krj@na.toronto.edu (Ken Jackson) (06/02/89)
NAgMAg Thursday, June 1 1989 Volume 89 Issue 13 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %% The official electronic digest of the NAG Users Association %% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Today's Topics Re: [NAgMAg V89 #12] NAGMAG contributions from Jeremy % % Welcome to the incredible shrinking digest -- I feel a bit % like the newspaper editor in the silly season!! % % So -- to fill out the digest here is a moan-- % % I just read through the NAGUA "Invitation to NAGUA '89" % (no I'm not really short of reading material) and happened % upon an article advertising two important new services for % NAGUA members. One of these was -- you guessed it -- NAGMAG. % So what's the problem? The problem is the poxy wordprocessor % that's used has a special command denoted by @ which says % throw away the next two characters!! This does wonders for % email addresses :-) % % The question is how do I undo this? Suggestions via e-mail % with a @ in the address to ......... % % Tim % --------------------------- Date: Thu, 25 May 89 20:50:59 BST From: MRO2@phoenix.cambridge.ac.uk Subject: Re: [NAgMAg V89 #12] I have to agree with David Rhead's comment about the change in size of workspace arguments in D02 routines at Mark 13. Although the change of specification is very minor, it does cause considerable havoc when documentation is out of step with the latest Mark. Also, where there is no IFAIL error associated with incorrect workspace size, the user is inclined to deduce that the library routine is defective when it is not. --------------------------- Date: Thu, 1 JUN 89 17:16:50 GMT From: DAVID@vax.nag.co.uk Subject: NAGMAG contributions from Jeremy Re: error in F02AGF and F02AKF Here is advance warning of an error which has recently been discovered in two important routines in the NAG Library. They are the black box routines for computing all the eigenvalues and eigenvectors of an unsymmetric matrix - either real (F02AGF) or complex (F02AKF). These routines were revised at Mark 13 in order to incorporate balancing - in response to some reports from users of poor results on badly scaled problems. Unfortunately we introduced an error into the code. To be as precise as we can, if the original matrix has at least two non- zero elements in each row and column, the routines perform correctly - and we assume that such matrices account for the great majority of the problems presented to these routines; otherwise the eigenvectors are almost always wrong. To avoid the problem, either revert to the Mark 12 version of the routines, or imlement the following simple correction. In F02AGF change the line: CALL F02AQF(N,K,L,MACHEP,A,IA,VR,IVR,RR,RI,INTGER,IFAIL) - - to: CALL F02AQF(N,1,N,MACHEP,A,IA,VR,IVR,RR,RI,INTGER,IFAIL) In F02AKF change the line: CALL F02AKZ(N,K,L,ACHEPS,AR,IAR,AI,IAI,WR,WI,VR,IVR,VI,IVI,IERR) - - to: CALL F02AKZ(N,1,N,ACHEPS,AR,IAR,AI,IAI,WR,WI,VR,IVR,VI,IVI,IERR) (Only two characters need to be changed in each case.) All this will be publicized in our next Error Bulletin on the Fortran Library. Thanks to Thomas Archibald of the Department of Mathematics, University of Edinburgh, who reported the error to us. Yes, our test programs should have been sufficiently thorough to detect the error, but no, they weren't. Yes, we should improve our test programs, and yes, we are: since the beginning of this year, we have been engaged on a project to revise all the test programs that implementors use to test the Library, and are aiming to cover as much as possible of chapters F01 to F05 in time for Mark 14. A reply to Peter Endebrock's item on usage statistics: For many years we have been glad to receive statistics from sites on the frequency of use of different routines in the Library. Peter Endebrock has faithfully sent us each year the statistics which he has gathered at Hannover. The University of Manchester used to send us similar figures, but gave up when they got rid of their CDC 7600 machines. Is there anyone collecting figures but not sending them to us? If so, we'd be delighted to learn more: how you do it, and what the figures are. --------------------------- %% For further information about the NAG Users Association please contact: %% Janet Bentley, Administrator NAGUA, %% Shore Lane Farm, Blackstone Edge Old Road, %% LITTLEBOROUGH, Lancashire, OL15 0LQ, UK. %% %% Replies or submissions to nagmag@uk.ac.ukc %% Distribution changes to nagmag-request@uk.ac.ukc %% %% END OF ISSUE ---------------------------- Reposted by -- Kenneth R. Jackson, krj@na.toronto.edu (on Internet, CSNet, Computer Science Dept., ARPAnet, BITNET) University of Toronto, krj@na.utoronto.ca (CDNnet and other Toronto, Canada M5S 1A4 X.400 nets (Europe)) (Phone: 416-978-7075) ...!{uunet,pyramid,watmath,ubc-cs}!utai!krj