1 Introduction

Let p>1 and a n >0 (n=1,2,) with n = 1 a n p <+, then Hardy’s well-known inequality [1] is given by

n = 1 ( 1 n k = 1 n a k ) p < ( p p 1 ) p n = 1 a n p .
(1.1)

Recently, the refinement, improvement, generalization, extension, and application for Hardy’s inequality have attracted the attention of many researchers [210].

Yang and Zhu [11] presented an improvement of Hardy’s inequality (1.1) for p=2 as follows:

n = 1 ( 1 n k = 1 n a k ) 2 <4 n = 1 ( 1 1 3 n + 5 ) a n 2 .

For 7/6p2, Huang [12] proved that

n = 1 ( 1 n k = 1 n a k ) p < ( p p 1 ) p n = 1 ( 1 15 196 ( n 1 1 / p + 3 , 436 ) ) a n p .

In [13], Wen and Zhang proved that the inequality

n = 1 ( 1 n k = 1 n a k ) p < ( p p 1 ) p n = 1 ( 1 C p 2 n 1 1 / p ) a n p
(1.2)

holds for p>1 if a n >0 (n=1,2,), with n = 1 a n p <+, where C p =1 ( 1 1 / p ) p 1 for p2 and C p =11/p for 1<p2.

Xu et al. [14] gave a further improvement of the inequality (1.2):

n = 1 ( 1 n k = 1 n a k ) p < ( p p 1 ) p n = 1 ( 1 Z p 2 ( n 1 ) 1 1 / p ) a n p ,

where Z p =p1 ( p 1 ) 2 p 2 1 / p for 1<p2 and Z p =1 ( p 1 p ) p 1 2 p 1 p for p>2.

For the special parameter p=5/4, Deng et al. [15] established

n = 1 ( 1 n k = 1 n a k ) 5 / 4 5 5 / 4 n = 1 ( 1 1 10 ( n 1 / 5 + η 5 / 4 ) ) a n 5 / 4 ,

where η 5 / 4 = 5 5 / 4 /[10( 5 5 / 4 ( n = 1 1 n 5 / 4 ( m = 1 n m 4 / 5 ) 1 / 4 ))1]=0.46 .

In [16], Long and Linh discussed Hardy’s inequality with the parameter p<0, and proved that

n = 1 ( 1 n k = 1 n a k ) p < ( p p 1 ) p n = 1 a n p
(1.3)

for p1 and

n = 1 ( 1 n k = 1 n a k ) p < 2 1 p 1 p n = 1 a n p

for 1<p<0 if a n >0 (n=1,2,) with n = 1 a n p <+.

It is the aim of this paper to present an improvement of inequality (1.3) for the parameter p1. Our main result is Theorem 1.1.

Theorem 1.1 Let p1, d(p)=(1+( 2 1 / p 1)p)/[8(1+( 2 1 / p 1)p)+2] and a n >0 (n=1,2,) with n = 1 a n p <+, then

n = 1 ( 1 n k = 1 n a k ) p ( p p 1 ) p n = 1 ( 1 d ( p ) ( n 1 / 2 ) 1 1 / p ) a n p .

2 Lemmas

In order to establish our main result we need several lemmas, which we present in this section.

Lemma 2.1 (see [[17], Corollary 1.3])

Suppose that a,bR with a<b, f: [ a , b ] n R has continuous partial derivatives and

D m = { x = ( x 1 , x 2 , , x n ) | a min 1 k n { x k } < x m = max 1 k n { x k } b } ,m=1,2,,n.

If f ( x ) x m >0 holds for all x=( x 1 , x 2 ,, x n ) D m and m=1,2,,n, then

f( x 1 , x 2 ,, x n )f( x min , x min ,, x min )

for all x m [a,b] (m=1,2,,n), where x min = min 1 k n { x k }.

Lemma 2.2 Let nR be a positive natural number and rR with r1. Then

k = 1 n ( k 1 2 ) 1 / r r r + 1 ( n 1 + 1 / r 1 ) + 2 1 / r .
(2.1)

Proof We use mathematical induction to prove inequality (2.1). We clearly see that inequality (2.1) becomes equality for n=1. We assume that inequality (2.1) holds for n=i (iN, i1), namely

k = 1 i ( k 1 2 ) 1 / r r r + 1 ( i 1 + 1 / r 1 ) + 2 1 / r .

Then for n=i+1 we have

k = 1 i + 1 ( k 1 2 ) 1 / r = k = 1 i ( k 1 2 ) 1 / r + ( i + 1 2 ) 1 / r r r + 1 ( i 1 + 1 / r 1 ) + 2 1 / r + ( i + 1 2 ) 1 / r = r r + 1 [ ( i + 1 ) 1 + 1 / r 1 ] + 2 1 / r + ( i + 1 2 ) 1 / r i i + 1 x 1 / r d x .
(2.2)

Note that x 1 / r (r1) is concave on (0,+), therefore Hermite-Hadamard’s inequality implies that

( i + 1 2 ) 1 / r i i + 1 x 1 / r dx.
(2.3)

From (2.2) and (2.3) we know that inequality (2.1) holds for n=i+1. □

Remark 2.1 The inequality

2 1 / r r r + 1
(2.4)

holds for all r1 with equality if and only if r=1.

Proof We clearly see that inequality (2.4) becomes equality for r=1.

If r>1, then it is well known that the function ( 1 + 1 / r ) r is strictly increasing on (1,+), so we get

( 1 + 1 r ) r >2.
(2.5)

Therefore, inequality (2.4) follows from (2.5). □

Lemma 2.3 The inequality

( 3 r + 1 ) 2 1 1 / r r + 1 > ( 2 1 / r r r + 1 ) 3 r 2 + 1 r
(2.6)

holds for all r1.

Proof Let r1, then we clearly see that

(6log23) r 2 +r+2log224(2log21)>0.
(2.7)

Inequality (2.7) leads to

e ( 3 r 2 r + 2 ) / ( 6 r 2 + 2 ) <2.
(2.8)

It follows from the well-known inequality ( 1 + x ) 1 / x <e (x>0) that

e> ( 1 + 3 r 2 r + 2 6 r 3 + 2 r ) ( 6 r 3 + 2 r ) / ( 3 r 2 r + 2 ) .
(2.9)

From (2.8) and (2.9) we have

2 1 / r < 6 r 3 + 2 r 6 r 3 + 3 r 2 + r + 2 .
(2.10)

Therefore, inequality (2.6) follows easily from (2.10). □

Lemma 2.4 Let r1 and

f(x)= x r ( r r + 1 x 1 + 1 / r + 2 1 / r r r + 1 ) r + 1 .
(2.11)

Then f is convex on [1/2,+).

Proof From (2.11) we have

f ( x ) = 2 r + 1 r + 1 x r + 1 / r + ( 2 1 / r r r + 1 ) r x r 1 ( r r + 1 x 1 + 1 / r + 2 1 / r r r + 1 ) r + 2 , f ( x ) = ( 3 r + 1 ) ( 2 r + 1 ) ( r + 1 ) 2 x 2 + 2 / r ( 2 1 / r r r + 1 ) ( 3 r 2 + 1 ) ( 2 r + 1 ) r ( r + 1 ) x 1 + 1 / r + r ( r 1 ) ( 2 1 / r r r + 1 ) 2 ( r r + 1 x 1 + 1 / r + 2 1 / r r r + 1 ) r + 3 x r 2 .
(2.12)

It follows from Lemma 2.3 and (2.12) that

f (x) ( 3 r + 1 ) ( 2 r + 1 ) ( r + 1 ) 2 2 1 1 / r ( 2 1 / r r r + 1 ) ( 3 r 2 + 1 ) ( 2 r + 1 ) r ( r + 1 ) ( r r + 1 x 1 + 1 / r + 2 1 / r r r + 1 ) r + 3 x r + 1 / r 1 >0
(2.13)

for all x[1/2,+).

Therefore, Lemma 2.4 follows from inequality (2.13). □

Lemma 2.5 Let r1, 0t4 and c=(r+1 2 1 / r r)/[8(r+1 2 1 / r r)+2], then

(r+1) ( 2 1 / r r r + 1 ) (1ct)t1 [ 1 + ( 1 + r 2 1 / r r 1 ) t ] r .
(2.14)

Proof If r=1, then we clearly see that inequality (2.14) becomes equality. Next, we assume that r>1. Let

f(t)=(r+1) ( 2 1 / r r r + 1 ) (1ct)t1+ [ 1 + ( 1 + r 2 1 / r r 1 ) t ] r .
(2.15)

Then simple computations lead to

f(0)=0,
(2.16)
f (t)= ( r + 1 ) ( 2 1 / r r r + 1 ) [ 1 + ( r + 1 2 1 / r r 1 ) t ] r + 1 { ( 1 2 c t ) [ 1 + ( r + 1 2 1 / r r 1 ) t ] r + 1 1 } .
(2.17)

Note that

12ct18c= 1 4 ( r + 1 2 1 / r r ) + 1 >0,
(2.18)
[ 1 + ( r + 1 2 1 / r r 1 ) t ] r + 1 1+(r+1) ( r + 1 2 1 / r r 1 ) t.
(2.19)

It follows from Remark 2.1 and (2.17)-(2.19) that

f (t) ( r + 1 ) ( 2 1 / r r r + 1 ) [ 1 + ( r + 1 2 1 / r r 1 ) t ] r + 1 { ( 1 2 c t ) [ 1 + ( r + 1 ) ( r + 1 2 1 / r r 1 ) t ] 1 } .
(2.20)

Let

g(t)=(12ct) [ 1 + ( r + 1 ) ( r + 1 2 1 / r r 1 ) t ] 1.
(2.21)

Then from g(0)=0 and g(4)=4 ( r + 1 2 1 / r r ) 2 /[ 2 1 / r r(4(r+1 2 1 / r r)+1)]0 together with the fact that g(t) is a concave parabola we know that

g(t)0
(2.22)

for t[0,4].

Therefore, Lemma 2.5 follows easily from (2.15) and (2.16) together with (2.20)-(2.22). □

Lemma 2.6 Let r1, c=(r+1 2 1 / r r)/[8(r+1 2 1 / r r)+2], N is a positive natural number, a k >0 (k=1,2,,N) and B N = min 1 k N { ( k 1 / 2 ) 1 / r a k }, then

( r + 1 r ) r n = 1 N ( 1 c ( n 1 / 2 ) 1 + 1 / r ) a n r n = 1 N ( n k = 1 n 1 / a k ) r B N r [ ( r + 1 r ) r n = 1 N ( 1 c ( n 1 / 2 ) 1 + 1 / r ) 1 n 1 / 2 n = 1 N ( n k = 1 n ( k 1 / 2 ) 1 / r ) r ] .
(2.23)

Proof Let a k = b k / ( k 1 / 2 ) 1 / r (k=1,2,,N), then B N = min 1 k N { b k } and inequality (2.23) becomes

( r + 1 r ) r n = 1 N ( 1 c ( n 1 / 2 ) 1 + 1 / r ) b n r n 1 / 2 n = 1 N ( n k = 1 n ( k 1 / 2 ) 1 / r b k ) r B N r [ ( r + 1 r ) r n = 1 N ( 1 c ( n 1 / 2 ) 1 + 1 / r ) 1 n 1 / 2 n = 1 N ( n k = 1 n ( k 1 / 2 ) 1 / r ) r ] .
(2.24)

Let D m ={b=( b 1 , b 2 ,, b N )| b m = max 1 k N { b k }> min 1 k N { b k }} (m=1,2,,N), and

f ( b 1 , b 2 , , b N ) = ( r + 1 r ) r n = 1 N ( 1 c ( n 1 / 2 ) 1 + 1 / r ) b n r n 1 / 2 n = 1 N ( n k = 1 n ( k 1 / 2 ) 1 / r b k ) r .
(2.25)

Then for any b D m (m=1,2,,N) we have

f ( b ) b m = ( 1 c ( m 1 / 2 ) 1 + 1 / r ) ( r + 1 ) r b m r 1 ( m 1 / 2 ) r r 1 r ( m 1 / 2 ) 1 / r b m 2 n = m N n r ( k = 1 n ( k 1 / 2 ) 1 / r b k ) r + 1 > ( 1 c ( m 1 / 2 ) 1 + 1 / r ) ( r + 1 ) r b m r 1 ( m 1 / 2 ) r r 1 r ( m 1 / 2 ) 1 / r b m r 1 n = m + n r ( k = 1 n ( k 1 / 2 ) 1 / r ) r + 1 .
(2.26)

From Lemma 2.2 and (2.26) one has

1 r ( m 1 / 2 ) 1 / r b m r 1 f ( b ) b m > ( r + 1 r ) r ( 1 c ( m 1 / 2 ) 1 + 1 / r ) 1 ( m 1 / 2 ) 1 + 1 / r n = m + n r ( r r + 1 n 1 + 1 / r + 2 1 / r r r + 1 ) r + 1 .
(2.27)

It clearly follows from Lemma 2.4 and the Hermite-Hadamard inequality that

m 1 / 2 m + 1 / 2 x r ( r r + 1 x 1 + 1 / r + 2 1 / r r r + 1 ) r + 1 m r ( r r + 1 m 1 + 1 / r + 2 1 / r r r + 1 ) r + 1

and

m 1 / 2 + x r ( r r + 1 x 1 + 1 / r + 2 1 / r r r + 1 ) r + 1 n = m + n r ( r r + 1 n 1 + 1 / r + 2 1 / r r r + 1 ) r + 1 .
(2.28)

Note that

m 1 / 2 + x r ( r r + 1 x 1 + 1 / r + 2 1 / r r r + 1 ) r + 1 = ( 1 + r r ) r 2 1 / r r + 1 2 1 / r r { 1 [ 1 + ( r + 1 2 1 / r r 1 ) ( m 1 / 2 ) 1 1 / r ] r } ,
(2.29)
0< ( m 1 / 2 ) 1 1 / r 2 1 + 1 / r 4.
(2.30)

From Lemma 2.5 and (2.30) one has

( r + 1 r ) r ( 1 c ( m 1 / 2 ) 1 + 1 / r ) 1 ( m 1 / 2 ) 1 + 1 / r ( 1 + r r ) r 2 1 / r r + 1 2 1 / r r { 1 [ 1 + ( r + 1 2 1 / r r 1 ) ( m 1 / 2 ) 1 1 / r ] r } .
(2.31)

Inequalities (2.27), (2.28), and (2.31) together with (2.29) lead to the conclusion that

f ( b ) b m >0
(2.32)

for any b=( b 1 , b 2 ,, b N ) D m and m=1,2,,N.

It follows from Lemma 2.1 and (2.32) that

f( b 1 , b 2 ,, b N )f( B N , B N ,, B N ).
(2.33)

Therefore, inequality (2.24) follows from (2.25) and (2.33). □

Lemma 2.7 Let r1, c=(r+1 2 1 / r r)/[8(r+1 2 1 / r r)+2], then

( r + 1 r ) ( 1 2 1 + 1 / r c ) >1.
(2.34)

Proof We clearly see that inequality (2.34) holds for r=1. Next, we assume that r>1, let t= 2 1 + 1 / r , then 0<t<4 and Lemma 2.5 leads to

( r + 1 r ) r ( 1 2 1 + 1 / r c ) 1 ( r + 1 r ) r 2 ( r + 1 2 1 / r r ) [ 1 ( 1 + 2 ( r + 1 2 1 / r r ) r ) r ] 1 ( r + 1 r ) r 2 ( r + 1 2 1 / r r ) 2 ( r + 1 2 1 / r r ) 1 + 2 ( r + 1 2 1 / r r ) 1 = ( r + 1 r ) r 1 + 2 ( r + 1 2 1 / r r ) 1 .
(2.35)

Note that

r+1 2 1 / r r<1log2
(2.36)

for all r1. In fact, let x1 and

f(x)=x 2 1 / x x+1.
(2.37)

Then

f (x)=1+ ( log 2 x 1 ) 2 1 / x ,
(2.38)
f (x)= ( log 2 ) 2 x 3 2 1 / x <0.
(2.39)

It follows from (2.38) and (2.39) that

f (x)> lim x + [ 1 + ( log 2 x 1 ) 2 1 / x ] =0.
(2.40)

Equation (2.37) and inequality (2.40) lead to the conclusion that

f(x)< lim x + ( x 2 1 / x x + 1 ) =1log2.
(2.41)

From (2.35) and (2.36) together with the fact that [ ( r + 1 ) / r ] r 2 we have

( r + 1 r ) r ( 1 2 1 + 1 / r c ) 1> 2 1 + 2 ( 1 log 2 ) 1= 2 log 2 1 3 2 log 2 >0.
(2.42)

Therefore, inequality (2.34) follows from (2.42). □

Lemma 2.8 Let r1, c=(r+1 2 1 / r r)/[8(r+1 2 1 / r r)+2], N is a positive natural number, a k >0 (k=1,2,,N) and B N = min 1 k N { ( k 1 / 2 ) 1 / r a k }, then

( r + 1 r ) r n = 1 N ( 1 c ( n 1 / 2 ) 1 + 1 / r ) a n r n = 1 N ( n k = 1 n 1 / a k ) r 2 B N r [ ( r + 1 r ) r ( 1 2 1 + 1 / r c ) 1 ] .
(2.43)

Proof Let m{1,2,,N}, f(0)=0 and

f(m)= ( r + 1 r ) r n = 1 m ( 1 c ( n 1 / 2 ) 1 + 1 / r ) 1 n 1 / 2 n = 1 m ( n k = 1 n ( k 1 / 2 ) 1 / r ) r .
(2.44)

Then

f(1)=2 [ ( 1 + r r ) r ( 1 2 1 + 1 / r c ) 1 ] ,
(2.45)
f(m)f(m1)= ( 1 + r r ) r m 1 / 2 ( 1 c ( m 1 / 2 ) 1 + 1 / r ) ( m k = 1 m ( k 1 / 2 ) 1 / r ) r .
(2.46)

It follows from Lemma 2.2 and (2.46) together with Remark 2.1 that

f ( m ) f ( m 1 ) ( 1 + r r ) r m 1 / 2 ( 1 c ( m 1 / 2 ) 1 + 1 / r ) ( m r r + 1 ( m 1 + 1 / r 1 ) + 2 1 / r ) r ( 1 + r r ) r m 1 / 2 ( 1 c ( m 1 / 2 ) 1 + 1 / r ) ( m r r + 1 m 1 + 1 / r ) r = ( 1 + r r ) r [ ( 4 ( r + 1 2 1 / r r ) + 1 ) ( m 1 / 2 ) 1 + 1 / r m ( r + 1 2 1 / r r ) ] m ( m 1 / 2 ) 2 + 1 / r [ 8 ( r + 1 2 1 / r r ) + 2 ] .
(2.47)

Let

g(t)= [ 4 ( r + 1 2 1 / r r ) + 1 ] ( t 1 / 2 ) 1 + 1 / r ( r + 1 2 1 / r r ) t.
(2.48)

Then

g ( 1 ) = [ 4 ( r + 1 2 1 / r r ) + 1 ] 2 1 1 / r ( r + 1 2 1 / r r ) > ( 2 1 1 / r 1 ) ( r + 1 2 1 / r r ) 0 ,
(2.49)
g ( t ) = ( 1 + 1 r ) [ 4 ( r + 1 2 1 / r r ) + 1 ] ( t 1 / 2 ) 1 / r ( r + 1 2 1 / r r ) > ( 2 2 1 / r 1 ) ( r + 1 2 1 / r r ) 0
(2.50)

for t1.

From (2.47)-(2.50) we get

f(1)<f(2)<<f(N1)<f(N).
(2.51)

Therefore, Lemma 2.8 follows easily from Lemma 2.6, (2.44), (2.45), and (2.51). □

3 Proof of Theorem 1.1

Let r=p, c=c(r)=d(r) and b n =1/ a n (n=1,2,), then r1, c=(r+1 2 1 / r r)/[8(r+1 2 1 / r r)+2], b n >0 and n = 1 b n r <+.

It follows from Lemmas 2.7 and 2.8 that one has

n = 1 N ( n k = 1 n 1 / b k ) r ( r + 1 r ) r n = 1 N ( 1 c ( n 1 / 2 ) 1 + 1 / r ) b n r .
(3.1)

Letting n+, (3.1) leads to

n = 1 ( n k = 1 n 1 / b k ) r ( r + 1 r ) r n = 1 ( 1 c ( n 1 / 2 ) 1 + 1 / r ) b n r .
(3.2)

Therefore, Theorem 1.1 follows immediately from (3.2) and r=p together with b n =1/ a n .