I am interested in the singular value decomposition of a matrix: $\mathbf{M} = \mathbf{U} \mathbf{S} \mathbf{V}^T$.
Suppose $\mathbf{M} = \mathbf{0}$ (zero matrix) and square. Clearly, $\mathbf{S} = \mathbf{0}$, but what about $\mathbf{U}$ and $\mathbf{V}^T$ — Do they have defined values?
When use MATLAB, $\mathbf{U} = \mathbf{V} = \mathbf{I}$, but is this a definition or pure luck? If so, why is it defined as such, since any matrix for $\mathbf{U}$ or $\mathbf{V}$ will satisfy the decomposition – it does not even need to be orthogonal. Is there some sort of proof that it must indeed fall to a specific value?
>> [U, S, V] = svd(zeros(3))
U =
1 0 0
0 1 0
0 0 1
S =
0 0 0
0 0 0
0 0 0
V =
1 0 0
0 1 0
0 0 1
Best Answer
A matrix has not necessarily only one decomposition in singular value, so $U$, $V$, and $S$ aren't well defined, Matlab give you one of the solutions...
For example, if $(U,S,V)$ is a solution, $(-U, S, -V)$ is one too.