"-0.000" is more than zero?
in [Matlab]
|
Prev: rgb2ycbcr
Next: rgb2hsv functions mathmatical formula
From: Tamas on 28 Dec 2009 10:56 hi, I am using this line to test some properties of a matrix: (all((eig(COV_MARTIX)) >= 0)) but this expression get evaluated false because matlab returns -0.000 for some of the eigenvalues. is that a way around that? thanks a lot
From: Tamas on 28 Dec 2009 11:05 sorry, my mistake. even if it dispalyed -0.0000 is in fact not a zero...
From: Husam Aldahiyat on 28 Dec 2009 11:11 "Tamas " <t.jambor(a)ucl.ac.uk> wrote in message <hhakvg$5ei$1(a)fred.mathworks.com>... > sorry, my mistake. even if it dispalyed -0.0000 is in fact not a zero... format short g
From: Dave Berger on 28 Dec 2009 11:17 "Tamas " <t.jambor(a)ucl.ac.uk> wrote in message <hhakek$1rr$1(a)fred.mathworks.com>... > hi, > > I am using this line to test some properties of a matrix: > > (all((eig(COV_MARTIX)) >= 0)) > > but this expression get evaluated false because matlab returns -0.000 for some of the eigenvalues. is that a way around that? > > thanks a lot Tamas, Check out the 'eps' function. Your -0.000 is probably not a hard 0, but just below zero. The formatting doesn't have enough precision. You could try changing your formatting to "format long eng". -Dave
From: John D'Errico on 28 Dec 2009 11:43
"Tamas " <t.jambor(a)ucl.ac.uk> wrote in message <hhakek$1rr$1(a)fred.mathworks.com>... > hi, > > I am using this line to test some properties of a matrix: > > (all((eig(COV_MARTIX)) >= 0)) > > but this expression get evaluated false because matlab returns -0.000 for some of the eigenvalues. is that a way around that? > It returns false because some of the eigenvalues are negative (as they are computed.) Just because the formatting shows the numbers as -0.000, they are still negative numbers. So what do you want to get around? The matrix is apparently not positive definite, at least not to within a reasonable tolerance. It is possible (even likely) that the matrix is actually non-negative definite, but one or more of the eigenvalues ends up negative due to numerical slop. John |