arXiv:math/0606765v3 [math.

NT] 25 Jul 2006

A REMARK ON THE MANDLS INEQUALITY

MEHDI HASSANI
Abstract. In this note, first we refine Mandls inequality. Then, we consider the product
p1 p2 pn and we refine some known lower bounds for it, and we find some upper bounds
for it by using Mandls inequality and its refinement and the AGM-Inequality.

1. Introduction
As usual, let pn be the nth prime. The Mandls inequality [2, 11] asserts that for every
n 9 we have
(1.1)

n
X

pi <

i=1

n
pn .
2

Considering the AGM Inequality [9] and (1.1), for every n 9 we obtain
!n
n
 p n
1X
n
.
p1 p2 pn <
pi
<
n i=1
2
So, we have
(1.2)

p1 p2 pn <

 p n
n

(n 9),

where also holds true by computation for 5 n 8. In other hand, one can get a trivial lower
bound for that product using Euclids proof of infinity of primes; Letting En = p1 p2 pn 1
for every n 2, it is clear that pn < En . So, if pn < En < pn+1 then En should has a prime
factor among p1 , p2 , , pn which isnt possible. Thus En pn+1 and for every n 2 we
have
p1 p2 pn > pn+1 .
In 1957 in [6], Bonse used elementary methods to show that
p1 p2 pn > p2n+1

(n 4),

2000 Mathematics Subject Classification. 11A41.

Key words and phrases. Primes, Inequalities, AGM-Inequality.
1

MEHDI HASSANI

and
p1 p2 pn > p3n+1

(n 5).

In 1960 Posa [5] proved that for every k > 1 there exists an nk such that for all n nk we
have
p1 p2 pn > pkn+1 .
In 1988, J. Sandor found some inequalities of similar type; For example he showed that for
every n 24 we have
p1 p2 pn > p2n+5 + p2[n ] .
2

In 2000 Panaitopol [4] showed that in Posas result we can get nk = 2k. More precisely, he
proved that for every n 2 we have
n(n)

p1 p2 pn > pn+1

in which (x) = the number of primes x. In this paper, first we refine Mandls inequality
P
2
by showing n2 pn ni=1 pi > n14 for every n 10. This refinement helps us to sharpen (1.2).
Also, we refine Panaitopols result by proving
(1

p1 p2 pn > pn+1log n

)(n(n))

(n 101).

During proofs we will need some known results which we review them briefly; we have the
following known bounds for the function (x), [2]:


x
1
(1.3)
(x)
1+
log x
log x

(x 599),

and
(1.4)

1.2762 
x 
1+
(x)
log x
log x

(x 2).

For every n 53, we have [4]

(1.5)

log pn+1 < log n + log log n +

log log n 0.4

.
log n

Also, for every n 3, we have [7]



log log n 2.1454
(1.6)
(pn ) > n log n + log log n 1 +
,
log n
P
in which (x) = px log p. Specially, (pn ) = log(p1 p2 pn ) and this will act as a key for

approximating p1 p2 pn . Finally, just for insisting we note that base of all logarithms are
e.

A REMARK ON THE MANDLS INEQUALITY

2. Refinement of Mandls Inequality

To prove Mandls inequality, Dusart ([2], page 50) uses the following inequality


Z pn
3
p2n
1+
(n 109),
(t)dt c +
(2.1)
2 log pn
2 log pn
2
in which
c = 35995 3Li(5992 ) +

5992
47.06746,
log 599

and
Li(x) = lim+
0

Z

1
0

dt
+
log t

x
1+

dt
log t

is logarithmic integral [1]. Note that he has got (2.1) using (1.3). Also, for using (2.1) to
prove Mandls inequality we note that
Z pn
n
n
n
n
X
X
X

X


(t)dt =
pi pi1 (i 1) =
ipi (i 1)pi1
pi = npn
pi .
2

i=2

i=2

i=2

i=1

Therefore, we have
(2.2)

npn

n
X
i=1

p2n
pi c +
2 log pn

3
1+
2 log pn

(n 109).

Considering (1.4) and (2.2), for every n 109 we obtain





n
X
p2n
p2n
1.2762
0.2238
npn
pi c +
+
1+
2 log pn log pn
2 log pn
log pn
i=1

pn
n
p2n
p2n
c + 0.1119 2 + (pn ) = c + 0.1119 2 + pn .
2
log pn
log pn 2

So, for every n 109 we have

n

(2.3)

X
p2
n
pn
pi c + 0.1119 2n .
2
log pn
i=1

In other hand, we have the following bounds for pn ([10], page 69)
n log n pn n(log n + log log n)

(n 6).

Combining these bounds with (2.3), for every n 109 we yield that
n

X
n
0.1119(n log n)2

.
pi c +
pn
2
log2 n(log n + log log n)
i=1

MEHDI HASSANI
2

log n)
>
Now, for every n 21152 we have c + log20.0119(n
(n(log n+log log n))

n2
,
14

and so we obtain the following

inequality for every n 21152

n

X
n
n2
pn
pi > .
2
14
i=1

By computation we observe that it holds also for 10 n 21151. Thus, we get the following
refinement of Mandls inequality
(2.4)

n
X
i=1

pi <

n2
n
pn
2
14

(n 10).

3. Approximation of the Product p1 p2 pn

Using (2.4) and the AGM inequality we have
p
n n
n
(3.1)
p1 p2 pn <

2
14

(n 10).

Note that (3.1) holds also for 5 n 9. This yields an upper bound for the product
p1 p2 pn . About lower bound, as mentioned in introduction we show that
(3.2)

(1

p1 p2 pn > pn+1log n

)(n(n))

(n 101).

To prove this considering (1.3), (1.5) and (1.6) it is enough to prove that
!

1 log1 n
1 log1 n
log log n 0.4
1

log n + log log n +

log n
log n
log2 n
< log n + log log n 1 +

log log n 2.1454

log n

(n 599),

which by putting x = log n, is equivalent with:

1.7454x3 + 1.4x2 0.4
< log x
x3 + x2 x 1

x log 599,

and trivially this holds true; because for x log 599 we have

1.7454x3 +1.4x2 0.4

x3 +x2 x1

< 1.7454 and

1.85 < log x. Therefore, we yield (3.2) for all n 599. For 101 n 598 computation
verifies it.
Finally, we use a refinement of the AGM inequality to get some better bounds. In [8] Rooin
shows that for any non-negative real numbers x1 x2 xn we have
n

(3.3)

1
1
1 X nk
n
n
An1
(xnn An1
)k 0,
An Gn
n k=2

A REMARK ON THE MANDLS INEQUALITY

Pn

xi and Gnn =

Qn

xi . For using this refinements we need Robins

P
inequality (see [2], page 51) which gives a lower bound for the average n1 ni=1 pi ; for every
in which nAn =

i=1

i=1

n 2 it asserts

1X
pi .
n i=1

p n2

(3.4)

Applying (3.3) on p1 < p2 < < pn and using relations (3.4) and (2.4), for every n 10
we obtain
(3.5)

n p

p1 p2 pn <

in which

on
n
(n) ,
14

k

n
p
1
1 X nk
n  n1
n
n
n
(n) =
p n1 pn
> 0.

n k=2 2
2
14

In fact, all members under summation are positive. So


 n1 n
p
n
1
pn  1
1
n
n
n
n
>
pn
2 1 .
(n) >

n
2
14
2n
Using this bound for (n) and considering (3.5), for every n 10 we obtain
n
 1
n

p
2n 1
n
n
.
1
p1 p2 pn <
2
n
14
4. On a Limit Concerning the Product p1 p2 pn

Some people believe that e is The Master of All [3]. There are some reasons, which
Q
Q
one of them is the the result limn n = e with ( n )pn = p1 p2 pn = e(pn ) (see [12]). In
(pn )
Q
fact, considering the Prime Number Theorem that is n = e pn = e + o(1), when n .
Q
In this section, we prove that n = e + O( log4 (n1log n) ), when n . It is known [2] that for

x > 1, we have
(4.1)

|(x) x| < d

where d = 1717433. Using this and

d
log4 pn

<

=e

(pn )
pn

x
,
log4 x

, we obtain
d

< e log4 pn

(n 1).

MEHDI HASSANI

We have e

d
log 4 pn

> 1

d
.
log4 pn

d
log 4 pn

Also, for pn > 5270747586811033 a geometric approximation

2

d
, and so
< 1 + logd4 pn + 2 log4 pn (log
4
pn d)
Q
d2
d
d
n
<
+
1
<
1
+
e
log4 pn
log4 pn 2 log4 pn (log4 pn d)

yields e

(pn > 5270747586811033).

It is known [10] that pn > n log n for every n 1. Using this, for every pn > 5270747586811033
we obtain
Q
d
d
d2
n
<
1
+
<
+
.
1
e
log4 (n log n)
log4 (n log n) 2 log4 (n log n)(log4 (n log n) d)
Q
Q
This describes limn n = e explicitly and also yields that n = e + O( log4 (n1log n) ), as we
claimed.

References
[1] M. Abramowitz and I.A. Stegun, HANDBOOK OF MATHEMATICAL FUNCTIONS: with Formulas,
Graphs, and Mthematical Tables, Dover Publications, 1972.
[2] Pierre Dusart, Autour de la fonction qui compte le nombre de nombres premiers, PhD. Thesis, 1998.
[3] B.J. McCartin, e: The Master of All, Mathematical Intelligencer, Volume 28, Number 2, 2006, 10-21.
[4] Laurentiu Panaitopol, An Ineqiality involving Prime Numbers, Univ. Beograd. Publ. Elektrotehn. Fak,
Ser. Mat. 11 (2000), 33-35.

[5] L. P
osa, Uber
eine Eigenschaft der Primzahlen (Hungarian), Mat. Lapok, 11(1960), 124-129.
[6] H. Rademacher and O. Toeplitz, The enjoyment of mathematics, Princeton Univ. Press, 1957.
[7] G. Robin, Estimation de la fonction de Tschebyshev sur le ki`eme nombre premier et grandes valeurs
de la fonction (n), nombre des diviseurs premier de n, Acta. Arith, 43(1983), 367-389.
[8] J. Rooin, AGM inequality with binomial expansion, Elemente der Mathematik, 58 (2003) 115-117.
[9] J. Rooin, Some New Proofs for the AGM Inequality, Mathematical Inequalities & Applications, Vol. 7,
No. 4, (2004)517-521.
[10] J. Barkley Rosser & L. Schoenfeld, Approximate Formulas for Some Functions of Prime Numbers,
Illinois Journal Math., 6 (1962) pp. 64-94.
[11] J. Barkley Rosser & L. Schoenfeld, Sharper Bounds for the Chebyshev Functions (x) and (x), Math.
Of Computation, Vol. 29, Number 129 (January 1975) pp. 243-269.
[12] S.M. Ruiz, A Result on Prime Numebrs, Mathematical Gazette, 81 (1997), 269-270.

[13] J. Sandor, Uber

die Folge der Primzahlen, Mathematica (Cluj), 30(53)(1988), 67-74.
Mehdi Hassani,
Department of Mathematics, Institute for Advanced Studies in Basic Sciences, P.O. Box
45195-1159, Zanjan, Iran
E-mail address: mmhassany@yahoo.com