a==b and FixedPoint

*To*: mathgroup at yoda.ncsa.uiuc.edu*Subject*: a==b and FixedPoint*From*: jdelany at sargas.ACS.CalPoly.EDU (Jim Delany)*Date*: Wed, 28 Nov 90 06:03:00 PST

Can anyone tell me when two real numbers are Equal? Experiments seem to indicate that the ratio must be sufficiently close to 1, within 2^-44 on the machines I use. On the same topic, when does FixedPoint converge? I get the feeling == is used. This would explain why underflow occurs before convergence when the limit is 0. An approximate equality based on ratios makes sense to me, but it can produce bizarre results when used in conjunction with a coprocessor that uses other criteria. See below. Jim Delany ( jdelany at nike.calpoly.edu ) In[1]:= Log[2, .1^Precision[1.] ] Out[1]= -53.1508 In[2]:= r=1; While[ 1. != 1.+r, r/=2]; Log[2, r] Out[2]= -44 In[3]:= Log[2, -1.+FixedPoint[ Sqrt, 2.] ] Out[3]= -43.5283 In[4]:= tf[x_]:= Cos[3x]==4Cos[x]^3-3Cos[x] In[5]:= tf[2.^100] Out[5]= True In[6]:= tf[1.57] Out[6]= False