Closed mforets closed 5 years ago
We've worked out the reference values for those benchmarks using SDP, since i didn't find them in that paper, see TaylorModels_Benchmark_RefValues. In a PR to come we've evaluated the range of each polynomial with and without normalization, to measure accuracy and speed. This sets up the ground to compare with other approaches, such as brand and bound, interval arithmetic, etc.
CC: @Aadesha
Great! I suggest adding timings in the notebook, and saving the results to a DataFrame or similar.
Excellent! Looking forward to the PR
See eg. Tables 3-6 in AGK18.