Measure of apporoximation
kohmoto
zbi74583 at boat.zero.ad.jp
Mon May 16 12:25:53 CEST 2005
>"RE : Almost Integer"
Members who responded me.
Thanks for your good explanations.
I understood well.
>Ed Pegg Jr
I read your column on MAA site.
It is nice!!
I think that my "nega entropy" for an algorithm is equivalent as your
"keenness" for an approximation.
Definition of E(a) :
If a=algorithm then,
E(a)=1-log{search area of a}/log{object of a}
Comment : If you want, put "-" for "nega"
Example1: Prime test
Divide N by k , k=0 to N^(1/2)
E(P)=1-log(N^(1/2)/logN=1-1/2=0.5
Example2 : Unitary Amicable pair's record.
It has 317 digits.
I searched 10^8 candidates.
E(UA)=1-8/317=0.975
The relationship between E and K.
If K(a)=keenness of an approximation then,
E(a)=1-1/K(a)=1-{digits used}/{digits of accuracy}
Example1 :
If digits of accuracy is infinity, then it has exact information.
Indeed, E(a)=1-n/infinity=1
Example2 :
If digits of accuracy is the same as digits used, then it has no
information.
The entropy is calculated as follows. E(a)=1-1/1=0 .
Yasutoshi
More information about the SeqFan
mailing list