On Mon, Dec 19, 2011 at 1:00 PM, Modhurita Mitra <span dir="ltr"><<a href="mailto:modhurita@gmail.com">modhurita@gmail.com</a>></span> wrote:<br><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
I am trying to express the radiation pattern of an antenna in terms of spherical harmonic basis functions. I have radiation pattern data at N=324360 points. Therefore, my matrix is spherical harmonic basis functions evaluated till order sqrt(N) (which makes up at total of N elements), evaluated at N data points. So this is a linear least squares problem, and I have been trying to solve it by finding its pseudoinverse which uses SVD. The elements of the matrix are complex, and the matrix is non-sparse. I have solved it in MATLAB using a subsample of the data, but MATLAB runs out of memory while calculating the SVD at input matrix size 2500 X 2500. I need to solve this problem using the entire data.<br>
<br>I was thinking of using SLEPc because I found a ready-to-use code which computes the SVD of a complex-valued matrix ( <a href="http://www.grycap.upv.es/slepc/documentation/current/src/svd/examples/tutorials/ex14.c.html" target="_blank">http://www.grycap.upv.es/slepc/documentation/current/src/svd/examples/tutorials/ex14.c.html</a> ). I don't know how to use either PETSc or SLEPc (or Elemental) yet, so I am trying to figure out where to start and what I should learn. <br>
</blockquote><div><br></div><div>1) You could try Elemental in parallel (would probably use QR)</div><div><br></div><div>2) You could try LSQR in PETSc (preconditioner isa mystery)</div><div><br></div><div>3) The "right" solution for large LS problems is probably this compressed sensing</div>
<div> approach (there is a great review on arXiv by Martinsson).</div><div><br></div><div> Matt</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Thanks,<br>
Modhurita<br><br><br><div class="gmail_quote">On Mon, Dec 19, 2011 at 12:31 PM, Matthew Knepley <span dir="ltr"><<a href="mailto:knepley@gmail.com" target="_blank">knepley@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0pt 0pt 0pt 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
<div>On Mon, Dec 19, 2011 at 12:21 PM, Modhurita Mitra <span dir="ltr"><<a href="mailto:modhurita@gmail.com" target="_blank">modhurita@gmail.com</a>></span> wrote:<br></div><div class="gmail_quote"><div>
<blockquote class="gmail_quote" style="margin:0pt 0pt 0pt 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
Hi,<br><br>I have to compute the pseudoinverse of a 324360 X 324360 matrix. Can PETSc compute the SVD of this matrix without parallelization? If parallelization is needed, do I need to use SLEPc?<br></blockquote><div><br>
</div></div><div>With enough memory, yes. However, I am not sure you want to wait. I am not sure how SLEPc would help here.</div><div>From the very very little detail you have given, you would need parallel linear algebra, like Elemental. However,</div>
<div>I would start out from a more fundamental viewpoint. Such as replacing "compute the psuedoinverse" with</div><div>"solve a least-squares problem" if that is indeed the case.</div><div><br></div><div>
Matt</div><div> </div><blockquote class="gmail_quote" style="margin:0pt 0pt 0pt 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><br>Thanks,<br>Modhurita<br>
</blockquote></div><font color="#888888"><br><br clear="all"><span class="HOEnZb"><font color="#888888"><div><br></div>-- <br>What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.<br>
-- Norbert Wiener<br>
</font></span></font></blockquote></div><br>
</blockquote></div><br><br clear="all"><div><br></div>-- <br>What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.<br>
-- Norbert Wiener<br>