[R] nlm doesn't detect a gradient in my function - odd solution (bug?)

[Andrew Robinson]

Dear R community,

I sent a question to the list a few days ago.  Using nlm, I was unable to detect a gradient in a four-parameter function that I had written in the region that I was searching, even though the function surface clearly varied in the fourth significant figure there.

I stumbled across a peculiar solution: the nlm function was able to detect a non-zero gradient when the ndigit value was *reduced*.  The default value is 12, I was able to detect a gradient in one of the four dimensions if I used 10, and in all of the four if I used 8.

Does that make any sense to anyone?

Thanks,

Andrew