Theorem 1.1 (Conjecture 1.3)
8FDBFor n \geq 45, we have
\dfrac{p(n-1)}{p(n)}\Bigl(1+\dfrac{\pi}{\sqrt{24}n^{3/2}}\Bigr)>\dfrac{p(n)}{p(n+1)}.
Higher order log-concavity of the overpartition function and its consequences
- arXiv
- 2204.07961v1 · PDF · 17 April 2022
- Categories
- math.NT, math.CO
- Licence
- CC BY 4.0 The site shows the text.
Abstract
Let \bar{p}(n) denote the overpartition function. In this paper, we study the asymptotic higher order \log-concavity property of the overpatition function in a similar framework done by Hou and Zhang for the partition function. This will enable us to move on further in order to prove \log-concavity of overpartitions, explicitly by studying the asymptotic expansion of the quotient \bar{p}(n-1)\bar{p}(n+1)/\bar{p}(n)^2 upto a certain order so that one can finally ends up with the phenomena of 2-\log-concavity and higher order Turán property of \bar{p}(n) by following a sort of unified approach.
Items 20
Theorem 1.2
MSBKThe sequence \{{\overline{p}}(n)\}_{n\geq1} is asymptotically r-log-concave for any positive integer r.
Theorem 1.3
RI80For all n \geq 37, we have
s_n-\dfrac{15}{n^4}<{\overline{u}}_n<s_n+\dfrac{20}{n^4},
where
s_n=1-\dfrac{\pi}{4n^{3/2}}+\dfrac{1}{n^2}-\dfrac{3}{4\pi n^{5/2}}-\dfrac{32+\pi^4}{32\pi^2}\dfrac{1}{n^3}-\Bigl(\dfrac{5}{4\pi^3}+\dfrac{21\pi}{64}\Bigr)\dfrac{1}{n^{7/2}}.
Corollary 1.4
SY9J[Engel-2017, Theorem 1.2] \{{\overline{p}}(n)\}_{n \geq 4} is \log-concave.
Corollary 1.5
XCO1[LiuZhang-2021, Equation (1.6)] For n \geq 2,
\dfrac{{\overline{p}}(n-1)}{{\overline{p}}(n)}\Bigl(1+\dfrac{\pi}{4n^{3/2}}\Bigr)>\dfrac{{\overline{p}}(n)}{{\overline{p}}(n+1)}.
Theorem 1.6
K5AMFor all n \geq 27,
t_n-\dfrac{120}{n^{5/2}}< \dfrac{(1-{\overline{u}}_n)^2}{{\overline{u}}^2_n(1-{\overline{u}}_{n-1})(1-{\overline{u}}_{n+1})}<t_n+\dfrac{120}{n^{5/2}},
where
t_n=1+\dfrac{\pi}{2n^{3/2}}-\dfrac{7}{2n^2}.
Corollary 1.7
LBN0[Mukherjee-2022, Theorem 1.7] \{{\overline{p}}(n)\}_{n \geq 42} is 2-\log-concave.
Corollary 1.8
HYQZFor n \geq 52,
{\overline{u}}^2_n (1-{\overline{u}}_{n-1})(1-{\overline{u}}_{n+1}) \Bigl(1+\dfrac{\pi}{2n^{3/2}}\Bigr)>(1-{\overline{u}}_n)^2.
Theorem 1.9
V1W1For all n \geq 2,
v_n-\dfrac{120}{n^{5/2}}< \dfrac{4(1-{\overline{u}}_n)(1-{\overline{u}}_{n+1})}{(1-{\overline{u}}_n{\overline{u}}_{n+1})^2}<v_n+\dfrac{101}{n^{5/2}},
where
v_n=1+\dfrac{\pi}{4n^{3/2}}-\dfrac{25}{16n^2}.
Corollary 1.10
VQTY[LiuZhang-2021, Theorem 1.2] \{{\overline{p}}(n)\}_{n \geq 16} satisfies higher order Turán inequality.
Corollary 1.11
BR4QFor n \geq 2,
(1-{\overline{u}}_n{\overline{u}}_{n+1})^2 \Bigl(1+\dfrac{\pi}{4n^{3/2}}\Bigr)>4(1-{\overline{u}}_n)(1-{\overline{u}}_{n+1}).
Lemma 2.1
C9QQFor any integer m\geq1, there exists a real number
N_0(m) :=
\begin{cases}
1, &\quad \text{if}\ m=1,\\
2m \log m-m \log \log m, & \quad \text{if}\ m \geq 2,
\end{cases}
such that
\begin{aligned}
x^me^{-x}<1,~~~\text{for}~~x\geq N_0(m).
\end{aligned}
Used by Proof
Proof of Lemma 2.1
For m=1, it is immediate that N_0(m)=1. For m \geq 2, rewrite the inequality x^me^{-x}<1 as f(x):=x-m\log x>0. Now f(x) is strictly increasing for x>m. In order to show f(x)>0 for x\geq N_0(m), first we show that N_0(m)>m and then it is enough to show f(N_0(m))>0. To prove N_0(m)>m, it is equivalent to show m^2>e. \log m which holds for m \geq 2. Next, we observe that
f(N_0(m))>0 \Leftrightarrow\log m > \log 2+2 \log \log m+\log \Bigl(1-\dfrac{\log \log m}{2\log m}\Bigr)\\
For m \geq 3, \log \Bigl(1-\dfrac{\log \log m}{2\log m}\Bigr) <0 and hence, it is sufficient to prove
\log m > \log 2+2 \log \log m \Leftrightarrow m>2\ (\log m)^2,
which holds for m \geq 14. Therefore, f(N_0(m))>0 for all m \geq 14 and we conclude the proof by checking numerically that f(N_0(m))>0 for 2\leq m \leq 13.
Theorem 2.2
VYRYFor any integer m\geq2, there exist an integer N_1(m) with
\begin{aligned}
N_1(m)=\max\left\{184,\dfrac{9}{4\pi^2} N^2_0(m)\right\},
\end{aligned}
such that
\begin{aligned}
\left|{\widehat{y}}_n\right|<\left(\frac{3}{2}\right)^{m+1}{\widehat{\mu}}(n)^{-m},
\end{aligned}
where {\widehat{y}}_n={\widehat{R}}(n)/{\widehat{T}}(n).
Used by Proof
Proof of Theorem 2.2
By (?) and (?), we have
\begin{aligned}
{\widehat{y}}_{n}\leq e^{-\frac{2{\widehat{\mu}}(n)}{3}}\left({\widehat{T}}_1(n)+{\widehat{R}}_1(n)\right),
\end{aligned}
where
\begin{aligned}
{\widehat{T}}_1(n)=\frac{{\widehat{\mu}}(n)+1}{{\widehat{\mu}}(n)-1}e^{-\frac{4{\widehat{\mu}}(n)}{3}},
~~~~~~
{\widehat{R}}_1(n)=4\cdot3^{\frac{5}{2}}\frac{1}{{\widehat{\mu}}(n)-1}.
\end{aligned}
From (Proof), it follows that
\begin{aligned}
{\widehat{T}}_1(184)+{\widehat{R}}_1(184)<\frac{3}{2}.
\end{aligned}
Therefore,
\begin{aligned}
\left|{\widehat{y}}_n\right|&<\frac{3}{2}e^{-\frac{2{\widehat{\mu}}(n)}{3}},~~~\text{for}~~n\geq 184.
\end{aligned}
According to Lemma 2.1, there exists the integer N_0(m) such that for \frac{2}{3}{\widehat{\mu}}(n)\geq N_0(m)
\begin{aligned}
e^{-\frac{2{\widehat{\mu}}(n)}{3}}<\left(\frac{3}{2}\right)^m{\widehat{\mu}}(n)^{-m}.
\end{aligned}
On the other side,
\begin{aligned}
n^{\frac{1}{2}}=\frac{{\widehat{\mu}}(n)}{\pi}.
\end{aligned}
Therefore, when
\begin{aligned}
n\geq\max\left\{184, \dfrac{9}{4\pi^2} N^2_0(m)\right\}=N_1(m),
\end{aligned}
we have
\begin{aligned}
{\widehat{\mu}}(n)>\dfrac{3}{2}N_0(m),
\end{aligned}
which completes the proof. 4pt7pt
Uses Lemma 2.1
Lemma 2.3
Z8BOLet {\widehat{\nu}}(n)={\widehat{\mu}}(n)({\widehat{\mu}}_1(n)-1) and
\begin{aligned}
{\widehat{\mu}}_1(n)&=\sum_{k=0}^{m'}{1/2\choose k}\pi^{2k}{\widehat{\mu}}(n)^{-2k},\qquad
{\widehat{\varepsilon}}_1(n)=\left|{1/2\choose m'+1}\right|\pi^{2(m'+1)}{\widehat{\mu}}(n)^{-2(m'+1)},
\\[3pt]
{\widehat{\mu}}_2(n)&=\sum_{k=0}^{m'}{-3/2\choose k}\pi^{2k}{\widehat{\mu}}(n)^{-2k}, \qquad
{\widehat{\varepsilon}}_2(n)=\left|{-3/2\choose m'+1}\right|\pi^{2(m'+1)}{\widehat{\mu}}(n)^{-2(m'+1)},
\\[3pt]
{\widehat{\nu}}_1(n)&=\left({\widehat{\mu}}_1(n)-{\widehat{\varepsilon}}_1(n)-\frac{1}{{\widehat{\mu}}(n)}\right)\sum_{k=0}^m{\widehat{\mu}}(n)^{-k},
\\[3pt]
{\widehat{\nu}}_2(n)&=\left({\widehat{\mu}}_1(n)+{\widehat{\varepsilon}}_1(n)-\frac{1}{{\widehat{\mu}}(n)}\right)
\left(\sum_{k=0}^m{\widehat{\mu}}(n)^{-k}+2{\widehat{\mu}}(n)^{-m-1}\right),
\end{aligned}
then
\begin{aligned}
\frac{{\widehat{T}}(n+1)}{{\widehat{T}}(n)}&>{\widehat{\nu}}_1(n)\left({\widehat{\mu}}_2(n)-{\widehat{\varepsilon}}_2(n)\right)
(1-{\widehat{\mu}}(n){\widehat{\varepsilon}}_1(n))\sum_{k=0}^m\frac{{\widehat{\nu}}(n)^k}{k!},
\end{aligned}
and
\begin{aligned}
\frac{{\widehat{T}}(n+1)}{{\widehat{T}}(n)}&<{\widehat{\nu}}_2(n)({\widehat{\mu}}_2(n)+{\widehat{\varepsilon}}_2(n))
(1+2{\widehat{\mu}}(n){\widehat{\varepsilon}}_1(n))\left(\sum_{k=0}^m\frac{{\widehat{\nu}}(n)^k}{k!}
+e^{{\widehat{\nu}}(n)}\frac{{\widehat{\nu}}(n)^{m+1}}{(m+1)!}\right)
\end{aligned}
where {\widehat{T}}(n) is defined as in (?).
Used by Proof
Proof of Lemma 2.3
By (?), we have
\begin{aligned}
\frac{{\widehat{T}}(n+1)}{{\widehat{T}}(n)}=\frac{{\widehat{\mu}}(n+1)-1}{{\widehat{\mu}}(n)-1}\cdot\frac{{\widehat{\mu}}(n)^3}{{\widehat{\mu}}(n+1)^3}\cdot e^{{\widehat{\mu}}(n+1)-{\widehat{\mu}}(n)}.
\end{aligned}
Now we consider the above ratio term by term. For the first factor, we have
\begin{aligned}
\frac{{\widehat{\mu}}(n+1)-1}{{\widehat{\mu}}(n)-1}=\frac{\frac{{\widehat{\mu}}(n+1)}{{\widehat{\mu}}(n)}-\frac{1}{{\widehat{\mu}}(n)}}{1-\frac{1}{{\widehat{\mu}}(n)}}.
\end{aligned}
By Taylor’s Theorem, we have
\begin{aligned}
\left(1-\frac{1}{{\widehat{\mu}}(n)}\right)^{-1}
=\sum_{k=0}^\infty {\widehat{\mu}}(n)^{-k},
\end{aligned}
which implies that
\begin{aligned}
\sum_{k=0}^m{\widehat{\mu}}(n)^{-k}<\left(1-\frac{1}{{\widehat{\mu}}(n)}\right)^{-1}
<\sum_{k=0}^m{\widehat{\mu}}(n)^{-k}+2{\widehat{\mu}}(n)^{-m-1}.
\end{aligned}
Noting that
\begin{aligned}
{\widehat{\mu}}(n+1)={\widehat{\mu}}(n)\left(1+\frac{\pi^2}{{\widehat{\mu}}(n)^2}\right)^{\frac{1}{2}}.
\end{aligned}
For any integer m, let m'=\lfloor\frac{m}{2}\rfloor. Since
\begin{aligned}
\left(1+\frac{\pi^2}{{\widehat{\mu}}(n)^2}\right)^{\frac{1}{2}}=\sum_{k=0}^{m'}{1/2\choose k}\pi^{2k}{\widehat{\mu}}(n)^{-2k}
+{1/2\choose m'+1}\left(\frac{\pi^2}{{\widehat{\mu}}(n)^2}\right)^{m'+1}\left(1+\xi\right)^{\frac{1}{2}-m'-1},
\end{aligned}
where 0<\xi<\frac{\pi^2}{{\widehat{\mu}}(n)^2}. We have
\begin{aligned}
{\widehat{\mu}}_1(n)-{\widehat{\varepsilon}}_1(n)<\frac{{\widehat{\mu}}(n+1)}{{\widehat{\mu}}(n)}<{\widehat{\mu}}_1(n)+{\widehat{\varepsilon}}_1(n).
\end{aligned}
Combining (Proof) and (Proof), we deduce that
\begin{aligned}
{\widehat{\nu}}_1(n)<\frac{{\widehat{\mu}}(n+1)-1}{{\widehat{\mu}}(n)-1}<{\widehat{\nu}}_2(n).
\end{aligned}
For the second factor, we have
\begin{aligned}
\frac{{\widehat{\mu}}(n)^3}{{\widehat{\mu}}(n+1)^3}=\left(1+\frac{\pi^2}{{\widehat{\mu}}(n)^2}\right)^{-3/2}.
\end{aligned}
Since
\begin{aligned}
\left(1+\frac{\pi^2}{{\widehat{\mu}}(n)^2}\right)^{-\frac{3}{2}}=\sum_{k=0}^{m'}{-3/2\choose k}\pi^{2k}{\widehat{\mu}}(n)^{-2k}
+{-3/2\choose m'+1}\left(\frac{\pi^2}{{\widehat{\mu}}(n)^2}\right)^{m'+1}\left(1+\xi\right)^{-\frac{3}{2}-m'-1},
\end{aligned}
where 0<\xi<\frac{\pi^2}{{\widehat{\mu}}(n)^2}. We have
\begin{aligned}
{\widehat{\mu}}_2(n)-{\widehat{\varepsilon}}_2(n)<\frac{{\widehat{\mu}}(n)^3}{{\widehat{\mu}}(n+1)^3}<{\widehat{\mu}}_2(n)+{\widehat{\varepsilon}}_2(n).
\end{aligned}
For the last factor, using (Proof), then
\begin{aligned}
e^{{\widehat{\nu}}(n)-{\widehat{\mu}}(n){\widehat{\varepsilon}}_1(n)}<e^{{\widehat{\mu}}(n+1)-{\widehat{\mu}}(n)}<e^{{\widehat{\nu}}(n)+{\widehat{\mu}}(n){\widehat{\varepsilon}}_1(n)}.
\end{aligned}
Evidently, for 0<x<\frac{1}{2},
e^{-x}>1-x,\qquad e^x<1+2x
and for x>0
\sum_{k=0}^m\frac{x^k}{k!}<e^x<\sum_{k=0}^m\frac{x^k}{k!}+e^x\frac{x^{m+1}}{(m+1)!},
so that
\begin{aligned}
e^{{\widehat{\mu}}(n+1)-{\widehat{\mu}}(n)}>(1-{\widehat{\mu}}(n){\widehat{\varepsilon}}_1(n))\sum_{k=0}^m\frac{{\widehat{\nu}}(n)^k}{k!}
\end{aligned}
and
\begin{aligned}
e^{{\widehat{\mu}}(n+1)-{\widehat{\mu}}(n)}<(1+2{\widehat{\mu}}(n){\widehat{\varepsilon}}_1(n))\left(\sum_{k=0}^m\frac{{\widehat{\nu}}(n)^k}{k!}
+e^{{\widehat{\nu}}(n)}\frac{{\widehat{\nu}}(n)^{m+1}}{(m+1)!}\right).
\end{aligned}
Applying the estimates (Proof), (Proof), (Proof) and (Proof) to (Proof), we reach (2.3) and (2.3). This completes the proof. 4pt7pt
Theorem 2.4
G2LPFor any positive integer m, there exist integer N, real numbers a_k and C_1, C_2>0 such that for n\geq N
\begin{aligned}
\sum_{k=0}^ma_k{\widehat{\mu}}(n)^{-k}-C_1{\widehat{\mu}}(n)^{-m-1}<\frac{{\overline{p}}(n+1)}{{\overline{p}}(n)}<\sum_{k=0}^ma_k{\widehat{\mu}}(n)^{-k}+C_2{\widehat{\mu}}(n)^{-m-1}.
\end{aligned}
Proof of Theorem 2.4
By (?) and Theorem 2.2, for any m\geq2 there exists the integer N_1(m) such that
|{\overline{p}}(n)/{\widehat{T}}(n)-1|<\left(\frac 32\right)^{m+1}{\widehat{\mu}}(n)^{-m},~~\forall n\geq N_1(m).
We have
{\widehat{T}}(n)\left(1-\left(\frac 32\right)^{m+1}{\widehat{\mu}}(n)^{-m}\right)<{\overline{p}}(n)<{\widehat{T}}(n)\left(1+\left(\frac 32\right)^{m+1}{\widehat{\mu}}(n)^{-m}\right).
Since {\widehat{\mu}}(n) is a increasing function of n, we derive that
\frac {{\widehat{T}}(n+1)}{{\widehat{T}}(n)} \frac {1-\left(\frac 32\right)^{m+1}{\widehat{\mu}}(n)^{-m}}{1+\left(\frac 32\right)^{m+1}{\widehat{\mu}}(n)^{-m}}<\frac {{\overline{p}}(n+1)}{{\overline{p}}(n)}<\frac {{\widehat{T}}(n+1)}{{\widehat{T}}(n)} \frac {1+\left(\frac 32\right)^{m+1}{\widehat{\mu}}(n)^{-m}}{1-\left(\frac 32\right)^{m+1}{\widehat{\mu}}(n)^{-m}}.
We find that for 0<\lambda<1/3,
\frac {1+\lambda}{1-\lambda}<1+3\lambda~~~ and~~~\frac {1-\lambda}{1+\lambda}>1-2\lambda.
According to
0<\frac {\left(\frac 32\right)^{m+1}}{{\widehat{\mu}}(n)^m}<1/3,
we have that for all n\geq N_1(m)
\frac {{\widehat{T}}(n+1)}{{\widehat{T}}(n)}\left(1-4\cdot2^m{\widehat{\mu}}(n)^{-m}\right)<\frac {{\overline{p}}(n+1)}{{\overline{p}}(n)}<\frac {{\widehat{T}}(n+1)}{{\widehat{T}}(n)}\left(1+6\cdot2^m{\widehat{\mu}}(n)^{-m}\right).
By Lemma 2.3, we can see that {\widehat{T}}(n+1)/{\widehat{T}}(n) is bounded by a pair of polynomials in {\widehat{\mu}}(n)^{-1} whose difference is a polynomials in {\widehat{\mu}}(n)^{-1} with degree at least m+1. Combining (Proof) and \lim_{n\rightarrow+\infty}{\widehat{\mu}}(n)=+\infty, we have (2.4). This completes the proof. 4pt7pt
Lemma 2.5
5XBKLet \{a_n\}_{n\geq0} be a positive sequence such that \mathscr{R}^2a_n=a_na_{n+2}/a_{n+1}^2 has the following asymptotic expression
\mathscr{R}^2a_n=1+\frac c{n^\alpha}+\cdots+o\left(\frac c{n^\beta}\right),~~n\rightarrow \infty,
where 0<\alpha\leq\beta. If c<0 and \alpha <2, then \{a_n\}_{n\geq0} is asymptotically [\beta /\alpha]-log-concave.
Source and licence
The text on this page is from arXiv:2204.07961v1 by Gargi Mukherjee, Helen W. J. Zhang and Ying Zhong, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2204/2204.07961.jsonl.