Homepage Solution manuals Ivan Niven An Introduction to the Theory of Numbers Exercise 2.4.22* (Estimation of $\prod_{j=1}^{k-1} \left( 1 - \frac{j}{n} \right)$)

Exercise 2.4.22* (Estimation of $\prod_{j=1}^{k-1} \left( 1 - \frac{j}{n} \right)$)

Show that the product in Lemma 2.21 is smaller than exp ( k 2 2 n + k 2 n ) , but larger than exp ( k 2 2 n k 3 3 n 2 )

Hint: Show that 1 v e v for all real numbers v . Derive a corresponding lower bound from inequality (1.9) in Section 1.3.

Answers

Note: There is a missing hypothesis. We must suppose that k n 2 to prove the second inequality (see below).

We can give numerical counterexamples. For k = 90 , n = 100

exp ( k 2 2 n k 3 3 n 2 ) > 7.20 1 0 29 > 2.58 1 0 29 > j = 1 k 1 ( 1 j n ) .

With Sage:

def left_member(k,n):
    return exp(-k^2/(2*n) - k^3/(3*n^2)).n()

def right_member(k,n):
    return prod([1- j/n for j in range(1,k)]).n()

left_member(90,100)

                        7.20638686090140e-29

right_member(90,100)

                        2.57182031095525e-29

Proof. The product in lemma 21, giving the probability that numbers u 1 , , u k independently chosen from the set { 1 , 2 , , n } are distinct, is

P = ( 1 1 n ) ( 1 2 n ) ( 1 k 1 n ) ,

where k n .

The function exp is convex, since exp ( x ) = exp ( x ) > 0 for all x . Therefore the graph of f is above the tangent at 0 , which gives

x , 1 + x e x .

Substituting x to x , we obtain

x , 1 x e x .

Therefore

P = j = 1 k 1 ( 1 j n ) j = 1 k 1 exp ( j n ) = exp ( 1 n j = 1 k 1 j ) = exp ( 1 n ( k 1 ) k 2 ) = exp ( k 2 2 n + k 2 n ) .

To prove the other inequality, we use inequality (1.9), page 27:

0 < 1 1 x e x + x 2 , if  0 x 1 2 .

Then

exp ( x x 2 ) 1 x , if  0 x 1 2 .

We must suppose here that k n 2 , since this inequality is not true on [ 0 , 1 [ (for instance, if x = 0.7 , exp ( x x 2 ) > 0.304 > 0.3 = 1 x ) .

We apply this inequality to x = j n , where 0 j n 1 2 , for 1 j k 1 (because k n 2 ), so

P = j = 1 k 1 ( 1 j n ) j = 1 k 1 exp ( j n j 2 n 2 ) = exp ( 1 n j = 1 k 1 j 1 n 2 j = 1 k 1 j 2 ) = exp ( 1 n ( k 1 ) k 2 1 n 2 ( k 1 ) k ( 2 k 1 ) 6 ) = exp ( k 2 2 n k 3 3 n 2 + k 2 n + k 2 2 n 2 k 6 n 2 ) = exp ( k 2 2 n k 3 3 n 2 + k 2 n + 3 k 2 k 6 n 2 ) exp ( k 2 2 n k 3 3 n 2 ) ,

because k 2 n + 3 k 2 k 6 n 2 0 .

To conclude, j = 1 k 1 ( 1 j n ) is smaller than exp ( k 2 2 n + k 2 n ) , and if k 2 n , is larger than exp ( k 2 2 n k 3 3 n 2 ) . □

User profile picture
2024-08-25 12:26
Comments