Well, yes. Exactly. For 2D input, the space between the function name and the brackets is parsed as a multiplication, while in 1D input it's parsed as a function call. Having different parsers behave in different ways on things like this is not good. acer

Well, yes. Exactly. For 2D input, the space between the function name and the brackets is parsed as a multiplication, while in 1D input it's parsed as a function call. Having different parsers behave in different ways on things like this is not good. acer

Another trap: If a space is inserted between the `int` and the open-bracket then the parsed meaning is different in 1D and 2D input. acer

Another trap: If a space is inserted between the `int` and the open-bracket then the parsed meaning is different in 1D and 2D input. acer

Yes, and there are connections between the authors. Aside from papers on calculating Pi, etc, there also seems to have been some contention in the past, though I know nothing about the truth of it all. acer

You have described close to the usual way in which LaTeX output, and Maple's latex() command, are supposed to work. But, instead of having to copy the .sty files all over, it should be possible to just append the TEXINPUTS environment variable with /etc (or \ETC I guess, if you use Windows). That environment variable is used by latex and tex as a search path. acer

I hardly ever use any CAS other than Maple. I sometimes use Matlab. But even then, it is mostly for Simulink. I usually do that from Maple's own evalM() command. On occasion, I will benchmark parts of the two systems. I only very rarely use Mathematica. I do not like Mathematica's model for numerical computations. acer

Thanks, but I was hoping that finer, low-level detail would be instructional for us all. For example given a Maple set S, whose elements are ordered by memory address, adding a new element could be fast since no comparison with the current order is necessary. (That seems to me to be what you were driving at.) But I also wonder about the mechanics of accessing elements. What precisely happens when one issues, `for i in S do...end do`? Does memory address ordering help there, or hinder, and why? acer

It's just a guess, but I suspect that some internal routine with option `remember,system` may be caching some of the the partial sum computations. But then, if a garbage collection takes place and its remember table is cleared, then the full computation might take a different path. Just a wild guess, though. The bug doesn't seem intermittent, no? Are both 0 and -infinity wrong results, even if they alternate? acer

Relying on creation order to be the same as memory address order would of course be misguided. Once memory management kicks in, and garbage is collected and memory freed, new objects might have addresses all over the place. I wanted to give the original poster (a Maple newcomer) a simple explanation for the behaviour being seen by the students. I believe those examples might illustrate what was going on. Perhaps you might expound a bit, Bryan, on why Maple behaves this way. Ie, for performance in accessing objects, or for memory by holding unique representations of some objects no matter how often they are input or arise, etc. We generally know some of the drawbacks of session dependent ordering of results, but it'd be refreshing to hear some of the benefits. acer

Maple can sort objects (in sets, or sums) by memory address or by the order in which they first appear in the session. Consider, restart: seq( addressof(t), t in op(expand((x+y+z)^6)) ); restart: 60*x*y^2*z^3:30*x*y*z^4:6*y^5*z: # notice now which terms appear first expand((x+y+z)^6); acer

What was the method that you used? It might be possible to speed up a high precision "software" floating-point Matrix calculation by increasing garbage collection frequency. See ?gc . acer

For the original poster's benefit: Values which approximate exact zero, but which have small nonzero components due to floating-point evaluation under fixed precision, can also be handled by judicious use of Maple's `fnormal` routine. For example, `+`( seq(`if`(Im(x)=0.0,signum(Re(x)),NULL),x in map(fnormal,evals)) ); Additional optional arguments to fnormal() allow one to fine-tune what is taken to be "close enough" to zero. acer

For the original poster's benefit: Values which approximate exact zero, but which have small nonzero components due to floating-point evaluation under fixed precision, can also be handled by judicious use of Maple's `fnormal` routine. For example, `+`( seq(`if`(Im(x)=0.0,signum(Re(x)),NULL),x in map(fnormal,evals)) ); Additional optional arguments to fnormal() allow one to fine-tune what is taken to be "close enough" to zero. acer

That's very nice. It makes me wonder, about usability and nice defaults for students. The default value of the discont option to plot() could be reconsidered. An option to show asymptotes easily would be nice, in Maple's own plotting routines. acer