- #1
gnome
- 1,041
- 1
I understand that the "singular values" used to form the diagonal matrix D in a singular value decomposition of a matrix A are the square roots of the eigenvalues of [itex]A^TA[/itex].
But I'm having trouble getting a grip on the concept of "singularity" in the context of a matrix. Previously I've only come across singularity as a point where a function is non-continuous and therefore not differentiable. I don't see how that definition applies here. I read in at least one book that "a square matrix A is nonsingular if and only if all its singular values are different from zero." So it sounds like that means that 0 must NOT be an eigenvalue of [itex]A^TA[/itex]. (On second thought, this author is probably not saying that at all; rather just using "A" to represent several different matrices in the course of his discussion.)
But in another book, I got the distinct impression that an SVD for a matrix A can be formed just by using the nonzero singular values of A; i.e. if 0 is an eigenvalue of [itex]A^TA[/itex], just don't use 0; use the nonzero eigenvalues to form a smaller matrix D.
I also read that the condition number [itex]\frac{\sigma_1}{\sigma_n}[/itex] is a measure of the degree of singularity of A.
But I have no feel at all for what is meant by "degree of singularity" of A, and what it means qualitatively for a matrix to be singular or nonsingular.
Can anybody give a qualitative explanation of these terms?
But I'm having trouble getting a grip on the concept of "singularity" in the context of a matrix. Previously I've only come across singularity as a point where a function is non-continuous and therefore not differentiable. I don't see how that definition applies here. I read in at least one book that "a square matrix A is nonsingular if and only if all its singular values are different from zero." So it sounds like that means that 0 must NOT be an eigenvalue of [itex]A^TA[/itex]. (On second thought, this author is probably not saying that at all; rather just using "A" to represent several different matrices in the course of his discussion.)
But in another book, I got the distinct impression that an SVD for a matrix A can be formed just by using the nonzero singular values of A; i.e. if 0 is an eigenvalue of [itex]A^TA[/itex], just don't use 0; use the nonzero eigenvalues to form a smaller matrix D.
I also read that the condition number [itex]\frac{\sigma_1}{\sigma_n}[/itex] is a measure of the degree of singularity of A.
But I have no feel at all for what is meant by "degree of singularity" of A, and what it means qualitatively for a matrix to be singular or nonsingular.
Can anybody give a qualitative explanation of these terms?
Last edited by a moderator: