(^648) Series
and k are quite small, it is advisable to substitute in (3) and simplify the result
before making numerical calculations. Thus we can put (3) in the form
1 k)
(5) log 2n (k/ = log 2 - `k +
1
- log
2k
n
-(n-k+1)log2(1 )+En.k,
where En,k = En - Ek - En_k Much progress in probability and statistics is
based upon the idea that when k is near n/2, we can represent it in the form
(6) k=2+xv,
where X is a number that depends upon k and n. When (6) holds, we find that
(7) 2k =1+2(1-n)=1-
and we can put (5) in the form
(8) log 2n (k) =
log
nar
and hence
(9)
-(?+a-v/,-
+1)log(1+
-(2-A V1- +Y1) log 1 2X
log Zn (k) =log
n
2 1 log (1 4n/
-Ay7GLlog(1+2X-)-log(1 2 +E_,k.
Vn
Now let a be a positive number and suppose that k is close enough to n/2 to
make
(10) 2-a'<k2 +X /n52+aVn.
Then Jai < a and we will have
4X2<
n
2X = <1
n
provided n is sufficiently great, say n > na. It can be shown that there is a
constant MI such that, when jx( < I,
(12)
(13) log (1 + x) - log (1 - x) = 2x + F2(x)lxIa,
log (1 - x) = -x + FI(x)Ix12
where JFI(x)l 9 MI and JF2(x)l MI. These facts and (9) yield the conclusion