| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1583.1 | see 1560.0? | SGOUTL::BELDIN_R | Pull us together, not apart | Fri Mar 20 1992 15:28 | 15 |
| Re: <<< Note 1583.0 by ISIDRO::RODRIGUEZ >>>
Elena,
According to Note 1560.0, there is apparently a software
package named "MULTIGRID". Is it possible that what your
customer is asking is whether we sell or support that package
on our platforms? Personally, I don't know of any
methodology for linear equations that is called "multigrid"
(but it has been a long time). Maybe the author of 1560 will
have some more references.
Good luck,
Dick
|
| 1583.2 | | WSE169::KAMATH | | Fri Mar 20 1992 16:39 | 31 |
|
A quick summary off the top of my head:
Multigrid solvers have recently become popular as solvers for linear systems
of equations. It essentially requires a series of problems to be solved on a
hierarchy of grids with different mesh sizes. You start with a fine grid in
the computational domain and use a relaxation technique to smooth the error -
this takes care of the high frequency components of the error. The low
frequency components are removed by considering the error on a coarser grid
(a subset of the fine grid) and applying a relaxation technique on the coarse
grid. The error correction is then interpolated back onto the fine grid.
This process can be done with grids of different mesh sizes - starting with a
fine grid and moving towards increasingly coarser grids. This gives rise to
what are referred to as 'V' cycles and 'W' cycles. In the V cycle, you move
from a fine grid to increasingly coarser grids and then reverse back to the
fine grid. In a W cycle, you start with a fine grid, move towards a coarser
grid, then back part way towards the fine grid, again move towards the coarser
grids and finally reverse back to the fine grid. The cycles are named V and W
as that is the appearance of the cycle if you assume the top of the character
(V or W) to represent the fine grid and the bottom to represent the coarsest
grid. The idea is that finally the grid will be coarse enough that the system
can be solved exactly with a direct method.
To the author of .0 :
Could you please provide some additional information about the requirements
of the customer, specifically the kind of application they are interested
in solving using multigrids?
thanks,
chandrika
|
| 1583.3 | | ISIDRO::RODRIGUEZ | | Tue Mar 24 1992 12:08 | 9 |
| The customer want's the package for multiple applications (He is the
university), not just one. The point is tha he is going to buy some
machines and a multigrid package to run on it.
Can you tell me if there is any running on VAX?
Regards.
Elena.
|
| 1583.4 | try netlib | WSE169::KAMATH | | Wed Mar 25 1992 02:58 | 17 |
|
There is no software from DEC that contains the multigrid solver. You might
have better luck browsing through netlib. Send a mail message to
netlib@ornl.gov
with the message
send index
It will return a list of software that you can obtain from netlib plus
instructions on how to obtain the codes. There is a package on netlib
called PLTMG (this is Banks multigrid code), but I believe it is too
large to be sent by ordinary mail. You might be able to find other codes
or instructions on how to get PLTMG.
chandrika
|
| 1583.5 | netlib? | ISIDRO::RODRIGUEZ | | Wed Mar 25 1992 07:34 | 15 |
|
Chendrika:
Is netlib@ornl.gov the right direction? It returns me a wrong message,
even in the all-in-1 mail.
mail> send
to: netlib@ornl.gov
%MAIL-E-USERSPEC, invalid user specification '@ORNL.GOV'
mail>
Regards.
Elena.
|
| 1583.6 | the format | STAR::ABBASI | i^(-i) = SQRT(exp(PI)) | Wed Mar 25 1992 07:52 | 22 |
| Elena,
That Address is for a differnt Network with different addressing
format, you cant use Digital Easy Net addressing, to access please type
>Send foo.txt DECWRL::"Netlib@ornl.gov"
This goes to DECWRL gateway and from there bounced off to the outside world.
If You NM installed on your node, you ca also do
>Send foo.txt NM%DECWRL::"Netlib@ornl.gov"
This is if DECWRL is down, NM will Keep Trying sending your message
for you untill it sends it, so you dont have to do it yourself.
As mentioned befor, foo.txt should have the SEND INDEX message in it or
whatever you'r asking for, after you send it, the reply will come back
usually in less than a day or so.
/Nasser
|
| 1583.7 | MORE INFO !/? | BIS6::KREATSOULAS | | Mon Mar 30 1992 12:31 | 36 |
| From: John KREATSOULAS
Dept: Edu & Science / Europe
Tel No: DTN 856-8277,
FAX 856-8203
TO: ELENA RODRIGUEZ @SQO
Subject: RE: a: solving linear systems?
Elena,
My apologies for the delayed response!
I have placed calls to IMSL and NAG (library vendors) to find out if their
libraries contain such a solver.
I have also made a query to NETlib re MULTIGRID algorithms. Typically, what
is in there is FORTRAN 77 and should run on the VAX.
They are in use in a number of applications (including Computational Fluid
Dynamics) They are the solvers for some of Tony Jameson's FLOxx codes and
a Superconvergent Multigrid Algorithm has been developed for use on
Massively Parallel Machines. And I could provide pointers to papers
regading multigrid algorithms. But I have not heard of any PACKAGE marketed
under that name.
I will keep you posted on the responses I expect to receive shortly.
But if your customer is looking for something specific, is it possible
to give us some additional information?
Best Regards,
John
PS. Chandrika, greetings from Belgium to you and Sisira!
|
| 1583.8 | Some more info | BIS6::KREATSOULAS | | Mon Mar 30 1992 15:53 | 47 |
| From: John KREATSOULAS
Dept: Edu & Science / Europe
Tel No: DTN 856-8277,
FAX 856-8203
TO: ELENA RODRIGUEZ @SQO
Subject: More info
Elena,
IMSL does not have anything on multigrid algorithms in their library.
NAG still needs to get me an answer.
Here is the info I 've got from the net, just in case you are still trying:
from the index:
...
pltmg - Bank's multigrid code; too large for ordinary mail
...
from the descriptions:
...
lib pltmg
for elliptic partial differential equations in general regions of the plane
# It features adaptive local mesh
# refinement, multigrid iteration, and a pseudo-arclength
# continuation option for parameter dependencies. The package
# includes an initial mesh generator and several graphics
# packages.
ref PLTMG User's Guide, SIAM publications
by Randy Bank
editor Bill Coughran
master research.att.com
...
So far so good. But how do we get the code, since it is too big for ordinary
mail.
Two Suggestions:
1. Send the request anyway and see what happens
send pltmg
2. Call Office for Scientific and Technical Information (OSTI) in the US
Julia H. Daniel TEL 615-576-2139 and ask for further information
Elena, please let us all know how it works out
Regards,
John
|
| 1583.9 | Reply from NAG | BIS6::KREATSOULAS | | Tue Mar 31 1992 12:03 | 42 |
| <<< Note 1583.8 by BIS6::KREATSOULAS >>>
-< Some more info >-
From: John KREATSOULAS
Dept: Edu & Science / Europe
Tel No: DTN 856-8277,
FAX 856-8203
TO: ELENA RODRIGUEZ @SQO
Subject: NAG Response
Elena,
Just got the response from NAG, and I copy from the FAX:
...
"The routine D03EDF in the Partial Differential Equations Chapter will solve an
elliptic PDE for a finite difference equation using a multigrid technique. This
is a generalized product, and may not suit your client's purpose exactly. I
would therefore recommend that you endeavour to find out further details of the
problem so that we can advise more precisely. Nag can build soutions to clients
problems either based upon routines in the library, or by writing new routines
to specification. The Library routine may be built into the customers own
program, or we can produce a program to specification or interface the solution
to an existing package."
...
The person to contact at NAG if you decide to pursue this route is
John Piper,
TEL: +44 865-511245
FAX: +44 865-310139
My suggestion:
1. Ask the customer for more information (I could talk to them, if you would
like to arrange a call)
2. See if public domain SW from netlib or other sources is what they need.
3. Pursue the NAG avenue if this proves to be what they want, but bear in mind
that this will involve some cost.
Please let me nkow how I can be of further assistance
Regards,
John
|