Rational Bounds for the Logarithm Function

with Applications
Robert Bosch

Abstract
We find rational bounds for the logarithm function and we show applications to problem-solving.

1 Introduction
Let  n
1
an = 1 + ,
n
solving the problem U385 from the journal Mathematical Reflections I realized that I needed good
lower and upper bounds for an depending on n. The classical an < e was not enough. The problem

U385 was proposed by Angel Plaza, Universidad de Las Palmas de Gran Canaria, Spain:

Evaluate r s !
(n + 1)n nn1
lim n .
n nn1 (n 1)n2
More precisely, I conjectured that r
an 1
1 + 2.
an1 n
To prove this conjecture I used the double inequality
2n 2n + 1
e < an < e, (1)
2n + 1 2n + 2
that I found in [1]. Exactly, problem 170, page 38 with solution in page 216. The source of this
problem is old, from 1872 in Nouvelles Annales de Mathematiques, [2], the proposer is unknown and
was solved by C. Moreau in [3]. Here we present a new proof. Later we will show how this result
helps to find the limit. Our proof of (1) is based on non-standard bounds for the logarithm function,
these bounds are rational functions (quotient of two polynomials) and the seed inequalities (the proof
is iterative) are the well-known

ex x + 1, for x R, (2)
1
ln x > 1 , for x > 1. (3)
x

2 Main Theorem
Let us begin with the proof of (2) and (3).

ex x + 1 for all real x. Equality holds if and only if x = 0.

Proof: Let f (x) = ex x 1. Note that f 0 (x) = ex 1 0 for x 0, therefore f is an

Mathematical Reflections 1 (2017) 1

 increasing function for x 0. So, f (x) f (0) or equivalently ex x + 1. If x 1 then
x + 1 is negative and ex is positive. When 1 < x < 0 the function f (x) is also positive
because f (1) = e1 > 0 and f (0) = 0. Suppose by contradiction that exist x0 (1, 0) such
that f (x0 ) < 0, then since f (x) is clearly continuous by Bolzanos theorem there is a value
x1 (1, x0 ) with f (x1 ) = 0, contradiction since f (x) = 0 only for x = 0. To prove this last
b (x, 0) such that f 0 (b
result suppose that f (x) = 0 and x 6= 0. By Rolles theorem exist x x) = 0,
x
or e = 1, hence x
b
b = 0, contradiction.
1
ln x > 1 x
for x > 1.

Proof: Z x Z x
1 1
dt > dt.
1 t 1 t2

Now we are ready to find the rational bounds.

Theorem:

I1 : ln(1 + x) x, x > 1,
x(x + 2)
I2 : ln(1 + x) , x 0,
2(x + 1)
x(x + 6)
I3 : ln(1 + x) , x 0,
2(2x + 3)
2(x 1)
I4 : ln x > , x > 1.
x+1
Proof:

I1 is a direct consequence of (2). Now we integrate I1 to obtain I2 ,

Z x Z x
ln(1 + t)dt tdt.
0 0

Integrating I2 the result is I3 ,

Z x Z x
t(t + 2)
ln(1 + t)dt dt.
0 0 2(t + 1)

Finally I4 follows from integrate (3),

Z x Z x  
1
ln tdt > 1 dt.
1 1 t
Notice that in each step the new inequalities are refinements of the previous ones. Holds
x(x + 6) x(x + 2)
x,
2(2x + 3) 2(x + 1)
for x 0. Also
2(x 1) x1
,
x+1 x
for x > 1. The result of continuing this algorithm are not rational bounds. To see how to find new
refinements that are rational bounds look at [4].

Mathematical Reflections 1 (2017) 2

3 Applications
We show a proof of the Arithmetic-Geometric Mean inequality using inequality (2).

x1 +x2 ++xn
Let = n
. By (2),
x1 x2 xn x1 x2 xn
e( 1) e( 1) e( 1) .

After simple transformations this inequality is equivalent to AM-GM:

n x1 x2 xn .

Let us see the proof of the double inequality (1). Taking logarithms on both sides we need to
prove      
2n 1 2n + 1
ln + 1 < n ln 1 + < ln + 1.
2n + 1 n 2n + 2
The left hand inequality is
   
1 2n + 1
n ln 1 + + ln > 1.
n 2n

The lower bound provided by (3) is not effective here, but I4 it is,
2n 2
+ > 1 2 > 1.
2n + 1 4n + 1
The right hand inequality becomes
   
1 1
n ln 1 + + ln 1 + < 1.
n 2n + 1

I1 and I2 are not good enough, but I3 it is,

6n + 1 12n + 7
+ < 1 3n + 1 > 0.
2(3n + 2) 2(6n + 5)(2n + 1)

A nice consequence of the double inequality (1) is

lim an = e.
n

The problem U373 from Mathematical Reflections is

Prove the following inequality holds for all positive integers n 2,

    
1 1 1
1+ 1+ 1 + < 3.
1+2 1+2+3 1 + 2 + 3 + + n

Proposed by Nguyen Viet Hung, Hanoi University of Science, Vietnam.

Mathematical Reflections 1 (2017) 3

 The published solution is due to Albert Stadler, Herrliberg, Switzerland using the inequality
I1 .
n   n  
Y 1 Y 2
1+ = 1+ ,
k=2
1 + 2 + + k k=2
k(k + 1)
n  !
X 2
= exp ln 1 + ,
k=2
k(k + 1)
n
!
X 1
exp 2 ,
k=2
k(k + 1)
n  !
X 1 1
= exp 2 ,
k k + 1
 k=2 
2
= exp 1 e < 3.
n+1

Now, we want to show how applying I2 we obtain a refinement.

We need to show that n  

Y 2
1+ < 6.
k=1
k(k + 1)
Or equivalently
n  
X 2
ln 1 + < ln 6.
k=1
k(k + 1)
By I2 it is enough to prove that

X 1 X 1
+ 2
ln 6.
k=1
k(k + 1) k=1 k + k + 2

The first series telescopes to 1 and the second one can be found from the series for z cot(z) for
an appropriate z, the value using Wolfram Alpha is
!
7

tanh
X 1 2 1
2
= 0.686827.
k=1
k +k+2 7 2

Finally 1.686827 < 1.791759 = ln 6.

Notice that e1.686827 5.4023 < 5.4365 2e.

To finish let us see our solution to problem U385, the before mentioned limit.

e
The value of the limit is . Denoting
2
 n
1
an = 1 + e,
n

Mathematical Reflections 1 (2017) 4

 the expression to find the limit is
r 
an p
an1 n n(n 1) .
an1

1
It only remains to show that the expression inside parenthesis tend to . We shall use the
2
double inequality r
an 1
1 1 + 2.
an1 n
The lower bound is because the sequence an is increasing, the upper bound is hard and is true
because the double inequality
2n 2n + 1
e < an < e.
2n + 1 2n + 2
We obtain s
4n2 1
r
an 1
< 2
< 1 + 2.
an1 4n 4 n
To finish, note that p
p 1 1 1/n
lim n n(n 1) = lim ,
n n 1/n
and doing x = 1/n, we have

1 1x 1 1
lim = lim = .
x0 x x0 2 1 x 2

By LHopitals rule.

References
[1] G. Po . Problems and Theorems in Analysis. Vol. I. Springer, (1998).
lya - G. Szego

[2] Problem 1098. Nouv. Annls. Math. Ser. 2, Vol. 11, p. 480. (1872)

[3] C. Moreau. Nouv. Annls. Math. Ser. 2, Vol. 13, p. 61. (1874)

[4] Flemming Topsoe. Some bounds for the logarithmic function. University of Copenhagen. (avail-
able on the Internet.)

Robert Bosch
USA.
bobbydrg@gmail.com

Mathematical Reflections 1 (2017) 5