OK, I have recreated the worksheet and set it up to be able to compare results with and without units for different values of CTOL with the exact solutions derived symbolically. The solutions depend rather heavy on the guess value, better guesses give better solutions. If we look at the 3D-plot we see that the area around the minimum is rather flat, so small steps will not change the surface area much and the algorithm stops (too soon) according to the convergence tolerance set. Given that 1cm^3 = 10^-6 m^3 I would have expected the same results for (no_units & CTOL=10^-3) and (units cm & CTOL=10^-9). This is pretty much true, but not exactly. Also for (no_units & CTOL=10^+3) I get similar results as with (cm units & CTOL=10^-3) which is also as expected. So far no bug, just a word of caution when using those numeric approximation algorithms with units different from the base units.
But .... from the previous posts it looks like different people get different results from the very same sheet. Can anybody confirm that with this sheet and post screenshots if he/she gets different results? The screenshots below are the result of using Mathcad 15 M030 under Win XP.
