1 Introduction and preliminary results

Batir and Cancan [[1], Theorem 2.5] presented the following sharp inequalities:

exp ( 1 n 2 ( n + c ) 2 ) ( 1 + 1 n ) n <exp ( 1 n 2 ( n + d ) 2 ) ,n1,
(1)

where c= 1 2 ln 4 1=0.27649 and d= 1 3 . The proof is based on the fact that the function

ω(x)= 1 2 ( 1 x ln ( 1 + 1 x ) ) x

is strictly increasing on (0,), while inequalities (1) follow from ω(1)ω(n)<ω(). Such an approach of the problem does not offer good results in the left-hand side inequality (1), when n approaches infinity. As we wish to see (1) as an accurate approximation of the form

( 1 + 1 n ) n exp ( 1 n 2 ( n + δ ( n ) ) 2 ) ,
(2)

we are interested in finding δ(n) which gives the best such approximation for large values of n. Moreover, numerical computations show us that for large values of n, the expression ( 1 + 1 / n ) n gets closer to the right-hand side of (1). This fact suggests us that the best approximation (2) is obtained when δ(n) tends to 1/3, as n. For δ(n)=1/3, we deduce

( 1 + 1 n ) n exp ( 1 n 2 ( n + 1 3 ) 2 ) ,
(3)

but a better result is

( 1 + 1 n ) n exp ( 1 n 2 ( n + 1 3 1 12 n ) 2 ) .

The rigorous argument is the following theorem, which is also an improvement of the Batir and Cancan inequality (1).

Theorem 1 For every real number x1, we have

exp ( 1 x 2 ( x + 1 3 1 12 x ) 2 ) < ( 1 + 1 x ) x <exp ( 1 x 2 ( x + 1 3 1 12 x + 23 540 x 2 ) 2 ) .

Proof Let

f(x)=1 x 2 ( x + 1 3 1 12 x ) 2 xln ( 1 + 1 x )

and

g(x)=1 x 2 ( x + 1 3 1 12 x + 23 540 x 2 ) 2 xln ( 1 + 1 x ) .

We have f(1)= 17 25 ln2=0.013<0, g(1)= 85 , 351 121 , 801 ln2=0.0075>0 and

f ( x ) = 16 x + 384 x 2 + 544 x 3 + 15 , 440 x 4 + 17 , 664 x 5 1 x ( x + 1 ) 2 ( 2 x + 1 ) 4 ( 6 x 1 ) 4 , g ( x ) = A ( x 1 ) x ( x + 1 ) 2 ( 180 x 2 45 x + 540 x 3 + 23 ) 4 ,

where

A ( x ) = 1 , 031 , 414 , 482 , 080 x + 3 , 366 , 173 , 125 , 800 x 2 + 6 , 243 , 358 , 941 , 720 x 3 + 7 , 196 , 770 , 443 , 825 x 4 + 5 , 277 , 779 , 208 , 000 x 5 + 2 , 403 , 240 , 062 , 400 x 6 + 620 , 618 , 112 , 000 x 7 + 69 , 480 , 990 , 000 x 8 + 137 , 506 , 401 , 616 .

Evidently, f is concave, g is convex on [1,), with f()=g()=0, so f<0 and g>0 on [1,). The proof is completed. □

The same remarks we make on Batir and Cancan’s assertion [[1], Theorem 2.6], which is proven to have some computation errors, since the expression ( 1 + 1 / n ) n + 1 can be approximated for large values of n as

( 1 + 1 n ) n + 1 exp ( 1 + n 2 ( n + 1 6 ) 2 ) ,
(4)

but a better result is

( 1 + 1 n ) n + 1 exp ( 1 + n 2 ( n + 1 6 1 24 n ) 2 ) ,

as we can see from the following.

Theorem 2 For every real number x1, we have

exp ( 1 + x 2 ( x + 1 6 1 24 x + 43 2 , 160 x 2 ) 2 ) < ( 1 + 1 x ) x + 1 <exp ( 1 + x 2 ( x + 1 6 1 24 x ) 2 ) .

Proof Let

u ( x ) = 1 + x 2 ( x + 1 6 1 24 x ) 2 ( x + 1 ) ln ( 1 + 1 x ) , v ( x ) = 1 + x 2 ( x + 1 6 1 24 x + 43 2 , 160 x 2 ) 2 ( x + 1 ) ln ( 1 + 1 x ) .

We have u(1)= 113 81 2ln2=0.00876>0, v(1)= 8 , 448 , 529 6 , 115 , 729 2ln2=0.00485<0 and

u ( x ) = 16 x + 832 x 3 + 2 , 624 x 4 + 132 , 096 x 5 1 x 2 ( x + 1 ) ( 4 x + 24 x 2 1 ) 4 , v ( x ) = B ( x 1 ) x 2 ( x + 1 ) ( 360 x 2 90 x + 2 , 160 x 3 + 43 ) 4 ,

where

B ( x ) = 63 , 717 , 283 , 707 , 480 x + 222 , 339 , 093 , 374 , 520 x 2 + 443 , 153 , 651 , 482 , 080 x 3 + 551 , 897 , 650 , 861 , 200 x 4 + 439 , 841 , 844 , 403 , 200 x 5 + 219 , 097 , 061 , 222 , 400 x 6 + 62 , 377 , 579 , 008 , 000 x 7 + 7 , 772 , 423 , 040 , 000 x 8 + 7 , 984 , 133 , 521 , 441 .

Evidently, v is concave, u is convex, with u()=v()=0, so v<0 and u>0 on [1,). The proof is completed. □

2 Some extensions

In this section we discuss the problem of approximating ( 1 + 1 / n ) n + a , a[0,1], in the form

( 1 + 1 n ) n + a exp ( 1 n p ( n + q ) 2 ) ,p,qR.

More precisely, we propose as an open problem the following approximation formula:

( 1 + 1 n ) n + a exp ( 1 ( 1 2 a ) n ( n + 3 a 2 6 ( 2 a 1 ) ) 2 ) ,a[0,1] { 1 2 } .
(5)

Particular values a=0 and a=1 in (5) lead again to (3) and (4).

The special case a=1/2 is treated at the final part of this section.

By now, we proved the approximation formula (5) for

a[0, 4 2 7 ) ( 1 2 , 4 + 2 7 ) (θ,1],

where θ=0.82462 is the unique real root of 558a1,098 a 2 +675 a 3 92.

This assertion is sustained by the following three theorems.

Theorem 3 Let a[0, 4 2 7 ). Then there exists x 1 >0 (depending on a) such that for every real number x x 1 , the following inequalities hold true:

exp ( 1 ( 1 2 a ) x ( x + 3 a 2 6 ( 2 a 1 ) 7 a 2 8 a + 2 24 ( 2 a 1 ) 2 x ) 2 ) < ( 1 + 1 x ) x + a <exp ( 1 ( 1 2 a ) x ( x + 3 a 2 6 ( 2 a 1 ) ) 2 ) .
(6)

Theorem 4 Let a( 1 2 , 4 + 2 7 ). Then there exists x 2 >0 (depending on a) such that, for every real number x x 2 , the following inequalities hold true:

exp ( 1 + ( a 1 2 ) x ( x + 3 a 2 6 ( 2 a 1 ) + 8 a 7 a 2 2 24 ( 2 a 1 ) 2 x ) 2 ) < ( 1 + 1 x ) x + a <exp ( 1 + ( a 1 2 ) x ( x + 3 a 2 6 ( 2 a 1 ) ) 2 ) .
(7)

Theorem 5 Let a(θ,1]. Then there exists x 3 >0 (depending on a) such that, for every real number x x 3 , the following inequalities hold true:

exp ( 1 + ( a 1 2 ) x ( x + 3 a 2 6 ( 2 a 1 ) ) 2 ) < ( 1 + 1 x ) x + a <exp ( 1 + ( a 1 2 ) x ( x + 3 a 2 6 ( 2 a 1 ) 7 a 2 8 a + 2 24 ( 2 a 1 ) 2 x ) 2 ) .
(8)

Inequalities (6)-(8) are closely related to the functions

s(x)=1+ ( a 1 2 ) x ( x + 3 a 2 6 ( 2 a 1 ) ) 2 (x+a)ln ( 1 + 1 x )

and

t(x)=1+ ( a 1 2 ) x ( x + 3 a 2 6 ( 2 a 1 ) 7 a 2 8 a + 2 24 ( 2 a 1 ) 2 x ) 2 (x+a)ln ( 1 + 1 x ) .

We have

s (x)= P ( x ) x 2 ( x + 1 ) 2 ( 3 a 6 x + 12 a x 2 ) 4 ,
(9)

where

P ( x ) = 648 ( 7 a 2 8 a + 2 ) ( 2 a 1 ) 3 x 3 + 48 ( 18 a 11 ) ( 3 a 2 ) ( 3 a 1 ) ( 2 a 1 ) 2 x 2 + ( 27 a 2 ) ( 2 a 1 ) ( 3 a 2 ) 3 x + a ( 3 a 2 ) 4

and

t (x)= Q ( x ) x 2 ( x + 1 ) 2 ( 24 ( 2 a 1 ) 2 x 2 + 4 ( 2 a 1 ) ( 3 a 2 ) x ( 7 a 2 8 a + 2 ) ) 4 ,
(10)

where

Q ( x ) = 3 , 072 ( 558 a 1 , 098 a 2 + 675 a 3 92 ) ( 2 a 1 ) 6 x 6 + 64 ( 94 , 140 a 2 31 , 392 a 122 , 256 a 3 + 57 , 753 a 4 + 3 , 860 ) ( 2 a 1 ) 5 x 5 + 128 ( 816 a 3 , 556 a 2 + 7 , 164 a 3 6 , 813 a 4 + 2 , 484 a 5 68 ) ( 2 a 1 ) 4 x 4 + 64 ( 7 a 2 8 a + 2 ) ( 942 a 2 356 a 1 , 086 a 3 + 465 a 4 + 48 ) ( 2 a 1 ) 3 x 3 + 16 ( 10 a 14 a 2 + 9 a 3 4 ) ( 7 a 2 8 a + 2 ) 2 ( 2 a 1 ) 2 x 2 + ( 41 a 2 24 a 2 ) ( 7 a 2 8 a + 2 ) 3 ( 2 a 1 ) x a ( 7 a 2 8 a + 2 ) 4 .

Proofs of Theorems 3 and 4 For a[0, 4 2 7 )( 1 2 , 4 + 2 7 ), the leading coefficients of the polynomials P and Q are negative. We are in a position to consider j>0 (depending on a) such that P(x)<0 and Q(x)<0, for every x[j,). By (9) and (10), s is concave and t is convex. But s()=t()=0, so s<0 and t>0 on [j,).

Now inequalities s(x)<0 and t(x)>0, for every x[j,) are (6) and (7) and we are done. □

Proof of Theorem 5 For a(θ,1], the leading coefficients of polynomials P and Q are positive. We are in a position to consider l>0 (depending on a) such that P(x)>0 and Q(x)>0, for every x[l,). By (9) and (10), now s is convex and t is concave. But s()=t()=0, so s>0 and t<0 on [l,).

Now inequalities s(x)>0 and t(x)<0, for every x[l,) are (6) and (7) and we are done. □

The special case a=1/2 provides the approximation formula

( 1 + 1 n ) n + 1 2 exp ( 1 + n 12 ( n + 1 3 ) 3 ) .

It is sustained by the following.

Theorem 6 For every real number x1, the following inequalities hold:

exp ( 1 + x 12 ( x + 1 3 ) 3 ) < ( 1 + 1 x ) x + 1 2 <exp ( 1 + x 12 ( x + 1 3 7 90 x ) 3 ) .
(11)

Proof Let us define the functions

b(x)=1+ x 12 ( x + 1 3 ) 3 ( x + 1 2 ) ln ( 1 + 1 x )

and

c(x)=1+ x 12 ( x + 1 3 7 90 x ) 3 ( x + 1 2 ) ln ( 1 + 1 x ) .

We have

b (x)= 15 x + 171 x 2 + 189 x 3 + 1 2 x 2 ( x + 1 ) 2 ( 3 x + 1 ) 5

and

c (x)= R ( x 1 ) 2 x 2 ( x + 1 ) 2 ( 30 x + 90 x 2 7 ) 5 ,

where

R ( x ) = 44 , 117 , 710 , 950 x + 122 , 222 , 582 , 550 x 2 + 187 , 066 , 692 , 000 x 3 + 170 , 643 , 105 , 000 x 4 + 92 , 646 , 666 , 000 x 5 + 27 , 671 , 382 , 000 x 6 + 3 , 499 , 200 , 000 x 7 + 6 , 793 , 216 , 207 .

Evidently, b is concave, c is convex, with b()=c()=0, so b<0 and c>0 on [1,). The proof is completed. □

In case a=1/2, the entire asymptotic representation

( 1 + 1 x ) x + 1 / 2 exp ( 1 + 1 12 x ( x + j = 0 a j x j ) 3 ) ,x,
(12)

can be constructed. In this sense, we write (12) as

( f ( x ) ) 1 / 3 x+ j = 0 a j x j ,

where

f(x)= ( x + 1 2 ) ln ( 1 + 1 x ) 1 1 12 x .

By using the Maclaurin expansion of ln(1+t), with t= x 1 , we deduce that

f(x)= j = 3 ( 1 ) j 1 6 ( j 2 ) j ( j 1 ) x j .

Now the coefficients a j in (12) can be inductively obtained by equating the following relation:

{ j = 3 ( 1 ) j 1 6 ( j 2 ) j ( j 1 ) x j } × { x + j = 0 a j x j } 3 =1.

The first coefficients are a 0 = 1 3 , a 1 = 7 90 (see (11)), then a 2 = 16 405 , a 3 = 2 , 141 85 , 050 , ….

Further research in the problem of approximating the constant e can be found in [14].

Finally, we leave as an open problem the approximation formula (5) for values of a[0,1] other than those discussed in this paper.