Re: FullSimplify on Hyperbolic Functions

*To*: mathgroup at smc.vnet.net*Subject*: [mg53394] Re: [mg53383] FullSimplify on Hyperbolic Functions*From*: Andrzej Kozlowski <akoz at mimuw.edu.pl>*Date*: Sun, 9 Jan 2005 23:03:52 -0500 (EST)*References*: <BE06D85D.C981%carlos@titan.colorado.edu>*Sender*: owner-wri-mathgroup at wolfram.com

On 9 Jan 2005, at 20:47, Carlos Felippa wrote: > on 1/9/05 1:46 AM, Andrzej Kozlowski at akoz at mimuw.edu.pl wrote: > >> The reasons why some of these do not work have already been explained >> before: basically they are that FullSimplify when using certain >> transformations functions, cannot use assumptions during the >> intermediate steps and also that certain transformation functions only >> "proceed in one direction". In particular this means that you often >> need to use explicitly ComplexExpand or TrigToExp with FullSimplify. >> This is one of such cases: >> >> In[1]:= >> r1 = ArcCosh[1 + b^2/2]; >> r2 = 2*ArcSinh[b/2]; >> >> In[3]:= >> FullSimplify[ComplexExpand[ >> r1 - r2], b > 0] >> >> Out[3]= >> 0 >> >> However, this is more tricky: >> >> In[4]:= >> FullSimplify[ComplexExpand[ >> r1^2 - r2^2], b > 0] >> >> Out[4]= >> -4*Log[(1/2)*(b + Sqrt[ >> 4 + b^2])]^2 + >> Log[(1/2)*(2 + >> b*(b + Sqrt[4 + b^2]))]^2 >> >> >> The problem seems to be again that FullSimplify does not try certain >> transformations that go in the "opposite direction" to what it >> normally >> does, but without which it can't see the cancellations. You can make >> it >> see them by forcing expansion of terms: >> >> >> FullSimplify[FunctionExpand[ComplexExpand[r1^2-r2^2],b>0],b>0] >> >> 0 >> >> This is actually an interesting case because it shows a use of >> FunctionExpand with assumptions in cases where FullSimplify with >> assumptions is alone not sufficient. Also note that ComplexExpand >> performs better than TrigToExp here. >> >> Of course if you make the expression even more complicated it will get >> harder and harder for Mathematica to reduce it and at some point it >> will become completely impossible. This isn't really surprising, is >> it? >> > True. But the expression I sent is fairly trivial. On a scale > of 0-10, its a 2. Seems like humans have nothing to fear yet. >>> > Certainly. Also it means you should use computers for what computers are good at and human intelligence for what it is good at. "Simplification" of expressions is not really algorithmic but requires judgement and computers are not very good at using judgement. This is example maybe fairly trivial from the human viewpoint but it is far from trivial from the algorithmic viewpoint. Before the various parts of the expression cancel out and thus become simpler they have to first be made more "complex". In fact the above example can be done even without FunctionExpand, the simpler ExpandAll (without any assumptions) works just as well: FullSimplify[ExpandAll[ComplexExpand[r1^2-r2^2]],b>0] 0 Of course ExpandAll and FunctionExpand do not belong to the usual transformation functions used by FullSimplify and for good reason, since expanding generally makes things more complex. In most cases therefore applying ExpandAll would result in, at best, a substantial waste of time. But in this kind of case it seems to be the only way to obtain the required cancellation. In such a situation a human being will use judgement and intuition that tells one to try expanding the expression to see if cancellations might occur. But since a computer program can't use intuition or judgement it has to either to include ExpandAll always among the transformation functions, which seems like a very bad idea, or in some other algorithmic way recognize those cases when expanding is useful and those when it will be counterproductive. Of course if you can describe such an algorithmic criterion with acceptable complexity I am sure that Adam Strzebonski will consider using it in Simplify. Andrzej Kozlowski