenvectors of a Hermitian matrix A using the 2-stage technique for the reduction to tridiagonal. More...

int64_t	lapack::heevr_2stage (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, std::complex< float > A, int64_t lda, float vl, float vu, int64_t il, int64_t iu, float abstol, int64_t m, float W, std::complex< float > Z, int64_t ldz, int64_t *isuppz)

int64_t	lapack::heevx (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, std::complex< double > A, int64_t lda, double vl, double vu, int64_t il, int64_t iu, double abstol, int64_t nfound, double W, std::complex< double > Z, int64_t ldz, int64_t *ifail)
	Computes selected eigenvalues and, optionally, eigenvectors of a Hermitian matrix A. More...

int64_t	lapack::heevx (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, std::complex< float > A, int64_t lda, float vl, float vu, int64_t il, int64_t iu, float abstol, int64_t m, float W, std::complex< float > Z, int64_t ldz, int64_t *ifail)

int64_t	lapack::heevx_2stage (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, std::complex< double > A, int64_t lda, double vl, double vu, int64_t il, int64_t iu, double abstol, int64_t nfound, double W, std::complex< double > Z, int64_t ldz, int64_t *ifail)
	Computes selected eigenvalues and, optionally, eigenvectors of a Hermitian matrix A using the 2-stage technique for the reduction to tridiagonal. More...

int64_t	lapack::heevx_2stage (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, std::complex< float > A, int64_t lda, float vl, float vu, int64_t il, int64_t iu, float abstol, int64_t m, float W, std::complex< float > Z, int64_t ldz, int64_t *ifail)

int64_t	lapack::syev (lapack::Job jobz, lapack::Uplo uplo, int64_t n, double A, int64_t lda, double W)

int64_t	lapack::syev (lapack::Job jobz, lapack::Uplo uplo, int64_t n, float A, int64_t lda, float W)

int64_t	lapack::syev_2stage (lapack::Job jobz, lapack::Uplo uplo, int64_t n, double A, int64_t lda, double W)

int64_t	lapack::syev_2stage (lapack::Job jobz, lapack::Uplo uplo, int64_t n, float A, int64_t lda, float W)

int64_t	lapack::syevd (lapack::Job jobz, lapack::Uplo uplo, int64_t n, double A, int64_t lda, double W)

int64_t	lapack::syevd (lapack::Job jobz, lapack::Uplo uplo, int64_t n, float A, int64_t lda, float W)

int64_t	lapack::syevd_2stage (lapack::Job jobz, lapack::Uplo uplo, int64_t n, double A, int64_t lda, double W)

int64_t	lapack::syevd_2stage (lapack::Job jobz, lapack::Uplo uplo, int64_t n, float A, int64_t lda, float W)

int64_t	lapack::syevr (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, double A, int64_t lda, double vl, double vu, int64_t il, int64_t iu, double abstol, int64_t m, double W, double Z, int64_t ldz, int64_t *isuppz)

int64_t	lapack::syevr (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, float A, int64_t lda, float vl, float vu, int64_t il, int64_t iu, float abstol, int64_t m, float W, float Z, int64_t ldz, int64_t *isuppz)

int64_t	lapack::syevr_2stage (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, double A, int64_t lda, double vl, double vu, int64_t il, int64_t iu, double abstol, int64_t m, double W, double Z, int64_t ldz, int64_t *isuppz)

int64_t	lapack::syevr_2stage (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, float A, int64_t lda, float vl, float vu, int64_t il, int64_t iu, float abstol, int64_t m, float W, float Z, int64_t ldz, int64_t *isuppz)

int64_t	lapack::syevx (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, double A, int64_t lda, double vl, double vu, int64_t il, int64_t iu, double abstol, int64_t m, double W, double Z, int64_t ldz, int64_t *ifail)

int64_t	lapack::syevx (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, float A, int64_t lda, float vl, float vu, int64_t il, int64_t iu, float abstol, int64_t m, float W, float Z, int64_t ldz, int64_t *ifail)

int64_t	lapack::syevx_2stage (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, double A, int64_t lda, double vl, double vu, int64_t il, int64_t iu, double abstol, int64_t m, double W, double Z, int64_t ldz, int64_t *ifail)

int64_t	lapack::syevx_2stage (lapack::Job jobz, lapack::Range range, lapack::Uplo uplo, int64_t n, float A, int64_t lda, float vl, float vu, int64_t il, int64_t iu, float abstol, int64_t m, float W, float Z, int64_t ldz, int64_t *ifail)

Detailed Description

Function Documentation

◆ heev()

int64_t lapack::heev	(	lapack::Job	jobz,
		lapack::Uplo	uplo,
		int64_t	n,
		std::complex< double > *	A,
		int64_t	lda,
		double *	W
	)

Computes all eigenvalues and, optionally, eigenvectors of a Hermitian matrix A.

Overloaded versions are available for float, double, std::complex<float>, and std::complex<double>. For real matrices, this is an alias for lapack::syev.

Parameters

[in]	jobz	lapack::Job::NoVec: Compute eigenvalues only; lapack::Job::Vec: Compute eigenvalues and eigenvectors.
[in]	uplo	lapack::Uplo::Upper: Upper triangle of A is stored; lapack::Uplo::Lower: Lower triangle of A is stored.
[in]	n	The order of the matrix A. n >= 0.
[in,out]	A	The n-by-n matrix A, stored in an lda-by-n array. On entry, the Hermitian matrix A. If uplo = Upper, the leading n-by-n upper triangular part of A contains the upper triangular part of the matrix A. If uplo = Lower, the leading n-by-n lower triangular part of A contains the lower triangular part of the matrix A. On exit, if jobz = Vec, then if successful, A contains the orthonormal eigenvectors of the matrix A. If jobz = NoVec, then on exit the lower triangle (if uplo=Lower) or the upper triangle (if uplo=Upper) of A, including the diagonal, is destroyed.
[in]	lda	The leading dimension of the array A. lda >= max(1,n).
[out]	W	The vector W of length n. If successful, the eigenvalues in ascending order.

Returns: = 0: successful exit; > 0: if return value = i, the algorithm failed to converge; i off-diagonal elements of an intermediate tridiagonal form did not converge to zero.

◆ heev_2stage()

int64_t lapack::heev_2stage	(	lapack::Job	jobz,
		lapack::Uplo	uplo,
		int64_t	n,
		std::complex< double > *	A,
		int64_t	lda,
		double *	W
	)

Computes all eigenvalues and, optionally, eigenvectors of a Hermitian matrix A using the 2-stage technique for the reduction to tridiagonal.

Overloaded versions are available for float, double, std::complex<float>, and std::complex<double>. For real matrices, this is an alias for lapack::syev_2stage.

Parameters

[in]	jobz	lapack::Job::NoVec: Compute eigenvalues only; lapack::Job::Vec: Compute eigenvalues and eigenvectors. Not yet available (as of LAPACK 3.8.0).
[in]	uplo	lapack::Uplo::Upper: Upper triangle of A is stored; lapack::Uplo::Lower: Lower triangle of A is stored.
[in]	n	The order of the matrix A. n >= 0.
[in,out]	A	The n-by-n matrix A, stored in an lda-by-n array. On entry, the Hermitian matrix A. If uplo = Upper, the leading n-by-n upper triangular part of A contains the upper triangular part of the matrix A. If uplo = Lower, the leading n-by-n lower triangular part of A contains the lower triangular part of the matrix A. On exit, if jobz = Vec, then if successful, A contains the orthonormal eigenvectors of the matrix A. If jobz = NoVec, then on exit the lower triangle (if uplo=Lower) or the upper triangle (if uplo=Upper) of A, including the diagonal, is destroyed.
[in]	lda	The leading dimension of the array A. lda >= max(1,n).
[out]	W	The vector W of length n. If successful, the eigenvalues in ascending order.

Returns: = 0: successful exit; > 0: if return value = i, the algorithm failed to converge; i off-diagonal elements of an intermediate tridiagonal form did not converge to zero.

Further Details

All details about the 2-stage techniques are available in:

Azzam Haidar, Hatem Ltaief, and Jack Dongarra. Parallel reduction to condensed forms for symmetric eigenvalue problems using aggregated fine-grained and memory-aware kernels. In Proceedings of 2011 International Conference for High Performance Computing, Networking, Storage and Analysis (SC '11), New York, NY, USA, Article 8, 11 pages. http://doi.acm.org/10.1145/2063384.2063394

A. Haidar, J. Kurzak, P. Luszczek, 2013. An improved parallel singular value algorithm and its implementation for multicore hardware, In Proceedings of 2013 International Conference for High Performance Computing, Networking, Storage and Analysis (SC '13). Denver, Colorado, USA, 2013. Article 90, 12 pages. http://doi.acm.org/10.1145/2503210.2503292

A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra. A novel hybrid CPU-GPU generalized eigensolver for electronic structure calculations based on fine-grained memory aware tasks. International Journal of High Performance Computing Applications. Volume 28 Issue 2, Pages 196-209, May 2014. http://hpc.sagepub.com/content/28/2/196

◆ heevd()

int64_t lapack::heevd	(	lapack::Job	jobz,
		lapack::Uplo	uplo,
		int64_t	n,
		std::complex< double > *	A,
		int64_t	lda,
		double *	W
	)

Computes all eigenvalues and, optionally, eigenvectors of a.

divide and conquer algorithm.

The divide and conquer algorithm makes very mild assumptions about floating point arithmetic. It will work on machines with a guard digit in add/subtract, or on those binary machines without guard digits which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or Cray-2. It could conceivably fail on hexadecimal or decimal machines without guard digits, but we know of none.

Overloaded versions are available for float, double, std::complex<float>, and std::complex<double>. For real matrices, this is an alias for lapack::syevd.

Parameters

[in]	jobz	lapack::Job::NoVec: Compute eigenvalues only; lapack::Job::Vec: Compute eigenvalues and eigenvectors.
[in]	uplo	lapack::Uplo::Upper: Upper triangle of A is stored; lapack::Uplo::Lower: Lower triangle of A is stored.
[in]	n	The order of the matrix A. n >= 0.
[in,out]	A	The n-by-n matrix A, stored in an lda-by-n array. On entry, the Hermitian matrix A. If uplo = Upper, the leading n-by-n upper triangular part of A contains the upper triangular part of the matrix A. If uplo = Lower, the leading n-by-n lower triangular part of A contains the lower triangular part of the matrix A. On exit, if jobz = Vec, then if successful, A contains the orthonormal eigenvectors of the matrix A. If jobz = NoVec, then on exit the lower triangle (if uplo=Lower) or the upper triangle (if uplo=Upper) of A, including the diagonal, is destroyed.
[in]	lda	The leading dimension of the array A. lda >= max(1,n).
[out]	W	The vector W of length n. If successful, the eigenvalues in ascending order.

Returns: = 0: successful exit; > 0: if return value = i and jobz = NoVec, then the algorithm failed to converge; i off-diagonal elements of an intermediate tridiagonal form did not converge to zero; if return value = i and jobz = Vec, then the algorithm failed to compute an eigenvalue while working on the submatrix lying in rows and columns info/(n+1) through mod(info,n+1).

◆ heevd_2stage()

int64_t lapack::heevd_2stage	(	lapack::Job	jobz,
		lapack::Uplo	uplo,
		int64_t	n,
		std::complex< double > *	A,
		int64_t	lda,
		double *	W
	)

Computes all eigenvalues and, optionally, eigenvectors of a Hermitian matrix A using the 2-stage technique for the reduction to tridiagonal.

If eigenvectors are desired, it uses a divide and conquer algorithm.

The divide and conquer algorithm makes very mild assumptions about floating point arithmetic. It will work on machines with a guard digit in add/subtract, or on those binary machines without guard digits which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or Cray-2. It could conceivably fail on hexadecimal or decimal machines without guard digits, but we know of none.

Overloaded versions are available for float, double, std::complex<float>, and std::complex<double>. For real matrices, this is an alias for lapack::syevd_2stage.

Parameters

[in]	jobz	lapack::Job::NoVec: Compute eigenvalues only; lapack::Job::Vec: Compute eigenvalues and eigenvectors. Not yet available (as of LAPACK 3.8.0).
[in]	uplo	lapack::Uplo::Upper: Upper triangle of A is stored; lapack::Uplo::Lower: Lower triangle of A is stored.
[in]	n	The order of the matrix A. n >= 0.
[in,out]	A	The n-by-n matrix A, stored in an lda-by-n array. On entry, the Hermitian matrix A. If uplo = Upper, the leading n-by-n upper triangular part of A contains the upper triangular part of the matrix A. If uplo = Lower, the leading n-by-n lower triangular part of A contains the lower triangular part of the matrix A. On exit, if jobz = Vec, then if successful, A contains the orthonormal eigenvectors of the matrix A. If jobz = NoVec, then on exit the lower triangle (if uplo=Lower) or the upper triangle (if uplo=Upper) of A, including the diagonal, is destroyed.
[in]	lda	The leading dimension of the array A. lda >= max(1,n).
[out]	W	The vector W of length n. If successful, the eigenvalues in ascending order.

Returns: = 0: successful exit; > 0: if return value = i and jobz = NoVec, then the algorithm failed to converge; i off-diagonal elements of an intermediate tridiagonal form did not converge to zero; if return value = i and jobz = Vec, then the algorithm failed to compute an eigenvalue while working on the submatrix lying in rows and columns info/(n+1) through mod(info,n+1).

Further Details

All details about the 2-stage techniques are available in:

Azzam Haidar, Hatem Ltaief, and Jack Dongarra. Parallel reduction to condensed forms for symmetric eigenvalue problems using aggregated fine-grained and memory-aware kernels. In Proceedings of 2011 International Conference for High Performance Computing, Networking, Storage and Analysis (SC '11), New York, NY, USA, Article 8, 11 pages. http://doi.acm.org/10.1145/2063384.2063394

A. Haidar, J. Kurzak, P. Luszczek, 2013. An improved parallel singular value algorithm and its implementation for multicore hardware, In Proceedings of 2013 International Conference for High Performance Computing, Networking, Storage and Analysis (SC '13). Denver, Colorado, USA, 2013. Article 90, 12 pages. http://doi.acm.org/10.1145/2503210.2503292

A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra. A novel hybrid CPU-GPU generalized eigensolver for electronic structure calculations based on fine-grained memory aware tasks. International Journal of High Performance Computing Applications. Volume 28 Issue 2, Pages 196-209, May 2014. http://hpc.sagepub.com/content/28/2/196

◆ heevr()

int64_t lapack::heevr	(	lapack::Job	jobz,
		lapack::Range	range,
		lapack::Uplo	uplo,
		int64_t	n,
		std::complex< double > *	A,
		int64_t	lda,
		double	vl,
		double	vu,
		int64_t	il,
		int64_t	iu,
		double	abstol,
		int64_t *	nfound,
		double *	W,
		std::complex< double > *	Z,
		int64_t	ldz,
		int64_t *	isuppz
	)

Computes selected eigenvalues and, optionally, eigenvectors of a Hermitian matrix A.

Eigenvalues and eigenvectors can be selected by specifying either a range of values or a range of indices for the desired eigenvalues.

heevr first reduces the matrix A to tridiagonal form T with a call to lapack::hetrd. Then, whenever possible, heevr calls lapack::stemr to compute eigenspectrum using Relatively Robust Representations. lapack::stemr computes eigenvalues by the dqds algorithm, while orthogonal eigenvectors are computed from various "good" \(L D L^T\) representations (also known as Relatively Robust Representations). Gram-Schmidt orthogonalization is avoided as far as possible. More specifically, the various steps of the algorithm are as follows.

For each unreduced block (submatrix) of T,

(a) Compute \(T - \sigma I = L D L^T\), so that L and D define all the wanted eigenvalues to high relative accuracy. This means that small relative changes in the entries of D and L cause only small relative changes in the eigenvalues and eigenvectors. The standard (unfactored) representation of the tridiagonal matrix T does not have this property in general.

(b) Compute the eigenvalues to suitable accuracy. If the eigenvectors are desired, the algorithm attains full accuracy of the computed eigenvalues only right before the corresponding vectors have to be computed, see steps c) and d).

(c) For each cluster of close eigenvalues, select a new shift close to the cluster, find a new factorization, and refine the shifted eigenvalues to suitable accuracy.

(d) For each eigenvalue with a large enough relative separation compute the corresponding eigenvector by forming a rank revealing twisted factorization. Go back to (c) for any clusters that remain.

The desired accuracy of the output can be specified by the input parameter abstol.

For more details, see lapack::stemr documentation and:

Inderjit S. Dhillon and Beresford n. Parlett: "Multiple representations to compute orthogonal eigenvectors of symmetric tridiagonal matrices," Linear Algebra and its Applications, 387(1), pp. 1-28, August 2004.
Inderjit Dhillon and Beresford Parlett: "Orthogonal Eigenvectors and Relative Gaps," SIAM Journal on Matrix Analysis and Applications, Vol. 25,
1. Also LAPACK Working Note 154.
Inderjit Dhillon: "A new O(n^2) algorithm for the symmetric tridiagonal eigenvalue/eigenvector problem", Computer Science Division Technical Report No. UCB/CSD-97-971, UC Berkeley, May 1997.

Note 1 : heevr calls lapack::stemr when the full spectrum is requested on machines which conform to the ieee-754 floating point standard. heevr calls lapack::stebz and lapack::stein on non-IEEE machines and when partial spectrum requests are made.

Normal execution of lapack::stemr may create NaNs and infinities and hence may abort due to a floating point exception in environments which do not handle NaNs and infinities in the IEEE standard default manner.

Overloaded versions are available for float, double, std::complex<float>, and std::complex<double>. For real matrices, this is an alias for lapack::syevr.

Parameters

[in]	jobz	lapack::Job::NoVec: Compute eigenvalues only; lapack::Job::Vec: Compute eigenvalues and eigenvectors.
[in]	range	lapack::Range::All: all eigenvalues will be found. lapack::Range::Value: all eigenvalues in the half-open interval (vl,vu] will be found. lapack::Range::Index: the il-th through iu-th eigenvalues will be found. For range = Value or Index and iu - il < n - 1, `lapack::stebz` and `lapack::stein` are called.
[in]	uplo	lapack::Uplo::Upper: Upper triangle of A is stored; lapack::Uplo::Lower: Lower triangle of A is stored.
[in]	n	The order of the matrix A. n >= 0.
[in,out]	A	The n-by-n matrix A, stored in an lda-by-n array. On entry, the Hermitian matrix A. If uplo = Upper, the leading n-by-n upper triangular part of A contains the upper triangular part of the matrix A. If uplo = Lower, the leading n-by-n lower triangular part of A contains the lower triangular part of the matrix A. On exit, the lower triangle (if uplo=Lower) or the upper triangle (if uplo=Upper) of A, including the diagonal, is destroyed.
[in]	lda	The leading dimension of the array A. lda >= max(1,n).
[in]	vl	If range=Value, the lower bound of the interval to be searched for eigenvalues. vl < vu. Not referenced if range = All or Index.
[in]	vu	If range=Value, the upper bound of the interval to be searched for eigenvalues. vl < vu. Not referenced if range = All or Index.
[in]	il	If range=Index, the index of the smallest eigenvalue to be returned. 1 <= il <= iu <= n, if n > 0; il = 1 and iu = 0 if n = 0. Not referenced if range = All or Value.
[in]	iu	If range=Index, the index of the largest eigenvalue to be returned. 1 <= il <= iu <= n, if n > 0; il = 1 and iu = 0 if n = 0. Not referenced if range = All or Value.
[in]	abstol	The absolute error tolerance for the eigenvalues. An approximate eigenvalue is accepted as converged when it is determined to lie in an interval [a,b] of width less than or equal to abstol + eps * max( \|a\|,\|b\| ), where eps is the machine precision. If abstol is less than or equal to zero, then eps*\|T\| will be used in its place, where \|T\| is the 1-norm of the tridiagonal matrix obtained by reducing A to tridiagonal form. See "Computing Small Singular Values of Bidiagonal Matrices with Guaranteed High Relative Accuracy," by Demmel and Kahan, LAPACK Working Note #3. If high relative accuracy is important, set abstol to DLAMCH( 'Safe minimum' ). Doing so will guarantee that eigenvalues are computed to high relative accuracy when possible in future releases. The current code does not make any guarantees about high relative accuracy, but future releases will. See J. Barlow and J. Demmel, "Computing Accurate Eigensystems of Scaled Diagonally Dominant Matrices", LAPACK Working Note #7, for a discussion of which matrices define their eigenvalues to high relative accuracy.
[out]	nfound	The total number of eigenvalues found. 0 <= nfound <= n. If range = All, nfound = n; if range = Index, nfound = iu-il+1.
[out]	W	The vector W of length n. The first nfound elements contain the selected eigenvalues in ascending order.
[out]	Z	The n-by-nfound matrix Z, stored in an ldz-by-zcol array. If jobz = Vec, then if successful, the first nfound columns of Z contain the orthonormal eigenvectors of the matrix A corresponding to the selected eigenvalues, with the i-th column of Z holding the eigenvector associated with W(i). If jobz = NoVec, then Z is not referenced. Note: the user must ensure that zcol >= max(1,nfound) columns are supplied in the array Z; if range = Value, the exact value of nfound is not known in advance and an upper bound must be used.
[in]	ldz	The leading dimension of the array Z. ldz >= 1, and if jobz = Vec, ldz >= max(1,n).
[out]	isuppz	The vector isuppz of length 2max(1,nfound). The support of the eigenvectors in Z, i.e., the indices indicating the nonzero elements in Z. The i-th eigenvector is nonzero only in elements isuppz( 2i-1 ) through isuppz( 2*i ). This is an output of `lapack::stemr` (tridiagonal matrix). The support of the eigenvectors of A is typically 1:n because of the unitary transformations applied by `lapack::unmtr`. Implemented only for range = All or Index and iu - il = n - 1

Returns: = 0: successful exit; > 0: Internal error

◆ heevr_2stage()

int64_t lapack::heevr_2stage	(	lapack::Job	jobz,
		lapack::Range	range,
		lapack::Uplo	uplo,
		int64_t	n,
		std::complex< double > *	A,
		int64_t	lda,
		double	vl,
		double	vu,
		int64_t	il,
		int64_t	iu,
		double	abstol,
		int64_t *	nfound,
		double *	W,
		std::complex< double > *	Z,
		int64_t	ldz,
		int64_t *	isuppz
	)

Computes selected eigenvalues and, optionally, eigenvectors of a Hermitian matrix A using the 2-stage technique for the reduction to tridiagonal.

Eigenvalues and eigenvectors can be selected by specifying either a range of values or a range of indices for the desired eigenvalues.

heevr_2stage first reduces the matrix A to tridiagonal form T with a call to lapack::hetrd. Then, whenever possible, heevr_2stage calls lapack::stemr to compute eigenspectrum using Relatively Robust Representations. lapack::stemr computes eigenvalues by the dqds algorithm, while orthogonal eigenvectors are computed from various "good" \(L D L^T\) representations (also known as Relatively Robust Representations). Gram-Schmidt orthogonalization is avoided as far as possible. More specifically, the various steps of the algorithm are as follows.

For each unreduced block (submatrix) of T,

(a) Compute \(T - \sigma I = L D L^T\), so that L and D define all the wanted eigenvalues to high relative accuracy. This means that small relative changes in the entries of D and L cause only small relative changes in the eigenvalues and eigenvectors. The standard (unfactored) representation of the tridiagonal matrix T does not have this property in general.

(b) Compute the eigenvalues to suitable accuracy. If the eigenvectors are desired, the algorithm attains full accuracy of the computed eigenvalues only right before the corresponding vectors have to be computed, see steps c) and d).

(c) For each cluster of close eigenvalues, select a new shift close to the cluster, find a new factorization, and refine the shifted eigenvalues to suitable accuracy.

(d) For each eigenvalue with a large enough relative separation compute the corresponding eigenvector by forming a rank revealing twisted factorization. Go back to (c) for any clusters that remain.

The desired accuracy of the output can be specified by the input parameter abstol.

For more details, see lapack::stemr documentation and:

Inderjit S. Dhillon and Beresford n. Parlett: "Multiple representations to compute orthogonal eigenvectors of symmetric tridiagonal matrices," Linear Algebra and its Applications, 387(1), pp. 1-28, August 2004.
Inderjit Dhillon and Beresford Parlett: "Orthogonal Eigenvectors and Relative Gaps," SIAM Journal on Matrix Analysis and Applications, Vol. 25,
1. Also LAPACK Working Note 154.
Inderjit Dhillon: "A new O(n^2) algorithm for the symmetric tridiagonal eigenvalue/eigenvector problem", Computer Science Division Technical Report No. UCB/CSD-97-971, UC Berkeley, May 1997.

Note 1 : heevr_2stage calls lapack::stemr when the full spectrum is requested on machines which conform to the ieee-754 floating point standard. heevr_2stage calls lapack::stebz and lapack::stein on non-IEEE machines and when partial spectrum requests are made.

Normal execution of lapack::stemr may create NaNs and infinities and hence may abort due to a floating point exception in environments which do not handle NaNs and infinities in the IEEE standard default manner.

Overloaded versions are available for float, double, std::complex<float>, and std::complex<double>. For real matrices, this is an alias for lapack::syevr_2stage.

Parameters

[in]	jobz	lapack::Job::NoVec: Compute eigenvalues only; lapack::Job::Vec: Compute eigenvalues and eigenvectors. Not yet available (as of LAPACK 3.8.0).
[in]	range	lapack::Range::All: all eigenvalues will be found. lapack::Range::Value: all eigenvalues in the half-open interval (vl,vu] will be found. lapack::Range::Index: the il-th through iu-th eigenvalues will be found. For range = Value or Index and iu - il < n - 1, `lapack::stebz` and `lapack::stein` are called.
[in]	uplo	lapack::Uplo::Upper: Upper triangle of A is stored; lapack::Uplo::Lower: Lower triangle of A is stored.
[in]	n	The order of the matrix A. n >= 0.
[in,out]	A	The n-by-n matrix A, stored in an lda-by-n array. On entry, the Hermitian matrix A. If uplo = Upper, the leading n-by-n upper triangular part of A contains the upper triangular part of the matrix A. If uplo = Lower, the leading n-by-n lower triangular part of A contains the lower triangular part of the matrix A. On exit, the lower triangle (if uplo=Lower) or the upper triangle (if uplo=Upper) of A, including the diagonal, is destroyed.
[in]	lda	The leading dimension of the array A. lda >= max(1,n).
[in]	vl	If range=Value, the lower bound of the interval to be searched for eigenvalues. vl < vu. Not referenced if range = All or Index.
[in]	vu	If range=Value, the upper bound of the interval to be searched for eigenvalues. vl < vu. Not referenced if range = All or Index.
[in]	il	If range=Index, the index of the smallest eigenvalue to be returned. 1 <= il <= iu <= n, if n > 0; il = 1 and iu = 0 if n = 0. Not referenced if range = All or Value.
[in]	iu	If range=Index, the index of the largest eigenvalue to be returned. 1 <= il <= iu <= n, if n > 0; il = 1 and iu = 0 if n = 0. Not referenced if range = All or Value.
[in]	abstol	The absolute error tolerance for the eigenvalues. An approximate eigenvalue is accepted as converged when it is determined to lie in an interval [a,b] of width less than or equal to abstol + eps * max( \|a\|,\|b\| ), where eps is the machine precision. If abstol is less than or equal to zero, then eps*\|T\| will be used in its place, where \|T\| is the 1-norm of the tridiagonal matrix obtained by reducing A to tridiagonal form. See "Computing Small Singular Values of Bidiagonal Matrices with Guaranteed High Relative Accuracy," by Demmel and Kahan, LAPACK Working Note #3. If high relative accuracy is important, set abstol to DLAMCH( 'Safe minimum' ). Doing so will guarantee that eigenvalues are computed to high relative accuracy when possible in future releases. The current code does not make any guarantees about high relative accuracy, but future releases will. See J. Barlow and J. Demmel, "Computing Accurate Eigensystems of Scaled Diagonally Dominant Matrices", LAPACK Working Note #7, for a discussion of which matrices define their eigenvalues to high relative accuracy.
[out]	nfound	The total number of eigenvalues found. 0 <= nfound <= n. If range = All, nfound = n; if range = Index, nfound = iu-il+1.
[out]	W	The vector W of length n. The first nfound elements contain the selected eigenvalues in ascending order.
[out]	Z	The n-by-nfound matrix Z, stored in an ldz-by-zcol array. If jobz = Vec, then if successful, the first nfound columns of Z contain the orthonormal eigenvectors of the matrix A corresponding to the selected eigenvalues, with the i-th column of Z holding the eigenvector associated with W(i). If jobz = NoVec, then Z is not referenced. Note: the user must ensure that zcol >= max(1,nfound) columns are supplied in the array Z; if range = Value, the exact value of nfound is not known in advance and an upper bound must be used.
[in]	ldz	The leading dimension of the array Z. ldz >= 1, and if jobz = Vec, ldz >= max(1,n).
[out]	isuppz	The vector isuppz of length 2max(1,nfound). The support of the eigenvectors in Z, i.e., the indices indicating the nonzero elements in Z. The i-th eigenvector is nonzero only in elements isuppz( 2i-1 ) through isuppz( 2*i ). This is an output of `lapack::stemr` (tridiagonal matrix). The support of the eigenvectors of A is typically 1:n because of the unitary transformations applied by `lapack::unmtr`. Implemented only for range = All or Index and iu - il = n - 1

Returns: = 0: successful exit; > 0: Internal error

Further Details

All details about the 2-stage techniques are available in:

Azzam Haidar, Hatem Ltaief, and Jack Dongarra. Parallel reduction to condensed forms for symmetric eigenvalue problems using aggregated fine-grained and memory-aware kernels. In Proceedings of 2011 International Conference for High Performance Computing, Networking, Storage and Analysis (SC '11), New York, NY, USA, Article 8, 11 pages. http://doi.acm.org/10.1145/2063384.2063394

A. Haidar, J. Kurzak, P. Luszczek, 2013. An improved parallel singular value algorithm and its implementation for multicore hardware, In Proceedings of 2013 International Conference for High Performance Computing, Networking, Storage and Analysis (SC '13). Denver, Colorado, USA, 2013. Article 90, 12 pages. http://doi.acm.org/10.1145/2503210.2503292

A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra. A novel hybrid CPU-GPU generalized eigensolver for electronic structure calculations based on fine-grained memory aware tasks. International Journal of High Performance Computing Applications. Volume 28 Issue 2, Pages 196-209, May 2014. http://hpc.sagepub.com/content/28/2/196

◆ heevx()

int64_t lapack::heevx	(	lapack::Job	jobz,
		lapack::Range	range,
		lapack::Uplo	uplo,
		int64_t	n,
		std::complex< double > *	A,
		int64_t	lda,
		double	vl,
		double	vu,
		int64_t	il,
		int64_t	iu,
		double	abstol,
		int64_t *	nfound,
		double *	W,
		std::complex< double > *	Z,
		int64_t	ldz,
		int64_t *	ifail
	)

Computes selected eigenvalues and, optionally, eigenvectors of a Hermitian matrix A.

Eigenvalues and eigenvectors can be selected by specifying either a range of values or a range of indices for the desired eigenvalues.

Overloaded versions are available for float, double, std::complex<float>, and std::complex<double>.

Parameters

[in]	jobz	lapack::Job::NoVec: Compute eigenvalues only; lapack::Job::Vec: Compute eigenvalues and eigenvectors.
[in]	range	lapack::Range::All: all eigenvalues will be found. lapack::Range::Value: all eigenvalues in the half-open interval (vl,vu] will be found. lapack::Range::Index: the il-th through iu-th eigenvalues will be found.
[in]	uplo	lapack::Uplo::Upper: Upper triangle of A is stored; lapack::Uplo::Lower: Lower triangle of A is stored.
[in]	n	The order of the matrix A. n >= 0.
[in,out]	A	The n-by-n matrix A, stored in an lda-by-n array. On entry, the Hermitian matrix A. If uplo = Upper, the leading n-by-n upper triangular part of A contains the upper triangular part of the matrix A. If uplo = Lower, the leading n-by-n lower triangular part of A contains the lower triangular part of the matrix A. On exit, the lower triangle (if uplo=Lower) or the upper triangle (if uplo=Upper) of A, including the diagonal, is destroyed.
[in]	lda	The leading dimension of the array A. lda >= max(1,n).
[in]	vl	If range=Value, the lower bound of the interval to be searched for eigenvalues. vl < vu. Not referenced if range = All or Index.
[in]	vu	If range=Value, the upper bound of the interval to be searched for eigenvalues. vl < vu. Not referenced if range = All or Index.
[in]	il	If range=Index, the index of the smallest eigenvalue to be returned. 1 <= il <= iu <= n, if n > 0; il = 1 and iu = 0 if n = 0. Not referenced if range = All or Value.
[in]	iu	If range=Index, the index of the largest eigenvalue to be returned. 1 <= il <= iu <= n, if n > 0; il = 1 and iu = 0 if n = 0. Not referenced if range = All or Value.
[in]	abstol	The absolute error tolerance for the eigenvalues. An approximate eigenvalue is accepted as converged when it is determined to lie in an interval [a,b] of width less than or equal to abstol + eps * max(\|a\|, \|b\|), where eps is the machine precision. If abstol is less than or equal to zero, then eps\|T\| will be used in its place, where \|T\| is the 1-norm of the tridiagonal matrix obtained by reducing A to tridiagonal form. Eigenvalues will be computed most accurately when abstol is set to twice the underflow threshold 2DLAMCH('S'), not zero. If this routine returns with return value > 0, indicating that some eigenvectors did not converge, try setting abstol to 2*DLAMCH('S'). See "Computing Small Singular Values of Bidiagonal Matrices with Guaranteed High Relative Accuracy," by Demmel and Kahan, LAPACK Working Note #3.
[out]	nfound	The total number of eigenvalues found. 0 <= nfound <= n. If range = All, nfound = n; if range = Index, nfound = iu-il+1.
[out]	W	The vector W of length n. On normal exit, the first nfound elements contain the selected eigenvalues in ascending order.
[out]	Z	The n-by-nfound matrix Z, stored in an ldz-by-zcol array. If jobz = Vec, then if successful, the first nfound columns of Z contain the orthonormal eigenvectors of the matrix A corresponding to the selected eigenvalues, with the i-th column of Z holding the eigenvector associated with W(i). If an eigenvector fails to converge, then that column of Z contains the latest approximation to the eigenvector, and the index of the eigenvector is returned in ifail. If jobz = NoVec, then Z is not referenced. Note: the user must ensure that zcol >= max(1,nfound) columns are supplied in the array Z; if range = Value, the exact value of nfound is not known in advance and an upper bound must be used.
[in]	ldz	The leading dimension of the array Z. ldz >= 1, and if jobz = Vec, ldz >= max(1,n).
[out]	ifail	The vector ifail of length n. If jobz = Vec, then if successful, the first nfound elements of ifail are zero. If return value > 0, then ifail contains the indices of the eigenvectors that failed to converge. If jobz = NoVec, then ifail is not referenced.

Returns: = 0: successful exit; > 0: if return value = i, then i eigenvectors failed to converge. Their indices are stored in array ifail.

◆ heevx_2stage()

int64_t lapack::heevx_2stage	(	lapack::Job	jobz,
		lapack::Range	range,
		lapack::Uplo	uplo,
		int64_t	n,
		std::complex< double > *	A,
		int64_t	lda,
		double	vl,
		double	vu,
		int64_t	il,
		int64_t	iu,
		double	abstol,
		int64_t *	nfound,
		double *	W,
		std::complex< double > *	Z,
		int64_t	ldz,
		int64_t *	ifail
	)

Computes selected eigenvalues and, optionally, eigenvectors of a Hermitian matrix A using the 2-stage technique for the reduction to tridiagonal.

Eigenvalues and eigenvectors can be selected by specifying either a range of values or a range of indices for the desired eigenvalues.

Overloaded versions are available for float, double, std::complex<float>, and std::complex<double>.

Parameters

[in]	jobz	lapack::Job::NoVec: Compute eigenvalues only; lapack::Job::Vec: Compute eigenvalues and eigenvectors. Not yet available (as of LAPACK 3.8.0).
[in]	range	lapack::Range::All: all eigenvalues will be found. lapack::Range::Value: all eigenvalues in the half-open interval (vl,vu] will be found. lapack::Range::Index: the il-th through iu-th eigenvalues will be found.
[in]	uplo	lapack::Uplo::Upper: Upper triangle of A is stored; lapack::Uplo::Lower: Lower triangle of A is stored.
[in]	n	The order of the matrix A. n >= 0.
[in,out]	A	The n-by-n matrix A, stored in an lda-by-n array. On entry, the Hermitian matrix A. If uplo = Upper, the leading n-by-n upper triangular part of A contains the upper triangular part of the matrix A. If uplo = Lower, the leading n-by-n lower triangular part of A contains the lower triangular part of the matrix A. On exit, the lower triangle (if uplo=Lower) or the upper triangle (if uplo=Upper) of A, including the diagonal, is destroyed.
[in]	lda	The leading dimension of the array A. lda >= max(1,n).
[in]	vl	If range=Value, the lower bound of the interval to be searched for eigenvalues. vl < vu. Not referenced if range = All or Index.
[in]	vu	If range=Value, the upper bound of the interval to be searched for eigenvalues. vl < vu. Not referenced if range = All or Index.
[in]	il	If range=Index, the index of the smallest eigenvalue to be returned. 1 <= il <= iu <= n, if n > 0; il = 1 and iu = 0 if n = 0. Not referenced if range = All or Value.
[in]	iu	If range=Index, the index of the largest eigenvalue to be returned. 1 <= il <= iu <= n, if n > 0; il = 1 and iu = 0 if n = 0. Not referenced if range = All or Value.
[in]	abstol	The absolute error tolerance for the eigenvalues. An approximate eigenvalue is accepted as converged when it is determined to lie in an interval [a,b] of width less than or equal to abstol + eps * max(\|a\|, \|b\|), where eps is the machine precision. If abstol is less than or equal to zero, then eps\|T\| will be used in its place, where \|T\| is the 1-norm of the tridiagonal matrix obtained by reducing A to tridiagonal form. Eigenvalues will be computed most accurately when abstol is set to twice the underflow threshold 2DLAMCH('S'), not zero. If this routine returns with return value > 0, indicating that some eigenvectors did not converge, try setting abstol to 2*DLAMCH('S'). See "Computing Small Singular Values of Bidiagonal Matrices with Guaranteed High Relative Accuracy," by Demmel and Kahan, LAPACK Working Note #3.
[out]	nfound	The total number of eigenvalues found. 0 <= nfound <= n. If range = All, nfound = n; if range = Index, nfound = iu-il+1.
[out]	W	The vector W of length n. On normal exit, the first nfound elements contain the selected eigenvalues in ascending order.
[out]	Z	The n-by-nfound matrix Z, stored in an ldz-by-zcol array. If jobz = Vec, then if successful, the first nfound columns of Z contain the orthonormal eigenvectors of the matrix A corresponding to the selected eigenvalues, with the i-th column of Z holding the eigenvector associated with W(i). If an eigenvector fails to converge, then that column of Z contains the latest approximation to the eigenvector, and the index of the eigenvector is returned in ifail. If jobz = NoVec, then Z is not referenced. Note: the user must ensure that zcol >= max(1,nfound) columns are supplied in the array Z; if range = Value, the exact value of nfound is not known in advance and an upper bound must be used.
[in]	ldz	The leading dimension of the array Z. ldz >= 1, and if jobz = Vec, ldz >= max(1,n).
[out]	ifail	The vector ifail of length n. If jobz = Vec, then if successful, the first nfound elements of ifail are zero. If return value > 0, then ifail contains the indices of the eigenvectors that failed to converge. If jobz = NoVec, then ifail is not referenced.

Returns: = 0: successful exit; > 0: if return value = i, then i eigenvectors failed to converge. Their indices are stored in array ifail.

Further Details

All details about the 2-stage techniques are available in:

Azzam Haidar, Hatem Ltaief, and Jack Dongarra. Parallel reduction to condensed forms for symmetric eigenvalue problems using aggregated fine-grained and memory-aware kernels. In Proceedings of 2011 International Conference for High Performance Computing, Networking, Storage and Analysis (SC '11), New York, NY, USA, Article 8, 11 pages. http://doi.acm.org/10.1145/2063384.2063394

A. Haidar, J. Kurzak, P. Luszczek, 2013. An improved parallel singular value algorithm and its implementation for multicore hardware, In Proceedings of 2013 International Conference for High Performance Computing, Networking, Storage and Analysis (SC '13). Denver, Colorado, USA, 2013. Article 90, 12 pages. http://doi.acm.org/10.1145/2503210.2503292

A. Haidar, R. Solca, S. Tomov, T. Schulthess and J. Dongarra. A novel hybrid CPU-GPU generalized eigensolver for electronic structure calculations based on fine-grained memory aware tasks. International Journal of High Performance Computing Applications. Volume 28 Issue 2, Pages 196-209, May 2014. http://hpc.sagepub.com/content/28/2/196

◆ syev()

int64_t lapack::syev	(	lapack::Job	jobz,
		lapack::Uplo	uplo,
		int64_t	n,
		double *	A,
		int64_t	lda,
		double *	W
	)

See also: lapack::heev

◆ syev_2stage()

int64_t lapack::syev_2stage	(	lapack::Job	jobz,
		lapack::Uplo	uplo,
		int64_t	n,
		double *	A,
		int64_t	lda,
		double *	W
	)

See also: lapack::heev_2stage

◆ syevd()

int64_t lapack::syevd	(	lapack::Job	jobz,
		lapack::Uplo	uplo,
		int64_t	n,
		double *	A,
		int64_t	lda,
		double *	W
	)

See also: lapack::heevd

◆ syevd_2stage()

int64_t lapack::syevd_2stage	(	lapack::Job	jobz,
		lapack::Uplo	uplo,
		int64_t	n,
		double *	A,
		int64_t	lda,
		double *	W
	)

See also: lapack::heevd_2stage

◆ syevr()

int64_t lapack::syevr	(	lapack::Job	jobz,
		lapack::Range	range,
		lapack::Uplo	uplo,
		int64_t	n,
		double *	A,
		int64_t	lda,
		double	vl,
		double	vu,
		int64_t	il,
		int64_t	iu,
		double	abstol,
		int64_t *	nfound,
		double *	W,
		double *	Z,
		int64_t	ldz,
		int64_t *	isuppz
	)

See also: lapack::heevr

◆ syevr_2stage()

int64_t lapack::syevr_2stage	(	lapack::Job	jobz,
		lapack::Range	range,
		lapack::Uplo	uplo,
		int64_t	n,
		double *	A,
		int64_t	lda,
		double	vl,
		double	vu,
		int64_t	il,
		int64_t	iu,
		double	abstol,
		int64_t *	nfound,
		double *	W,
		double *	Z,
		int64_t	ldz,
		int64_t *	isuppz
	)

See also: lapack::heevr_2stage

◆ syevx()

int64_t lapack::syevx	(	lapack::Job	jobz,
		lapack::Range	range,
		lapack::Uplo	uplo,
		int64_t	n,
		double *	A,
		int64_t	lda,
		double	vl,
		double	vu,
		int64_t	il,
		int64_t	iu,
		double	abstol,
		int64_t *	nfound,
		double *	W,
		double *	Z,
		int64_t	ldz,
		int64_t *	ifail
	)

See also: lapack::heevx

◆ syevx_2stage()

int64_t lapack::syevx_2stage	(	lapack::Job	jobz,
		lapack::Range	range,
		lapack::Uplo	uplo,
		int64_t	n,
		double *	A,
		int64_t	lda,
		double	vl,
		double	vu,
		int64_t	il,
		int64_t	iu,
		double	abstol,
		int64_t *	nfound,
		double *	W,
		double *	Z,
		int64_t	ldz,
		int64_t *	ifail
	)

See also: lapack::heevx_2stage

Functions

Detailed Description

Function Documentation

◆ heev()

◆ heev_2stage()

◆ heevd()

◆ heevd_2stage()

◆ heevr()

◆ heevr_2stage()

◆ heevx()

◆ heevx_2stage()

◆ syev()

◆ syev_2stage()

◆ syevd()

◆ syevd_2stage()

◆ syevr()

◆ syevr_2stage()

◆ syevx()

◆ syevx_2stage()