<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<font face="Ubuntu">Thank you.<br>
I will ask to update to 3.4 but I am working on a supercomputer
thus I do not have control on the installed </font><font
face="Ubuntu"><font face="Ubuntu"> software</font>.<br>
So you suggest to use conjugate gradient + multigrid as
preconditioner, correct?<br>
If so, I retain </font><font face="Ubuntu"> <br>
</font>
<pre wrap="">-ksp_type cg
</pre>
<pre wrap=""><font face="Ubuntu">instead of</font>
</pre>
-ksp_type richardson
<br>
<br>
correct?<br>
<br>
Michele<br>
<font face="Ubuntu"><br>
<br>
</font>
<div class="moz-cite-prefix">On 05/17/2013 12:45 PM, Jed Brown
wrote:<br>
</div>
<blockquote cite="mid:87r4h5pezo.fsf@mcs.anl.gov" type="cite">
<pre wrap="">Michele Rosso <a class="moz-txt-link-rfc2396E" href="mailto:mrosso@uci.edu"><mrosso@uci.edu></a> writes:
</pre>
<blockquote type="cite">
<pre wrap="">Hi,
I am successfully using PETSc (v3.3) in parallel to solve the Poisson
equation in 3D .
</pre>
</blockquote>
<pre wrap="">
Please upgrade to petsc-3.4 when you get a chance.
</pre>
<blockquote type="cite">
<pre wrap="">The discretization is done by using finite difference on a uniform
structured grid. So far I used the conjugate gradient method, but I
would like to give a try to multigrid. The documentation describes
multigrid as a preconditioner only, thus I would like to know if it is
possible to use multigrid as a solver
</pre>
</blockquote>
<pre wrap="">
-ksp_type richardson
will not accelerate your multigrid. Krylov with preconditioning is
almost never slower, and a lot more robust.
</pre>
<blockquote type="cite">
<pre wrap="">and, if so, if you could give my some tips to start.
</pre>
</blockquote>
<pre wrap="">
-pc_type mg -pc_mg_levels 3
-pc_type gamg
-pc_type ml
-pc_type hypre
</pre>
</blockquote>
<br>
</body>
</html>