<div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><DIV>Thanks. I get it.<BR><BR><BR><BR><BR></DIV>
<DIV></DIV>
<DIV id="divNeteaseMailCard"></DIV>
<DIV><BR></DIV>>ÔÚ 2012-10-31 21:24:03£¬"Hong Zhang" <hzhang@mcs.anl.gov> дµÀ£º<BR>
<BLOCKQUOTE style="BORDER-LEFT: #ccc 1px solid; MARGIN: 0px 0px 0px 0.8ex; PADDING-LEFT: 1ex" id="isReplyContent">>Jim,<BR><BR>
<DIV class="gmail_quote">
<BLOCKQUOTE style="BORDER-LEFT: #ccc 1px solid; MARGIN: 0px 0px 0px 0.8ex; PADDING-LEFT: 1ex" class="gmail_quote">
<DIV style="LINE-HEIGHT: 1.7; FONT-FAMILY: arial; FONT-SIZE: 14px">
<DIV>>> In my view, the intent of parallel computing is sovling large system. As you said, MUMPS is mainly used in</DIV></DIV></BLOCKQUOTE>
<DIV> </DIV>
<DIV>>" <SPAN style="LINE-HEIGHT: 23px; FONT-SIZE: 14px">the intent" does not automatically become the reality.</SPAN></DIV>
<DIV><SPAN style="LINE-HEIGHT: 23px; FONT-SIZE: 14px">>All parallel direct solvers are intended to solve ill-conditioned problems as large as possible.</SPAN></DIV>
<DIV><SPAN style="LINE-HEIGHT: 23px; FONT-SIZE: 14px">>But the algorithms (full matrix factorization) consume large memories and inter-processor </SPAN></DIV>
<DIV><SPAN style="LINE-HEIGHT: 23px; FONT-SIZE: 14px">>communications. Unless your matrix has special data structure, the direct solvers cannot be scalable</SPAN></DIV>
<DIV><SPAN style="LINE-HEIGHT: 23px; FONT-SIZE: 14px">>in general.</SPAN></DIV>
<DIV><SPAN style="LINE-HEIGHT: 23px; FONT-SIZE: 14px"><BR></SPAN></DIV>
<BLOCKQUOTE style="BORDER-LEFT: #ccc 1px solid; MARGIN: 0px 0px 0px 0.8ex; PADDING-LEFT: 1ex" class="gmail_quote">
<DIV style="LINE-HEIGHT: 1.7; FONT-FAMILY: arial; FONT-SIZE: 14px">
<DIV></DIV>
<DIV>>>small problems. So are they contradictory. Or, is the "Small problems and moderate-size 2D problems" just relative,</DIV>
<DIV>>>compared with the large system?</DIV></DIV></BLOCKQUOTE>
<DIV>>Yes. </DIV>
<DIV><BR></DIV>
<DIV>>Hong</DIV>
<BLOCKQUOTE style="BORDER-LEFT: #ccc 1px solid; MARGIN: 0px 0px 0px 0.8ex; PADDING-LEFT: 1ex" class="gmail_quote">
<DIV style="LINE-HEIGHT: 1.7; FONT-FAMILY: arial; FONT-SIZE: 14px">
<DIV> </DIV>
<DIV><BR><BR><BR><BR></DIV>
<DIV>
<DIV class="h5">
<DIV></DIV>
<DIV></DIV>
<DIV><BR></DIV>>>At 2012-10-30 23:47:59,"Jed Brown" <<A href="mailto:jedbrown@mcs.anl.gov" target="_blank">jedbrown@mcs.anl.gov</A>> wrote:<BR>
<BLOCKQUOTE style="BORDER-LEFT: #ccc 1px solid; MARGIN: 0px 0px 0px 0.8ex; PADDING-LEFT: 1ex">
<P>>>Simple: it's better when it's faster. Small problems and moderate-size 2D problems often work well with direct solvers.</P>
<DIV class="gmail_quote">>>On Oct 30, 2012 8:29 AM, "w_ang_temp" <<A href="mailto:w_ang_temp@163.com" target="_blank">w_ang_temp@163.com</A>> wrote:<BR type="attribution">
<BLOCKQUOTE style="BORDER-LEFT: #ccc 1px solid; MARGIN: 0px 0px 0px 0.8ex; PADDING-LEFT: 1ex" class="gmail_quote">
<DIV style="LINE-HEIGHT: 1.7; FONT-FAMILY: arial; FONT-SIZE: 14px">
<DIV style="LINE-HEIGHT: 1.7; FONT-FAMILY: arial; FONT-SIZE: 14px">
<DIV>>>Hello,</DIV>
<DIV> >>As is know, MUMPS is based on a direct method. When the system is big, the parallel direct solver is </DIV>
<DIV>>>not effective compared with the iterative solver. So when is it better to use MUMPS?</DIV>
<DIV> >> Thanks.</DIV>
<DIV> >> Jim</DIV></DIV><BR><BR><SPAN title="neteasefooter"><SPAN></SPAN></SPAN></DIV><BR><BR><SPAN title="neteasefooter"><SPAN></SPAN></SPAN></BLOCKQUOTE></DIV></BLOCKQUOTE></DIV></DIV></DIV><BR><BR><SPAN title="neteasefooter"><SPAN></SPAN></SPAN></BLOCKQUOTE></DIV><BR></BLOCKQUOTE></div><br><br><span title="neteasefooter"><span id="netease_mail_footer"></span></span>