scipy.sparse.linalg.eigsh

scipy.sparse.linalg.eigsh(A, k=6, M=None, sigma=None, which='LM', v0=None, ncv=None, maxiter=None, tol=0, return_eigenvectors=True, Minv=None, OPinv=None, mode='normal')

Find k eigenvalues and eigenvectors of the real symmetric square matrix or complex Hermitian matrix A.

Solves A * x[i] = w[i] * x[i], the standard eigenvalue problem for w[i] eigenvalues with corresponding eigenvectors x[i].

If M is specified, solves A * x[i] = w[i] * M * x[i], the generalized eigenvalue problem for w[i] eigenvalues with corresponding eigenvectors x[i].

Note that there is no specialized routine for the case when A is a complex Hermitian matrix. In this case, eigsh() will call eigs() and return the real parts of the eigenvalues thus obtained.

Parameters

Andarray, sparse matrix or LinearOperator

A square operator representing the operation A * x, where A is real symmetric or complex Hermitian. For buckling mode (see below) A must additionally be positive-definite.

kint, optional

The number of eigenvalues and eigenvectors desired. k must be smaller than N. It is not possible to compute all eigenvectors of a matrix.

Returns

warray

Array of k eigenvalues.

varray

An array representing the k eigenvectors. The column v[:, i] is the eigenvector corresponding to the eigenvalue w[i].

Other Parameters

MAn N x N matrix, array, sparse matrix, or linear operator representing

the operation M @ x for the generalized eigenvalue problem

A @ x = w * M @ x.

M must represent a real symmetric matrix if A is real, and must represent a complex Hermitian matrix if A is complex. For best results, the data type of M should be the same as that of A. Additionally:

If sigma is None, M is symmetric positive definite.

If sigma is specified, M is symmetric positive semi-definite.

In buckling mode, M is symmetric indefinite.

If sigma is None, eigsh requires an operator to compute the solution of the linear equation M @ x = b. This is done internally via a (sparse) LU decomposition for an explicit matrix M, or via an iterative solver for a general linear operator. Alternatively, the user can supply the matrix or operator Minv, which gives x = Minv @ b = M^-1 @ b.

sigmareal

Find eigenvalues near sigma using shift-invert mode. This requires an operator to compute the solution of the linear system [A - sigma * M] x = b, where M is the identity matrix if unspecified. This is computed internally via a (sparse) LU decomposition for explicit matrices A & M, or via an iterative solver if either A or M is a general linear operator. Alternatively, the user can supply the matrix or operator OPinv, which gives x = OPinv @ b = [A - sigma * M]^-1 @ b. Note that when sigma is specified, the keyword ‘which’ refers to the shifted eigenvalues w'[i] where:

if mode == ‘normal’, w'[i] = 1 / (w[i] - sigma).

if mode == ‘cayley’, w'[i] = (w[i] + sigma) / (w[i] - sigma).

if mode == ‘buckling’, w'[i] = w[i] / (w[i] - sigma).

(see further discussion in ‘mode’ below)

v0ndarray, optional

Starting vector for iteration. Default: random

ncvint, optional

The number of Lanczos vectors generated ncv must be greater than k and smaller than n; it is recommended that ncv > 2*k. Default: min(n, max(2*k + 1, 20))

whichstr [‘LM’ | ‘SM’ | ‘LA’ | ‘SA’ | ‘BE’]

If A is a complex Hermitian matrix, ‘BE’ is invalid. Which k eigenvectors and eigenvalues to find:

‘LM’ : Largest (in magnitude) eigenvalues.

‘SM’ : Smallest (in magnitude) eigenvalues.

‘LA’ : Largest (algebraic) eigenvalues.

‘SA’ : Smallest (algebraic) eigenvalues.

‘BE’ : Half (k/2) from each end of the spectrum.

When k is odd, return one more (k/2+1) from the high end. When sigma != None, ‘which’ refers to the shifted eigenvalues w'[i] (see discussion in ‘sigma’, above). ARPACK is generally better at finding large values than small values. If small eigenvalues are desired, consider using shift-invert mode for better performance.

maxiterint, optional

Maximum number of Arnoldi update iterations allowed. Default: n*10

tolfloat

Relative accuracy for eigenvalues (stopping criterion). The default value of 0 implies machine precision.

MinvN x N matrix, array, sparse matrix, or LinearOperator

See notes in M, above.

OPinvN x N matrix, array, sparse matrix, or LinearOperator

See notes in sigma, above.

return_eigenvectorsbool

Return eigenvectors (True) in addition to eigenvalues. This value determines the order in which eigenvalues are sorted. The sort order is also dependent on the which variable.

For which = ‘LM’ or ‘SA’:

If return_eigenvectors is True, eigenvalues are sorted by algebraic value.

If return_eigenvectors is False, eigenvalues are sorted by absolute value.

For which = ‘BE’ or ‘LA’:

eigenvalues are always sorted by algebraic value.

For which = ‘SM’:

If return_eigenvectors is True, eigenvalues are sorted by algebraic value.

If return_eigenvectors is False, eigenvalues are sorted by decreasing absolute value.

modestring [‘normal’ | ‘buckling’ | ‘cayley’]

Specify strategy to use for shift-invert mode. This argument applies only for real-valued A and sigma != None. For shift-invert mode, ARPACK internally solves the eigenvalue problem OP * x'[i] = w'[i] * B * x'[i] and transforms the resulting Ritz vectors x’[i] and Ritz values w’[i] into the desired eigenvectors and eigenvalues of the problem A * x[i] = w[i] * M * x[i]. The modes are as follows:

‘normal’ :

OP = [A - sigma * M]^-1 @ M, B = M, w’[i] = 1 / (w[i] - sigma)

‘buckling’ :

OP = [A - sigma * M]^-1 @ A, B = A, w’[i] = w[i] / (w[i] - sigma)

‘cayley’ :

OP = [A - sigma * M]^-1 @ [A + sigma * M], B = M, w’[i] = (w[i] + sigma) / (w[i] - sigma)

The choice of mode will affect which eigenvalues are selected by the keyword ‘which’, and can also impact the stability of convergence (see [2] for a discussion).

Raises

ArpackNoConvergence

When the requested convergence is not obtained.

The currently converged eigenvalues and eigenvectors can be found as eigenvalues and eigenvectors attributes of the exception object.

See also

eigs

eigenvalues and eigenvectors for a general (nonsymmetric) matrix A

svds

singular value decomposition for a matrix A

Notes

This function is a wrapper to the ARPACK [1] SSEUPD and DSEUPD functions which use the Implicitly Restarted Lanczos Method to find the eigenvalues and eigenvectors [2].

References

1

ARPACK Software, http://www.caam.rice.edu/software/ARPACK/

2

R. B. Lehoucq, D. C. Sorensen, and C. Yang, ARPACK USERS GUIDE: Solution of Large Scale Eigenvalue Problems by Implicitly Restarted Arnoldi Methods. SIAM, Philadelphia, PA, 1998.

Examples

>>> from scipy.sparse.linalg import eigsh
>>> identity = np.eye(13)
>>> eigenvalues, eigenvectors = eigsh(identity, k=6)
>>> eigenvalues
array([1., 1., 1., 1., 1., 1.])
>>> eigenvectors.shape
(13, 6)