Cannot compute the SVD of an empty matrix.

[Lishen1]

let m = DMatrix::<f64>::identity(0, 1); // any zero dimension
let _ = nalgebra::linalg::SVD::try_new_unordered(m, true, true, 0.0, 10); // panic here

probably, from try_ interface expected panic-free behavior independent on input. just return None if input is incorrect.

[Andlon]

I will chime in here to say that I think the correct behavior here is for the SVD to succeed, not return None. Handling empty/zero-sized matrices is an important corner case that sometimes happens in applications. For example, say that your matrix is an mxn matrix with n observations. In this case it may still make sense for the case of n == 0 observations.

My understanding of try_... in the context of SVD is that it will only fail if the iterative process fails (which technically should not happen but can still happen in practice, in rare instances).

[tpdickso]

I think that aligns with my intuitions; a matrix with a 0 dimension does have a SVD since you can always express it with valid UΣV* matrices. I think all the function needs to do here is return (identity, original matrix, identity) with at least one of those identity matrices being 0x0 matching the 0-dimension of the input matrix.