<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
</head>
<body>
<p>Hello,<br>
<br>
I represent a group of researchers who are working towards
infusing the SHAKE algorithm in GROMACS with a parallel
implementation of Newton's method.<br>
We have the necessary expertise to write the required kernels from
scratch without the overhead associated with using a general
purpose library for solving sparse linear systems.<br>
<br>
An obstacle remains. We need the ability to redistribute the
coordinates and velocities of the atoms of a large molecule as
described below.<br>
<br>
Consider a GROMACS simulation that includes of a large molecule
with bond constraints.<br>
<b>Given an arbitrary coloring of the vertices of the molecular
graph, is there in GROMACS a mechanism that will allow us to
collect the data of all atoms with the same color on the same
MPI rank?</b><br>
If the answer is yes, then how can this be done?<br>
<br>
We make no assumptions on the number of atoms, bonds, ranks or
colors.<br>
The ideal mechanism will work regardless of how GROMACS has
distributed the data at the current time.<br>
The ideal mechanism will allow us to use a subcommunicator of the
MPI communicator used by GROMACS.<br>
<br>
Kind regards<br>
<br>
Carl Christian.</p>
<pre class="moz-signature" cols="72">--
Carl Christian Kjelgaard Mikkelsen, Ph.D
Department of Computing Science and HPC2N
Umeå University
90187 Sweden
Telephone: +46(0)90-786 5759
Email: <a class="moz-txt-link-abbreviated" href="mailto:spock@cs.umu.se">spock@cs.umu.se</a> / <a class="moz-txt-link-abbreviated" href="mailto:carl.c.kjelgaard.mikkelsen@umu.se">carl.c.kjelgaard.mikkelsen@umu.se</a>
By sending an email to Umeå University, the University processes your
personal data. For more information, please read: <a class="moz-txt-link-freetext" href="https://www.umu.se/en/gdpr">https://www.umu.se/en/gdpr</a></pre>
</body>
</html>