|
From: | K. Michael Salerno |
Subject: | Confidence in eigs answers |
Date: | Sat, 08 May 2010 21:28:42 -0400 |
User-agent: | Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.6; en-US; rv:1.9.1.9) Gecko/20100317 Thunderbird/3.0.4 |
Hi,
I'm trying to calculate the lowest N=10 eigenvalues of a 8196x8196 sparse matrix. the eigs(d,10,'sm') gives the following eigenvalues: 1.6049e+00 1.3749e+00 1.3460e+00 1.1004e+00 9.7992e-01 7.1926e-01 7.1131e-01 1.5208e-14 5.2319e-15 -2.9884e-13 the eig command gives these lowest 10 eigenvalues: 1.6877e+00 1.6046e+00 1.3748e+00 1.3459e+00 1.1002e+00 9.7953e-01 7.1918e-01 7.1118e-01 3.6485e-12 -3.8858e-13 I do not have access to matlab, but Mathematica gives the lowest 10 eigenvalues identical to the eig command output, leading me to believe that this answer is correct. Have there been other reports of this? There was mention of possible problems with the underlying lapack or arpack commands on the scipy list: http://projects.scipy.org/scipy/ticket/1159 While waiting for the moderator to approve the original message I found that the command eigs(d,10,1), finding eigenvalues near the value of 1 gives the correct lowest eigenvalues. Somehow the sm option is creating an incorrect eigenvalue. Regards, Mike S |
[Prev in Thread] | Current Thread | [Next in Thread] |