Incorrect condition for stopping the LevenbergMarquardtMinimizer's Minimum() by parameter tolerance.

[anotherche]

mathnet-numerics/src/Numerics/Optimization/LevenbergMarquardtMinimizer.cs

Line 161 in f196418

if (Pstep.L2Norm() <= stepTolerance * (stepTolerance + P.DotProduct(P)))

Why is the vector length (Pstep.L2Norm()) compared to the square of the vector length (P.DotProduct(P))? Looks like a bug. Of course, if the parameters were scaled correctly (|Pi|~1) this does not cause a problem. But in general, this is not correct. Wouldn't it be more correct to compare Pstep.L2Norm() to P.L2Norm(), or their squares?

[diluculo]

You are correct, It's a bug.
LINE 160: // if ||ΔP|| <= xTol * (||P|| + xTol), found and stop