Rounded Newton's method

[davidweichiang]

Is the method in this paper practical? https://arxiv.org/pdf/1302.3741.pdf

[davidweichiang]

My reason for interest: I have occasionally seen cases where Newton's method overshoots the LFP just slightly due to rounding, and this breaks convergence (the convergence proof says that if an iterate is less than the LFP, then the next iterate will be less than the LFP).

[davidweichiang]

• [ ] Rounding as described above
• [x] Take the max with the previous iterate and/or the fixed-point iterate