My proposal for the test kernel.

[Don A. Marshall]

Jeffry,

The attached Word document gives a revised description of the test problem 
to make it more specific.  As I originally proposed it during the closing 
hours of the meeting, it left a number of details to the reader, who if he 
was on the code project would have understood implicitly what I meant.  I 
have attached a "bmp" file showing an example of the kind of mesh I am 
talking about.

I tried sending most of this email on 3/8/01, but I guess there were still 
problems with the email address.  At that time I tried to use 
address@hidden  The next day we got a message from Mark saying to use 
pooma-dev, but then Dave Nystrom tried to use it only to have his email 
bounce.  My message never came back, so I assumed you had it.

The test problem contains a number of interesting "features" of the hydro 
that we modeled using a fine mesh to contain subcell information.  It makes 
use of cell centered, vertex centered, and face centered arrays.  Missing 
are the edge centered arrays; however, in our implementation of the 
geometry calculations, we also made use of edge centered arrays.  In other 
words, this test problem requires the use of arrays having more than one 
size as well as different centerings.  No matter how it is coded, it 
contains lots of special cases which we took care of in "R1" by using 
"NDIndexes."  The kernel can be handled without resorting to scalar code.

The Word document also suggests using the artificial viscosity routine as a 
kernel for scalar code.

Our version of the artificial viscosity code right now has an error in 
it.  If I am lucky, I will be able to fix it tomorrow.  This afternoon we 
got our first stripped down version of the code running in POOMA R2.  The 
movies looked good.  When John Hall gets back on Monday or Tuesday, we will 
send you a copy of the working routines for the predictor step as well as 
for the artificial viscosity.  I just talked to Jean, she said we need to 
send more than just those 2 routines.

My real intent when I proposed the kernels was more for you to see how many 
arrays we have to deal with routinely and how much complexity arises 
because of the mixture of different centerings.  I just finished talking to 
John Hall; he has a different agenda.  We will thrash it all out early next 
week.

Hope this helps.

I am anxious to get the collaboration going.  If there is anything I can do 
to help speed it along, please don't hesitate to give me a call, or better 
yet drop me an email message.  I respond to email better than to the phone.

The numbers that we gave you over the phone are the best bet.  We rotate 
around through the offices.  Right now we are spending most of our time in 
John Hall's office.  Again the numbers are:
        505 667-8996    Don Marshall
        505 665-4488    Jean Marshall
        505-667-7568    John Hall

        505-667-7913    William "Dave" Nystrom

We all have the same fax number:  505-667-3726

I certainly prefer email  because it can contain snips of listings and code 
examples which can be compared with data or existing code without having to 
type very much.

Good luck, and glad to have you "on board".    :)

Don

POOMA_TP.doc
Description: MS-Word document

mesh.bmp
Description: Binary data