1. Introduction
Given a real number $\omega \geqslant 1$
and integers $m,\,r$
satisfying $0\leqslant r\leqslant m$
, set
where the binomial coefficient $\binom{m}{i}$
equals $\prod _{k=1}^i\frac {m-k+1}{k}$
for $i>0$
and $\binom{m}{0}=1$
. The weighted binomial sum $g_{\omega,m}(r)$
and the partial binomial sum $s_m(r)=g_{1,m}(r)$
appear in many formulas and inequalities, e.g. the cumulative distribution function $2^{-m}s_m(r)$
of a binomial random variable with $p=q=\tfrac {1}{2}$
as in remark 5.3, and the Gilbert–Varshamov bound [Reference Ling and Xing6, Theorem 5.2.6] for a code $C\subseteq \{0,\,1\}^n$
. Partial sums of binomial coefficients are found in probability theory, coding theory, group theory, and elsewhere. As $s_m(r)$
cannot be computed exactly for most values of $r$
, it is desirable for certain applications to find simple sharp upper and lower bounds for $s_m(r)$
. Our interest in bounding $2^{-r}s_m(r)$
was piqued in [Reference Glasby and Paseman4] by an application to Reed–Muller codes $\textrm{RM}(m,\,r)$
, which are linear codes of dimension $s_m(r)$
.
Our main result is a generalized continued fraction $a_0 + \mathcal{K}_{i=1}^{r} \frac {b_i}{a_i}$
(using Gauss’ Kettenbruch notation) for $Q:=\frac {(r+1)}{s_m(r)}\binom{m}{r+1}$
. From this we derive useful approximations to $Q,\, 2+\frac {Q}{r+1}$
, and $s_m(r)$
, and with these find a maximizing input $r_0$
for $g_{\omega,m}(r)$
.
The $j$
th tail of the generalized continued fraction $\mathcal{K}_{i=1}^{r} \frac {b_i}{a_i}$
is denoted by ${\mathcal {T}}_j$
where
If ${\mathcal {T}}_j=\frac {B_j}{A_j}$
, then ${\mathcal {T}}_j=\frac {b_j}{a_j+{\mathcal {T}}_{j+1}}$
shows $b_jA_j-a_jB_j={\mathcal {T}}_{j+1}B_j$
. By convention we set ${\mathcal {T}}_{r+1}=0$
.
It follows from $\binom{m}{r-i}=\binom{m}{r}\prod _{k=1}^i\frac {r-k+1}{m-r+k}$
that $x^i\binom{m}{r}\leqslant \binom{m}{r-i}\le y^i\binom{m}{r}$
for $0\leqslant i\leqslant r$
where $x:=\frac {1}{m}$
and $y:=\frac {r}{m-r+1}$
. Hence $\frac {1-x^{r+1}}{1-x}\binom{m}{r}\leqslant s_m(r)\leqslant \frac {1-y^{r+1}}{1-y}\binom{m}{r}$
. These bounds are close if $\frac {r}{m}$
is near 0. If $\frac {r}{m}$
is near $\tfrac {1}{2}$
then better approximations involve the Berry–Esseen inequality [Reference Nagaev and Chebotarev7] to estimate the binomial cumulative distribution function $2^{-m}s_m(r)$
. The cumulative distribution function $\Phi (x)=\frac {1}{\sqrt {2\pi }}\int _{-\infty }^xe^{-t^2/2}\,dt$
is used in remark 5.3 to show that $|2^{-m}s_m(r)-\Phi (\frac {2r-m}{\sqrt {m}})|\leqslant \frac {0.4215}{\sqrt {m}}$
for $0\leqslant r\leqslant m$
and $m\ne 0$
. Each binomial $\binom{m}{i}$
can be estimated using Stirling's approximation as in [Reference Stănică10, p. 2]: $\binom{m}{i}=\frac {C_i^m}{\sqrt {2\pi p(1-p)m}}\left (1+O\left (\frac {1}{m}\right )\right )$
where $C_i=\frac {1}{p^p(1-p)^{1-p}}$
and $p=p_i=i/m$
. However, the sum $\sum _{i=0}^r\binom{m}{i}$
of binomials is harder to approximate. The preprint [Reference Worsch11] discusses different approximations to $s_m(r)$
.
Sums of binomial coefficients modulo prime powers, where $i$
lies in a congruence class, can be studied using number theory, see [Reference Granville5, p. 257]. Theorem 1.1 below shows how to find excellent rational approximations to $s_m(r)$
via generalized continued fractions.
Theorem 1.1 Fix $r,\,m\in {\mathbb {Z}}$
where $0\leqslant r\leqslant m$
and recall that $s_m(r)=\sum _{i=0}^r\binom{m}{i}$
.
(a) If $b_i=2i(r+1-i)$
, $a_i=m-2r+3i$ for $0\leqslant i\leqslant r$, then
\[ Q:=\frac{(r+1)\binom{m}{r+1}}{s_m(r)}=a_0+{\vcenter{\hbox{ $\mathcal{K}$}}}_{i=1}^{r} \frac{b_i}{a_i}. \]
(b) If $1\leqslant j\leqslant r$
, then ${\mathcal {T}}_j=R_j/R_{j-1}>0$ where the sum $R_j:= 2^j j!\sum _{k=0}^{r-j}\binom{r-k}{j}\binom{m}{k}$ satisfies $b_j R_{j-1} - a_j R_j = R_{j+1}$. Also, $(m-r)\binom{m}{r}-a_0R_0=R_1$.
Since $s_m(m)=2^m$
, it follows that $s_m(m-r)=2^m-s_m(r-1)$
so we focus on values of $r$
satisfying $0\leqslant r\leqslant \lfloor \frac {m}{2}\rfloor$
. Theorem 1.1 allows us to find a sequence of successively sharper upper and lower bounds for $Q$
(which can be made arbitrarily tight), the coarsest being $m-2r\leqslant Q\leqslant m-2r+\frac {2r}{m-2r+3}$
for $1\leqslant r<\frac {m+3}{2}$
, see proposition 2.3 and corollary 2.4.
The fact that the tails ${\mathcal {T}}_1,\,\ldots,\,{\mathcal {T}}_r$
are all positive is unexpected as $b_i/a_i$
is negative if $\frac {m+3i}{2}< r$
. This fact is crucial for approximating ${\mathcal {T}}_1=\mathcal{K}_{i=1}^{r} \frac {b_i}{a_i}$
, see theorem 1.3. Theorem 1.1 implies that ${\mathcal {T}}_1{\mathcal {T}}_2\cdots {\mathcal {T}}_r=R_r/R_0$
. Since $R_0=s_m(r)$
, $R_r=2^rr!$
, ${\mathcal {T}}_j=\frac {b_j}{a_j+{\mathcal {T}}_{j+1}}$
and $\prod _{j=1}^rb_j=2^r(r!)^2$
, the surprising factorizations below follow c.f. remark 2.1.
Corollary 1.2 We have $s_m(r)\prod _{j=1}^r{\mathcal {T}}_j=2^rr!$
and $r!s_m(r)=\prod _{j=1}^r(a_j+{\mathcal {T}}_{j+1})$
.
Suppose that $\omega >1$
and write $g(r)=g_{\omega,m}(r)$
. We extend the domain of $g(r)$
by setting $g(-1)=0$
and $g(m+1)=\frac {g(m)}{\omega }$
in keeping with (1.1). It is easy to prove that $g(r)$
is a unimodal function c.f. [Reference Byun and Poznanović2, § 2]. Hence there exists some $r_0\in \{0,\,1,\,\ldots,\,m\}$
that satisfies
As $g(-1)< g(0)=1$
and $\left (\frac {2}{\omega }\right )^m=g(m)>g(m+1)=\frac {2^m}{\omega ^{m+1}}$
, both chains of inequalities are non-empty. The chains of inequalities (1.3) serve to define $r_0$
.
We use theorem 1.1 to show that $r_0$
is commonly close to $r':=\lfloor \frac {m+2}{\omega +1}\rfloor$
. We always have $r'\leqslant r_0$
(by lemma 3.3) and though $r_0-r'$
approaches $\frac {m}{2}$
as $\omega$
approaches 1 (see remark 4.4), if $\omega \geqslant \sqrt {3}$
then $0\leqslant r_0 - r'\leqslant 1$
by the next theorem.
Theorem 1.3 If $\omega \geqslant \sqrt {3}$
, $m\in \{0,\,1,\,\ldots \}$
and $r':=\lfloor \frac {m+2}{\omega +1}\rfloor$
, then $r_0\in \{r',\,r'+1\}$
, that is
Sharp bounds for $Q$
seem powerful: they enable short and elementary proofs of results that previously required substantial effort. For example, our proof in [Reference Glasby and Paseman4, Theorem 1.1] for $\omega =2$
of the formula $r_0=\lfloor \frac {m}{3}\rfloor +1$
involved a lengthy argument, and our first proof of theorem 1.4 below involved a delicate induction. By this theorem there is a unique maximum, namely $r_0=r'=\lfloor \frac {m+2}{\omega +1}\rfloor$
when $\omega \in \{3,\,4,\,5,\,\ldots \}$
and $\omega \ne m+1$
, c.f. remark 4.2. In particular, strict inequality $g_{\omega,m}(r'-1)< g_{\omega,m}(r')$
holds.
Theorem 1.4 Suppose that $\omega \in \{3,\,4,\,5,\,\ldots \}$
and $r'=\lfloor \frac {m+2}{\omega +1}\rfloor$
. Then
with equality if and only if $\omega =m+1$
.
Our motivation was to analyse $g_{\omega,m}(r)$
by using estimates for $Q$
given by the generalized continued fraction in theorem 1.1. This gives tighter estimates than the method involving partial sums used in [Reference Glasby and Paseman4]. The plots of $y=g_{\omega,m}(r)$
for $0\leqslant r\leqslant m$
are highly asymmetrical if $\omega -1$
and $m$
are small. However, if $m$
is large the plots exhibit an ‘approximate symmetry’ about the vertical line $r=r_0$
(see figure 1). Our observation that $r_0$
is close to $r'$
for many choices of $\omega$
was the starting point of our research.
Figure 1. Plots of $y=g_{\omega,24}(r)$
for $0\leqslant r\leqslant 24$
with $\omega \in \{1,\,\frac 32,\,2,\,3\}$
, and $y=g_{3,12}(r)$
Byun and Poznanović [Reference Byun and Poznanović2, Theorem 1.1] compute the maximizing input, call it $r^*$
, for the function $f_{m,a}(r):=(1+a)^{-r}\sum _{i=0}^r\binom{m}{i}a^i$
where $a\in \{1,\,2,\,\ldots \}$
. Their function equals $g_{\omega,m}(r)$
only when $\omega =1+a=2$
. Some of their results and methods are similar to those in [Reference Glasby and Paseman4] which studied the case $\omega =2$
. They prove that $r^*=\lfloor \frac {a(m+1)+2}{2a+1}\rfloor$
provided $m\not \in \{3,\,2a+4,\,4a+5\}$
or $(a,\,m)\ne (1,\,12)$
when $r^*=\lfloor \frac {a(m+1)+2}{2a+1}\rfloor -1$
.
In Section 2 we prove theorem 1.1 and record approximations to our generalized continued fraction expansion. When $m$
is large, the plots of $y=g_{\omega,m}(r)$
are reminiscent of a normal distribution with mean $\mu \approx \frac {m}{\omega +1}$
. Section 3 proves key lemmas for estimating $r_0$
, and applies theorem 1.1 to prove theorem 1.4. Non-integral values of $\omega$
are considered in Section 4 where theorem 1.3 is proved. In Section 5 we estimate the maximum height $g(r_0)$
using elementary methods and estimations, see lemma 5.1. A ‘statistical’ approximation to $s_m(r)$
is given in remark 5.3, and it is compared in remark 5.4 to the ‘generalized continued fraction approximations’ of $s_m(r)$
in proposition 2.3.
2. Generalized continued fraction formulas
In this section we prove theorem 1.1, namely that $Q:=\frac {r+1}{s_m(r)}\binom{m}{r+1}=a_0+{\mathcal {T}}_1$
where ${\mathcal {T}}_1=\mathcal{K}_{i=1}^r\frac {b_i}{a_i}$
. The equality $s_m(r)=\frac {r+1}{a_0+{\mathcal {T}}_1}\binom{m}{r+1}$
is noted in corollary 2.2.
A version of theorem 1.1(a) was announced in the SCS2022 Poster room, created to run concurrently with vICM 2022, see [Reference Paseman9].
Proof Proof of theorem 1.1
Set $R_{-1}=Q\,s_m(r)=(r+1)\binom{m}{r+1}=(m-r)\binom{m}{r}$
and
Clearly $R_0=s_m(r)$
, $R_{r+1}=0$
and $R_j>0$
for $0\leqslant j\leqslant r$
. We will prove in the following paragraph that the quantities $R_j$
, $a_j=m-2r+3j$
, and $b_j=2j(r+1-j)$
satisfy the following $r+1$
equations, where the first equation (2.1) is atypical:
Assuming (2.2) is true, we prove by induction that ${\mathcal {T}}_{j}=R_{j}/R_{j-1}$
holds for $r+1\geqslant j\geqslant 1$
. This is clear for $j=r+1$
since ${\mathcal {T}}_{r+1}=R_{r+1}=0$
. Suppose that $1\leqslant j\leqslant r$
and ${\mathcal {T}}_{j+1}=R_{j+1}/R_{j}$
holds. We show that ${\mathcal {T}}_{j}=R_{j}/R_{j-1}$
holds. Using (2.2) and $R_j>0$
we have $b_jR_{j-1}/R_j-a_j=R_{j+1}/R_j={\mathcal {T}}_{j+1}$
. Hence $R_j/R_{j-1}=b_j/(a_j+{\mathcal {T}}_{j+1})={\mathcal {T}}_j$
, completing the induction. Equation (2.1) gives $Q=R_{-1}/R_0=a_0+R_1/R_{0}=a_0+{\mathcal {T}}_1$
as claimed. Since $R_j>0$
for $0\leqslant j\leqslant r$
, we have ${\mathcal {T}}_{j}=R_{j}/R_{j-1}>0$
for $1\leqslant j\leqslant r$
. This proves the first half of theorem 1.1(b), and the recurrence ${\mathcal {T}}_j=b_j/(a_j+{\mathcal {T}}_{j+1})$
for $1\leqslant j\leqslant r$
, proves part (a).
We now show that (2.1) holds. The identity $R_0=2^00!\sum _{k=0}^r\binom{m}{k}=s_m(r)$
gives
As $(i+1)\binom{m}{i+1}=(m-i)\binom{m}{i}$
, we get $R_{-1}-a_0R_0=2\sum _{k=0}^{r-1}\binom{r-k}{1}\binom{m}{k}=R_1$
.
We next show that (2.2) holds. In order to simplify our calculations, we divide by $C_{j}:= 2^j j!$
. Using $(j+1)\binom{r-k}{j+1}=(r-k-j)\binom{r-k}{j}$
gives
noting that the term with $k=r-j+1$
in the first sum is zero as $\binom{j-1}{j}=0$
. Using the abbreviation $L=\sum _{k=0}^{r-j}(k-2r+3j)\binom{r-k}{j}\binom{m}{k}$
and using the identity $j\binom{r-k}{j}=(r+1-j-k)\binom{r-k}{j-1}$
gives
However, $k\binom{m}{k}=(m-k+1)\binom{m}{k-1}$
, and therefore,
Thus
Hence $\frac {R_{j+1}}{C_{j}}=\frac {b_{j}R_{j-1}}{C_{j}}-\frac {a_{j}R_{j}}{C_{j}}$
for $1\leqslant j\leqslant r$
. When $j=r$
, our convention gives $R_{r+1}=0$
. This proves part (b) and completes the proof of part (a).
Remark 2.1 View $m$
as an indeterminant, so that $r!s_m(r)$
is a polynomial in $m$
over ${\mathbb {Z}}$
of degree $r$
. The factorization $r!s_m(r)=\prod _{j=1}^r(a_j+{\mathcal {T}}_{j+1})$
in corollary 1.2 involves the rational functions $a_j+{\mathcal {T}}_{j+1}$
. However, theorem 1.1(b) gives ${\mathcal {T}}_{j+1}=\frac {R_{j+1}}{R_{j}}$
, so that $a_j+{\mathcal {T}}_{j+1}=\frac {a_{j}R_{j}+R_{j+1}}{R_j}=\frac {b_{j}R_{j-1}}{R_j}$
. This determines the numerator and denominator of the rational function $a_j+{\mathcal {T}}_{j+1}$
, and explains why we have $\prod _{j=1}^r(a_j+{\mathcal {T}}_{j+1})= \frac {R_0}{R_r}\prod _{j=1}^rb_{j}=r!s_m(r)$
. This is different from, but reminiscent of, the ratio $p_{j+1}/p_j$
described on p. 26 of [Reference Olds8]. $\diamond$
Corollary 2.2 If $r,\,m\in {\mathbb {Z}}$
and $0< r< m$
, then
If $r=0$
, then $s_m(r)=\frac {(r+1)\binom{m}{r+1}}{m-2r+{\mathcal {T}}_1}$
is true, but ${\mathcal {T}}_1=\mathcal{K}_{i=1}^r\frac {2i(r+1-i)}{m-2r+3i}=0$
.
We will need some additional tools such as proposition 2.3 and corollary 2.4 below in order to prove theorem 1.3.
Since $s_m(m-r)=2^m-s_m(r-1)$
approximating $s_m(r)$
for $0\leqslant r\leqslant m$
reduces to approximating $s_m(r)$
for $0\leqslant r\leqslant \lfloor \frac {m}{2}\rfloor$
. Hence the hypothesis $r<\frac {m+3}{2}$
in proposition 2.3 and corollary 2.4 is not too restrictive. Proposition 2.3 generalizes [Reference Olds8, Theorem 3.3].
Let ${\mathcal {H}}_j:=\mathcal{K}_{i=1}^j\frac {b_i}{a_i}$
denote the $j$
th head of the fraction $\mathcal{K}_{i=1}^r\frac {b_i}{a_i}$
, where ${\mathcal {H}}_0=0$
.
Proposition 2.3 Let $b_i=2i(r+1-i)$
and $a_i=m-2r+3i$
for $0\leqslant i\leqslant r$
. If $r<\frac {m+3}{2}$
, then $a_0+{\mathcal {H}}_r=\frac {(r+1)\binom{m}{r+1}}{s_m(r)}$
can be approximated using the following chain of inequalities
Proof. Note that $r$
equals either $2\lfloor r/2\rfloor$
or $2\lfloor (r-1)/2\rfloor +1$
, depending on its parity.
We showed in the proof of theorem 1.1 that $\frac {(r+1)\binom{m}{r+1}}{s_m(r)}=a_0+{\mathcal {H}}_r=a_0+\mathcal{K}_{i=1}^r\frac {b_i}{a_i}$
. Since $r<\frac {m+3}{2}$
, we have $a_i>0$
and $b_i>0$
for $1\leqslant i\leqslant r$
and hence $\frac {b_i}{a_i}>0$
. A straightforward induction (which we omit) depending on the parity of $r$
proves that ${\mathcal {H}}_0<{\mathcal {H}}_2<\cdots <{\mathcal {H}}_{2\lfloor r/2\rfloor }<{\mathcal {H}}_{2\lfloor (r-1)/2\rfloor +1}<\cdots <{\mathcal {H}}_3<{\mathcal {H}}_1$
. For example, if $r=3$
, then
proves ${\mathcal {H}}_0<{\mathcal {H}}_2<{\mathcal {H}}_3<{\mathcal {H}}_1$
as the tails are positive. Adding $a_0$
proves the claim.
In asking whether $g_{\omega,m}(r)$
is a unimodal function, it is natural to consider the ratio $g_{\omega,m}(r+1)/g_{\omega,m}(r)$
of successive terms. This suggests defining
We will prove in lemma 3.1 that $t(r)$
is a strictly decreasing function that determines when $g_{\omega,m}(r)$
is increasing or decreasing, and $t_{m}(r_0-1)\geqslant \omega >t_{m}(r_0)$
determines $r_0$
.
Corollary 2.4 We have $m-2r\leqslant \frac {(r+1)\binom{m}{r+1}}{s_m(r)}$
for $r\geqslant 0$
, and
Hence $\frac {m+2}{r+1}\leqslant t_m(r)+1$
for $r\geqslant 0$
, and
Also $\frac {m+2}{r+1}< t_m(r)+1$
for $r>0$
, and the above upper bound is strict for $1< r<\frac {m+3}{2}$
.
Proof. We proved $Q=\frac {(r+1)\binom{m}{r}}{s_m(r)}=(m-2r)+\mathcal{K}_{i=1}^r\frac {2i(r+1-i)}{m-2r+3i}$
in theorem 1.1. Hence $m-2r=\frac {(r+1)\binom{m}{r+1}}{s_m(r)}$
if $r=0$
and $m-2r<\frac {(r+1)\binom{m}{r+1}}{s_m(r)}$
if $1\leqslant r<\frac {m+3}{2}$
by proposition 2.3. Clearly $m-2r<0\le \frac {(r+1)\binom{m}{r+1}}{s_m(r)}$
if $\frac {m+3}{2}\le r\le m$
. Similarly $\frac {(r+1)\binom{m}{r+1}}{s_m(r)}= m-2r+\frac {2r}{m-2r+3}$
if $r=0,\,1$
, and again proposition 2.3 shows that ${\frac {(r+1)\binom{m}{r+1}}{s_m(r)}< m-2r+\frac {2r}{m-2r+3}}$
if $1< r<\frac {m+3}{2}$
. The remaining inequalities (and equalities) follow similarly since $t_m(r)+1=2+\frac {\binom{m}{r+1}}{s_m(r)}$
and $2+\frac {m-2r}{r+1}=\frac {m+2}{r+1}$
.
3. Estimating the maximizing input $r_0$
Fix $\omega >1$
. In this section we consider the function $g(r)=g_{\omega,m}(r)$
given by (1.1). As seen in table 1, it is easy to compute $g(r)$
if $r$
is near $0$
or $m$
. For $m$
large and $r$
near $0$
, we have ‘sub-exponential’ growth $g(r)\approx \frac {m^r}{r!\omega ^r}$
. Similarly for $r$
near $m$
, we have exponential decay $g(r)\approx \frac {2^m}{\omega ^r}$
. The middle values require more thought.
Table 1. Values of $g_{w,m}(r)$
On the other hand, the plots $y=g(r)$
, $0\leqslant r\leqslant m$
, exhibit a remarkable visual symmetry when $m$
is large. The relation $s_m(m-r)=2^m-s_m(r-1)$
and the distorting scale factor of $\omega ^{-r}$
shape the plots. The examples in figure 1 show an approximate left–right symmetry about a maximizing input $r\approx \frac {m}{\omega +1}$
. It surprised the authors that in many cases there exists a simple exact formula for the maximizing input (it is usually unique as corollary 3.2 suggests). In figure 1 we have used different scale factors for the $y$
-axes. The maximum value of $g_{\omega,m}(r)$
varies considerably as $\omega$
varies (c.f. lemma 5.1), so we scaled the maxima (rounded to the nearest integer) to the same height.
Lemma 3.1 Recall that $g(r)=\omega ^{-r} s_m(r)$
by (1.1) and $t(r)=\frac {s_m(r+1)}{s_m(r)}$
by (2.3).
(a) $t(r-1)>t(r)>\frac {m-r}{r+1}$
for $0\leqslant r\leqslant m$ where $t(-1):=\infty$;(b) $g(r)< g(r+1)$
if and only if $t(r)>\omega$;(c) $g(r)\leqslant g(r+1)$
if and only if $t(r)\geqslant \omega$;(d) $g(r)> g(r+1)$
if and only if $\omega > t(r)$;(e) $g(r)\geqslant g(r+1)$
if and only if $\omega \geqslant t(r)$;(f) if $\omega >1$
then some $r_0\in \{0,\,\ldots,\,m\}$ satisfies $t(r_0-1)\geqslant \omega > t(r_0)$, and this condition is equivalent to
\[ g(0)<\cdots < g(r_0-1)\leqslant g(r_0) \quad {\rm and}\quad g(r_0)>\cdots >g(m). \]
Proof. (a) We prove, using induction on $r$
, that $t(r-1)>t(r)>\binom{m}{r+1}/\binom{m}{r}$
holds for $0\leqslant r\leqslant m$
. These inequalities are clear for $r=0$
as $\infty >m+1>m$
. For real numbers $\alpha,\,\beta,\,\gamma,\,\delta >0$
, we have $\alpha \delta -\beta \gamma >0$
if and only if $\frac {\alpha }{\beta }>\frac {\alpha +\gamma }{\beta +\delta }>\frac {\gamma }{\delta }$
; that is, the mediant $\frac {\alpha +\gamma }{\beta +\delta }$
of $\frac {\alpha }{\beta }$
and $\frac {\gamma }{\delta }$
lies strictly between $\frac {\alpha }{\beta }$
and $\frac {\gamma }{\delta }$
. If $0< r\leqslant m$
, then by induction
Applying the ‘mediant sum’ to $t(r-1)=\frac {s_m(r)}{s_m(r-1)}>\frac {\binom{m}{r+1}}{\binom{m}{r}}$
gives
Therefore $t(r-1)>t(r)>\binom{m}{r+1}/\binom{m}{r}=\frac {m-r}{r+1}$
completing the induction, and proving (a).
(b,c,d,e) The following are equivalent: $g(r)< g(r+1)$
; $\omega s_m(r)< s_m(r+1)$
; and $\omega < t(r)$
. The other claims are proved similarly by replacing < with $\leqslant$
, >, $\geqslant$
.
(f) Observe that $t(m)=\frac {s_m(m+1)}{s_m(m)}=\frac {2^m}{2^m}=1$
. By part (a), the function $y=t(r)$
is decreasing for $-1\leqslant r\leqslant m$
. Since $\omega >1$
, there exists an integer $r_0\in \{0,\,\ldots,\,m\}$
such that $\infty =t(-1)>\cdots >t(r_0-1)\geqslant \omega > t(r_0)>\cdots > t(m)=1$
. By parts (b,c,d,e) an equivalent condition is $g(0)<\cdots < g(r_0-1)\leqslant g(r_0)$
and $g(r_0)>\cdots >g(m)$
.
The following is an immediate corollary of lemma 3.1(f).
Corollary 3.2 If $t(r_0-1)>\omega$
, then the function $g(r)$
in (1.1) has a unique maximum at $r_0$
. If $t(r_0-1)=\omega$
, then $g(r)$
has two equal maxima, one at $r_0-1$
and one at $r_0$
.
As an application of theorem 1.1 we show that the largest maximizing input $r_0$
for $g_{\omega,m}(r)$
satisfies $\lfloor \frac {m+2}{\omega +1}\rfloor \leqslant r_0$
. There are at most two maximizing inputs by corollary 3.2.
Lemma 3.3 Suppose that $\omega >1$
and $m\in {\mathbb {Z}}$
, $m\geqslant 0$
. If $r':=\lfloor \frac {m+2}{\omega +1}\rfloor$
, then
if $r'>1$
or $\omega \ne m+1$
.
Proof. The result is clear when $r'=0$
. If $r'=1$
, then $r'\leqslant \frac {m+2}{\omega +1}$
gives $\omega \leqslant m+1$
or $g(0)\leqslant g(1)$
. Hence $g(0)< g(1)$
if $\omega \ne m+1$
. Suppose that $r'>1$
. By lemma 3.1(c,f) the chain $g(0)<\dots < g(r')$
is equivalent to $g(r'-1)< g(r')$
, that is $t(r'-1)> \omega$
. However, $t(r'-1)+1>\frac {m+2}{r'}$
by corollary 2.4 and $r'\leqslant \frac {m+2}{\omega +1}$
implies $\frac {m+2}{r'}\geqslant \omega +1$
. Hence $t(r'-1)+1>\omega +1$
, so that $t(r'-1)> \omega$
as desired.
Proof Proof of theorem 1.4
Suppose that $\omega \in \{3,\,4,\,\ldots \}$
. Then $g(0)<\dots < g(r'-1)\leqslant g(r')$
by lemma 3.3 with strictness when $\omega \ne m+1$
. If $\omega =m+1$
, then $r'=\lfloor \frac {m+2}{\omega +1}\rfloor =1$
and $g(0)=g(1)$
as claimed. It remains to show that $g(r')>g(r'+1)>\cdots >g(m)$
. However, we need only prove that $g(r')> g(r'+1)$
by lemma 3.1(f), or equivalently $\omega > t(r')$
by lemma 3.1(d).
Clearly $\omega \geqslant 3$
implies $r'\leqslant \frac {m+2}{\omega +1}\leqslant \frac {m+2}{4}$
. As $0\leqslant r'<\frac {m+3}{2}$
, corollary 2.4 gives
Hence $\omega +1> t(r')+1$
holds if $\omega +1> \frac {m+2}{r'+1}+\frac {2r'}{(r'+1)(m-2r'+3)}$
. Since $\omega +1$
is an integer, we have $m+2=r'(\omega +1)+c$
where $0\leqslant c\leqslant \omega$
. It follows from $0\leqslant r'\leqslant \frac {m+2}{4}$
that $\frac {2r'}{m-2r'+3}<1$
. This inequality and $m+2\leqslant r'(\omega +1)+\omega$
gives
Thus $\omega +1>\frac {m+2}{r'+1}+\frac {2r'}{(r'+1)(m-2r'+3)}\geqslant t(r')+1$
, so $\omega >t(r')$
as required.
Remark 3.4 The proof of theorem 1.4 can be adapted to the case $\omega =2$
. If $m+2=3r'+c$
where $c\leqslant \omega -1=1$
, then $\frac {2r'}{m-2r'+3}=\frac {2r'}{r'+c+1}<2$
, and if $c=\omega =2$
, then a sharper ${\mathcal {H}}_2$
-bound must be used. This leads to a much shorter proof than [Reference Glasby and Paseman4, Theorem 1.1]. $\diamond$
4. Non-integral values of $\omega$
In this section, we prove that the maximum value of $g(r)$
is $g(r')$
or $g(r'+1)$
if $\omega \geqslant \sqrt {3}$
. Before proving this result (theorem 1.3), we shall prove two preliminary lemmas.
Lemma 4.1 Suppose that $\omega >1$
and $r':=\lfloor \frac {m+2}{\omega +1}\rfloor$
. If $\frac {m+2}{r'+1}\geqslant \sqrt {3}+1$
, then
Proof. It suffices, by lemma 3.1(f) and lemma 3.3 to prove that $g(r'+1)>g(r'+2)$
. The strategy is to show $\omega >t(r'+1)$
, that is $\omega +1>t(r'+1)+1$
. However, $\omega +1>\frac {m+2}{r'+1}$
, so it suffices to prove that $\frac {m+2}{r'+1}\geqslant t(r'+1)+1$
. Since $r'+1\leqslant \frac {m+2}{\sqrt {3}+1}<\frac {m+2}{2}$
, we can use corollary 2.4 and just prove that $\frac {m+2}{r'+1}\geqslant \frac {m+2}{r'+2}+\frac {2r'+2}{(r'+2)(m-2r'+1)}$
. This inequality is equivalent to $\frac {m+2}{r'+1}\geqslant \frac {2r'+2}{m-2r'+1}$
. However, $\frac {m+2}{r'+1}\geqslant \sqrt {3}+1$
, so we need only show that $\sqrt {3}+1\geqslant \frac {2(r'+1)}{m-2r'+1}$
, or equivalently $m-2r'+1\geqslant (\sqrt {3}-1)(r'+1)$
. This is true since $\frac {m+2}{r'+1}\geqslant \sqrt {3}+1$
implies $m-2r'+1\geqslant (\sqrt {3}-1)r'+\sqrt {3}>(\sqrt {3}-1)(r'+1)$
.
Remark 4.2 The strict inequality $g(r'-1)< g(r')$
holds by lemma 3.3 if $r'>1$
or $\omega \ne m+1$
. It holds vacuously for $r'=0$
. Hence adding the additional hypothesis that $\omega \ne m+1$
if $r'=1$
to lemma 4.1 (and theorem 1.3), we may conclude that the inequality $g(r'-1)\leqslant g(r')$
is strict.
Remark 4.3 In lemma 4.1, the maximum can occur at $r'+1$
. If $\omega =2.5$
and $m=8$
, then $r'=\lfloor \frac {10}{3.5}\rfloor =2$
and $\frac {m+2}{r'+1}=\frac {10}{3}\geqslant \sqrt {3}+1$
however $g_{2.5,8}(2)=\frac {740}{125}< \frac {744}{125}=g_{2.5,8}(3)$
. $\diamond$
Remark 4.4 The gap between $r'$
and the largest maximizing input $r_0$
can be arbitrarily large if $\omega$
is close to 1. For $\omega >1$
, we have $r'=\lfloor \frac {m+2}{\omega +1}\rfloor <\frac {m+2}{2}$
. If $1<\omega \leqslant \frac {1}{1-2^{-m}}$
, then $g(m-1)\leqslant g(m)$
, so $r_0=m$
. Hence $r_0-r'>\frac {m-2}{2}$
.
Remark 4.5 Since $r'\leqslant \lfloor \frac {m+2}{\omega +1}\rfloor < r'+1$
, we see that $r'+1\approx \frac {m+2}{\omega +1}$
, so that $\frac {m+2}{r'+1}\approx \omega +1$
. Thus lemma 4.1 suggests that if $\omega \gtrsim \sqrt {3}$
, then $g_{\omega,m}(r)$
may have a maximum at $r'$
or $r'+1$
. This heuristic reasoning is made rigorous in theorem 1.3. $\diamond$
Remark 4.6 Theorem 1.1 can be rephrased as $t_m(r)=\frac {s_m(r+1)}{s_m(r)}=\frac {m-r+1}{r+1}+\frac {{\mathcal {K}}_m(r)}{r+1}$
where
The following lemma repeatedly uses the expression $\omega >t_m(r+1)$
. This is equivalent to $\omega >\frac {m-r}{r+2}+\frac {{\mathcal {K}}_m(r+1)}{r+2}$
, that is $(\omega +1)(r+2)>m+2+{\mathcal {K}}_m(r+1)$
. $\diamond$
Lemma 4.7 Let $m\in \{0,\,1,\,\ldots \}$
and $r'=\lfloor \frac {m+2}{\omega +1}\rfloor$
. If any of the following three conditions are met, then $g_{\omega,m}(r'+1)>\cdots >g_{\omega,m}(m)$
holds: (a) $\omega \geqslant 2$
, or (b) $\omega \geqslant \frac {1+\sqrt {97}}{6}$
and $r'\ne 2$
, or (c) $\omega \geqslant \sqrt {3}$
and $r'\not \in \{2,\,3\}$
.
Proof. The conclusion $g_{\omega,m}(r'+1)>\cdots >g_{\omega,m}(m)$
holds trivially if $r'+1\geqslant m$
. Suppose henceforth that $r'+1< m$
. Except for the excluded values of $r',\,\omega$
, we will prove that $g_{\omega,m}(r'+1)>g_{\omega,m}(r'+2)$
holds, as this implies $g_{\omega,m}(r'+1)>\cdots >g_{\omega,m}(m)$
by lemma 3.1(f). Hence we must prove that $\omega >t_m(r'+1)$
by lemma 3.1(d).
Recall that $r'\leqslant \frac {m+2}{\omega +1}< r'+1$
. If $r'=0$
, then $m+2<\omega +1$
, that is $\omega >m+1>t(1)$
as desired. Suppose now that $r'=1$
. There is nothing to prove if $m=r'+1=2$
. Assume that $m>2$
. Since $m+2<2(\omega +1)$
, we have $2< m<2\omega$
. The last line of remark 4.6 and (4.1) give the desired inequality:
In summary, $g_{\omega,m}(r'+1)>\cdots >g_{\omega,m}(m)$
holds for all $\omega >1$
if $r'\in \{0,\,1\}$
.
We next prove $g_{\omega,m}(r'+1)>g_{\omega,m}(r'+2)$
, or equivalently $\omega >t_m(r'+1)$
for $r'$
large enough, depending on $\omega$
. We must prove that $(\omega +1)(r'+2)>m+2+\mathcal{K}_m(r'+1)$
by remark 4.6. Writing $m+2=(\omega +1)(r'+\varepsilon )$
where $0\leqslant \varepsilon <1$
, our goal, therefore, is to show $(\omega +1)(2-\varepsilon )>{\mathcal {K}}_m(r'+1)$
. Using (4.1) gives
where ${\mathcal {T}}>0$
by theorem 1.1 as $r'>0$
. Rewriting the denominator using
our goal $(\omega +1)(2-\varepsilon )>{\mathcal {K}}_m(r'+1)$
becomes
Dividing by $(2-\varepsilon )(r'+1)$
and rearranging gives
This inequality may be written $(\omega ^2-1)+\lambda >\frac {2}{2-\varepsilon }+\mu (1-\varepsilon )$
where $\lambda =\frac {(\omega +1)(1+{\mathcal {T}})}{r'+1}>0$
and $\mu =\frac {(\omega +1)^2}{r'+1}>0$
. We view $f(\varepsilon ):=\frac {2}{2-\varepsilon }+\mu (1-\varepsilon )$
as a function of a real variable $\varepsilon$
where $0\leqslant \varepsilon <1$
. However, $f(\varepsilon )$
is concave as the second derivative $f''(\varepsilon )=\frac {4}{(2-\varepsilon )^3}$
is positive for $0\leqslant \varepsilon <1$
. Hence the maximum value occurs at an end point: either $f(0)=1+\mu$
or $f(1)=2$
. Therefore, it suffices to prove that $(\omega ^2-1)+\lambda >\max \{2,\,1+\mu \}$
.
If $2\geqslant 1+\mu$
, then the desired bound $(\omega ^2-3)+\lambda >0$
holds as $\omega \geqslant \sqrt {3}$
. Suppose now that $2<1+\mu$
. We must show $(\omega ^2-1)+\lambda >1+\mu$
, that is $\omega ^2-2>\mu -\lambda =\frac {(\omega +1)(\omega -{\mathcal {T}})}{r'+1}$
. Since ${\mathcal {T}}>0$
, a stronger inequality (that implies this) is $\omega ^2-2\geqslant \frac {(\omega +1)\omega }{r'+1}$
. The (equivalent) quadratic inequality $r'\omega ^2-\omega -2(r'+1)\geqslant 0$
in $\omega$
is true provided $\omega \geqslant \frac {1+\sqrt {1+8r'(r'+1)}}{2r'}$
. This says $\omega \geqslant 2$
if $r'=2$
, and $\omega \geqslant \frac {1+\sqrt {97}}{6}$
if $r'=3$
. If $r'\geqslant 4$
, we have
The conclusion now follows from the fact that $2>\frac {1+\sqrt {97}}{6} >\sqrt {3}$
.
Proof Proof of theorem 1.3
By lemma 4.1 it suffices to show that $g(r'+1)>g(r'+2)$
holds when $r'+1< m$
and $\omega \geqslant \sqrt {3}$
. By lemma 4.7(a), we can assume that $\sqrt {3}\leqslant \omega <2$
and $r'\in \{2,\,3\}$
. For these choices of $\omega$
and $r'$
, we must show that $\omega >t_m(r'+1)$
by lemma 3.1 for all permissible choices of $m$
. Since $(\omega +1)r'\leqslant m+2<(\omega +1)(r'+1)$
, when $r'=2$
we have $5<2(\sqrt {3}+1)\leqslant m+2<9$
so that $4\leqslant m\leqslant 6$
. However, $t_m(3)$
equals $\tfrac {16}{15},\,\tfrac {31}{26},\,\tfrac {19}{14}$
for these values of $m$
. Thus $\sqrt {3}>t_m(3)$
holds as desired. Similarly, if $r'=3$
, then $8<3(\sqrt {3}+1)\leqslant m+2<12$
so that $7\leqslant m\leqslant 9$
. In this case $t_m(4)$
equals $\tfrac {40}{33},\,\tfrac {219}{163},\,\tfrac {191}{128}$
for these values of $m$
. In each case $\sqrt {3}>t_m(4)$
, so the proof is complete.
Remark 4.8 We place remark 4.4 in context. The conclusion of theorem 1.3 remains true for values of $\omega$
smaller than $\sqrt {3}$
and not ‘too close to 1’ and $m$
is ‘sufficiently large’. Indeed, by adapting the proof of lemma 4.7 we can show there exists a sufficiently large integer $d$
such that $m > d^4$
and $\omega > 1 + \frac {1}{d}$
implies $g(r'+d) > g(r'+d+1)$
. This shows that $r' \leqslant r_0 \leqslant r' +d$
, so $r_0-r'\leqslant d$
. We omit the technical proof of this fact. $\diamond$
Remark 4.9 The sequence, $a_0+{\mathcal {H}}_1,\,\ldots,\,a_0+{\mathcal {H}}_r$
terminates at $\frac {r+1}{s_m(r)}\binom{m}{r+1}$
by theorem 1.1. We will not comment here on how quickly the alternating sequence in proposition 2.3 converges when $r<\frac {m+3}{2}$
. If $r=m$
, then $a_0=-m$
and $\frac {m+1}{s_m(m+1)}\binom{m}{m+1}=0$
, so theorem 1.1 gives the curious identity ${\mathcal {H}}_m=\mathcal{K}_{i=1}^m\frac {2i(m+1-i)}{3i-m}=m$
. If $\omega$
is less than $\sqrt {3}$
and ‘not too close to 1’, then we believe that $r_0$
is approximately $\lfloor \frac {m+2}{\omega +1}+ \frac {2}{\omega ^2 -1}\rfloor$
, c.f. remark 4.8.
5. Estimating the maximum value of $g_{\omega,m}(r)$
In this section we relate the size of the maximum value $g_{\omega,m}(r_0)$
to the size of the binomial coefficient $\binom{m}{r_0}$
. In the case that we know a formula for a maximizing input $r_0$
, we can readily estimate $g_{\omega,m}(r_0)$
using approximations, such as [Reference Stănică10], for binomial coefficients.
Lemma 5.1 The maximum value $g_{\omega,m}(r_0)$
of $g_{\omega,m}(r)$
, $0\leqslant r\leqslant m$
, satisfies
Proof. Since $g(r_0)$
is a maximum value, we have $g(r_0-1)\leqslant g(r_0)$
. This is equivalent to $(\omega -1)s_m(r_0-1)\leqslant \binom{m}{r_0}$
as $s_m(r_0)=s_m(r_0-1)+\binom{m}{r_0}$
. Adding $(\omega -1)\binom{m}{r_0}$
to both sides gives the equivalent inequality $(\omega -1) s_m(r_0)\leqslant \omega \binom{m}{r_0}$
. This proves the upper bound.
Similar reasoning shows that the following are equivalent: (a) $g(r_0)> g(r_0+1)$
; (b) $(\omega -1)s_m(r_0)> \binom{m}{r_0+1}$
; and (c) $g_{\omega,m}(r_0)>\frac {1}{(\omega -1)\omega ^{r_0}}\binom{m}{r_0+1}$
.
In theorem 1.3 the maximizing input $r_0$
satisfies $r_0=r'+d$
where $d\in \{0,\,1\}$
. In such cases when $r_0$
and $d$
are known, we can bound the maximum $g_{\omega,m}(r_0)$
as follows.
Corollary 5.2 Set $r' := \lfloor \frac {m+2}{\omega +1}\rfloor$
and $k := m+2 - (\omega +1)r'$
. Suppose that $r_0 = r'+d$
and $G=\frac {1}{(\omega -1)\omega ^{r_0-1}}\binom{m}{r_0}$
. Then
Proof. By lemma 3.3, $r_0 = r'+d$
where $d\in \{0,\,1,\,\ldots \}$
. Since $r' = \lfloor \frac {m+2}{\omega +1}\rfloor$
, we have $m+2 = (\omega +1)r'+k$
where $0\leqslant k< \omega +1$
. The result follows from lemma 5.1 and $m = (\omega +1)(r_0-d)+k-2$
as $\binom{m}{r_0+1}=\frac {m-r_0}{r_0+1}\binom{m}{r_0}$
and $\frac {m-r_0}{r_0+1}$
equals
The following remark is an application of the Chernoff bound, c.f. [Reference Worsch11, Section 4]. Unlike theorem 1.1, it requires the cumulative distribution function $\Phi (x)$
, which is a non-elementary integral, to approximate $s_m(r)$
. It seems to give better approximations only for values of $r$
near $\frac {m}{2}$
, see remark 5.4.
Remark 5.3 We show how the Berry–Esseen inequality for a sum of binomial random variables can be used to approximate $s_m(r)$
. Let $B_1,\,\ldots,\,B_m$
be independent identically distributed binomial variables with a parameter $p$
where $0< p<1$
, so that $P(B_i=1)=p$
and $P(B_i=0)=q:= 1-p$
. Let $X_i:= B_i-p$
and $X:=\frac {1}{\sqrt {mpq}}(\sum _{i=1}^m X_i)$
. Then
Hence $E(X)=\frac {1}{\sqrt {mpq}}(\sum _{i=1}^m E(X_i))=0$
and $E(X^2)=\frac {1}{mpq}(\sum _{i=1}^m E(X_i^2))=1$
. By [Reference Nagaev and Chebotarev7, Theorem 2] the Berry–Esseen inequality applied to $X$
states that
where the constant $C:= 0.4215$
is close to the lower bound $C_0=\frac {10+\sqrt {3}}{6\sqrt {2\pi }}=0.4097\cdots$
and $\Phi (x)=\frac {1}{\sqrt {2\pi }}\int _{-\infty }^x e^{-t^2/2}\,dt=\frac {1}{2}\left (1+{\rm erf}\left (\frac {x}{\sqrt {2}}\right )\right )$
is the cumulative distribution function for standard normal distribution.
Writing $B=\sum _{i=1}^mB_i$
we have $P(B\leqslant b)=\sum _{i=0}^{\lfloor b\rfloor }\binom{m}{i}p^iq^{m-i}$
for $b\in {\mathbb {R}}$
. Thus $X=\frac {B-mp}{\sqrt {mpq}}$
and $x=\frac {b-mp}{\sqrt {mpq}}$
satisfy
Setting $p=q=\tfrac {1}{2}$
, and taking $b=r\in \{0,\,1,\,\ldots,\,m\}$
shows
Remark 5.4 Let $a_0+{\mathcal {H}}_k$
be the generalized continued fraction approximation to $\frac {(r+1)\binom{m}{r+1}}{s_m(r)}$
suggested by theorem 1.1, where ${\mathcal {H}}_k:=\mathcal{K}_{i=1}^k\frac {b_i}{a_i}$
, and $k$
is the depth of the generalized continued fraction. We compare the following two quantities:
The sign of $e_{m,r,k}$
is governed by the parity of $k$
by proposition 2.3. We shall assume that $r\leqslant \frac {m}{2}$
. As $\frac {2^m}{s_m(r)}$
ranges from $2^m$
to about 2 as $r$
ranges from 0 to $\lfloor \frac {m}{2}\rfloor$
, it is clear that the upper bound for $E_{m,r}$
will be huge unless $r$
satisfies $\frac {m-\varepsilon }{2}\le r\le \frac {m}{2}$
where $\varepsilon$
is ‘small’ compared to $m$
. By contrast, the computer code [Reference Glasby3] verifies that the same is true for $E_{m,r}$
, and shows that $|e_{m,r,k}|$
is small, even when $k$
is tiny, when $0\le r<\frac {m-\varepsilon }{2}$
, see table 2. Hence the ‘generalized continued fraction’ approximation to $s_m(r)$
is complementary to the ‘statistical’ approximation, as shown in table 2. The reader can extend table 2 by running the code [Reference Glasby3] written in the Magma [Reference Bosma, Cannon and Playoust1] language, using the online calculator http://magma.maths.usyd.edu.au/calc/, for example.
Table 2. Upper bounds for $|e_{m,r,k}|$
and $E_{m,r}$
for $m=10^4$
Acknowledgements
SPG received support from the Australian Research Council Discovery Project Grant DP190100450. GRP thanks his family, and SPG thanks his mother.
