Bug in nlgebra?

[shinaoka]

https://github.com/dimforge/nalgebra/blob/main/src/linalg/svd.rs#L98

Better to use T::EPSILON rather than T::RealField::default_epsilon()?

[mwallerb]

Yes, the problem is that AbsDiffEq::default_epsilon() is terribly named: one would never use epsilon to denote a absolute diffence in floating point. I suppose that the authors of ndalgebra assumed that it referred to something sensible ...

[shinaoka]

Could you make an issue or a PR for nalgebra?