1 Introduction and the main results

We first recall some definitions and basic results on or related to completely monotonic sequences and completely monotonic functions.

Definition 1 [1]

A sequence { μ n } n = 0 is called a moment sequence if there exists a function α(t) of bounded variation on the interval [0,1] such that

μ n = 0 1 t n dα(t),n N 0 .
(1)

Here, in Definition 1 and throughout the paper,

N 0 :={0}N,

and ℕ is the set of all positive integers.

Definition 2 [1]

A sequence { μ n } n = 0 is called completely monotonic if

( 1 ) k Δ k μ n 0,n,k N 0 ,
(2)

where

Δ 0 μ n = μ n
(3)

and

Δ k + 1 μ n = Δ k μ n + 1 Δ k μ n .
(4)

Such a sequence is called totally monotone in [2].

From Definition 2, using mathematical induction, we can prove, for a completely monotonic sequence { μ n } n = 0 , that the sequence { ( 1 ) m Δ m μ n } n = 0 is non-increasing for any fixed m N 0 , and that the sequence { ( 1 ) m Δ m μ n } m = 0 is non-increasing for any fixed n N 0 . The difference equation (4) plays an important role in the proofs of these properties and our main results of this paper.

In [3], the authors showed that for a completely monotonic sequence { μ n } n = 0 , we always have

( 1 ) k Δ k μ n >0,n,k N 0 ,
(5)

unless μ n =c, a constant for all nN.

Let

λ k , m := ( k m ) ( 1 ) k m Δ k m μ m ,k,m N 0 .
(6)

It was shown (see [1]) as follows.

Theorem 1 A sequence { μ n } n = 0 is a moment sequence if and only if there exists a constant L such that

m = 0 k | λ k , m |<L,k N 0 ,
(7)

where in (7), λ k , m is defined by (6).

For completely monotonic sequences, the following is the well-known Hausdorff’s theorem (see [1]).

Theorem 2 A sequence { μ n } n = 0 is completely monotonic if and only if there exists a non-decreasing and bounded function α(t) on [0,1] such that

μ n = 0 1 t n dα(t),n N 0 .
(8)

From this theorem, we know (see [1]) that a completely monotonic sequence is a moment sequence and is as follows.

Theorem 3 A necessary and sufficient condition that the sequence { μ n } n = 0 should be a moment sequence is that it should be the difference of two completely monotonic sequences.

We also recall the following definition.

Definition 3 [1]

A function f is said to be completely monotonic on an interval I if f is continuous on I has derivatives of all orders on I o (the interior of I) and for all n N 0 ,

( 1 ) n f ( n ) (x)0,x I o .
(9)

Some mathematicians use the terminology completely monotone instead of completely monotonic. The class of all completely monotonic functions on the interval I is denoted by CM(I).

The completely monotonic functions and completely monotonic sequences have remarkable applications in probability and statistics [410], physics [11, 12], numerical and asymptotic analysis [2], etc.

For the completely monotonic functions on the interval [0,), Widder proved (see [1]).

Theorem 4 A function f on the interval [0,) is completely monotonic if and only if there exists a bounded and non-decreasing function α(t) on [0,) such that

f(x)= 0 e x t dα(t).
(10)

There is rich literature on completely monotonic functions. For more recent works, see, for example, [1326].

There exists a close relationship between completely monotonic functions and completely monotonic sequences. For example, Widder [27] showed the following.

Theorem 5 Suppose that fCM[a,), then for any δ0, the sequence { f ( a + n δ ) } n = 0 is completely monotonic.

This result was generalized in [28] as follows.

Theorem 6 Suppose that fCM[a,). If the sequence { Δ x k } k = 0 is completely monotonic and x 0 a, then the sequence { f ( x k ) } k = 0 is also completely monotonic.

For the meaning of Δ x k , k N 0 in Theorem 6, see (3) and (4).

Suppose that fCM[0,). By Theorem 5, we know that { f ( n ) } n = 0 is completely monotonic.

The following result was obtained in [16].

Theorem 7 Suppose that the sequence { μ n } n = 0 is completely monotonic, then for any ε(0,1), there exists a continuous interpolating function f(x) on the interval [0,) such that f | [ 0 , ε ] and f | [ ε , ) are both completely monotonic and

f(n)= μ n ,n N 0 .

From this result or Theorem 2, we can get the following.

Theorem 8 Suppose that the sequence { μ n } n = 0 is completely monotonic. Then there exists a completely monotonic interpolating function g(x) on the interval [1,) such that

g(n)= μ n ,nN.

It should be noted that (see [[1], Chapter IV]) under the condition of Theorem 8, we cannot guarantee that there exists a completely monotonic interpolating function g(x) on the interval [0,) such that

g(n)= μ n ,n N 0 .

In this article, we shall further investigate the properties of the completely monotonic sequences. We shall give some necessary conditions, a sufficient condition and a necessary and sufficient condition for sequences to be completely monotonic. More precisely we have the following results.

Theorem 9 Suppose that the sequence { μ n } n = 0 is completely monotonic. Then, for any m N 0 , the series

j = 0 ( 1 ) j Δ j μ m + 1

converges and

μ m j = 0 ( 1 ) j Δ j μ m + 1 .

Corollary 1 Suppose that the sequence { μ n } n = 0 is completely monotonic. Then for m,k N 0 ,

μ m = ( 1 ) k + 1 Δ k + 1 μ m + i = 0 k ( 1 ) i Δ i μ m + 1 .
(11)

Remark 1 Although from the complete monotonicity of the sequence { μ n } n = 0 , we can deduce that for any m N 0 , the series

j = 0 ( 1 ) j Δ j μ m + 1

converges, it cannot guarantee the convergence of the series

j = 0 ( 1 ) j Δ j μ 0 .

For example, let

μ n = 1 n + 1 ,n N 0 .

Since the function

f(x)= 1 x + 1

is completely monotonic on the interval [0,), by Theorem 5, we see that the sequence

{ μ n } n = 0 := { f ( n ) } n = 0 = { 1 n + 1 } n = 0

is completely monotonic. This conclusion can also be obtained by setting

α(t)=t

in Theorem 2.

We can verify that

Δ j μ 0 = ( 1 ) j j + 1 .

Hence,

j = 0 ( 1 ) j Δ j μ 0 = j = 0 1 j + 1

is the famous harmonic series, which is divergent.

Theorem 10 Suppose that the sequence { μ n } n = 0 is completely monotonic. Then for any k,m N 0 ,

( 1 ) k Δ k μ m j = k ( 1 ) j Δ j μ m + 1 .
(12)

Theorem 11 Suppose that the sequence { μ n } n = 1 is completely monotonic and that the series

j = 0 ( 1 ) j Δ j μ 1

converges. Let μ 0 be such that

μ 0 j = 0 ( 1 ) j Δ j μ 1 .

Then the sequence { μ n } n = 0 is completely monotonic.

Theorem 12 A necessary and sufficient condition for the sequence { μ n } n = 0 to be completely monotonic is that the sequence { μ n } n = 1 is completely monotonic, the series

j = 0 ( 1 ) j Δ j μ 1

converges and

μ 0 j = 0 ( 1 ) j Δ j μ 1 .

2 Proofs of the main results

Now, we are in a position to prove the main results.

Proof of Theorem 9 Since { μ n } n = 0 is completely monotonic, by Theorem 2, there exists a non-decreasing and bounded function α(t) on the interval [0,1] such that

μ n = 0 1 t n dα(t),n N 0 .
(13)

From (3), (4) and (13), we can prove that

( 1 ) i Δ i μ n = 0 1 ( 1 t ) i t n dα(t),i,n N 0 .
(14)

Now, for kN, we have

i = 0 k 1 ( 1 ) i Δ i μ m + 1 = i = 0 k 1 0 1 ( 1 t ) i t m + 1 d α ( t ) = 0 1 t m + 1 i = 0 k 1 ( 1 t ) i d α ( t ) = 0 1 t m ( 1 ( 1 t ) k ) d α ( t ) = 0 1 t m d α ( t ) 0 1 ( 1 t ) k t m d α ( t ) = μ m ( 1 ) k Δ k μ m , m N 0 .

Hence, for kN,

μ m = ( 1 ) k Δ k μ m + i = 0 k 1 ( 1 ) i Δ i μ m + 1 ,m N 0 .
(15)

Since

( 1 ) i Δ i μ n 0,i,n N 0 ,
(16)

from (15), we get, for k1,

μ m i = 0 k 1 ( 1 ) i Δ i μ m + 1 ,m N 0 .
(17)

From (16), we also know that

j = 0 ( 1 ) j Δ j μ m + 1 ,m N 0

is a positive series. Then by (17), we obtain that

j = 0 ( 1 ) j Δ j μ m + 1 ,m N 0

converges and that

μ m j = 0 ( 1 ) j Δ j μ m + 1 ,m N 0 .
(18)

The proof of Theorem 9 is thus completed. □

Proof of Corollary 1 This corollary can be obtained from (15). □

Proof of Theorem 10 Let m be a fixed non-negative integer.

From Theorem 9, we see that

μ m j = 0 ( 1 ) j Δ j μ m + 1 ,
(19)

which means that (12) is valid for k=0.

Suppose that (12) is valid for k=r. Then

( 1 ) r + 1 Δ r + 1 μ m = ( 1 ) r + 1 ( Δ r μ m + 1 Δ r μ m ) = ( 1 ) r ( Δ r μ m Δ r μ m + 1 ) = ( 1 ) r Δ r μ m ( 1 ) r Δ r μ m + 1 j = r ( 1 ) j Δ j μ m + 1 ( 1 ) r Δ r μ m + 1 = j = r + 1 ( 1 ) j Δ j μ m + 1 ,
(20)

which means that (12) is valid for k=r+1. Therefore, by mathematical induction, (12) is valid for all k N 0 . The proof of Theorem 10 is completed. □

Proof of Theorem 11 By the definition of completely monotonic sequence, we only need to prove that

( 1 ) k Δ k μ 0 0,k N 0 .
(21)

We first prove that

( 1 ) k Δ k μ 0 j = k ( 1 ) j Δ j μ 1 ,k N 0 .
(22)

From the condition of Theorem 11, (22) is valid for k=0.

Suppose that (22) is valid for k=m. Then we have

( 1 ) m + 1 Δ m + 1 μ 0 = ( 1 ) m + 1 ( Δ m μ 1 Δ m μ 0 ) = ( 1 ) m ( Δ m μ 0 Δ m μ 1 ) = ( 1 ) m Δ m μ 0 ( 1 ) m Δ m μ 1 j = m ( 1 ) j Δ j μ 1 ( 1 ) m Δ m μ 1 = j = m + 1 ( 1 ) j Δ j μ 1 ,
(23)

which means that (22) is valid for k=m+1. Therefore, by mathematical induction, (22) is valid for all k N 0 .

Since

j = 0 ( 1 ) j Δ j μ 1

is a convergent positive series, we know that

j = k ( 1 ) j Δ j μ 1 0,k N 0 .
(24)

From (22) and (24), we obtain that

( 1 ) k Δ k μ 0 0,k N 0 .

The proof of Theorem 11 is completed. □

Proof of Theorem 12 By Definition 2 and by setting m=0 in Theorem 9, we see that the condition is necessary. By Theorem 11, we know that the condition is sufficient. The proof of Theorem 12 is completed. □