NAG Library Routine Document
f12abf (real_iter)
Note: this routine uses optional parameters to define choices in the problem specification. If you wish to use default
settings for all of the optional parameters, then the option setting routine f12adf need not be called.
If, however, you wish to reset some or all of the settings please refer to Section 11 in f12adf for a detailed description of the specification of the optional parameters.
1
Purpose
f12abf is an iterative solver used to find some of the eigenvalues (and optionally the corresponding eigenvectors) of a standard or generalized eigenvalue problem defined by real nonsymmetric matrices. This is part of a suite of routines that also includes
f12aaf,
f12acf,
f12adf and
f12aef. It is
2
Specification
Fortran Interface
Integer, Intent (In)  ::  ldv  Integer, Intent (Inout)  ::  irevcm, icomm(*), ifail  Integer, Intent (Out)  ::  nshift  Real (Kind=nag_wp), Intent (Inout)  ::  resid(*), v(ldv,*), x(*), mx(*), comm(*) 

3
Description
The suite of routines is designed to calculate some of the eigenvalues, $\lambda $, (and optionally the corresponding eigenvectors, $x$) of a standard eigenvalue problem $Ax=\lambda x$, or of a generalized eigenvalue problem $Ax=\lambda Bx$ of order $n$, where $n$ is large and the coefficient matrices $A$ and $B$ are sparse, real and nonsymmetric. The suite can also be used to find selected eigenvalues/eigenvectors of smaller scale dense, real and nonsymmetric problems.
f12abf is a
reverse communication routine, based on the ARPACK routine
dnaupd, using the Implicitly Restarted Arnoldi iteration method. The method is described in
Lehoucq and Sorensen (1996) and
Lehoucq (2001) while its use within the ARPACK software is described in great detail in
Lehoucq et al. (1998). An evaluation of software for computing eigenvalues of sparse nonsymmetric matrices is provided in
Lehoucq and Scott (1996). This suite of routines offers the same functionality as the ARPACK software for real nonsymmetric problems, but the interface design is quite different in order to make the option setting clearer and to simplify the interface of
f12abf.
The setup routine
f12aaf must be called before
f12abf, the reverse communication iterative solver. Options may be set for
f12abf by prior calls to the option setting routine
f12adf and a postprocessing routine
f12acf must be called following a successful final exit from
f12abf.
f12aef, may be called following certain flagged, intermediate exits from
f12abf to provide additional monitoring information about the computation.
f12abf uses
reverse communication, i.e., it returns repeatedly to the calling program with the argument
irevcm (see
Section 5) set to specified values which require the calling program to carry out one of the following tasks:
 – compute the matrixvector product $y=\mathrm{OP}x$, where $\mathrm{OP}$ is defined by the computational mode;
 – compute the matrixvector product $y=Bx$;
 – notify the completion of the computation;
 – allow the calling program to monitor the solution.
The problem type to be solved (standard or generalized), the spectrum of eigenvalues of interest, the mode used (regular, regular inverse, shifted inverse, shifted real or shifted imaginary) and other options can all be set using the option setting routine
f12adf (see
Section 11.1 in
f12adf for details on setting options and of the default settings).
4
References
Lehoucq R B (2001) Implicitly restarted Arnoldi methods and subspace iteration SIAM Journal on Matrix Analysis and Applications 23 551–562
Lehoucq R B and Scott J A (1996) An evaluation of software for computing eigenvalues of sparse nonsymmetric matrices Preprint MCSP5471195 Argonne National Laboratory
Lehoucq R B and Sorensen D C (1996) Deflation techniques for an implicitly restarted Arnoldi iteration SIAM Journal on Matrix Analysis and Applications 17 789–821
Lehoucq R B, Sorensen D C and Yang C (1998) ARPACK Users' Guide: Solution of Largescale Eigenvalue Problems with Implicitly Restarted Arnoldi Methods SIAM, Philidelphia
5
Arguments
Note: this routine uses
reverse communication. Its use involves an initial entry, intermediate exits and reentries, and a final exit, as indicated by the argument
irevcm. Between intermediate exits and reentries,
all arguments other than x, mx and comm must remain unchanged.
 1: $\mathbf{irevcm}$ – IntegerInput/Output

On initial entry: ${\mathbf{irevcm}}=0$, otherwise an error condition will be raised.
On intermediate reentry: must be unchanged from its previous exit value. Changing
irevcm to any other value between calls will result in an error.
On intermediate exit:
has the following meanings.
 ${\mathbf{irevcm}}=1$
 The calling program must compute the matrixvector product $y=\mathrm{OP}x$, where $x$ is stored in x (by default) or in the array comm (starting from the location given by the first element of icomm) when the option ${\mathbf{Pointers}}=\mathrm{YES}$ is set in a prior call to f12adf. The result $y$ is returned in x (by default) or in the array comm (starting from the location given by the second element of icomm) when the option ${\mathbf{Pointers}}=\mathrm{YES}$ is set.
If $B$ is not symmetric semidefinite then the precomputed values in mx should not be used (see the explanation under ${\mathbf{irevcm}}=2$).
 ${\mathbf{irevcm}}=1$
 The calling program must compute the matrixvector product $y=\mathrm{OP}x$. This is similar to the case ${\mathbf{irevcm}}=1$ except that the result of the matrixvector product $Bx$ (as required in some computational modes) has already been computed and is available in mx (by default) or in the array comm (starting from the location given by the third element of icomm) when the option ${\mathbf{Pointers}}=\mathrm{YES}$ is set.
 ${\mathbf{irevcm}}=2$
 The calling program must compute the matrixvector product $y=Bx$, where $x$ is stored as described in the case ${\mathbf{irevcm}}=1$ and $y$ is returned in the location described by the case ${\mathbf{irevcm}}=1$.
This computation is requested when solving the Generalized problem using either Shifted Inverse Imaginary or Shifted Inverse Real; in these cases $B$ is used as an innerproduct space and requires that $B$ be symmetric semidefinite. If neither $A$ nor $B$ is symmetric semidefinite then the problem should be reformulated in a Standard form.
 ${\mathbf{irevcm}}=3$
 Compute the nshift real and imaginary parts of the shifts where the real parts are to be returned in the first nshift locations of the array x and the imaginary parts are to be returned in the first nshift locations of the array mx. Only complex conjugate pairs of shifts may be applied and the pairs must be placed in consecutive locations. This value of irevcm will only arise if the optional parameter Supplied Shifts is set in a prior call to f12adf which is intended for experienced users only; the default and recommended option is to use exact shifts (see Lehoucq et al. (1998) for details).
 ${\mathbf{irevcm}}=4$
 Monitoring step: a call to f12aef can now be made to return the number of Arnoldi iterations, the number of converged Ritz values, their real and imaginary parts, and the corresponding Ritz estimates.
On final exit:
${\mathbf{irevcm}}=5$:
f12abf has completed its tasks. The value of
ifail determines whether the iteration has been successfully completed, or whether errors have been detected. On successful completion
f12acf must be called to return the requested eigenvalues and eigenvectors (and/or Schur vectors).
Constraint:
on initial entry,
${\mathbf{irevcm}}=0$; on reentry
irevcm must remain unchanged.
Note: any values you return to f12abf as part of the reverse communication procedure should not include floatingpoint NaN (Not a Number) or infinity values, since these are not handled by f12abf. If your code does inadvertently return any NaNs or infinities, f12abf is likely to produce unexpected results.
 2: $\mathbf{resid}\left(*\right)$ – Real (Kind=nag_wp) arrayInput/Output

Note: the dimension of the array
resid
must be at least
${\mathbf{n}}$ (see
f12aaf).
On initial entry: need not be set unless the option
Initial Residual has been set in a prior call to
f12adf in which case
resid should contain an initial residual vector, possibly from a previous run.
On intermediate reentry: must be unchanged from its previous exit. Changing
resid to any other value between calls may result in an error exit.
On intermediate exit:
contains the current residual vector.
On final exit: contains the final residual vector.
 3: $\mathbf{v}\left({\mathbf{ldv}},*\right)$ – Real (Kind=nag_wp) arrayInput/Output

Note: the second dimension of the array
v
must be at least
$\mathrm{max}\phantom{\rule{0.125em}{0ex}}\left(1,{\mathbf{ncv}}\right)$ (see
f12aaf).
On initial entry: need not be set.
On intermediate reentry: must be unchanged from its previous exit.
On intermediate exit:
contains the current set of Arnoldi basis vectors.
On final exit: contains the final set of Arnoldi basis vectors.
 4: $\mathbf{ldv}$ – IntegerInput

On entry: the first dimension of the array
v as declared in the (sub)program from which
f12abf is called.
Constraint:
${\mathbf{ldv}}\ge {\mathbf{n}}$.
 5: $\mathbf{x}\left(*\right)$ – Real (Kind=nag_wp) arrayInput/Output

Note: the dimension of the array
x
must be at least
${\mathbf{n}}$ if
${\mathbf{Pointers}}=\mathrm{NO}$ (default) and at least
$1$ if
${\mathbf{Pointers}}=\mathrm{YES}$ (see
f12aaf).
On initial entry: need not be set, it is used as a convenient mechanism for accessing elements of
comm.
On intermediate reentry: if
${\mathbf{Pointers}}=\mathrm{YES}$,
x need not be set.
If
${\mathbf{Pointers}}=\mathrm{NO}$,
x must contain the result of
$y=\mathrm{OP}x$ when
irevcm returns the value
$1$ or
$+1$. It must return the real parts of the computed shifts when
irevcm returns the value
$3$.
On intermediate exit:
if
${\mathbf{Pointers}}=\mathrm{YES}$,
x is not referenced.
If
${\mathbf{Pointers}}=\mathrm{NO}$,
x contains the vector
$x$ when
irevcm returns the value
$1$ or
$+1$.
On final exit: does not contain useful data.
 6: $\mathbf{mx}\left(*\right)$ – Real (Kind=nag_wp) arrayInput/Output

Note: the dimension of the array
mx
must be at least
${\mathbf{n}}$ if
${\mathbf{Pointers}}=\mathrm{NO}$ (default) and at least
$1$ if
${\mathbf{Pointers}}=\mathrm{YES}$ (see
f12aaf).
On initial entry: need not be set, it is used as a convenient mechanism for accessing elements of
comm.
On intermediate reentry: if
${\mathbf{Pointers}}=\mathrm{YES}$,
mx need not be set.
If
${\mathbf{Pointers}}=\mathrm{NO}$,
mx must contain the result of
$y=Bx$ when
irevcm returns the value
$2$. It must return the imaginary parts of the computed shifts when
irevcm returns the value
$3$.
On intermediate exit:
if
${\mathbf{Pointers}}=\mathrm{YES}$,
mx is not referenced.
If
${\mathbf{Pointers}}=\mathrm{NO}$,
mx contains the vector
$Bx$ when
irevcm returns the value
$+1$.
On final exit: does not contain any useful data.
 7: $\mathbf{nshift}$ – IntegerOutput

On intermediate exit:
if the option
Supplied Shifts is set and
irevcm returns a value of
$3$,
nshift returns the number of complex shifts required.
 8: $\mathbf{comm}\left(*\right)$ – Real (Kind=nag_wp) arrayCommunication Array

Note: the dimension of the array
comm
must be at least
$\mathrm{max}\phantom{\rule{0.125em}{0ex}}\left(1,{\mathbf{lcomm}}\right)$ (see
f12aaf).
On initial entry: must remain unchanged following a call to the setup routine
f12aaf.
On exit: contains data defining the current state of the iterative process.
 9: $\mathbf{icomm}\left(*\right)$ – Integer arrayCommunication Array

Note: the dimension of the array
icomm
must be at least
$\mathrm{max}\phantom{\rule{0.125em}{0ex}}\left(1,{\mathbf{licomm}}\right)$ (see
f12aaf).
On initial entry: must remain unchanged following a call to the setup routine
f12aaf.
On exit: contains data defining the current state of the iterative process.
 10: $\mathbf{ifail}$ – IntegerInput/Output

On initial entry:
ifail must be set to
$0$,
$1\text{or}1$. If you are unfamiliar with this argument you should refer to
Section 3.4 in How to Use the NAG Library and its Documentation for details.
For environments where it might be inappropriate to halt program execution when an error is detected, the value
$1\text{or}1$ is recommended. If the output of error messages is undesirable, then the value
$1$ is recommended. Otherwise, because for this routine the values of the output arguments may be useful even if
${\mathbf{ifail}}\ne {\mathbf{0}}$ on exit, the recommended value is
$1$.
When the value $\mathbf{1}\text{or}1$ is used it is essential to test the value of ifail on exit.
On intermediate exit:
the value of
ifail is meaningless and should be ignored.
On final exit: (i.e., when
${\mathbf{irevcm}}=5$)
${\mathbf{ifail}}={\mathbf{0}}$, unless the routine detects an error or a warning has been flagged (see
Section 6).
6
Error Indicators and Warnings
If on entry
${\mathbf{ifail}}=0$ or
$1$, explanatory error messages are output on the current error message unit (as defined by
x04aaf).
Errors or warnings detected by the routine:
 ${\mathbf{ifail}}=1$

On initial entry, the maximum number of iterations
$\le 0$, the option
Iteration Limit has been set to a nonpositive value.
 ${\mathbf{ifail}}=2$

The options
Generalized and
Regular are incompatible.
 ${\mathbf{ifail}}=3$

The option
Initial Residual was selected but the starting vector held in
resid is zero.
 ${\mathbf{ifail}}=4$

The maximum number of iterations has been reached. Some Ritz values may have converged; a subsequent call to
f12acf will return the number of converged values and the converged values.
 ${\mathbf{ifail}}=5$

No shifts could be applied during a cycle of the implicitly restarted Arnoldi iteration. One possibility is to increase the size of
ncv relative to
nev (see
Section 5 in
f12aaf for details of these arguments).
 ${\mathbf{ifail}}=6$

Could not build an Arnoldi factorization. Consider changing
ncv or
nev in the initialization routine (see
Section 5 in
f12aaf for details of these arguments).
 ${\mathbf{ifail}}=7$

Unexpected error in internal call to compute eigenvalues and corresponding error bounds of the current upper Hessenberg matrix. Please contact
NAG.
 ${\mathbf{ifail}}=8$

Either the initialization routine
f12aaf has not been called prior to the first call of this routine or a communication array has become corrupted.
 ${\mathbf{ifail}}=9$

An unexpected error has occurred. Please contact
NAG.
 ${\mathbf{ifail}}=99$
An unexpected error has been triggered by this routine. Please
contact
NAG.
See
Section 3.9 in How to Use the NAG Library and its Documentation for further information.
 ${\mathbf{ifail}}=399$
Your licence key may have expired or may not have been installed correctly.
See
Section 3.8 in How to Use the NAG Library and its Documentation for further information.
 ${\mathbf{ifail}}=999$
Dynamic memory allocation failed.
See
Section 3.7 in How to Use the NAG Library and its Documentation for further information.
7
Accuracy
The relative accuracy of a Ritz value,
$\lambda $, is considered acceptable if its Ritz estimate
$\text{}\le {\mathbf{Tolerance}}\times \left\lambda \right$. The default
Tolerance used is the
machine precision given by
x02ajf.
8
Parallelism and Performance
f12abf is threaded by NAG for parallel execution in multithreaded implementations of the NAG Library.
f12abf makes calls to BLAS and/or LAPACK routines, which may be threaded within the vendor library used by this implementation. Consult the documentation for the vendor library for further information.
Please consult the
X06 Chapter Introduction for information on how to control and interrogate the OpenMP environment used within this routine. Please also consult the
Users' Note for your implementation for any additional implementationspecific information.
None.
10
Example
This example solves $Ax=\lambda x$ in shiftinvert mode, where $A$ is obtained from the standard central difference discretization of the convectiondiffusion operator $\frac{{\partial}^{2}u}{\partial {x}^{2}}+\frac{{\partial}^{2}u}{\partial {y}^{2}}+\rho \frac{\partial u}{\partial x}$ on the unit square, with zero Dirichlet boundary conditions. The shift used is a real number.
10.1
Program Text
Program Text (f12abfe.f90)
10.2
Program Data
Program Data (f12abfe.d)
10.3
Program Results
Program Results (f12abfe.r)