Is this a bug in eig?
2 次查看(过去 30 天)
显示 更早的评论
This issue came up with the roots function (see roots_ on File Exchange), where eig was failing on the companion matrix. Following are two test cases; the first one works and the second one doesn't. (I reported this to tech support, but it's not listed in the official Matlab bug reports.)
a = [-1,-1,0;1,0,0;0,1,0];
a(1,3) = 10e-32;
[v,d] = eig(a);
disp(abs(a*v-v*d))
% 1.0e-15 *
% 0.055511151231258 0.055511151231258 0.000000000000000
% 0.166533453693773 0.166533453693773 0.000000000000000
% 0.138777878078145 0.138777878078145 0
disp(num2str(diag(d)))
% -0.5+0.86603i
% -0.5-0.86603i
% 1e-31+0i
a(1,3) = 9e-32;
[v,d] = eig(a);
disp(abs(a*v-v*d))
% 0.000000000000000 0.000000000000000 0.000000000000000
% 0.000000000000000 0.000000000000000 0.000000000000000
% 0.707106781186548 0.707106781186548 0.000000000000000
disp(num2str(diag(d)))
% -0.5+0.86603i
% -0.5-0.86603i
% 0+0i
0 个评论
采纳的回答
Matt J
2021-3-30
编辑:Matt J
2021-3-30
It's fine. You just need to disable balancing:
a = [-1,-1,0;1,0,0;0,1,0];
for e=[10e-32, 9e-32]
a(1,3) = e;
[v,d] = eig(a,'nobalance');
Discrepancy = abs(a*v-v*d)
end
13 个评论
Bruno Luong
2021-3-31
编辑:Bruno Luong
2021-3-31
Google I found this page where there is a paper discuss issue with banancing. Someone claims (last post) he has developped a correct balancing method in Lapack 3.5.0.
更多回答(0 个)
另请参阅
产品
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!