Abstract
We improve the classical discrete Hardy inequality for \( 1<p<\infty \) for functions on the natural numbers. For integer values of p the Hardy weight is shown to have a series expansion with strictly positive coefficients. Notably, this weight is optimal, i.e. critical and null-critical.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and Main Result
In 1918 Hardy was looking for a simple and elegant proof of Hilbert’s theorem in the context of the convergence of double sums, [12]. Although it is not explicitly mentioned, the paper contains the essential argument for his then famous inequality. In a letter to Hardy in 1921, [20], Landau gave a proof with the sharp constant
for \(p>1\) where \((a_n)\) is an arbitrary sequence of non-negative real numbers. This inequality was first highlighted in [13] and is referred to as a p-Hardy inequality. Since then various proofs of this inequality were given, where short and elegant ones are due to Elliott [6] and Ingham, see [13, p. 243] and by Lefèvre [21]. See also [15] for a beautiful historical survey about the origins of Hardy’s inequality.
It is not hard to see that the inequality above can be derived from the following inequality for compactly supported \(\phi \in C_c(\mathbb {N})\) with \(\phi (0)=0\)
where
In this work, we show that the classical weight \(w_p^H\) can be replaced by a pointwise strictly larger weight \(w_p\), and the aforementioned Hardy inequality holds with \(w_p^H\) being replaced by \(w_p\). Recently, the concept of optimality was studied for general \(1< p <\infty \) on weighted graphs in [7]. Employing these results, we conclude that \(w_p\) is optimal, which means that
-
for every function w with \(w \ge w_p\) and \(w \ne w_p\), the p-Hardy inequality does not hold (criticality), and in addition
-
\(u \notin \ell ^p(\mathbb {N}, w_p)\), where \(u(n) = n^{(p-1)/p}\), \(n \in \mathbb {N}\) (null-criticality).
The first optimality criterion says that there cannot be a p-Hardy weight dominating \(w_p\). The second optimality criterion says that the underlying ground state u is not an eigenfunction and so, the Hardy inequality does not admit a minimizer.
Consequently, by [7, Theorem 2.6], the weight \(w_p\) is also optimal at infinity, which means that for every \(\lambda > 0\) and each finite set \(K \subset \mathbb {N}\), the weight \((1 + \lambda )w_p\) does not yield a p-Hardy inequality for functions \(\varphi \) supported outside of K. This criterion in particular shows that the constant \(((p-1)/p)^p\) is optimal, a fact already known for a century, cf. [15].
We formulate the main theorem of this paper.
Theorem 1
Let \(p> 1\). Then, for all \(\phi \in C_c(\mathbb {N})\) with \(\phi (0)=0\),
where \(w_{p}\) is a strictly positive function given by
Furthermore, \(w_p\) is optimal, and we have for all \( n\in \mathbb {N}\)
Moreover, for integer \( p\ge 2 \), we have \( w_{p}(n)=\sum _{k\in 2\mathbb {N}_{0}}c_{k} n^{-k-p} \) with \( c_{k}>0 \).
Example 2
The case \(p = 2\) was already covered in [16] (with optimality proven in [17]), and via a different method in [11, 19]. In this case one gets \( w_2(1) = 2 - \sqrt{2} \) and for \( n\ge 2 \)
In the case \(p=3\), one obtains \( w_3(1) = \,\, 1 - (2^{2/3}-1)^2 \) and for \( n\ge 2 \)
In the case \(p=4\), one gets \( w_4(1) = \,\, 1 - (2^{3/4}-1)^3 \) and for \( n\ge 2 \)
For general \( p>1 \) one obtains the asymptotics
where
From this formula it is clear that \( w_{p}(n) \) is strictly larger than the classical Hardy weight for large n. Note however that the theorem above states that \( \epsilon _{p}(n)>0 \) at all places \(n \in \mathbb {N}\). It is not hard to check that \( \epsilon _{p}(n) \) can be expanded into a power series with respect to 1/n where all odd coefficients vanish. Theorem 1 states that for integer \( p\ge 2 \) these coefficients are positive. We conjecture that all these coefficients are strictly positive for all \( p>1 \).
Remark 3
It is easy to see that for \(1< p < \infty \), our p-Hardy inequality can be stated as follows: for every real-valued sequence \(a=(a_n)\), one has
where the function \(\epsilon _p\) is as in the previous example. Denoting by C the Cesàro mean operator on \(\ell ^p(\mathbb {N})\), defined as \(C(a) = \frac{1}{n} \sum _{j=1}^n a_j\), the above inequality says that \(C:\ell ^p(\mathbb {N}) \rightarrow \ell ^p(\mathbb {N}, \rho )\) is bounded, where \(\rho = 1 + \epsilon _p\) is understood as a measure on \(\mathbb {N}\). Equivalently, one obtains \(\ell ^p\)-boundedness of a weighted version of the Cesàro mean operator. For certain weights, such boundedness phenomena were studied recently in [22].
2 Proof of the Hardy Inequality
The combinatorial p-Laplacian \(\Delta _p\) for real valued functions on \( \mathbb {N}_{0} \) is given by
for all functions f and \( n\ge 1 \), where \( \textrm{sgn} \) is the function which takes the value \( -1 \) on \( (-\infty ,0) \), the value 1 on \( (0,\infty ) \) and 0 at 0.
The following proposition shows that the existence of a suitable positive supersolution of \(\Delta _p u \ge 0\) implies the non-negativity of the corresponding energy functional. This is one of the implications of the so-called Allegretto-Piepenbrink-type theorem (see [1, 2, 23] for linear versions in the continuum, [4, 18] for a linear version in the discrete setting, [24] for a non-linear version in the continuum and [8] for a recent version in the quasi-linear discrete setting). This statement is used to show that the weight \(w_p\) is in fact a p-Hardy weight.
Proposition 4
Let \(p> 1\) and let \(u:\mathbb {N}_0\rightarrow [0,\infty )\) be strictly positive on \(\mathbb {N}\) and such that \(u(0) = 0\). Suppose that \(w:\mathbb {N}\rightarrow \mathbb {R}\) satisfies \(\Delta _p u=wu^{p-1}\) on \(\mathbb {N}\). Then for all \(\phi \in C_c(\mathbb {N})\) with \(\phi (0)=0\) we have
The proof follows along the lines of the proof of Proposition 2.2 in [10].
Proof
Let \(p> 1\). From Lemma 2.6 in [10], we obtain for all \(0\le t\le 1\) and \(a\in \mathbb {C}\)
Let w be such that \(\Delta _p u=wu^{p-1} \). For given \( \varphi \in C_{c}(\mathbb {N}) \), we can consider \(\psi = \varphi / u \in C_c(\mathbb {N})\), by strict positivity of u on \(\mathbb {N}\). We assume for a moment that \(m,n \in \mathbb {N}\) are such that \(u(n) \ge u(m)\) and \(\psi (m) \ne 0\). We apply the above inequality with the choice \(t=u(m)/u(n)\) and \(a=\psi (n)/\psi (m)\) in order to obtain
Further, since \(u^{p}(n) \ge |u(n) - u(m)|^{p-1} u(n)\), the above inequality remains true even if \(\psi (m) = 0\). Summing over \(\mathbb {N}\), we obtain
Note that the latter two equalities follow from rearranging the involved sums while recalling that \(u(0)= 0\), and using the assumption \(\Delta _p u=wu^{p-1} \). Recalling that \( \phi =u\psi \), we infer the statement. \(\square \)
Next we show that for the weight \( w_{p} \) on \( \mathbb {N}\) taken from Theorem 1
there is a suitable positive function u such that \( \Delta _{p}u =w_{p}u^{p-1}\).
Proposition 5
Let \(p> 1\). Then, the function \( u:\mathbb {N}_0 \rightarrow [0,\infty )\), \( u(n)=n^{(p-1)/p} \) satisfies
Proof
One directly checks that for all \(n\in \mathbb {N}\)
which immediately yields the statement. \(\square \)
The choice of the function u in the previous proposition is motivated by the so-called supersolution construction which yields optimal p-Hardy weights both in the continuum case for \(p> 1\), cf. [3, 5], as well as for graphs with \(p=2\), cf. [17]. Moreover, for \(p=2\), the function \(u(n) = n^{1/2}\) arises naturally in the method applied in [11, 19] for a proof of the optimal Hardy inequality on the line graph.
Combining the two propositions above already yields the p-Hardy inequality with the weight \( w_{p} \). Next we show that \( w_{p} \) is strictly larger than the classical Hardy weight \( w_{p}^{H}(n)=\bigl ({(p-1)}/{p }\bigr )^{p}{n^{-p}}\) for all \( n\in \mathbb {N}\).
3 Proof of \( w_p>w_{p}^{H}\)
In this section we show that the weight
from the main theorem, Theorem 1, is strictly larger than the classical p-Hardy weight
In fact, for fixed \( p\in (1,\infty ) \), we analyze the function \( w:[0,1]\rightarrow [0,\infty ) \)
for \(x \in [0,1/2]\) and \(x=1\), where \(q\in (1,\infty )\) is such that \(1/p + 1/q=1\). Specifically, we show
The case \( x=1 \) is simple and is treated at the end of the section. The proof for \( x\le 1/2 \) is also elementary but more involved. We proceed by bringing \( w_{p} \) into form for which we then analyze its parts. This will be eventually done by a case distinction depending on p.
Recall the binomial theorem for \(r \in [0,\infty )\) and \(0\le x \le 1\)
where \(\left( {\begin{array}{c}r\\ 0\end{array}}\right) =1, \left( {\begin{array}{c}r\\ 1\end{array}}\right) =r\) and \(\left( {\begin{array}{c}r\\ k\end{array}}\right) =r(r-1)\cdots (r-k+1)/k!\) for \(k\ge 2\) which is derived from the Taylor expansion of the function \( x\mapsto (1\pm x)^{r} \). Applying this formula to the function w from above we obtain
To streamline notation we set
Note that since \(q\left( {\begin{array}{c}1/q\\ 1\end{array}}\right) =1\) and \(q\left|\left( {\begin{array}{c}1/q\\ k\end{array}}\right) \right|<1\) for \(k\ge 2\), we have \({0< |g(\pm x)|<1}\) for \(0< x\le 1/2\). Thus, we can apply the binomial theorem to \(\bigl (1+g(\pm x)\bigr )^{p-1}\) in order to get
Thus, we have to show that the second factor on the right hand side is strictly larger than x/q. Using \( q=p/(p-1) \) we compute the first term in the parenthesis on the left hand side
with
since \( -2p \left( {\begin{array}{c}{1/q }\\ k+1\end{array}}\right) > 0\) for odd k and \(x > 0\). So, it remains to show that for the term
we have for \(0<x\le 1/2 \)
Specifically, we then get with the substitution \( x=1/n \)
for \( n\ge 2 \).
Remark 6
It is not hard to see that \( F_{p}\ge 0 \) whenever \( p\in \mathbb {N}\) is integer valued. Indeed, \( g(-x)\ge g(x) \) as all terms in the sum \( g(-x) \) are positive since \( - \left( {\begin{array}{c}{1/q }\\ k+1\end{array}}\right) \ge 0\) for odd k, while the terms in g(x) alternate, (they are positive for even k and negative for odd k). Moreover for positive integers p the binomial coefficients \( \left( {\begin{array}{c}p-1\\ n\end{array}}\right) \) are positive. Thus, the Hardy weight we computed is larger than the classical one for integer p.
Let us now turn to the proof of
for \( p \in (1,\infty )\) and \(0< x\le 1/2\).
We collect the following basic properties of the function g which were partially already discussed above and will be used subsequently.
Lemma 7
For \( p\in (1,\infty ) \) and \( 0<x\le 1/2 \), we have
Proof
The function g is given by \( g(x)=q\sum _{k=1}^{\infty }\left( {\begin{array}{c}{1/q }\\ k+1\end{array}}\right) x^{k} \). Since \( q>1 \), the coefficients \( b_k=q\left( {\begin{array}{c}{1/q }\\ k+1\end{array}}\right) \) are negative for odd k and positive for even k. Furthermore, the sequence \( (|b_{k} |)\) takes values strictly less than 1 and decays monotonically. Thus, the asserted inequalities follow easily. \(\square \)
We distinguish the following three cases depending on p for which the arguments are quite different:
-
p lies between an odd and an even number with the subcases:
-
\(\bullet \) \( p\in [3,\infty ) \)
-
\(\bullet \) \( p\in (1,2] \)
-
-
p lies between an even and an odd number.
Here, for \(a,b \in \mathbb {N}\), we say that p is between a and b if \(a \le p \le b\).
We start with investigating the case of p lying between an odd and an even number. To this end we consider two subsequent summands as they appear in the sum given by \( F_{p} \) and show that they are positive. (Indeed the sum in \( F_{p} \) starts at \( n=2 \) but we also consider the corresponding term for \( n=1 \).)
Lemma 8
Let p be between an odd and an even integer. Then, for all \(0< x\le 1/2\) and odd \( n\in 2\mathbb {N}-1 \)
Proof
Let p be between an odd and an even integer. We first consider \( n < p-1 \). In the case \( k\le p-1 \), we have \( \left( {\begin{array}{c}{p-1 }\\ k\end{array}}\right) \ge 0 \). So, the statement for \( n < p-1 \) follows directly from Lemma 7 as \( |g(x)|< 1 \) for \(0< x\le 1/2 \). (Observe that \( n+1\le p-1 \) for \( n < p-1 \) and \( n\in 2\mathbb {N}-1 \) as p is between an odd and an even integer.)
On the other hand, for odd \( n\in 2\mathbb {N}-1 \) with \( n\ge p-1 \),
From Lemma 7 we know that \(g^{n+1}(x)\ge 0 \ge g^n(x)\) for odd \(n\in 2\mathbb {N}-1 \) and \(0\le x \le 1/2\).
We obtain
where the last inequality follows from \( 0\le g(-x)< 1 \) for \(0\le x\le 1/2 \) thanks to Lemma 7. \(\square \)
With Lemma 8 we can treat the case of \( p\ge 3 \) lying between an odd and an even number. This is done in the next proposition.
Proposition 9
Let \( p \ge 3 \) be between an odd and an even integer. Then, for all \(0< x\le 1/2\) we have \( F_{p}(x)\ge 0 \) and
In particular, \( w_{p}(n)>w_{p}^{H}(n) \) for \( n\ge 2 \).
Proof
We can write \( F_{p}(x)=\sum _{n=2}^{\infty }\left( {\begin{array}{c}p-1\\ n\end{array}}\right) \bigl (g^{n}(-x)-g^{n}(x)\bigr )\) as
By Lemma 8 the terms in the sum on the right hand side are all positive. Furthermore, \( \left( {\begin{array}{c}p-1\\ 2\end{array}}\right) \ge 0 \) for \( p\ge 3 \) and \( g(-x)\ge |g(x) |\) by Lemma 7. Thus, also the first term on the right hand side is positive as well and \( F_{p}\ge 0 \) follows. From the discussion in the beginning in the section we take \( E_{p}(x)>0 \) for \( 0<x \le 1/2 \). The ”in particular“ follows from the discussion above Lemma 7.
\(\square \)
Note that we cannot treat the case \( 1\le p\le 2 \) in the same way since the sum in \( F_{p} \) starts at the index \( n=2 \). Hence, there is still a negative term \( \left( {\begin{array}{c}p-1\\ 2\end{array}}\right) \bigl (g^{2}(-x)-g^{2}(x)\bigr ).\) We deal with this case, \( 1\le p\le 2 \), next.
We denote the Taylor coefficients of \(x\mapsto g(-x) \) by \( a_{k} \), i.e.,
The function \( E_{p}(x)=-2p\sum _{k\in 2\mathbb {N}+1}\left( {\begin{array}{c}{1/q }\\ k+1\end{array}}\right) x^{k} \) is odd and, therefore, we have
Furthermore, recall that \(E_p(x) > 0\) for \(x > 0\), since \( -2p \left( {\begin{array}{c}{1/q }\\ k+1\end{array}}\right) > 0\) for odd k.
Lemma 10
Let \({p\ge 1}\) and \({0\le x\le 1/2}\). Then,
Proof
We calculate using \( a_{2}\ge a_{n} \) for \( n\ge 2\), the geometric series, \(x\le 1/2\) and the specific value of the Taylor coefficient \(a_2=q\left( {\begin{array}{c}1/q\\ 3\end{array}}\right) =\frac{(p+1)}{6p^2}\)
\(\square \)
With the help of this lemma and Lemma 8 we can treat the case \(p\in (1,2]\).
Proposition 11
Let \( p \in (1,2] \). Then, for all \(0< x\le 1/2\), we have
In particular, \( w_{p}(n)>w_{p}^{H}(n) \) for \( n\ge 2 \).
Proof
We show \( E_{p}+F_{p}>0 \) and deduce the “in particular” from the discussion above Lemma 7. By Lemma 8 we have for all \(0< x\le 1/2\)
where we used the definition of \( E_{p} \), i.e., \((p-1)\bigl (g(-x)-g(x)\bigr )=E_{p}(x)+ \frac{p-1}{p}x \) and Lemma 10 which is justified since \(E_p(x)>0\) and \( p-2<0 \). Moreover, in the last step we estimated the coefficient of the first term in its minimum in \( p=1 \) and \( x=1/2 \).
Now, we use the representation of \( E_{p} \) as a power series to estimate
Putting this together with the estimate on \( F_{p} \) above, we arrive at
Hence, it remains to consider the case of p between an even and an odd integer for which we need the following three lemmas.
Lemma 12
Let \( p, q \ge 1 \) such that \(1/p+1/q=1\) and \( k\ge 2 \). Then,
Proof
We calculate using \(1/p+1/q=1\)
Lemma 13
Let \( p, q\in (1,\infty ) \) such that \(1/p+1/q=1\) and \(k\in \mathbb {N}, k> p \). Then,
Proof
Let \( n\in \mathbb {N}\) be such that \( n-1\le p\le n \). Moreover, let \(\gamma = p-(n-1)\), i.e., \(1-\gamma =n-p\), so, \( \gamma \in [0,1] \). Since \( k>p \) and \(n,k\in \mathbb {N}\), we have that \( k\ge n \) and therefore,
Lemma 14
For \(0< x\le 1/2 \) and \( q>1 \), we get
Proof
We calculate using \(a_2\ge a_k\) for \(k\ge 2\)
With the help of these lemmas we can finally treat the case where p lies between an even and an odd number.
Proposition 15
Let \( p \in [2,\infty ) \) be between an even and an odd integer. Then, for all \(0< x\le 1/2\) we have
In particular, \( w_{p}(n)>w_{p}^{H}(n) \) for \( n\ge 2 \).
Proof
Clearly, we have \( \left( {\begin{array}{c}{p-1 }\\ n\end{array}}\right) \ge 0 \) for \( n\le p\) and for \( n\in 2\mathbb {N}\). Since we have \( g(-x)\ge |g(x)| \) by Lemma 7, we obtain for the first \( n\le p \) terms and the terms for even n in \( F_{p}(x) \) that
Note that \( E_{p}(x)= 2(p-1)\sum _{n=1}^{\infty }a_{2n+1}x^{2n+1}> 2(p-1)\sum _{n=k}^{\infty }a_{2n+1}x^{2n+1}\) since the coefficients \(a_k\) are positive. With the observation made at the beginning of the proof, this leads to
We continue to show that all the terms in the sum are strictly positive which finishes the proof. To this end note that for \( n\ge p \) with \( n\in 2\mathbb {N}+1 \), we use \( g(-x) \ge |g(x)| \), see Lemma 7, as well as \(\left( {\begin{array}{c}p-1\\ n\end{array}}\right) \le 0\) in the first step and the estimate on \( \left( {\begin{array}{c}p-1\\ n\end{array}}\right) \), see Lemma 13, and the estimate on \( g(-x) \), see Lemma 14, in the second step in order to get
We use the estimate on \( a_{n} \), Lemma 12,
Next, we put these two estimates together and find that the minimum in the coefficient is clearly assumed at \( q=2 \) since \( p\ge 2\ge q\)
where the positivity follows by a simple induction argument. This concludes the proof by noticing that the “in particular” part follows from the discussion above Lemma 7. \(\square \)
In summary, the above considerations yield
for \( p\in (1,\infty ) \) and \(0< x\le 1/2 \). By the discussion at the beginning of the section this yields \( w_{p}(n)>w_{p}^{H}(n) \) for \( n\ge 2 \).
We finish the section by treating the case \( n=1 \) which corresponds to \( x=1 \). With this we finally conclude that \( w_{p}(n)>w_{p}^{H}(n) \) for all \( n\ge 1 \) in the next section.
Proposition 16
Let \( p\in (1,\infty ) \). Then, \(w_p(1)> w^H_p(1)\).
Proof
Recall that \( w_p(1) =1-(2^{1-1/p}-1)^{p-1}\) and \( w_{p}^{H}=(1-1/p)^{p} \). By the mean value theorem applied to the function \( [1,2]\rightarrow [1,2^{1-1/p}] \), \( t\mapsto t^{1-1/p} \) we find
Therefore,
Now the function \(\psi :(1,\infty ) \rightarrow (0,\infty )\), \( p\mapsto \left( 1-{1}/{p}\right) ^{p-1}+\left( 1-{1}/{p}\right) ^{p} \) is strictly monotonically decreasing because
since \(\theta :p\mapsto (2p-1)\log (p-1)/p \) is strictly monotonically increasing and we have \( \lim _{p\rightarrow \infty }\theta (p) = -2 \). Hence, we conclude
4 Proof of Theorem 1
Proof of Theorem 1
Combining Propositions 4 and 5 from Sect. 2 yields that \( w_{p} \) satisfies the Hardy inequality. In Sect. 3, one obtains from Proposition 9, Proposition 11, Proposition 15 and Proposition 16 that \( w_{p}>w_{p}^{H} \) on \(\mathbb {N}\).
The optimality of the p-Hardy weight, i.e., for every \(w\gneq w_p\) the p-Hardy inequality fails (criticality), and in addition \(u \not \in \ell ^p(\mathbb {N},w_p)\), where \(u(n)=n^{(p-1)/p}, n\in \mathbb {N}\) (null-criticality) can be deduced from [7, Theorem 2.3]. To this end one has to check that the function \( v> 0 \) which gives rise to the Hardy weight via the supersolution construction, i.e., \( u=v^{(p-1)/p} \) and \( w_p = \Delta _{p}u / u^{(p-1)} \) is proper, i.e., the preimage of every compact set in \( (0,\infty ) \) is compact, and of bounded oscillation, i.e., \( C^{-1}\le v(n)/v(n+1)\le C \) for some \( C>0 \) and all n. These two properties are clearly satisfied for the identity function \( \textrm{id}(n)=n \) which was used in Proposition 5.
Note that given criticality, null-criticality is also a simple consequence of \(w_p> w_p^H\) on \(\mathbb {N}\) and since \( u(n)=n^{(p-1)/p} \) is not in \( \ell ^{p}( \mathbb {N}, w^H_{p}) \). Moreover, in the remark below, Remark 17, we give a sketch of the proof of criticality to give the reader an idea of the arguments involved.
To see the statement about the coefficients in the series expansion of \( w_{p} \) for integer \(p \ge 2\), recall the function
with \( 1/p+1/q=1 \) on [0, 1] from the previous section. It is easy to check that the function \(w_{+}:x\mapsto {\bigl (1-(1-x)^{1/q}\bigr )^{p-1}}\) is absolutely monotonic on [0, 1) for integer \( p\ge 2 \) with strictly positive derivatives. On the other hand, expanding the function \(w_{-}:x\mapsto {\bigl ((1+x)^{1/q}-1\bigr )^{p-1}} \) at \( x=0 \), we observe that it has the same Taylor coefficients in absolute value as \( w_{+} \). However, the signs alternate such that for the difference \( w_{+}-w_{-} \) of these two functions the even/odd coefficients cancel for odd/even p. Furthermore, in the Taylor expansion of w at \( x=0 \) the first non-zero coefficient is the one for \( x^{p} \) (confer Sect. 3). \(\square \)
Remark 17
(Sketch of the proof of criticality) Criticality of the weight \( w_{p} \) is equivalent to existence of a null sequence, i.e., existence of \( 0\le \varphi _N\in C_{c} (\mathbb {N})\) with \( \varphi (0)=0 \), \( \varphi (1)=1 \) and \( \varphi _{N} \) converging pointwise to \( u(n)=n^{1/q} \) and
cf. [8, Theorem 5.1]. We denote \( a\vee b =\max \{a,b\} \) for \( a,b\in \mathbb {R}\) and choose
for \( n,N\in \mathbb {N}\) and \( u(n)=n^{1/q} \). We now use the simplified energy from [9, Theorem 3.1] which serves as a substitute of a ground state transform for \( p\ne 2 \)
where \( \asymp \) means that there are two sided estimates with positive constants independent of \( \psi \). We employ \( u(n)=n^{1/q} \) and the definition of \(\varphi _{N}, \psi _{N}\) to obtain
where we used \( a(b+c)^{r}\le C( ab^{r}+ac^{r}) \) for all \( a,b,c\ge 0 \), \( r\in \mathbb {R}\) and some \( C=C(r) \) to split the sum into two sums, \( |(0\vee a)-(0\vee b)|\le |a-b| \) for all \( a,b\in \mathbb {R}\) and \( |n^r-(n+1)^r|\asymp n^{r-1} \) for \( r\in (0,\infty ) \) cf. e.g. [14, Lemma 2.28]. Now, using \( \log (1+1/n)\le 1/n \) and again \( a(b+c)^{r}\le C( ab^{r}+ac^{r}) \), we infer
which finishes the proof.
While we know that all Taylor coefficients of the Hardy weight \( w_{p}\) are strictly positive for integer \( p\ge 2 \), we only know \( w_{p}>w^{H}_p \) for non-integer \( p>1 \). This leads us to the following conjecture.
Conjecture
We conjecture that \(x\mapsto w(x)\) as defined in Sect. 3 is absolutely monotonic.
Data availability
Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
References
Allegretto, W.: On the equivalence of two types of oscillation for elliptic operators. Pac. J. Math. 55, 319–328 (1974)
de Monvel, A.B., Lenz, D., Stollmann, P.: Sch’nol’s theorem for strongly local forms. Israel J. Math. 173, 189–211 (2009)
Devyver, B., Fraas, M., Pinchover, Y.: Optimal Hardy weight for second-order elliptic operator: an answer to a problem of Agmon. J. Funct. Anal. 266(7), 4422–4489 (2014)
Dodziuk, J.: Difference equations, isoperimetric inequality and transience of certain random walks. Trans. Amer. Math. Soc. 284(2), 787–794 (1984)
Devyver, B., Pinchover, Y.: Optimal \(L^p\) Hardy-type inequalities. Ann. Inst. H. Poincaré Anal. Non Linéaire 33(1), 93–118 (2016)
Elliott, E.B.: A simple exposition of some recently proved facts as to convergency. J. London Math. Soc. 1(2), 93–96 (1926)
Fischer, F.: On the Optimality and Decay of \(p\)-Hardy Weights on Graphs (2022). arXiv:2212.07728
Fischer, F.: Quasi-Linear Criticality Theory and Green’s Functions on Graphs (2022). arXiv:2207.05445
Fischer, F.: A non-local quasi-linear ground state representation and criticality theory. Calc. Var. Partial Differ. Equ. 62(5), 33 (2023)
Frank, R.L., Seiringer, R.: Non-linear ground state representations and sharp Hardy inequalities. J. Funct. Anal. 255(12), 3407–3430 (2008)
Gerhat, B., Krejčiřík, D., Štampach, F.: An improved discrete Rellich inequality on the half-line. to appear in Israel J. Math (2022). arXiv:2206.11007
Hardy, G.H.: Note on a theorem of Hilbert. Math. Z. 6(3–4), 314–317 (1920)
Hardy, G.H., Littlewood, J.E., Pólya, G.: Inequalities. Cambridge University Press, Cambridge (1934)
Keller, M., Lenz, D., Wojciechowski, R.K: Graphs and discrete Dirichlet spaces, volume 358 of Grundlehren der mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer, Cham, [2021] \(\copyright \) (2021)
Kufner, A., Maligranda, L., Persson, L.E.: The prehistory of the Hardy inequality. Amer. Math. Monthly 113(8), 715–732 (2006)
Keller, Matthias, Pinchover, Yehuda, Pogorzelski, Felix: An improved discrete Hardy inequality. Amer. Math. Monthly 125(4), 347–350 (2018)
Keller, M., Pinchover, Y., Pogorzelski, F.: Optimal Hardy inequalities for Schrödinger operators on graphs. Comm. Math. Phys. 358(2), 767–790 (2018)
Keller, M., Pinchover, Y., Pogorzelski, F.: Criticality theory for Schrödinger operators on graphs. J. Spectr. Theory 10(1), 73–114 (2020)
Krejčiřík, D., Štampach, F.: A sharp form of the discrete Hardy inequality and the Keller–Pinchover–Pogorzelski inequality. Amer. Math. Monthly 129(3), 281–283 (2022)
Landau, E.: A letter to G. H. Hardy, June 21, 1921 (1921)
Lefèvre, P.: A short direct proof of the discrete Hardy’s inequality. Archiv der Mathematik, pp. 1–4, Oct (2019)
Lefèvre, P.: Weighted discrete Hardy’s inequalities. To appear in Ukr. J. Math (2020)
Piepenbrink, J.: Nonoscillatory elliptic equations. J. Differ. Equ. 15, 541–550 (1974)
Pinchover, Y., Psaradakis, G.: On positive solutions of the \((p, A)\)-Laplacian with potential in Morrey space. Anal. PDE 9(6), 1317–1358 (2016)
Acknowledgements
The authors gratefully acknowledge inspiring discussions with Y. Pinchover. Moreover, the authors are grateful for the financial support of the DFG. Furthermore, F. F. thanks the Heinrich-Böll-Stiftung for the support. Furthermore, we express our gratitude towards the anonymous referee for helpful suggestions.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors have no competing interests to declare that are relevant to the content of this article.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Fischer, F., Keller, M. & Pogorzelski, F. An Improved Discrete p-Hardy Inequality. Integr. Equ. Oper. Theory 95, 24 (2023). https://doi.org/10.1007/s00020-023-02743-6
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00020-023-02743-6