Skip to Content
AxiomsAxioms
  • Article
  • Open Access

29 September 2026

21 Pages

A Half-Discrete Mulholland-Type Inequality Involving One Multiple Upper Limit Function

,
,
and
1
School of Mathematics and Physics, Hechi University, Yizhou 546300, China
2
School of Mathematics, Guangdong University of Education, Guangzhou 510303, China
*
Authors to whom correspondence should be addressed.
This article belongs to the Special Issue Applications in Functional Analysis

Abstract

In this paper, by means of weight coefficient methods and the Euler–Maclaurin summation formula, a half-discrete Mulholland-type inequality involving one multiple upper limit function is established, with the kernel ( x + n β ) − λ (λ > 0, β ∈ (0, 1]). In comparison with the existing result with the kernel ( x + n β ) − λ , the homogeneous-type kernel is replaced by a logarithmic (Mulholland-type) one, the admissible parameter ranges are enlarged from λ ∈ (0, 5 − m], m ∈ {0, 1, 2, 3, 4} to λ > 0 with arbitrary m ∈ ℕ0, and the partial sum is removed. The best constant factor of the obtained inequality is Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) . Several equivalent conditions for the best possible constant are further provided, together with some particular cases and extended forms; in particular, for m = 0 with specific parameters, the classical Mulholland inequality is recovered.

1. Introduction

We begin by recalling three classical inequalities that serve as the background of our study. Let p > 1, 1 p + 1 q = 1 , and let {am},{bn} be nonnegative sequences with finite positive p- and q-norms, i.e., 0 < ∑ m = 1 ∞ a m p < ∞ , 0 < ∑ n = 1 ∞ b n q < ∞ . The Hardy–Hilbert inequality [cf. [1], Theorem 315] states
∑ m = 1 ∞ ∑ n = 1 ∞ a m b n m + n < π sin ( π / p ) ∑ m = 1 ∞ a m p 1 p ∑ n = 1 ∞ b n q 1 q
with the best constant factor π/sin(π/p). Its logarithmic counterpart, known as Mulholland’s inequality [cf. [1], Theorem 343], replaces m + n by ln(mn) and preserves the same best constant:
∑ m = 2 ∞ ∑ n = 2 ∞ a m b n ln mn < π sin ( π / p ) ∑ m = 2 ∞ a m p m 1 − p 1 p ∑ n = 2 ∞ b n q n 1 − q 1 q .
Moreover, setting λ = 1, ν(n) = lnn, r = q, s = p in Formula (12) of [2] yields the following half-discrete Mulholland-type inequality for b > 0:
∫ b ∞ ∑ n = 2 ∞ f ( x ) b n u ( x ) + ln n < π sin ( π / p ) ∫ b ∞ f p ( x ) ( u ′ ( x ) ) p − 1 dx 1 p ∑ n = 2 ∞ b n q n 1 − q 1 q .
These fundamental inequalities have been extended and refined in various directions over the past decades, as briefly reviewed in what follows.
In 2006, Krnić et al. [3] proposed a generalized form of inequality (1), which reads
∑ m = 1 ∞ ∑ n = 1 ∞ a m b n ( m + n ) λ < B ( λ 1 , λ 2 ) ∑ m = 1 ∞ m p ( 1 − λ 1 ) − 1 a m p 1 p ∑ n = 1 ∞ n q ( 1 − λ 2 ) − 1 b n q 1 q ,
where λ1 + λ2 = λ ∈ (0, 4] (λ1 ∈ (0, 2], λ2 ∈ (0, 2])), the constant B(λ1, λ2) is the best value, and B ( u , v ) : = ∫ 0 ∞ t u − 1 ( 1 + t ) u + v d t ( u , v > 0 ) is the Beta function. Further generalizations have been documented in references [4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19].
Hong et al. [20] investigated necessary and sufficient parameter conditions for sharp constants in Hilbert-type series inequalities in 2016. Follow-up relevant research can be found in works [21,22,23,24,25]. In 2019, following the approach of [3], Adiyasuren et al. [26] established an extended application of (3) involving two partial sums:
λi ∈ (0, 1] ∩ (0, λ) (λ ∈ (0, 2]; i = 1, 2), λ1 + λ2 = λ,
∑ m = 1 ∞ ∑ n = 1 ∞ a m b n ( m + n ) λ < λ 1 λ 2 B ( λ 1 , λ 2 ) ∑ m = 1 ∞ m − p λ 1 − 1 A m p 1 p ∑ n = 1 ∞ n − q λ 2 − 1 B n q 1 q ,
where λ1λ2B(λ1, λ2) is the best value, and am, bn ≥ 0, A m = ∑ i = 1 m a i and B n = ∑ k = 1 n b k (m, n ∈ N = {1, 2, …}), satisfying Am = o(etm), Bn = o(etn) (t > 0; m, n → ∞),
0 < ∑ m = 1 ∞ m − p λ 1 − 1 A m p < ∞   and   0 < ∑ n = 1 ∞ n − q λ 2 − 1 B n q < ∞ .
In recent years, employing Euler–Maclaurin summation formula, Yang et al. [27] gave a more accurate extension of (5) with the kernel as
1 [ ( m − ξ ) α + ( n − η ) β ] λ ( α , β ∈ ( 0 , 1 ] , ξ , η ∈ [ 0 , 1 5 ] ) .
In 2024, Peng et al. [28] gave an extension of (2) with the kernel as 1 ln λ mn ( λ > 0 ) involving one partial sum. In 2023, Peng et al. [29] obtained an extension of the reverse half-discrete Mulholland-type inequality with the kernel as 1 ( 1 + u ( x ) ln n ) λ ( λ > 0 ) . In 2024, utilizing the Euler–Maclaurin summation formula, Wang et al. [30] proved a half-discrete Hilbert-type inequality with the kernel as 1 ( x + n β ) λ (λ ∈ (0, 5 − m], m = 0, 1, 2, 3, 4, β ∈ (0, 1]) involving one multiple upper limit function and one partial sum as follows:
∫ 0 ∞ ∑ n = 1 ∞ b n f ( x ) ( x + n β ) λ dx < β 1 q Γ ( λ + m + 1 ) Γ ( λ ) B ( λ 1 + m , λ 2 + 1 ) × ∫ 0 ∞ x p ( 1 − m − λ 1 ) − 1 F m p ( x ) dx 1 p ∑ n = 1 ∞ n q ( 1 − β ( 1 + λ 2 ) − 1 B n q 1 q .
Hong et al. [31] generalized Hilbert’s integral inequality with two upper limit functions by relaxing the conditions imposed on F(x) and G(y). Moreover, He et al. [32] derived necessary and sufficient parameter conditions for multidimensional Hilbert-type integral inequalities with non-homogeneous kernels and gave the explicit expression of the best constant factor.
Despite these advances, to the best of our knowledge, no work has yet established a half-discrete Mulholland-type inequality with the kernel 1 / ( x + ln β n ) λ involving one multiple upper limit function, nor have the conditions on the multiple upper limit function been refined in this setting. Motivated by this gap, the aim of this paper is to (i) establish, via weight coefficient methods and the Euler–Maclaurin summation formula, a new half-discrete Mulholland-type inequality with one multiple upper limit function, which refines inequality (3) with u(x) = x and differs structurally from inequality (6) in [30]; (ii) provide equivalent parameter conditions for the best possible constant factor; and (iii) derive several particular and extended inequalities.
We note that, as shown in Remark 5 below, taking m = 0, β = 1, λ = 1, λ1 = 1/q, and λ2 = 1/p in our main result recovers inequality (3) with u(x) = x (and hence Mulholland’s inequality (2) via [2]); thus, the classical case corresponds exactly to the absence of the multiple upper limit function (m = 0).
In the present paper, we adopt weight coefficient techniques together with parameter-introducing strategies and apply the Euler–Maclaurin summation formula to modify the kernel in reference [30] to ( x + ln β n ) − λ (λ > 0, β ∈ (0, 1]). Following the framework proposed in [31] and with the aid of Lemma 4 and its supplementary note, we refine the hypotheses imposed on the multiple upper limit function Fm(x).
The closest preceding result is that of [30], and the differences are as follows: First, [30] employs the kernel ( x + n β ) − λ (inequality (6) above); in this paper we replace nβ by lnβn, thereby passing from a kernel homogeneous of degree −λ in the pair (x, nβ) to a logarithmic (Mulholland-type) one. Second, the admissible ranges are enlarged from λ ∈ (0, 5 − m], m ∈ {0, 1, 2, 3, 4} to λ > 0 with arbitrary m ∈ ℕ0, together with the new constraint λ 2 ∈ ( 0 , 2 β ] ∩ ( 0 , λ ) (see Remark 1 for the role of β ∈ (0, 1]). Third, the partial sum is removed, so that the weight falls directly on b n q . Fourth, the best constant factor of inequality (6) in [30] is β − 1 Γ ( λ + m + 1 ) Γ ( λ ) B ( λ 1 + m , λ 2 + 1 ) , whereas the best constant factor obtained in this paper is Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) ; Theorem 1 establishes the new inequality (20) with this best possible constant, whose optimality is characterized in Theorem 2 and Corollary 1.

2. Some Lemmas

In what follows, we suppose that
Assumption 1. 
The following conditions hold:
  • (i) Throughout this paper, p > 1, q := p/(p − 1) is the conjugate exponent of p, N0 := {0, 1, 2, …}, and the parameters satisfy
λ > 0 ,         β ∈ ( 0 , 1 ] ,     λ 1 ∈ ( 0 , λ ) ,   λ 2 ∈ ( 0 , 2 β ] ∩ ( 0 , λ ) ,     m ∈ N 0 ,
and we set
λ ^ 1 : = λ − λ 2 p + λ 1 q ,         λ ^ 2 : = λ − λ 1 q + λ 2 p .
(ii) Assumptions on the functions f and Fi.  F 0 ( x ) : = f ( x )  is a nonnegative function on R+ := (0, ∞), continuous except possibly at finitely many points. For m ∈ N := {1, 2, ⋯}, we define the multiple upper limit functions as follows:
F i ( x ) : = ∫ 0 x F i − 1 ( t ) dt ( x ≥ 0 ) ,
namely:
F i ( x ) = ∫ 0 x ( ∫ 0 t i − 1 ⋯ ∫ 0 t 1 F 0 ( t 0 ) d t 0 ⋯ d t i − 2 ) d t i − 1 ( i = 1 , ⋯ , m ) .
For every m ∈ N0, we assume that
0 < c i − 1 : = ∫ 0 ∞ t p ( 2 − i − λ ^ 1 ) − 1 F i − 1 p ( t ) dt < ∞ ( i = 1 , ⋯ , m + 1 ) ,
i.e., each Fi−1 belongs to the weighted Lp-space with weight t p ( 2 − i − λ ^ 1 ) − 1 (in particular, this is an integrability condition on f when i = 1).
  • (iii) Assumption on the sequence. The sequence { b n } n ∈ N 2 (N2 := {2, 3, ⋯}), bn ≥ 0, satisfies
0 < ∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b n q < ∞ .
Condition (7) is imposed for i = 1, …, m + 1. Among these, the case i = m + 1 (i.e., the finiteness of the constant cm) is used only in Lemma 3 and, through it, in Theorems 1–2, whereas Lemma 4 and the Note below require (7) only for i = 1, …, m; the statement of Lemma 4 below reflects this observation.
Remark 1. 
The restriction β ∈ (0, 1] in Assumption 1 is imposed for the following three reasons:
  • (I) If β = 0, then lnβn ≡ 1 and the weight series ∑ n = 2 ∞ 1 n ln n diverges, so that the weight coefficient method fails; hence, β > 0 is necessary. (II) In the proof of Lemma 2(ii) below, in order for the auxiliary function βs ln β s 2 − ( 3 − β ) t ( x + ln β t ) s + 1 lnt t 2 to be positive and decreasing on [2, ∞), one needs βs2 − (3 − β) ≤ 0; for s 2 ∈ ( 0 , 2 β ] , this condition holds automatically whenever β ≤ 1, which also yields βs2(βs2 + β) ≤ 6 and hence guarantees h0(s2) > 0 and the validity of estimate (11) below. For β > 1, the above argument breaks down for s2 close to 2 β , and different estimation techniques would be required. (III) The condition β ≤ 1 ensures that βλ2 ∈ (0, 2], consistent with the classical parameter ranges in [3,20,31], so that Theorem 1 is a genuine extension of inequality (3) under the setting u(x) = x.
Remark 2. 
The parameters  λ ^ 1  and  λ ^ 2 in Assumption 1(i) are not introduced arbitrarily; they are exactly the exponents for which the two weight functions in Lemma 3 attain their Beta function values simultaneously. Indeed, applying Hölder’s inequality to Iλ+m, the x-side and the n-side involve the weights ϖλ+m(λ2, x) and ωλ+m(λ1 + m, n); balancing the resulting exponents yields the weights x p ( 1 − m − λ ^ 1 ) − 1 and ln q ( 1 − β λ ^ 2 ) − 1 n in (13), and the convergence condition (7) is formulated precisely in terms of these exponents. Note that λ ^ 1 + λ ^ 2 = λ always holds, and under the natural constraint λ1 + λ2 = λ (see Theorems 1–2), one has λ ^ i = λ i (i = 1, 2), so that the best constant factor becomes the Beta value B(λ1 + m, λ2), exactly as in (3).
Lemma 1. 
(cf. [4], (2.2.13)) (i) If k ∈ N0, ( − 1 ) i d i d t i g ( t ) > 0 ,  t ∈ [k, ∞), g(i)(∞) = 0 (i = 0, 1), and P 1 ( t ) : = t − [ t ] − 1 2 ( t ∈ R = ( − ∞ , ∞ ) ) is the Bernoulli function of the first order, then we have
− 1 8 g ( k ) < ∫ k ∞ P 1 ( t ) g ( t ) dt < 0 .
(ii) (cf. [4], (2.1.13)) If k, n ∈ N0, f(t) ∈ C1[k, ∞), then we have the following Euler–Maclaurin summation formula:
∑ i = k n f ( i ) = ∫ k n f ( t ) dt + 1 2 ( f ( k ) + f ( n ) ) + ∫ k n P 1 ( t ) f ′ ( t ) d t ( n ≥ k ) .
Lemma 2. 
For s > 0, s 2 ∈ ( 0 , 2 β ] ∩ ( 0 , s ) ,  ks(s2) := B(s2, s − s2), we define the following weight coefficient:
ϖ s ( s 2 , x ) : = β x s − s 2 ∑ n = 2 ∞ ln β s 2 − 1 n ( x + ln β n ) s n ( x > 0 ) .
The following inequalities hold:
0 < k s ( s 2 ) ( 1 − O s ( 1 x s 2 ) ) < ϖ s ( s 2 , x ) < k s ( s 2 ) ( x > 0 ) ,
where  O s ( 1 x s 2 ) : = 1 k s ( s 2 ) ∫ 0 ln β 2 x u s 2 − 1 ( 1 + u ) s du > 0 .
Proof. 
For fixed x > 0, we define a positive function h(t) as follows:
h ( t ) : = ln β s 2 − 1 t ( x + ln β t ) s t ( t > 1 ) .
Note that h ∈ C1(1, ∞), h(t) > 0, and, since s2 ∈ (0, s) by Assumption 1:
h ( t ) = ln β s 2 − 1 t ( x + ln β t ) s t ≤ t − s ln β s 2 − 1 − βs t = t − 1 ln − 1 − δ t , δ : = β ( s − s 2 ) > 0
for t > e1/β; hence, ∫ 1 ∞ h ( t ) , dt < ∞ . Moreover, each of the three terms of h′(t) computed below is of the form C t − 2 ln a t ( x + ln β t ) − c with fixed a, c ∈ ℝ, hence absolutely integrable on [2, ∞); in particular, h(t) → 0 and h′(t) → 0 as t → ∞, and ∫ 2 ∞ | h ′ ( t ) | dt < ∞ . Finally, under the substitution u = lnβt/x used below:
∫ 0 ∞ u s 2 − 1 ( 1 + u ) s du = B ( s 2 , s − s 2 ) < ∞ ,
again by s2 ∈ (0, s).
  • (i) For s 2 ∈ ( 0 , 1 β ] ∩ ( 0 , s ) , β ∈ ( 0 , 1 ] , we find
h ′ ( t ) = − ( 1 − β s 2 ) ln β s 2 − 2 t ( x + ln β t ) s t 2 − βs ln β ( s 2 + 1 ) − 2 t ( x + ln β t ) s + 1 t 2 − ln β s 2 − 1 t ( x + ln β t ) s t 2 < 0 .
In view of h′(t) < 0, h(t) is strictly decreasing, and hence, by the integral test for positive decreasing functions (i.e., ∫ n n + 1 h ( t ) dt ≤ h ( n ) ≤ ∫ n − 1 n h ( t ) dt for n ≥ 3, which is the order-zero case of the Euler–Maclaurin comparison), and by the convergence of ∫ 1 ∞ h ( t ) dt just proved, summing over n gives
∫ 2 ∞ h ( t ) dt < ∑ n = 2 ∞ h ( n ) < ∫ 1 ∞ h ( t ) dt .
(ii) For s 2 ∈ ( 1 β , 2 β ] ∩ ( 0 , s ) , we use first-order Euler–Maclaurin summation Formula (9) with k = 2 and f = h for every n ≥ 2:
∑ i = 2 n h ( i ) = ∫ 2 n h ( t ) dt + 1 2 ( h ( 2 ) + h ( n ) ) + ∫ 2 n P 1 ( t ) h ′ ( t ) dt .
Letting n → ∞ is justified by the properties established above: h(n) → 0, and ∫ 2 n P 1 h ′ → ∫ 2 ∞ P 1 h ′ since | P 1 | ≤ 1 2 and ∫ 2 ∞ | h ′ | < ∞ . We thus obtain:
∑ n = 2 ∞ h ( n ) = ∫ 2 ∞ h ( t ) dt + 1 2 h ( 2 ) + ∫ 2 ∞ P 1 ( t ) h ′ ( t ) dt = ∫ 1 ∞ h ( t ) dt − h 0 ( s 2 ) ,
h 0 ( s 2 ) : = ∫ 1 2 h ( t ) dt − 1 2 h ( 2 ) − ∫ 2 ∞ P 1 ( t ) h ′ ( t ) dt .
We obtain − 1 2 h ( 2 ) = − ln β s 2 − 1 2 4 ( x + ln β 2 ) s . Setting u = lnβt, by integration by parts, we find
∫ 1 2 h ( t ) dt = 1 β ∫ 0 ln β 2 u s 2 − 1 ( x + u ) s du = 1 β ∫ 0 ln β 2 d u s 2 s 2 ( x + u ) s = u s 2 β s 2 ( x + u ) s | 0 ln β 2 − 1 β ∫ 0 ln β 2 u s 2 s 2 d 1 ( x + u ) s = ln β s 2 2 β s 2 ( x + ln β 2 ) s + s β s 2 ( s 2 + 1 ) ∫ 0 ln β 2 d u s 2 + 1 ( x + u ) s + 1 > ln β s 2 2 β s 2 ( x + ln β 2 ) s + s β s 2 ( s 2 + 1 ) ln β ( s 2 + 1 ) 2 ( x + ln β 2 ) s + 1 .
To estimate the remainder term − ∫ 2 ∞ P 1 ( t ) h ′ ( t ) d t in the definition of h0(s2), we decompose h′(t) according to the signs of its three terms. Since in Case (ii) 1 < βs2 ≤ 2, the first term of h′(t) is positive, and we may write h′(t) = G1(t) − G2(t) − G3(t), where
G 1 ( t ) : = ( β s 2 − 1 ) ln β s 2 − 2 t ( x + ln β t ) s t 2 ;     G 2 ( t ) : = βs ln β s 2 + β − 2 t ( x + ln β t ) s + 1 t 2 ;     G 3 ( t ) : = ln β s 2 − 1 t ( x + ln β t ) s t 2 .
We verify that G1, G2, G3 satisfy all hypotheses of (8) with k = 2. Each of them is positive. Writing any of them as G ( t ) = t − 2 ln a t ( x + ln β t ) − c with c ≥ 0 and a ∈ βs2 − 2, βs2 + β − 2, βs2 − 1, its logarithmic derivative equals
G ′ ( t ) G ( t ) = 1 t [ a lnt − 2 − cβ ln β − 1 t x + ln β t ] < 1 t [ a lnt − 2 ] < 0 ,   t ≥ 2 ,
since a ≤ max{βs2 − 1, βs2 + β − 2} ≤ max{1, β} = 1 < 2ln2 by βs2 ≤ 2 and β ≤ 1 from Assumption 1, and the last term in brackets is non-positive. Hence, G1, G2, G3 are positive and strictly decreasing on [2, ∞); moreover G(t), G′(t) = O(t−2lnat) → 0 as t → ∞, so G(∞) = G′(∞) = 0. Applying (8) to each of the three functions and combining the signs in h′ = F1 − G2 − G3, we obtain:
− 1 8 ( G 1 ( 2 ) + G 2 ( 2 ) + G 3 ( 2 ) ) < ∫ 2 ∞ P 1 ( t ) , h ′ ( t ) , dt < 1 8 ( G 2 ( 2 ) + G 3 ( 2 ) ) .
Together with the preceding estimates for − 1 2 h ( 2 ) and ∫ 1 2 h dt , this yields
h ′ ( t ) = ( β s 2 − 1 ) ln β s 2 − 2 t ( x + ln β t ) s t 2 − βs ln β s 2 − ( 3 − β ) t ( x + ln β t ) s + 1 lnt t 2 − ln β s 2 − 2 t ( x + ln β t ) s lnt t 2 , − ∫ 2 ∞ P 1 ( t ) ( β s 2 − 1 ) ln β s 2 − 2 t ( x + ln β t ) s t 2 dt > 0 ( 1 < β s 2 ≤ 2 ) ,
    ∫ 2 ∞ P 1 ( t ) [ βs ln β s 2 − ( 3 − β ) t ( x + ln β t ) s + 1 lnt t 2 + ln β s 2 − 2 t ( x + ln β t ) s lnt t 2 ] dt > − 1 8 [ βs ln β s 2 − ( 3 − β ) 2 ( x + ln β 2 ) s + 1 ln 2 2 2 + ln β s 2 − 2 2 ( x + ln β 2 ) s ln 2 2 2 ] ( β s 2 − ( 3 − β ) ≤ β s 2 − 2 ≤ 0 ) , − ∫ 2 ∞ P 1 ( t ) h ′ ( t ) dt > − βs ln β s 2 − ( 2 − β ) 2 32 ( x + ln β 2 ) s + 1 − ln β s 2 − 1 2 32 ( x + ln β 2 ) s .
Hence, for s 2 ∈ ( 1 β , 2 β ] ∩ ( 0 , s ) , β ∈ ( 0 , 1 ] , we obtain:
h 0 ( s 2 ) > ln β s 2 2 β s 2 ( x + ln β 2 ) s + s β s 2 ( s 2 + 1 ) ln β ( s 2 + 1 ) 2 ( x + ln β 2 ) s + 1 − ln β s 2 − 1 2 4 ( x + ln β 2 ) s − βs ln β ( s 2 + 1 ) − 1 2 32 ( x + ln β 2 ) s + 1 − β ln β s 2 − 1 2 32 ( x + ln β 2 ) s = ln β s 2 − 1 2 ( x + ln β 2 ) s ( ln 2 β s 2 − 1 4 − 1 32 ) + sβ ln β ( s 2 + 1 ) − 1 2 ( x + ln β 2 ) s + 1 [ ln 2 β s 2 ( β s 2 + β ) − 1 32 ] ≥ ln β s 2 − 1 2 ( x + ln β 2 ) s ( ln 2 2 − 1 4 − 1 32 ) + sβ ln β ( s 2 + 1 ) − 1 2 ( x + ln β 2 ) s + 1 ( ln 2 6 − 1 32 ) = ln β s 2 − 1 2 ( x + ln β 2 ) s 16 ln 2 − 9 32 + sβ ln β ( s 2 + 1 ) − 1 2 ( x + ln β 2 ) s + 1 16 ln 2 − 3 96 . > 0 .
We still have
∑ n = 2 ∞ h ( n ) = ∫ 2 ∞ h ( t ) dt + 1 2 h ( 2 ) + ∫ 2 ∞ P 1 ( t ) h ′ ( t ) dt = ∫ 2 ∞ h ( t ) dt + h 1 ( s 2 ) ,
h 1 ( s 2 ) : = 1 2 h ( 2 ) + ∫ 2 ∞ P 1 ( t ) h ′ ( t ) dt .
For s 2 ∈ ( 1 β , 2 β ] ∩ ( 0 , s ) , in view of (8), we find
∫ 2 ∞ P 1 ( t ) ( β s 2 − 1 ) ln β s 2 − 2 t ( x + ln β t ) s t 2 dt > − ( β s 2 − 1 ) ln β s 2 − 2 2 32 ( x + ln β 2 ) s ,
− ∫ 2 ∞ P 1 ( t ) [ βs ln β s 2 − ( 3 − β ) t ( x + ln β t ) s + 1 lnt t 2 + ln β s 2 − 2 t ( x + ln β t ) s lnt t 2 ] dt > 0 ,   ∫ 2 ∞ P 1 ( t ) h ′ ( t ) dt > − ( β s 2 − 1 ) ln β s 2 − 2 2 32 ( x + ln β 2 ) s .
h 1 ( s 2 ) > ln β s 2 − 1 2 4 ( x + ln β 2 ) s − ( β s 2 − 1 ) ln β s 2 − 2 2 32 ( x + ln β 2 ) s = ln β s 2 − 2 2 4 ( x + ln β 2 ) s ( ln 2 − β s 2 − 1 8 )   ≥ ln β s 2 − 2 2 4 ( x + ln β 2 ) s ( ln 2 − 1 8 ) = ln β s 2 − 2 2 4 ( x + ln β 2 ) s 8 ln 2 − 1 8 > 0 .
Hence, Estimate (12) remains valid in this case.
In view of (i) and (ii), for s 2 ∈ ( 0 , 2 β ] ∩ ( 0 , s ) , by (12), setting u = ln β t x , it follows that
ϖ s ( s 2 , x ) = β x s − s 2 ∑ n = 2 ∞ h ( n ) < β x s − s 2 ∫ 1 ∞ h ( t ) dt   = ∫ 0 ∞ u s 2 − 1 du ( 1 + u ) s = B ( s 2 , s − s 2 ) = k s ( s 2 ) ,
ϖ s ( s 2 , x ) > β x s − s 2 ∫ 2 ∞ h ( t ) dt = ∫ ln β 2 x ∞ u s 2 − 1 ( 1 + u ) s du = k s ( s 2 ) ( 1 − O s ( 1 x s 2 ) ) > 0 ,
where we indicate that O s ( 1 x s 2 ) = 1 k s ( s 2 ) ∫ 0 ln β 2 x u s 2 − 1 ( 1 + u ) s du , satisfying
0 < ∫ 0 ln β 2 x u s 2 − 1 ( 1 + u ) s du ≤ ∫ 0 ln β 2 x u s 2 − 1 du = 1 s 2 ( ln β 2 x ) s 2 .
Therefore, inequalities (11) follow.
This completes the proof. □
  • Note. All numerical constants appearing above are, in fact, exact rational combinations of ln2, namely 16 ln 2 − 9 32 ≈ 0.065 , 16 ln 2 − 3 96 ≈ 0.084 , and 8 ln 2 − 1 8 ≈ 0.568 ; the decimal values are displayed only for the reader’s convenience. We also point out that the estimate 16 ln 2 − 3 96 > 0 relies crucially on the assumption β ≤ 1.
Remark 3. 
The upper bound ks(s2) = B(s2, s − s2) in (11) is sharp as x → ∞. Indeed, by the lower estimate established in the proof of Lemma 2:
ϖ s ( s 2 , x ) ≥ k s ( s 2 ) ( 1 − O s ( 1 x s 2 ) ) ,   0 < O s ( 1 x s 2 ) ≤ 1 s 2 ( ln β 2 x ) s 2 → 0     ( x → ∞ ) ,
whence  lim x → ∞ ϖ s ( s 2 , x ) = k s ( s 2 ) . Therefore, neither the Beta-type main term nor the constant factor in (11) can be improved.
Lemma 3. 
Under Assumption 1, for every m ∈ N0, the following half-discrete Mulholland-type inequality holds:
I λ + m : = ∑ n = 2 ∞ ∫ 0 ∞ F m ( x ) b n ( x + ln β n ) λ + m dx < ( 1 β k λ + m ( λ 2 ) ) 1 p ( k λ + m ( λ 1 + m ) ) 1 q   × ∫ 0 ∞ x p ( 1 − m − λ ^ 1 ) − 1 F m p ( x ) dx 1 p ∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b n q 1 q .
Proof. 
Setting y = x ln β n , for s1 ∈ (0, s), we obtain the following weight function:
ω s ( s 1 , n ) : = ln β ( s − s 1 ) n ∫ 0 ∞ x s 1 − 1 ( x + ln β n ) s dx = ∫ 0 ∞ y s 1 − 1 ( y + 1 ) s dy = k s ( s 1 ) : = B ( s 1 , s − s 1 ) ( n ∈ N 2 ) .
Applying Hölder’s inequality (cf. [1], Theorem 210) to the right-hand side with respect to the product of Lebesgue measure on (0, ∞) and the counting measure on {2, 3, …} gives
I λ + m = ∑ n = 2 ∞ ∫ 0 ∞ 1 ( x + ln β n ) λ + m ln ( β λ 2 − 1 ) / p n n 1 / p x ( λ 1 + m − 1 ) / q F m ( x ) x ( λ 1 + m − 1 ) / q n 1 / p ln ( β λ 2 − 1 ) / p n b n dx ≤ ∫ 0 ∞ ∑ n = 2 ∞ 1 ( x + ln β n ) λ + m ln β λ 2 − 1 n n x ( λ 1 + m − 1 ) ( p − 1 ) F m p ( x ) dx 1 p × ∑ n = 2 ∞ ∫ 0 ∞ 1 ( x + ln β n ) λ + m x λ 1 + m − 1 n q − 1 ln ( β λ 2 − 1 ) ( q − 1 ) n b n q dx 1 q     = 1 β ∫ 0 ∞ ϖ λ + m ( λ 2 , x ) x p ( 1 − m − λ ^ 1 ) − 1 F m p ( x ) dx 1 p × ∑ n = 2 ∞ ω λ + m ( λ 1 + m , n ) ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b n q 1 q .
Since the integrand is nonnegative and measurable with respect to the product of Lebesgue measure on (0, ∞) and the counting measure on {2, 3, …} (being built from continuous functions of u and n), the interchange of the sum and the integral in the second step above, where the inner sum is evaluated as ϖλ+m(λ2, x), is justified by Tonelli’s theorem (cf. [33]). In the proof of Lemma 5 below, where the summands change sign, the corresponding interchange is instead justified by the Lebesgue term-by-term theorem.
By (11) and (14), for
s = λ + m   >   m , s 1 = λ 1 + m ∈ ( m , λ + m ) , s 2 = λ 2 ∈ ( 0 , 2 β ] ∩ ( 0 , λ ) ,
we have λ > 0, λ1 ∈ (0, λ), and by (7) (for i = m + 1) and the Assumption 1, we obtain (13).
This completes the proof. □
Lemma 4. 
Let  λ ^ 1 , λ ^ 2  be as in Assumption 1(i) and let Fi(i = 0,…, m) satisfy (7) for i = 1,…, m. For t > 0 and m ∈ N0, we have
∫ 0 ∞ e − tx f ( x ) dx = t m ∫ 0 ∞ e − tx F m ( x ) dx .
Proof. 
For m = 0, since F0(x) = f(x), Formula (15) holds trivially. It remains to consider the case m ∈ N.
  • We first establish the following recursion formula: for each i ∈ N:
∫ 0 ∞ e − tx F i − 1 ( x ) dx = t ∫ 0 ∞ e − tx F i ( x ) dx .
In fact, by Hölder’s inequality and (7) for i = 1,…, m, we have
F i ( x ) = ∫ 0 x F i − 1 ( t ) dt = ∫ 0 x ( t 1 q − i + 1 − λ ^ 1 F i − 1 ( t ) ) ( t − 1 q + i − 1 + λ ^ 1 ) dt   ≤ [ ∫ 0 x t p ( 2 − i − λ ^ 1 ) − 1 F i − 1 p ( t ) dt ] 1 p [ ∫ 0 x t q ( i − 1 + λ ^ 1 ) − 1 dt ] 1 q   ≤ ( c i − 1 ) 1 p [ 1 q ( i − 1 + λ ^ 1 ) ] 1 q x i − 1 + λ ^ 1 ( λ ^ 1 > 0 ) .
It follows that
0 ≤ lim x → ∞ e − tx F i ( x ) ≤ ( c i − 1 ) 1 p [ 1 q ( i − 1 + λ ^ 1 ) ] 1 q lim x → ∞ e − tx x i − 1 + λ ^ 1 = 0 ,
and then, since F i ( 0 ) = ∫ 0 0 F i − 1 ( t ) dt = 0 by the definition of Fi in Assumption 1(ii), we have e − tx F i ( x ) | 0 ∞ = 0 . By integration by parts, we obtain
∫ 0 ∞ e − tx F i − 1 ( x ) dx = ∫ 0 ∞ e − tx d F i ( x )   = e − tx F i ( x ) | 0 ∞ − ∫ 0 ∞ F i ( x ) d e − tx = t ∫ 0 ∞ e − tx F i ( x ) dx .
which proves (16).
  • Applying Formula (16) successively for i = 1, ⋯, m, we obtain
∫ 0 ∞ e − tx f ( x ) dx = ∫ 0 ∞ e − tx F 0 ( x ) dx = t ∫ 0 ∞ e − tx F 1 ( x ) dx
= t 2 ∫ 0 ∞ e − tx F 2 ( x ) dx = ⋯ = t m ∫ 0 ∞ e − tx F m ( x ) dx
which is exactly (15).
This completes the proof. □
  • Note. For i = 1, …, m (m ∈ N), if we change the assumption 0 < ∫ 0 ∞ t p ( 2 − i − λ ^ 1 ) − 1 F i − 1 p ( t ) dt < ∞ to F i ( x ) = o ( e tx ) ( t > 0 ; x → ∞ ) in this lemma, then Formula (15) remains valid in this case. In fact, the only place where the integral conditions (7) are invoked in the proof of Lemma 4 is in establishing the boundary estimate lim x → ∞ e − tx F i ( x ) = 0 . Under the present hypothesis, this follows directly: since Fi(x) = o(etx) as x → ∞ for every t > 0, we have lim x → ∞ e − tx F i ( x ) = 0 ; moreover, Fi(0) = 0 by the definition F i ( x ) = ∫ 0 x F i − 1 ( t ) dt . Hence, e − tx F i ( x ) | 0 ∞ = 0 , and the recursion (16) follows by integration by parts exactly as before. Chaining (16) for i = 1,…, m yields (15).
Lemma 5. 
Under Assumption 1, for every m ∈ N0, the following inequality holds:
I 0 : = ∫ 0 ∞ ∑ n = 2 ∞ b n f ( x ) ( x + ln β n ) λ dx ≤ Γ ( λ + m ) Γ ( λ ) ∫ 0 ∞ ∑ n = 2 ∞ b n F m ( x ) ( x + ln β n ) λ + m dx .
Proof. 
We employ the well-known formula for the Gamma function (cf. [34]):
1 ( x + ln β n ) λ = 1 Γ ( λ ) ∫ 0 ∞ t λ − 1 e − ( x + ln β n ) t dt ,
by Lebesgue term by term theorem (cf. [33]) and (15), it follows that:
I 0 = 1 Γ ( λ ) ∫ 0 ∞ ∑ n = 2 ∞ b n f ( x ) ∫ 0 ∞ t λ − 1 e − ( x + ln β n ) t dt dx   = 1 Γ ( λ ) ∫ 0 ∞ t λ − 1 ( ∫ 0 ∞ e − xt f ( x ) dx ) ( ∑ n = 2 ∞ e − t ln β n b n ) dt   = 1 Γ ( λ ) ∫ 0 ∞ t λ − 1 ( t m ∫ 0 ∞ e − xt F m ( x ) dx ) ( ∑ n = 2 ∞ e − t ln β n b n ) dt   = 1 Γ ( λ ) ∫ 0 ∞ ∑ n = 2 ∞ b n F m ( x ) ∫ 0 ∞ t ( λ + m ) − 1 e − ( x + ln β n ) t dt dx   = Γ ( λ + m ) Γ ( λ ) ∫ 0 ∞ ∑ n = 2 ∞ b n F m ( x ) ( x + ln β n ) λ + m dx .
Thus, we have (17).
This completes the proof. □

3. Main Results

In this section, we state and prove our main results, Theorems 1 and 2, which together establish a half-discrete Mulholland-type inequality involving one multiple upper limit function and provide an equivalent characterization of its best constant factor. The proof of Theorem 1 relies on Lemmas 2–5 from the preceding section and a standard ε-approximation argument, while Theorem 2 shows that the optimality of the constant forces the balancing condition λ1 + λ2 = λ. Since Theorem 1 involves a number of parameters and auxiliary quantities, we first give a brief overview of its structure for the reader’s convenience and refer the details to Assumption 1.
(i)
Parameters. Throughout this section we keep the standing hypotheses of Assumption 1: the free parameters are λ > 0, β ∈ (0, 1], p > 1 with conjugate exponent q := p/(p − 1), λ1 ∈ (0, λ), λ 2 ∈ ( 0 , 2 β ] ∩(0, λ), and m ∈ N0. The exponents λ ^ 1 , λ ^ 2 and the functions Fi are derived from these as defined in Assumption 1; note that λ ^ 1 + λ ^ 2 = λ always holds (see Remark 2).
(ii)
Left-hand side. The left-hand side of the inequality is the mixed integral–series quantity I 0 = ∫ 0 ∞ ∑ n = 2 ∞ b n f ( x ) ( x + ln β n ) λ dx , i.e., a Lebesgue integral in x of a series in n, with non-homogeneous kernel ( x + ln β n ) − λ .
(iii)
Right-hand side: two weighted norms and a constant factor. The right-hand side consists of the x-side weighted Lp-norm
‖ F m ‖ p , λ ^ 1 : = ∫ 0 ∞ x p ( 1 − m − λ ^ 1 ) − 1 F m p ( x ) dx 1 / p ,
the n-side weighted l q -norm
‖ b ‖ q , β , λ ^ 2 : = ∑ n = 2 ∞ ( lnn ) q ( 1 − β λ ^ 2 ) − 1 n 1 − q b n q 1 / q ,
and a constant factor C. The finiteness of these two norms is guaranteed by Assumption 1(ii) (condition (7)) and Assumption 1(iii) (the series condition).
(iv)
Best constant. Under the balancing condition λ1 + λ2 = λ (whose necessity, under a mild additional restriction on the parameters, is established in Theorem 2), one has λ ^ i = λ i ( i = 1 , 2 ) , the constant factor reduces to the explicit Beta value
C ∗ = Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) ,
and Theorem 1 shows that C* is the best possible constant.
With these conventions fixed, we state the main result.
Theorem 1. 
Under Assumption 1, for every m ∈ N0, we establish the following half-discrete Mulholland-type inequality involving one multiple upper limit function:
I 0 = ∫ 0 ∞ ∑ n = 2 ∞ b n f ( x ) ( x + ln β n ) λ dx < Γ ( λ + m ) Γ ( λ ) ( 1 β k λ + m ( λ 2 ) ) 1 p ( k λ + m ( λ 1 + m ) ) 1 q × ∫ 0 ∞ x p ( 1 − m − λ ^ 1 ) − 1 F m p ( x ) dx 1 p ∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b n q 1 q .
In particular, for λ1 + λ2 = λ, in view of the Assumption 1, we obtain
0 < ∫ 0 ∞ x p ( 1 − m − λ 1 ) − 1 F m p ( x ) dx < ∞ , 0 < ∑ n = 2 ∞ ln q ( 1 − β λ 2 ) − 1 n n 1 − q b n q < ∞ ,
and the following inequality:
∫ 0 ∞ ∑ n = 2 ∞ b n f ( x ) ( x + ln β n ) λ dx < Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) × ∫ 0 ∞ x p ( 1 − m − λ ^ 1 ) − 1 F m p ( x ) dx 1 p ∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b n q 1 q ,
where the constant factor Γ ( λ + m ) β 1 / p Γ ( λ ) B(λ1 + m, λ2) is the best possible.
Remark 4. 
Before the proof of Theorem 1, we explain the mechanism of the ε-approximation argument. When λ1 + λ2 = λ, one has λ ^ 1 = λ 1  and  λ ^ 2 = λ 2  by Remark 2, so the exponents of the test function  f ˜ ( x ) = x λ 1 − ε / p − 1  (for x ≥ 1) and of  b ˜ n = 1 n ln β ( λ 2 − ε / q ) − 1 n are matched exactly with the weight exponents p(1 − m − λ1) − 1 and q(1 − βλ2) − 1 in (20). Since the kernel ( x + ln β n ) − λ is homogeneous of degree −λ in the pair (x, lnβn), the change of variables u = x/lnβn reduces the weight ϖ λ ( λ ˜ 2 , x ) to the Beta integral k λ ( λ ˜ 2 ) (via Lemma 2), while both ∫ 1 ∞ x − ε − 1 dx and ∑ n 1 n ln − βε − 1 n blow up at the rate 1/ε. Letting ε → 0+ then forces M to be at least the constant in (20).
Proof. 
By (17) and (13) we obtain (19). It remains to prove that the constant factor in (20) is best possible. Suppose that λ1 + λ2 = λ; then, by Remark 2, λ ^ i = λ i (i = 1, 2). Assume, to the contrary, that there exists a constant M with 0 < M ≤ Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) such that (20) remains valid with the constant factor replaced by M for all admissible f and bn. We shall construct a family of admissible test objects, depending on a small parameter ε → 0+, for which the ratio of the two sides of (20) converges exactly to the claimed constant, which will force M to equal it.
  • For any 0 < ε < min { p λ 1 , q λ 2 2 } set
α : = λ 1 − ε p > 0 ,   λ ˜ 2 : = λ 2 − ε q ∈ ( 0 , 2 β ) ∩ ( 0 , λ ) ,
and define
F ˜ 0 ( x ) = f ˜ ( x ) : = 0 ,     0 < x < 1 , x λ 1 − ε p − 1 , x ≥ 1 ,   b ˜ n : = 1 n ln β ( λ 2 − ε q ) − 1 n ,         n ∈ N 2 ,
and, for m ∈ N:
F ˜ m ( x ) : = ∫ 0 x ( ∫ 0 t m − 1 ⋯ ∫ 0 t 1 f ˜ ( t 0 ) d t 0 ⋯ d t m − 2 ) d t m − 1 = 0 , 0 < x < 1 , x λ 1 + m − 1 − ε p − p m − 1 ( x ) ∏ i = 0 m − 1 ( λ 1 + i − ε p ) , x ≥ 1 ,
where pm−1(x) is a polynomial of degree m − 1 with positive coefficients satisfying
( x λ 1 + m − ε p − 1 − p m − 1 ( x ) ) ( i ) | x = 1 = 0         ( i = 0 , ⋯ , m − 1 ) ,
with the conventions p−1(x)≡0 and ∏ i = 0 − 1 ( a + i ) : = 1   ( a > 0 ) , so that the explicit formula remains valid also for m = 0, in which case F ˜ 0 = f ˜ .
  • (a) Verification of all hypotheses of the theorem for the test family.
  • (a1) Nonnegativity and regularity. Clearly f ˜ ≥ 0 , and f ˜ is continuous on (0, ∞) except possibly at the single point x = 1, as required in (ii). Since x λ 1 + m − ε / p − 1 − p m − 1 ( x ) and its first m − 1 derivatives vanish at x = 1, the zero extension on (0, 1) joins Cm−1-smoothly, so F ˜ m ∈ C m − 1 ( ℝ + ) is continuous. Moreover:
F ˜ m ( m ) ( x ) = ∏ i = 0 m − 1 ( λ 1 + i − ε / p ) x λ 1 − ε / p − 1 ∏ i = 0 m − 1 ( λ 1 + i − ε / p ) = f ˜ ( x )         ( x > 1 ) ,
and F ˜ m together with its first m − 1 derivatives vanishes at x = 1; hence, by successive applications of the Newton–Leibniz formula, the above expression is exactly the m-fold repeated integral of f ˜ . Since f ˜ > 0 on (1, ∞), we conclude that
F ˜ m ( x ) > 0     ( x > 1 ) ,   and   consequently   x λ 1 + m − 1 − ε p − p m − 1 ( x ) > 0     ( x ≥ 1 ) .
  • (a2) Verification of condition Assumption 1(ii), i.e., (7) for every m ∈ N0 and i = 1,⋯, m + 1. By (21), the positivity of the coefficients of pi−2 and t ≥ 1:
0 ≤ F ˜ i − 1 ( t ) ≤ t λ 1 + i − 2 − ε / p ∏ j = 0 i − 2 ( λ 1 + j − ε / p )         ( t ≥ 1 ) ,           F ˜ i − 1 ( t ) = 0         ( 0 < t < 1 ) .
Hence, since λ ^ 1 = λ 1 :
0 ≤ t p ( 2 − i − λ ^ 1 ) − 1 F i − 1 p ( t ) ≤ C i t p ( 2 − i − λ 1 ) − 1 + p ( λ 1 + i − 2 ) − ε = C i t − 1 − ε         ( t ≥ 1 ) ,
so that
0 < ∫ 0 ∞ t p ( 2 − i − λ ^ 1 ) − 1 F ˜ i − 1 p ( t ) dt ≤ C i ε < ∞         ( i = 1 , ⋯ , m + 1 ) ,
because F ˜ i − 1 > 0 on (1, ∞). Thus, (7) holds for all i = 1,⋯, m + 1 and for every m ∈ N0.
  • (a3) Verification of the sequence condition Assumption 1(iii) and monotonicity. Writing γ : = β ( λ 2 − ε q ) − 1 and using λ ^ 2 = λ 2 , we compute
∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q   b ˜ n q = ∑ n = 2 ∞ n − 1 ln q ( 1 − β λ 2 ) − 1 + qβ ( λ 2 − ε q ) − q n = ∑ n = 2 ∞ 1 n ln 1 + βε n ,
which converges by the integral test, since 1 + βε > 1; all terms being positive, condition Assumption 1(iii) is satisfied. Moreover, γ ≤ β · 2 β − 1 = 1 < ln x for x ≥ 3, whence
d dx ( ln γ x x ) = ln γ − 1 x ( γ − lnx ) x 2 ≤ 0         ( x ≥ 3 ) ,
so that { b ˜ n } n ≥ 3 is decreasing; this is the “decreasing property of the sequence” used below.
  • (a4) Consequently, for every ε ∈ ( 0 , min { p λ 1 , q λ 2 2 } ) the pair ( f ˜ , { b ˜ n } ) satisfies all the hypotheses Assumption 1(i)–(iii) of the theorem, and the assumed inequality with constant M applies to it:
I ˜ : = ∫ 0 ∞ ∑ n = 2 ∞ b ˜ n f ˜ ( x ) ( x + ln β n ) λ dx < M [ ∫ 0 ∞ x p ( 1 − m − λ ^ 1 ) − 1 F ˜ m p ( x ) dx ] 1 p [ ∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b ˜ n q ] 1 q .
  • (b) Asymptotic estimate of the left-hand side.
  • Since f ˜ is supported on [1, ∞) with f ˜ ( x ) = x λ 1 − ε / p − 1 there, and since −ε/p − 1 = ε/q − ε − 1:
I ˜ = ∫ 1 ∞ [ x λ 1 + ε q ∑ n = 2 ∞ ln β ( λ 2 − ε q ) − 1 n ( x + ln β n ) λ n ] x − ε − 1 dx = 1 β ∫ 1 ∞ ϖ λ ( λ ˜ 2 , x ) x − ε − 1 dx ,
where ϖ λ ( s , x ) : = β x λ − s ∑ n = 2 ∞ ln βs − 1 n ( x + ln β n ) λ n and we have used λ − λ ˜ 2 = λ 1 + ε / q , so that the exponents match exactly. By Assumption 1(i), λ 2 ≤ 2 β , and λ ˜ 2 = λ 2 − ε / q < λ 2 gives λ ˜ 2 ∈ ( 0 , 2 β ) ∩ ( 0 , λ ) , i.e., s = λ > 0 and s 2 = λ ˜ 2 satisfy precisely the hypotheses s > 0 ,   s 2 ∈ ( 0 , 2 β ) ∩ ( 0 , s ) of Lemma 2; hence, the left-hand inequality in (11) gives, for x ≥ 1:
1 β ϖ λ ( λ ˜ 2 , x ) > 1 β k λ ( λ ˜ 2 ) ( 1 − O λ ( x − λ ˜ 2 ) ) ,
where kλ(s) := B(s, λ − s) and O λ ( x − λ ˜ 2 ) > 0 is the remainder term defined in Lemma 2 (see the line following (11)).
  • Substituting and integrating, using ∫ 1 ∞ x − ε − 1 dx = 1 ε together with the following properties of the remainder term O λ ( x − λ ˜ 2 ) : it tends to 0 as x → ∞ for each fixed ε, and, uniformly for all x ≥ 1 (bounding (1 + u)−λ ≤ 1 in the explicit formula of Lemma 2):
0 < ∫ 1 ∞ O λ ( x − λ ˜ 2 ) x − ε − 1 dx ≤ ( ln β 2 ) λ ˜ 2 λ ˜ 2 k λ ( λ ˜ 2 ) ( λ ˜ 2 + ε ) ≤ O ( 1 )         as   ε → 0 + ,
where we have used λ ˜ 2 ≥ λ 2 2 (for 0 < ε < q λ 2 2 ) and the continuity and positivity of kλ(s) = B(s, λ − s) on [ λ 2 2 , λ 2 ] . We obtain
I ˜ > 1 βε B ( λ 1 + ε q ,   λ 2 − ε q ) ( 1 − εO ( 1 ) ) ,  
where O(1) is bounded as ε → 0+, so that εO(1) → 0. Hence:
liminf ε → 0 + ε I ˜   ≥   1 β B ( λ 1 , λ 2 ) .
  • (c) Asymptotic estimate of the right-hand side.
  • The two factors on the right-hand side of (22) are estimated as follows:
  • The p-factor. By (21) and the explicit formula for F ˜ m :
∫ 0 ∞ x p ( 1 − m − λ ^ 1 ) − 1 F ˜ m p ( x ) dx ≤ ∫ 1 ∞ x p ( 1 − m − λ 1 ) − 1 + p ( λ 1 + m − 1 ) − ε dx ∏ i = 0 m − 1 ( λ 1 + i − ε p ) p = ε − 1 ∏ i = 0 m − 1 ( λ 1 + i − ε p ) p .
The q-factor. By the computation in (a3) and the decreasing property, comparing the series from n ≥ 4 with the integral:
∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b ˜ n q = 1 2 ln − βε − 1 2 + ∑ n = 3 ∞ 1 n ln 1 + βε n   ≤ 1 2 ln − βε − 1 2 + ∫ 2 ∞ dy y ln 1 + βε y = 1 2 ln − βε − 1 2 + 1 βε ln − βε 2 .
Combining the two estimates in (22), and using ε ⋅ ε−1/p = ε1/q:
ε I ˜ < M ∏ i = 0 m − 1 ( λ 1 + i − ε p ) ( ε 2 ln − βε − 1 2 + 1 β ln − βε 2 ) 1 q = : U ( ε ) .  
Each factor of U(ε) has an exact limit:
ln − βε 2 → 1 ,         ε ln − βε − 1 2 → 0 ,         ∏ i = 0 m − 1 ( λ 1 + i − ε p ) → ∏ i = 0 m − 1 ( λ 1 + i ) ,
so that
lim ε → 0 + U ( ε ) = M β − 1 / q ∏ i = 0 m − 1 ( λ 1 + i ) .
  • (d) The limiting ratio and the best possible constant.
  • Linking (23) with (25) and multiplying by ε > 0 yields
1 β B ( λ 1 + ε q , λ 2 − ε q ) ( 1 − εO ( 1 ) ) < ε I ˜ < M ∏ i = 0 m − 1 ( λ 1 + i − ε p ) ( ε 2 ln − βε − 1 2 + 1 β ln − βε 2 ) 1 q .
Letting ε → 0+ and using the continuity of the Beta function together with (24) and (26), we obtain
1 β B ( λ 1 , λ 2 ) ≤ M β − 1 / q ∏ i = 0 m − 1 ( λ 1 + i ) ,
that is:
M ≥ β − 1 p B ( λ 1 , λ 2 ) ∏ i = 0 m − 1 ( λ 1 + i ) = β − 1 p Γ ( λ ) Γ ( λ 2 ) Γ ( λ 1 + m ) = Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) ,
where we have used λ1 + λ2 = λ and ∏ i = 0 m − 1 ( λ 1 + i ) = Γ ( λ 1 + m ) Γ ( λ 1 ) . Together with the standing assumption M ≤ Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) , this forces
M = Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) .
Hence, M = Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) is the best possible constant factor of (20), which completes the proof. □
Theorem 2. 
Under Assumption 1, for every m ∈ N0, if the constant factor
Γ ( λ + m ) β 1 / p Γ ( λ ) ( k λ + m ( λ 2 ) ) 1 p ( k λ + m ( λ 1 + m ) ) 1 q
in (19) is the best possible, then, provided that  λ − λ 1 − λ 2 ≤ q ( 2 β − λ 2 ) ,  we have λ1 + λ2 = λ.
Proof. 
Since λ ^ 1 = λ − λ 2 p + λ 1 q , λ ^ 2 = λ − λ 1 − λ 2 q + λ 2 , 0 < λ1, λ2 < λ, we have λ ^ 1 + λ ^ 2 = λ ,   0 < λ ^ 1 ,   λ ^ 2 < λ , and B ( λ ^ 1 + m , λ ^ 2 ) ∈ R + . If λ − λ 1 − λ 2 ≤ q ( 2 β − λ 2 ) , we have λ ^ 2 ≤ 2 β .
  • Since λ ^ 2 ≤ 2 β , Assumption 1 remains valid with ( λ ^ 1 , λ ^ 2 ) in place of (λ1, λ2) (in particular the series in Assumption 1(ii) converges), so that (20) applies with λ i = λ ^ i ( i = 1 , 2 ) , we have the following inequality:
∫ 0 ∞ ∑ n = 2 ∞ b n f ( x ) ( x + ln β n ) λ dx < Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ ^ 1 + m , λ ^ 2 ) × [ ∫ 0 ∞ x p ( 1 − m − λ ^ 1 ) − 1 F m p ( x ) dx ] 1 p [ ∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b n q ] 1 q .
Since Assumption 1 holds for the pair ( λ ^ 1 , λ ^ 2 ) (with λ ^ 2 ∈ ( 0 , 2 β ] ∩ ( 0 , λ ) ensuring Assumption 1(i)), Theorem 1 applies with (λ1, λ2) replaced by ( λ ^ 1 , λ ^ 2 ) , and, as λ ^ 1 + λ ^ 2 = λ , the left-hand side and the two weight functionals of the resulting inequality (27) coincide with those of (19); that is, (27) is precisely the inequality (19) with the constant factor Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ ^ 1 + m , λ ^ 2 ) . Since the constant factor in (19) is the best possible, it cannot exceed any constant with which the same inequality holds; that is:
Γ ( λ + m ) β 1 / p Γ ( λ ) ( k λ + m ( λ 2 ) ) 1 p ( k λ + m ( λ 1 + m ) ) 1 q ≤ Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ ^ 1 + m λ ^ 2 ) ,
Cancelling the common positive factor Γ ( λ + m ) β 1 / p Γ ( λ ) yields
B ( λ ^ 1 + m , λ ^ 2 ) ≥ ( k λ + m ( λ 2 ) ) 1 p ( k λ + m ( λ 1 + m ) ) 1 q .
Note that, since λ ^ 1 + λ ^ 2 = λ and Assumption 1 holds for ( λ ^ 1 , λ ^ 2 ) , the same ε-approximation argument as in the proof of Theorem 1 (with λi replaced by λ ^ i , i = 1, 2) shows that the constant B ( λ ^ 1 + m , λ ^ 2 ) in (27) is itself best possible.
  • By Hölder’s inequality, we further have
  B ( λ ^ 1 + m , λ ^ 2 ) = k λ + m ( λ 2 p + λ − λ 1 q ) = ∫ 0 ∞ 1 ( 1 + u ) λ + m u λ 2 p + λ − λ 1 q − 1 du = ∫ 0 ∞ 1 ( 1 + u ) λ + m ( u λ 2 − 1 p ) ( u λ − λ 1 − 1 q ) du ≤ [ ∫ 0 ∞ 1 ( 1 + u ) λ + m u λ 2 − 1 du ] 1 p [ ∫ 0 ∞ 1 ( 1 + u ) λ + m u λ − λ 1 − 1 du ] 1 q = [ ∫ 0 ∞ 1 ( 1 + u ) λ + m u λ 2 − 1 du ] 1 p [ ∫ 0 ∞ 1 ( 1 + v ) λ + m v ( λ 1 + m ) − 1 dv ] 1 q = ( k λ + m ( λ 2 ) ) 1 p ( k λ + m ( λ 1 + m ) ) 1 q .
Combining (28) and (29), we conclude that equality holds in (29). (The auxiliary estimates (20)–(27) use Hölder’s inequality only through its validity; its equality condition is invoked solely at (29).) We note that equality holds in (29) if and only if there exist constants A and B, not both vanishing, such that A u λ 2 − 1 = B u λ − λ 1 − 1 holds a.e. on R+ (cf. [1]). Necessarily A ≠ 0 (otherwise B u λ − λ 1 − 1 = 0 a.e. would force B = 0 as well, since u λ − λ 1 − 1 > 0 a.e.). Assuming that A ≠ 0, it follows that u λ 2 + λ 1 − λ = B / A a.e. in R+. Since u ↦ uc is strictly monotone (hence injective) on R+ whenever c ≠ 0, the above identity a.e. is possible only for c = λ2 + λ1 − λ = 0. Hence, we have λ1 + λ2 = λ.
This completes the proof. □
Corollary 1. 
Under Assumption 1 and the condition  λ − λ 1 − λ 2 ≤ q ( 2 β − λ 2 ) , the constant factor  Γ ( λ + m ) β 1 / p Γ ( λ ) ( k λ + m ( λ 2 ) ) 1 / p ( k λ + m ( λ 1 + m ) ) 1 / q in (19) is the best possible if and only if λ1 + λ2 = λ.
Proof. 
Sufficiency. If λ1 + λ2 = λ, then λ ^ 1 = λ 1 ,   λ ^ 2 = λ 2 by Remark 2, so (19) reduces to (20), whose constant factor is best possible by Theorem 1. Necessity. This is exactly Theorem 2. □
  • Note. Theorem 2 alone provides only the necessity of λ1 + λ2 = λ for the best possible constant in (19). In view of the ε-approximation argument in the proof of Theorem 1, this condition is also sufficient; the exact equivalence is stated in Corollary 1.
Remark 5. 
For m = 0 in (20), we have the following inequality with the best possible constant factor ( 1 β ) 1 p B ( λ 1 , λ 2 ) :
∫ 0 ∞ ∑ n = 2 ∞ b n f ( x ) ( x + ln β n ) λ dx < ( 1 β ) 1 p B ( λ 1 , λ 2 ) × ∫ 0 ∞ x p ( 1 − λ ^ 1 ) − 1 f p ( x ) dx 1 p ∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b n q 1 q .
Hence, inequality (20) is an extension of (30). In particular, taking m = 0, β = 1, λ = 1, λ1 = 1/q, λ2 = 1/p in (30), we have λ ^ 1 = 1 / q , λ ^ 2 = 1 / p , the kernel becomes 1 x + ln n , the weight exponents p ( 1 − λ ^ 1 ) − 1 = 0 and q ( 1 − β λ ^ 2 ) − 1 = 0 vanish, and the constant factor reduces to B ( 1 q , 1 p ) = Γ ( 1 p ) Γ ( 1 q ) = π sin ( π / p ) . Hence, (30) reduces exactly to inequality (3) with u(x) = x: the logarithmic weight exponents p ( 1 − λ ^ 1 ) − 1 and q ( 1 − β λ ^ 2 ) − 1 vanish, leaving precisely the factor nq−1 in the bn-weight; (3) is the half-discrete counterpart of Mulholland’s inequality (2).
Example 1. 
Taking m = 1, p = q = 2, β = 1, λ = 2, λ1 = λ2 = 1 in (20) (note that λ1 + λ2 = λ and λ 2 ∈ ( 0 , 1 β ] ∩ ( 0 , λ ) ), we have λ ^ 1 = λ ^ 2 = 1 and the best possible constant factor
Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) = Γ ( 3 ) Γ ( 2 ) B ( 2 , 1 ) = 2 × 1 2 = 1 .
Hence, the following half-discrete Mulholland-type inequality with the best possible constant factor 1 holds:
∫ 0 ∞ ∑ n = 2 ∞ b n f ( x ) ( x + ln 2 n ) 2 dx < ∫ 0 ∞ x − 3 F 1 2 ( x ) dx 1 2 ∑ n = 2 ∞ b n 2 n ln n 1 2 ,
where F 1 ( x ) = ∫ 0 x f ( t ) dt , bn ≥ 0 and ∑ n = 2 ∞ b n 2 / ( n ln n ) < ∞ . The weight exponents in condition (7) reduce to p ( 1 − m − λ ^ 1 ) − 1 = − 3 and q ( 1 − β λ ^ 2 ) − 1 = − 1 , which are exactly those appearing in the two integrals above. In contrast with the classical case (Remark 5 with m = 0), the presence of F1 shows concretely how the multiple upper limit function enters the inequality.
Remark 6. 
Foru(x) > 0, u′(x) > 0 (x ∈ (b,∞)), with u(b+) = 0, u(∞) = ∞, replacing x by t in (20), then setting t = u(x) (i.e., x = u−1(t)), after simplification, we have the following extended inequalities with the best possible constant factor Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) :
∫ b ∞ ∑ n = 2 ∞ b n f ( u ( x ) ) u ′ ( x ) ( u ( x ) + ln β n ) λ dx < Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) × ∫ b ∞ ( u ( x ) ) p ( 1 − m − λ ^ 1 ) − 1 u ′ ( x ) F m p ( u ( x ) ) dx 1 p ∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b n q 1 q .
Similarly, replacing x by t = u(x) in the equivalent form (27), we have
∫ b ∞ ∑ n = 2 ∞ b n f ( u − 1 ( x ) ) ( u ( x ) ) λ − 2 u ′ ( x ) ( u ( x ) + ln β n ) λ dx < Γ ( λ + m ) β 1 / p Γ ( λ ) B ( λ 1 + m , λ 2 ) × ∫ b ∞ ( u ( x ) ) p ( 1 − m − λ ^ 1 ) − 1 u ′ ( x ) F m p ( u − 1 ( x ) ) dx 1 p ∑ n = 2 ∞ ln q ( 1 − β λ ^ 2 ) − 1 n n 1 − q b n q 1 q ,
where b ≥ 0.

4. Conclusions

Novelty. The novelties of this work are threefold: (A) it establishes the first half-discrete Mulholland-type inequality involving one multiple upper limit function Fm(x), generalizing inequality (3) (the case m = 0, cf. Remark 5); (B) the kernel ( x + ln β n ) − λ is parametrized by (β, λ1, λ2, p, q), adjusting the kernel of [30] in a direction structurally different from (6); (C) Theorems 1 and 2 jointly provide the best possible constant factor together with an equivalent parameter condition λ1 + λ2 = λ for its optimality.
Summary. In this paper, we combine weight-coefficient methods, parameter analysis, real-analysis tools and the Euler–Maclaurin summation formula to derive Theorem 1. This theorem delivers a new half-discrete Mulholland-type inequality for kernel ( x + ln β n ) − λ ( λ > 0 , β ∈ ( 0 , 1 ] ) with a multiple upper limit function. It serves as an extension of the known inequality (3) and differs structurally from inequality (6) in [30]. Theorem 1 and Theorem 2 together yield equivalent conditions for the optimality of the constant factor depending on system parameters. Remark 5 and Remark 6 further supply concrete special-case outcomes and additional generalized inequalities.
Recommendations. Several directions merit further investigation: (A) the reverse (dual) analogue of Theorem 1 involving one multiple upper limit function, following the line of [29]; (B) combining multiple upper limit functions with one or two partial sums, along the lines of [26,28,30]; (C) extensions to multidimensional Hilbert-type integral inequalities with non-homogeneous kernels as in [32]; (D) relaxing condition (7) to more general growth conditions on Fi (x) beyond the o(etx) class discussed in the Note of Lemma 4.

Author Contributions

X.M. modified the kernel function, established a new half-discrete Mulholland-type inequality involving multiple upper limit functions, and participated in the revision and proofreading of the manuscript. R.L. supervised the research and validated the results. Using the Euler–Maclaurin summation formula, carried out key derivations, refined the validity conditions for the multiple upper limit function based on Reference [31] and Lemma 4 and its accompanying remarks, and participated in the revision and finalization of the paper. B.Y. conducted the overall conceptualization and methodological design, performed formal analysis and verification of theoretical rigor, derived equivalent statements for the optimal constants with respect to parameters, completed the core theoretical proofs, and drafted the original manuscript. Y.L. investigated extended inequalities, constructed several special and extended inequalities, assisted in literature review and result verification, and contributed to the editing and improvement of the manuscript. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (Nos. 11961021 and 11561019), the Guangxi Natural Science Foundation (No. 2020GXNSFAA159084), the Project for Enhancing Basic Scientific Research Competence of Young and Middle-aged Teachers of Hechi University (No. 2025XJZQN036), and the Research Program of Hechi University (No. 2023XJPT007).

Data Availability Statement

The original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding authors.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Hardy, G.H.; Littlewood, J.E.; Pólya, G. Inequalities, 1st ed.; Cambridge University Press: Cambridge, UK, 1934. [Google Scholar]
  2. Chen, Q.; Yang, B.C. Half-discrete Hardy–Hilbert’s inequality with two interval variables. J. Inequal. Appl. 2013, 2013, 485. [Google Scholar] [CrossRef] [Scilit]
  3. Krnić, M.; Pečarić, J. Extension of Hilbert’s inequality. J. Math. Anal. Appl. 2006, 324, 150–160. [Google Scholar] [CrossRef] [Scilit]
  4. Yang, B.C. The Norm of Operator and Hilbert-Type Inequalities; Science Press: Beijing, China, 2009. [Google Scholar]
  5. Krnić, M.; Pečarić, J. General Hilbert’s and Hardy’s inequalities. Math. Inequal. Appl. 2005, 8, 29–51. [Google Scholar]
  6. Perić, I.; Vuković, P. Multiple Hilbert’s type inequalities with a homogeneous kernel. Banach J. Math. Anal. 2011, 5, 33–43. [Google Scholar] [CrossRef] [Scilit]
  7. Huang, Q.L. A new extension of a Hardy-Hilbert-type inequality. J. Inequal. Appl. 2015, 2015, 397. [Google Scholar] [CrossRef] [Scilit]
  8. He, B. A multiple Hilbert-type discrete inequality with a new kernel and best possible constant factor. J. Math. Anal. Appl. 2015, 431, 990–992. [Google Scholar] [CrossRef] [Scilit]
  9. Xu, J.S. Hardy-Hilbert’s inequalities with two parameters. Adv. Math. 2007, 36, 63–76. [Google Scholar]
  10. Xie, Z.T.; Zeng, Z.; Sun, Y.F. A new Hilbert-type inequality with the homogeneous kernel of degree-2. Adv. Appl. Math. Sci. 2013, 12, 391–401. [Google Scholar]
  11. Zhen, Z.; Raja Rama Gandhi, K.; Xie, Z.T. A new Hilbert-type inequality with the homogeneous kernel of degree-2 and with the integral. Bull. Math. Sci. Appl. 2014, 3, 11–20. [Google Scholar] [CrossRef] [Scilit]
  12. Xin, D.M. A Hilbert-type integral inequality with the homogeneous kernel of zero degree. Math. Theory Appl. 2010, 30, 70–74. [Google Scholar]
  13. Azar, L.E. The connection between Hilbert and Hardy inequalities. J. Inequal. Appl. 2013, 2013, 452. [Google Scholar] [CrossRef] [Scilit]
  14. Adiyasuren, V.; Batbold, T.; Krnić, M. Hilbert–type inequalities involving differential operators, the best constants and applications. Math. Inequal. Appl. 2015, 18, 111–124. [Google Scholar] [CrossRef] [Scilit]
  15. Xie, Z.T. A new half-discrete Hilbert’s inequality with the homogeneous kernel of degree-4μ. J. Zhanjiang Norm. Univ. 2011, 32, 13–19. [Google Scholar]
  16. Zhong, W.Y. A mixed Hilbert-type inequality and its equivalent forms. J. Guangdong Univ. Educ. 2011, 31, 18–22. [Google Scholar]
  17. Xie, Z.; Gandhi, K.R.R.; Zeng, Z. A new Hilbert-type integral inequality with the homogeneous kernel of real degree form and the integral in whole plane. Bull. Math. Sci. Appl. 2013, 3, 71–82. [Google Scholar] [CrossRef] [Scilit]
  18. Saied, A.I.; Jadlovská, I. Novel dynamic inequalities of Hilbert -Pachpatte-type for a class of non-homogeneous kernels on time scales. J. Inequal. Appl. 2025, 2025, 140. [Google Scholar] [CrossRef] [Scilit]
  19. Saied, A.I. A study on reversed dynamic inequalities of Hilbert-type on time scales nabla calculus. J. Inequal. Appl. 2024, 2024, 75. [Google Scholar] [CrossRef] [Scilit]
  20. Hong, Y.; Wen, Y. A necessary and sufficient condition that Hilbert-type series inequality with homogeneous kernel has the best constant factor. Chin. Ann. Math. Ser. A 2016, 37A, 329–336. [Google Scholar]
  21. Hong, Y. On the structure character of Hilbert’s type integral inequality with homogeneous kernel and application. J. Jilin Univ. (Sci. Ed.) 2017, 55, 189–194. [Google Scholar]
  22. Chen, Q.; He, B.; Hong, Y.; Li, Z. Equivalent parameter conditions for the validity of half-discrete Hilbert-type multiple integral inequality with generalized homogeneous kernel. J. Funct. Spaces 2020, 2020, 7414861. [Google Scholar] [CrossRef] [Scilit]
  23. He, B.; Hong, Y.; Chen, Q. The equivalent parameter conditions for constructing multiple integral half-discrete Hilbert-type inequalities with a class of non-homogeneous kernels and their applications. Open Math. 2021, 19, 400–411. [Google Scholar] [CrossRef] [Scilit]
  24. Hong, Y.; Huang, Q.; Chen, Q. The parameter conditions for the existence of the Hilbert-type multiple integral inequality and its best constant factor. Ann. Funct. Anal. 2021, 12, 7. [Google Scholar] [CrossRef] [Scilit]
  25. Hong, Y.; Chen, Q. Equivalent parameter conditions for the construction of Hilbert-type integral inequalities with a class of non-homogeneous kernels. J. South China Norm. Univ. (Nat. Sci. Ed.) 2020, 52, 124–128. [Google Scholar]
  26. Adiyasuren, V.; Batbold, T.; Azar, L.E. A new discrete Hilbert-type inequality involving partial sums. J. Inequal. Appl. 2019, 2019, 127. [Google Scholar] [CrossRef] [Scilit]
  27. Yang, B.C.; Wu, S.H. An improved version of the parameterized Hardy-Hilbert inequality involving two partial sums. Mathematics 2025, 13, 1331. [Google Scholar] [CrossRef] [Scilit]
  28. Peng, L.; Yang, B.C. A new extended Mulholland’s inequality involving one partial sum. Open Math. 2024, 22, 20240039. [Google Scholar] [CrossRef] [Scilit]
  29. Peng, L.; Rahim, R.A.; Yang, B.C. A new reverse half-discrete Mulholland-type inequality with a nonhomogeneous kernel. J. Inequal. Appl. 2023, 2023, 114. [Google Scholar] [CrossRef] [Scilit]
  30. Wang, A.Z.; Yang, B.C. A new half-discrete Hilbert-type inequality involving one multiple upper limit function and one partial sums. Chin. Ann. Math. Ser. A 2024, 45A, 25–38. [Google Scholar]
  31. Hong, Y.; Zhang, L.J.; Xiao, H.S. Condition for the construction of a Hilbert-type integral inequality involving upper limit functions. Symmetry 2024, 16, 1682. [Google Scholar] [CrossRef] [Scilit]
  32. He, B.; Hong, Y.; Li, Z. Conditions for the Validity of a Class of Optimal Hilbert Type Multiple Integral Inequalities with Nonhomogeneous Kernels. J. Inequal. Appl. 2021, 2021, 64. [Google Scholar] [CrossRef] [Scilit]
  33. Royden, H.L.; Fitzpatrick, P.M. Real Analysis, 4th ed.; Pearson Prentice Hall: Hoboken, NJ, USA, 2010. [Google Scholar]
  34. Kuang, J.C. Introduction to Real Analysis; Hunan Education Press: Changsha, China, 1996. [Google Scholar]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.