Jim wrote: After going to see Matrix Reloaded with my wife tonight, I investigated the alleged "conjbug" a bit more.
In 30 years of programming, I've yet to come across a system that does not produce EXACTLY -5 from these values.
Jim, you are right about this. I instrumented the code a bit more and printed out the argument to the power function in hexadecimal. The result was identical in both the scenarios. My assumption that the two different ways of calculating -5 differed slightly was false. Current puzzle: passing the identical argument to Fractint's log function does indeed give different results in two different contexts, as you discovered. Very weird, and I don't know why. Here are the parts of my analysis that I still stand by: The argument (-5,0) is indeed a discontinuous point in the domain of the log function, so the behavior of the power function that you saw, where sometimes a complex value is returned and sometimes is the conjugate, is not so suprising as it seems. What I don't understand: Why the log function is not behaving deterministically. I am going to discuss the finer points of this in the fractdev forum. Tim