believed public domain and already widely distributed
From na-net.stanford.edu!moler Sun Nov 26 09:57:59 PST 1989
Received: by simplicity.stanford.edu (4.0/inc-1.5)
	id AA04644; Sun, 26 Nov 89 09:57:59 PST
Date: Sun, 26 Nov 89 09:57:59 PST
From: moler@na-net.stanford.edu (Cleve Moler)
Message-Id: <8911261757.AA04644@simplicity.stanford.edu>
To: dongarra@cs.utk.edu, ehg@research.att.com
Subject: Please fix EISPACK bugs in NETLIB
Cc: garbow@mcs.anl.gov

Jack and Eric--  Here is your own, personal, copy of a
blurb about EISPACK bugs that's going out in today's
NA Net News.  Please note the second paragraph.

Jack-- You haven't yet made the changes on COMLR2, COMQR2
and BANDR, and, as my previous mail indicates, you've
added a misleading comment with your changes to HQR.

   -- Cleve


------------------------------

From: Burt Garbow <garbow@mcs.anl.gov>
Date: Mon, 6 Nov 89 13:59:38 CST
Subject: Some Corrections to EISPACK

     We've recently come across a couple of minor bugs in EISPACK.
They only affect the results in rare situations.  The corrections
to HQR, COMLR2, and COMQR2 repair errors introduced in going to
Edition 3, while the correction to BANDR substitutes more robust
coding to avoid overflow.

[Editor's Note: The changes have not yet been made in the source
code available from netlib@ornl.gov or netlib@research.att.gov.
Give Jack and Eric a few days before you request corrected versions.
  --Cleve]

HQR (DO-loop indices)

  DO statements

         do 200 j = k, n
         do 210 i = 1, j
         do 230 j = k, n
         do 240 i = 1, j

  should be replaced, respectively, by

         do 200 j = k, EN
         do 210 i = L, j
         do 230 j = k, EN
         do 240 i = L, j



COMLR2 and COMQR2 (DO-loop indices)

  DO statements

         do 840 i = 1, enm1
         do 820 j = ip1, n
         do 880 jj = low, enm1

  should be replaced, respectively, by

         do 840 i = 1, N
         do 820 j = I, n
         do 880 jj = low, N



BANDR (Overflow protection)

  Line

         S2 = 1.0D0 / (1.0D0 + B1 * B2)

  should be replaced by

         IF (ABS(B1) .GT. 1.0D0) THEN
            U = 1.0D0 / B1
            S2 = U / (U + B2)
         ELSE
            S2 = 1.0D0 / (1.0D0 + B1 * B2)
         ENDIF



------m moler@na-net.stanford.edu garbow@mcs.anl.gov dongarra@cs.utk.edu Sun Nov 26 13:25:58 EST 1989
From-: Eric Grosse  ehg@research.att.com 201-582-5828
To-: moler@na-net.stanford.edu garbow@mcs.anl.gov dongarra@cs.utk.edu
Date-: Sun Nov 26 13:20:48 EST 1989
Subject: eispack bug fix

I have updated netlib's files at reearch.att.com and nac.no (in Oslo)
but not tried to test the change.  If you have programs that
exercise the new codes and would be willing to double-check my editing,
that would be great.

Best wishes
Eric

From na-net.stanford.edu!moler Sun Nov 26 11:40:26 PST 1989
Received: by simplicity.stanford.edu (4.0/inc-1.5)
	id AA04705; Sun, 26 Nov 89 11:40:26 PST
Date: Sun, 26 Nov 89 11:40:26 PST
From: moler@na-net.stanford.edu (Cleve Moler)
Message-Id: <8911261940.AA04705@simplicity.stanford.edu>
To: ehg@research.att.com
Subject: Re:  eispack bug fix

I'll take a look at your changes.  I don't have any test programs
handy.

From na-net.stanford.edu!moler Sun Nov 26 12:13:55 PST 1989
Received: by simplicity.stanford.edu (4.0/inc-1.5)
	id AA04782; Sun, 26 Nov 89 12:13:55 PST
Date: Sun, 26 Nov 89 12:13:55 PST
From: moler@na-net.stanford.edu (Cleve Moler)
Message-Id: <8911262013.AA04782@simplicity.stanford.edu>
To: dongarra@cs.utk.edu, ehg@research.att.com, garbow@mcs.anl.gov
Subject: EISPACK changes

Eric--  I asked your netlib for the 4 revised codes and checked the
source against Garbow's message.  They look OK.  Maybe Burt has
some test code.

Eric and Jack--  I revised the annoucement in the NA News to
indicate that research has already made the changes and that
ornl will soon.

Burt-- Do you have some test code?

   -- Cleve

From cs.utk.edu!dongarra%dasher Sun Nov 26 22:54:47 EST 1989
Received:  by dasher (4.0/2.3-UTK)
	id AA09823; Sun, 26 Nov 89 22:54:47 EST
Date: Sun, 26 Nov 89 22:54:47 EST
From: dongarra%dasher@cs.utk.edu
Message-Id: <8911270354.AA09823@dasher>
To: dongarra@cs.utk.edu, ehg@research.att.com, moler@patience.stanford.edu
Subject: Re:  Please fix EISPACK bugs in NETLIB
Cc: garbow@mcs.anl.gov

OK, OK, I've updated the ORNL version.

Burt -- Rather than do the editing here, do you have a single 
        precision version of the four codes?

Jack

From teak.cray.com!eca Wed Jun 10 13:40:34 CDT 1992
Date: Wed, 10 Jun 92 13:40:34 CDT
From: eca@teak.cray.com (Ed Anderson)
To: dongarra@cs.utk.edu, rosener@cs.utk.edu
Subject: corrupted eispack source?
Cc: eca@teak.cray.com, ehg@research.att.com

> From ehg@research.att.com Wed Jun 10 13:08:04 1992
> To: eca
> From: Eric Grosse  <ehg@research.att.com>  908-582-5828
> Subject: corrupted eispack source?
> 
> Ed,
> 
> An unhappy user has just pointed out that eispack/rsg.f used to contain
> subroutine rsg but now contains driver eispdrv4(rsgsummar) and
> subroutine rsgwzr.    There is a comment in the code
> C     4-28-92:  MODIFIED CALL TO TRED1 TO PASS SEPARATE ARRAYS
> C               TO THE DUMMY ARGUMENTS E AND E2.  (ECA)
> so I'm assuming that you made the change.  Did you do this in
> the wrong directory by accident, or was there some intentional change
> to eispack that I haven't grokked yet?
> 
> Best wishes, Eric


Jack and Bill,
   Eric Grosse sent me the above message, which sounds like the revised
test code I sent was put in the wrong directory.  The revised routines
should have overwritten chtest.f, rsbtest.f, rsgabtest.f, rsgbatest.f,
rsgtest.f, rsptest.f, and rstest.f in the directory of single precision
test routines (I don't recall the name).  Corresponding changes should
also be made to the double precision routines, which are probably in
eispack/ex.  A diff should show only the changes described in my
comments.  Let me know if you have any questions.

   --Ed

