Re: [Libmesh-users] Under Integration

[Kirk, B. (JSC-EG311) <ben...@na...>]

There are two possibilities here... You could declare two fe objects for the velocity components, each using a different quadrature rule. Then for integrals of the pressure space against the velocity space just choose the right fe.

The other possibility would be to use the optional argument to fe.reinit() to specify where you want the shape functions evaluated.

The first option is probably cleaner and more efficient anyway...

-Ben

________________________________
From: Andrea Hawkins
To: Kirk, Benjamin (JSC-EG311)
Cc: jwp...@gm... ; dk...@mi... ; lib...@li...
Sent: Wed Apr 08 19:39:39 2009
Subject: Re: [Libmesh-users] Under Integration

I guess I should explain what I'm trying to do. =)

I have an 8 variable system of which one variable is pressure. When the pressure variable is added, I get a very poorly conditioned matrix. A suggestion that was given to me was to under-integrate the pressure variable. So, I was thinking to just attach different quadrature rules to the different variables. But then, when the elem is getting updated, the basis functions would get updated only for their own quadrature point.  But for integrals of the equation for the pressure, I would need the basis functions of the other variables to also have values at the pressure quadrature points. Is there a way to do this? Or is there another way I am not thinking of?

Thanks!
Andrea



On Wed, Apr 8, 2009 at 6:54 PM, Kirk, Benjamin (JSC-EG311) <ben...@na...<mailto:ben...@na...>> wrote:
If you mean can the library properly under-integrate, say, the mass matrix for bilinear lagrange using 1pt quadrature no, but it should be an easy add...

That is where the problems come in, right?  For example you can solve the transient heat eqn under-integrated if you do something with the mass matrix?



----- Original Message -----
From: John Peterson <jwp...@gm...<mailto:jwp...@gm...>>
To: David Knezevic <dk...@mi...<mailto:dk...@mi...>>
Cc: libmesh-users <lib...@li...<mailto:lib...@li...>>
Sent: Wed Apr 08 18:49:58 2009
Subject: Re: [Libmesh-users] Under Integration

On Wed, Apr 8, 2009 at 6:27 PM, David Knezevic <dk...@mi...<mailto:dk...@mi...>> wrote:
> You could do:
>
> AutoPtr<QBase> qrule =
> fe_type.default_quadrature_rule(dim,extra_quad_order);
>
> where extra_quad_order is negative.

If you want to achieve "mass lumping" there are also Trapezoidal
(QTrap) rules available.  There's also Simpson's Rule (QSimpson) which
may lump the quadratics...?  I can't remember exactly how that works.

--
John

------------------------------------------------------------------------------
This SF.net email is sponsored by:
High Quality Requirements in a Collaborative Environment.
Download a free trial of Rational Requirements Composer Now!
http://p.sf.net/sfu/www-ibm-com
_______________________________________________
Libmesh-users mailing list
Lib...@li...<mailto:Lib...@li...>
https://lists.sourceforge.net/lists/listinfo/libmesh-users
------------------------------------------------------------------------------
This SF.net email is sponsored by:
High Quality Requirements in a Collaborative Environment.
Download a free trial of Rational Requirements Composer Now!
http://p.sf.net/sfu/www-ibm-com
_______________________________________________
Libmesh-users mailing list
Lib...@li...<mailto:Lib...@li...>
https://lists.sourceforge.net/lists/listinfo/libmesh-users