From research!csnet!CSNET-RELAY!ibm.com!CULLUMJ 7 July 1986, 13:46:36 EDT
From: "JANE K. CULLUM" <CULLUMJ%ibm.com@CSNET-RELAY>
Received: from su-score.arpa by CSNET-RELAY.ARPA id aa11967; 7 Jul 86 15:24 EDT
Return-Path: <CULLUMJ@IBM.COM>
Received: from IBM.COM by SU-SCORE.ARPA with TCP; Mon 7 Jul 86 11:08:17-PDT
To:   na.dis%su-score.arpa@CSNET-RELAY
Subject: LANCZOS CODE
ReSent-Date: Mon 7 Jul 86 11:28:16-PDT
ReSent-From: "Gene H. Golub" <GOLUB%su-score.arpa@CSNET-RELAY>
ReSent-To: 
MMDF-Warning:  Parse error in original version of preceding line at CSNET-RELAY.ARPA
ReSent-Message-ID: <12220830747.35.GOLUB@SU-SCORE.ARPA>
Via:  Csnet-Relay; 10 Jul 86 1:15-EDT

This message is addressed to anyone of you who has obtained
our LANCZOS code LEVAL from the Argonne NETLIB.
This is the main program for computing eigenvalues of real
symmetric matrices using our Lanczos procedure with no
reorthogonalization of the Lanczos vectors.
Somehow the LEVAL code which was sent
to Argonne acquired two code modifications that were used
for testing but which should not have been in the version
being distributed to other people.

One of these modifications is inconsequential
in that it only affects the number of messages a user would see.
However, the other modification
will cause real problems if you have saved
the Lanczos matrices from one run and then feed them back in
to Leval on a subsequent run and try to
make these matrices larger.

Please fix your LEVAL code as follows:

(1)  The important change occurs at the point in the main
program where the user calls the subroutine USPEC to initialize
the user-specified matrix arrays and parameters.  This statement
currently reads:

        if(istart.eq.0)  call USPEC (n,matno)

 It should read simply

        call USPEC(n,matno)

(2) The message level change: Two lines prior to the CALL INVERR
command which calls an inverse iteration subroutine to compute
error estimates on the computed eigenvalue approximations,
there is currently the statement

       Iwrite = 0

The flag iwrite controls the amount of information which the user
sees regarding the progress of the computations.  It is set by
the user initially being read in from file 5.  The user can in fact
freely change this flag throughout LEVAL
if desired.  However, if you want the message level to be controlled
by the value of iwrite read in from file 5, then you should remove
this iwrite = 0 statement.

Sorry for this problem.  Please keep us informed of any problems
which you encounter with this or any other of our Lanczos codes.
The Hermitian codes and the singular value codes are also
available from NETLIB.  The complex symmetric codes or any of
the other codes discussed in our book are also available but
currently from us and not NETLIB.

Thanks, Jane Cullum
Ralph Willoughby  914-945-1589


From research!csnet!CSNET-RELAY!ibm.com!CULLUMJ 10 July 1986, 12:43:26 EDT
From: "JANE K. CULLUM" <CULLUMJ%ibm.com@CSNET-RELAY>
Received: from su-score.arpa by CSNET-RELAY.ARPA id aa09648; 10 Jul 86 15:56 EDT
Received: from IBM.COM by SU-SCORE.ARPA with TCP; Thu 10 Jul 86 12:30:29-PDT
To:   na.grosse%su-score.arpa@CSNET-RELAY
Subject:
Via:  Csnet-Relay; 10 Jul 86 17:44-EDT

Hi,
Thanks very much for your interaction with Haldane
w.r.t. our Lanczos codes.  Please contact me if he
has any problems with these codes.  thanks, Jane Cullum


