1: % On the Number of Distinct Multinomial Coefficients
2: % George E. Andrews, Arnold Knopfmacher and Burkhard Zimmermann.
3:
4: % LaTeX detail: You need elsart.cls and elsart-num.bst to compile.
5:
6: % changes after refereeing:
7:
8: % according to the referees suggestion:
9: % 1. fixed typos detected by the referee.
10: % 2. changed the abstract.
11: % 3. added a paragraph in the introduction.
12:
13: % minor:
14: % 4. Updated B. Zimmermann's grant number.
15: % 5. CoCoA typeset and cited according to http://cocoa.dima.unige.it/citing.html
16:
17: \documentclass{elsart}
18:
19: \usepackage{amsmath}
20: \usepackage{amssymb}
21: \journal{Journal of Number Theory}
22:
23: \newcommand{\Primes}{{\mathbb P}}
24: \newcommand{\rmprime}{\mbox{\rm prime}}
25: % the integer part of a real number.
26: \newcommand{\floor}[1]{{\lfloor #1 \rfloor}}
27: % used for a homomorphism
28: \newcommand{\h}{h}
29:
30: % The CoCoA team prefers to typeset CoCoA as:
31: \def\cocoa{{\hbox{\rm C\kern-.13em o\kern-.07em C\kern-.13em o\kern-.15em A}}}
32:
33: \begin{document}
34: \begin{frontmatter}
35:
36: \title{On the Number of Distinct Multinomial Coefficients}
37:
38: \author{George E. Andrews\thanksref{GEA}\corauthref{cor}}
39: \address{Mathematics Department, 410 McAllister Building,
40: The Pennsylvania State University, University Park, PA 16802,
41: USA.}
42: \ead{andrews@math.psu.edu} \corauth[cor]{Corresponding author.}
43: \thanks[GEA]{Partially supported by
44: National Science Foundation Grant DMS-0200047.}
45:
46: \author{Arnold Knopfmacher\thanksref{AK}}
47: \address{The John Knopfmacher Centre for Applicable Analysis and Number Theory,
48: University of the Witwatersrand, Johannesburg, South Africa.}
49: \ead{arnoldk@cam.wits.ac.za}
50: \thanks[AK]{Partially supported by The John Knopfmacher Center for
51: Applicable Analysis and Number Theory of the University of
52: the Witwatersrand.}
53:
54: \author{Burkhard Zimmermann\thanksref{BZ}}
55: \address{Research Institute for Symbolic Computation, Johannes Kepler Universit\"{a}t Linz, A-4040 Linz,
56: Austria.} \ead{Zimmermann@risc.uni-linz.ac.at}
57: \thanks[BZ]{Supported by SFB grant F1301 of the
58: Austrian FWF.}
59: % Before refereeing, we had:
60: %%We study $M(n)$, the number of distinct values
61: %%taken by multinomial coefficients with upper entry $n$.
62: %%We show that both $p_{\Primes}(n)/M(n)$ and $M(n)/p(n)$ tend to zero as
63: %%$n$ goes to infinity, where $p_{\Primes}(n)$ is the number of
64: %%partitions of $n$ into primes and $p(n)$ is the total number of
65: %%partitions of $n$. We study some closely related sequences as well.
66: % The referee suggested to mention the link to commutative algebra.
67: % The phrase about the "closely related sequences" was moved up.
68: % now it is:
69: \begin{abstract}
70: We study $M(n)$, the number of distinct values
71: taken by multinomial coefficients with upper entry $n$,
72: and some closely related sequences.
73: We show that both $p_{\Primes}(n)/M(n)$ and $M(n)/p(n)$ tend to zero as
74: $n$ goes to infinity, where $p_{\Primes}(n)$ is the number of
75: partitions of $n$ into primes and $p(n)$ is the total number of
76: partitions of $n$. To use methods from commutative algebra,
77: we encode partitions and multinomial coefficients as monomials.
78: \end{abstract}
79:
80: \begin{keyword}
81: %05A10
82: Factorials, binomial coefficients, combinatorial functions,
83: %05A17
84: partitions of integers;
85: %11P81
86: %Elementary theory of partitions;
87: %11P82
88: %Analytic theory of partitions;
89: %11P83
90: %Partitions; congruences and congruential restrictions;
91: %13P10
92: polynomial ideals, Gr\"{o}bner bases.
93: \end{keyword}
94:
95: \end{frontmatter}
96:
97:
98: \section{Introduction}
99: The classical multinomial expansion is given by
100: \begin{equation} \label{eq:expansion}
101: (x_1 + x_2 + \cdots + x_k)^n =
102: \sum \binom{n}{i_1,i_2,\dots,i_k} x_1^{i_1}x_2^{i_2} \cdots x_k^{i_k}\,,
103: \end{equation}
104: where the sum
105: runs over all $(i_1,i_2,\dots,i_k)$ such that
106: $i_1 + i_2 + \cdots + i_k = n$ and $i_1,i_2,\dots,i_k \geq 0$.
107: Multinomial coefficients are defined by
108: \begin{equation} \binom{n}{i_1,i_2,\dots,i_k}
109: := \frac{n!}{i_1! i_2! \cdots i_k!}. \end{equation}
110:
111: It is natural to ask about $M_k(n)$,
112: the number of different values of
113: \begin{equation} \binom{n}{i_1,i_2,\dots,i_k}\end{equation}
114: where $i_1+i_2+\dots+i_k=n$. Obviously if the $i_1,i_2,\dots,i_k$
115: are merely permuted, then the value of $\binom{n}{i_1,i_2,\dots,
116: i_k}$ is unchanged. However identical values do not necessarily
117: arise only by permuting the $i_1,i_2,\dots,i_k$. For example,
118: \begin{equation} \label{eq:3224111}
119: \binom{7}{3,2,2} = \binom{7}{4,1,1,1}\end{equation}
120: and
121: \begin{equation}
122: \binom{236}{64,\,55,\,55,\,52,\,7,\,3}
123: = \binom{236}{62,\,56,\,54,\,51,\,13}.
124: \end{equation}
125:
126: We note that if $k \geq n$, then $M_k(n) = M_n(n)$, and we define
127: $M(n) := M_n(n)$ to be the total number of distinct multinomial
128: coefficients with upper entry~$n$.
129:
130: Since permuting its lower indices leaves the value of a
131: multinomial coefficient unchanged it is immediately clear that
132: \begin{equation} \label{ineq:Mkpk}
133: M_k(n) \leq p_k(n) \end{equation}
134: and
135: \begin{equation}
136: M(n) \leq p(n)\,, \end{equation} where $p_k(n)$ is the number of
137: partitions of $n$ into at most $k$ parts, and $p(n)$ is the total
138: number of partitions of $n$ respectively. Observing that the
139: binomial coefficients $\binom{n}{k,\,n-k}$ are strictly increasing
140: for $0 \leq k \leq \frac{n}{2}$, we deduce that, in fact,
141: \begin{equation} \label{eq:M2p2}
142: M_2(n) = p_2(n).
143: \end{equation}
144: However the inequality (\ref{ineq:Mkpk}) seems to be stronger for
145: large $k$. Indeed (Theorem~\ref{thm:Mp0}),
146: \begin{equation}
147: \lim_{n\rightarrow \infty} \frac{M(n)}{p(n)} = 0.
148: \end{equation}
149: Bounding $M(n)$ from below we will prove
150: (Theorem~\ref{lower-bound-thm}) that
151: \begin{equation}M(n) \geq p_{\Primes}(n)\end{equation}
152: where $p_{\Primes}(n)$
153: is the number of partitions of $n$ into
154: parts belonging to the set of primes
155: $\Primes$. Indeed
156: (Theorem~\ref{thm:pM0}),
157: \begin{equation}\lim_{n\rightarrow \infty}
158: \frac{p_{\Primes}(n)}{M(n)} = 0.\end{equation}
159:
160: It is natural to generalize the problem from $M(n)$ to $M_S(n)$,
161: the number of different multinomial coefficients with upper entry
162: $n$ whose lower entries belong to a given set $S$ of
163: natural numbers.
164: Let
165: \begin{equation}\label{eq:Mgen}
166: \mathcal{M}_S(q) := \sum_{n} M_S(n) q^n
167: \end{equation}
168: and
169: \begin{equation}\label{eq:pgen}
170: \mathcal{P}_S(q) := \sum_{n} p_S(n) q^n
171: \end{equation}
172: where $p_S(n)$ is the number of partitions of $n$ into elements
173: from $S$.
174: Define $[s] := \{1,2,\dots,s\}$. Results of numerical calculations
175: %up to $n=100$
176: such as
177: \begin{equation}\label{eq:45}
178: \mathcal{M}_{[4]}(q) \,/\, \mathcal{P}_{[4]}(q)
179: = 1 - q^7 + O(q^{100})
180: \end{equation}
181: and
182: \begin{equation}\label{eq:67}
183: \mathcal{M}_{[7]}(q) \,/\, \mathcal{P}_{[7]}(q)
184: = 1 - q^7 - q^8 -q^{10} + q^{12} + q^{13} + O(q^{100})
185: \end{equation}
186: suggest that
187: $\mathcal{M}_{S}(q) \,/\, \mathcal{P}_{S}(q)$
188: is a polynomial for any finite $S$.
189: This is indeed true (Theorem \ref{thm:form}) and leads
190: to an algorithm
191: for computing a closed form for the sequence $M_S(n)$
192: for a given finite set $S$ (Section~\ref{sec:finite}).
193:
194: Partitions and multinomial coefficients can be written as monomials in
195: a natural way:
196: For instance, the monomial $q_4 q_1^3$ represents the partition $4+1+1+1$, and
197: $x_7 x_5 x_3 x_2$ represents the multinomial coefficient $\binom{7}{4,1,1,1}$
198: whose factorization into primes is $7 \cdot 5 \cdot 3 \cdot 2$.
199: This encoding serves as a link between our counting problem
200: and Hilbert functions (Section \ref{sec:setting}). Sections \ref{sec:finite},
201: \ref{sec:uppers} and \ref{sec:lowers} are based on that link.
202:
203: We call a pair of partitions of $n$ that yield the same
204: multinomial coefficient but have no common parts an {\em
205: irreducible pair}.
206: For example, the partitions $4+1+1+1$ and
207: $3+2+2$ form an irreducible pair according to
208: Equation~(\ref{eq:3224111}).
209: In Section \ref{sec:in}, we study
210: $i(n)$ the total number of irreducible pairs of partitions of $n$,
211: and we prove (Theorem \ref{thm:n56}) that $i(n) > \frac{n}{56} -
212: 1$.
213:
214: \section{A Lower Bound for $M(n)$} \label{sec:lowerbound}
215: We relate $M(n)$ to $p_{\Primes}(n)$ whose asymptotics is
216: known by a theorem of Kerawala \cite{Ker69}:
217: \begin{equation} \log p_{\Primes}(n) \sim
218: \frac{2 \pi}{\sqrt{3}} \sqrt{\frac{n}{\log n}}.
219: \end{equation}
220: \begin{thm} \label{lower-bound-thm}
221: $M(n) \geq p_{\Primes}(n)$.\end{thm} Theorem~\ref{lower-bound-thm}
222: is implied by the following lemma:
223: \begin{lem}\label{NeverEquivalent}
224: Any two distinct partitions of the same natural number $n$
225: into primes yield different multinomial
226: coefficients.
227: \end{lem}
228: \begin{pf}[Proof of Lemma \ref{NeverEquivalent}]
229: It suffices to show that if
230: \begin{equation}
231: p_1! p_2! \cdots p_r! = q_1! q_2! \cdots q_s! \end{equation}
232: where $p_1 \leq p_2 \leq \cdots \leq p_r$ and $q_1 \leq q_2 \leq \cdots
233: \leq q_s$ are all primes, then $r = s$ and $p_i = q_i$ for $i = 1,
234: \ldots, s$. We proceed by mathematical induction on $r$.
235:
236: If $r = 1$, then $q_s$ must equal $p_1$ because if $q_s < p_1$
237: then $p_1$ divides the left side of the above equation but not the
238: right side. If $q_s > p_1$ then $q_s$ divides the right side but
239: not the left. Hence $q_s = p_1$, and dividing both sides by
240: $p_1!$ we see that there can be no other $q_i$. Hence $s=1$ and
241: $q_1 = p_1$.
242:
243: Assume now that our result holds up to but not including a
244: particular $r$. As in the case $r=1$, we must have $q_s = p_r$.
245: Cancel $p_r!$ from both sides and apply the induction hypothesis
246: to conclude that $s-1 = r-1$ and $p_i = q_i$ for $i = 1, \ldots,
247: s-1$. Hence the lemma follows by mathematical induction.
248: \qed\end{pf}
249: Some values of $p_{\Primes}(n)$ and $M(n)$ are listed on page
250: \pageref{numbers}.
251: We will refine Theorem~\ref{lower-bound-thm} in
252: Section \ref{sec:lowers}.
253:
254: \section{The Algebraic Setting}\label{sec:setting}
255:
256: Encoding partitions and multinomial coefficients as monomials
257: allows us to apply constructive methods from commutative algebra
258: to the problem of counting multinomial coefficients.
259: Let us assume that $S \subseteq \mathbb{N}$ throughout the paper.
260: We will see that $M_S(n)$ finds a natural interpretation as the Hilbert
261: function of a certain graded ring (Lemma~\ref{lem:hilbert}).
262: In the case of finite $S$, it can be computed by
263: the method of Gr\"{o}bner bases \cite{AL94,BB70,BB85,CLO92}.
264:
265: We represent the partition $\lambda_0 + \lambda_1 + \dots + \lambda_i$
266: of $n$ by the monomial
267: $q_{\lambda_0} q_{\lambda_1} \dots q_{\lambda_i}$
268: whose degree is $n$ if we define the degree of variables
269: suitably by $\deg q_j := j$.
270: For convenience, we will use the notions
271: ``partition of $n$'' and ``monomial of degree $n$''
272: interchangeably.
273:
274: Let $k$ be a field of characteristic zero. We abbreviate the ring
275: $k[q_i : i \in S]$ of polynomials in the variables $q_i$ for
276: $i \in S$ over $k$ by $k[S]$. Define the degree of monomials
277: by $\deg q_i := i$, and let $k[S]_n$ denote the subspace of all
278: homogeneous polynomials of degree $n$. In other words, $k[S]_n$ is
279: the $k$-vector space whose basis are the partitions of $n$ into
280: parts $S$. Note that $k[S]$ is graded by $k[S] = \bigoplus_n
281: k[S]_n$. For instance,
282: \begin{equation}
283: k[\{1,\, 3,\dots \}]_4 \,=\,
284: k \cdot q_3 q_1 \oplus\, k \cdot {q_1}^4
285: \end{equation}
286: corresponding to the partitions
287: $3+1$ and $1+1+1+1$
288: of $4$ into odd parts.
289:
290: The multinomial coefficients with upper entry $n$ into parts
291: belonging to $S$ are the numbers $n! / \prod_j j!^{a_j}$ where
292: $\prod_j q_j^{a_j}$ ranges over the monomials in $k[S]_n$. Since
293: the numerator $n!$ of these fractions is fixed, it suffices to
294: count the set of all denominators:
295: \begin{equation} \label{eq:prodj} M_S(n) = |\{
296: \prod_j j!^{a_j} \,:\, \prod_j q_j^{a_j} \in k[S]_n \}|.
297: \end{equation}
298: To count the values taken by $\prod_j j!^{a_j}$, we look at
299: their factorization into primes. Let $\h(q_j)$ be the
300: factorization of $j!$ into primes, written as a monomial in $k[x]
301: := k[x_p\,:\,p \; \text{prime}]$, multiplied by $q^j$. For
302: example, $ \h(q_5) = q^5 x_2^3 \, x_3 \, x_5$ corresponding to $5!
303: = 2^3 \cdot 3 \cdot 5$. An elementary counting argument
304: \cite{GKP94} shows that the prime $p$ occurs in the factorization
305: of $j!$ with exponent $\sum_{l=1}^\infty \floor{j/p^l}$, where
306: $\floor{x}$ denotes the largest integer that does not exceed
307: the real number $x$. Therefore,
308: \begin{equation}\label{eq:phidef} \h(q_j) =
309: q^j \prod_{p \; \text{prime}} {x_p}^{\sum_{l=1}^\infty \floor{j/p^l}}.\end{equation}
310: Since factorization into primes is unique, (\ref{eq:prodj})
311: can be written as
312: \begin{equation}\label{eq:MSphi} M_S(n) = |\{
313: \prod_j \h( q_j )^{a_j} \,:\, \prod_j q_j^{a_j} \in k[S]_n \}|.
314: \end{equation}
315: Extending $\h$ to a $k$-algebra homomorphism $k[S] \rightarrow
316: k[x,q]$ allows us to reformulate (\ref{eq:MSphi}) as
317: \begin{lem}\label{lem:dim}
318: \begin{equation} \label{eq:dimphi}
319: M_S(n) = \dim_k \h(k[S]_n). \end{equation}
320: \end{lem}
321:
322: {\em Example:} Since there are $10$ partitions of $7$ into parts $1,2,3$
323: and $4$, the dimension of
324: \begin{equation}k[\{1,2,3,4\}]_7 =
325: k \, q_4 q_3
326: \oplus k \, q_4 q_2 q_1
327: \oplus k \, q_4 q_1^3
328: \oplus k \, q_3 q_2^2
329: \oplus \dots
330: \oplus k \, q_1^7\end{equation}
331: over $k$ is $10$.
332: However, the dimension of its image
333: \begin{equation}
334: \h( k[\{1,2,3,4\}]_7 ) =
335: k \, q^7 x_2^4 x_3^2
336: \oplus k \, q^7 x_2^4 x_3
337: \oplus k \, q^7 x_2^3 x_3
338: \oplus \dots
339: \oplus k \, q^7
340: \end{equation}
341: under $\h$ is only $9$ and so $M_{[4]}(7)=9$. The defect is due to
342: $\h(q_4 q_1^3) = \h(q_3 q_2^2)$ which is nothing but a restatement
343: of (\ref{eq:3224111}).
344:
345: To use Lemma~\ref{lem:dim} for effective computation (in the case
346: of finite $S$), we express $\dim_k \h(k[S]_n)$ as the value (at
347: $n$) of the Hilbert function of a certain elimination ideal. This
348: method is taken from \cite{AL94}; the result in our case is
349: Lemma~\ref{lem:hilbert} below.
350:
351: First we make the map $\h$ degree-preserving (graded) by defining
352: $\deg q := 1$ and $\deg x_p := 0$ in the ring $k[x_p \,:\, p \;
353: \text{prime}][q]$. (This is why we introduced the extra factor of
354: $q^j$ in the defining equation (\ref{eq:phidef}) of $\h$.) Second,
355: note that
356: \begin{equation}
357: \h(k[S]_n) \cong k[S]_n / (k[S]_n \cap \,\ker \h)\end{equation} as
358: $k$-vector spaces, since $\h$ is a $k$-linear map on $k[S]_n$. In
359: particular, dimensions agree. Therefore,
360: \begin{equation} \label{eq:dimquot}
361: M_S(n) = \dim_k k[S]_n / (k[S]_n \cap \,\ker \h).
362: \end{equation}
363: Recall that the (projective) Hilbert function $H_R$ of a graded
364: $k$-algebra $R = \bigoplus_n R_n$ is defined by $H_R(n) := \dim_k
365: R_n$. Thus (\ref{eq:dimquot}) relates $M_S$ to the Hilbert
366: function of $k[S] / \ker \h$:
367: \begin{equation} \label{eq:hilbertker}
368: M_S(n) = H_{k[S] / \ker \h}(n).
369: \end{equation}
370: By Theorem 2.4.2 of \cite{AL94}, $\ker \h$ can be computed by
371: elimination:
372: \begin{equation} \ker \h = I \cap k[S]\end{equation}
373: where the ideal $I$ of $k[S][q][x_p \,:\, p \,\text{prime}]$
374: is defined by
375: \begin{equation} I :=
376: \langle q_j - \h(q_j) \,:\, j\in S \rangle.\end{equation}
377: Summarizing this section, we have proved the following Lemma:
378: \begin{lem}\label{lem:hilbert}
379: Let $k[S]$ be graded by $\deg q_i := i$. Define a $k$-algebra
380: homomorphism from $k[S]$ to $k[q,x]$ by
381: \begin{equation}\label{eq:phidef1} \h(q_j) :=
382: q^j \prod_p {x_p}^{\sum_{l=1}^\infty
383: \floor{j/p^l}}.\end{equation} Let the ideal $I$ of $k[S,q,x]$
384: be defined by
385: \begin{equation} I :=
386: \langle q_j - \h(q_j) \,:\, j\in S \rangle\end{equation} and let
387: \begin{equation} J := I \cap k[S].\end{equation}
388: Then $M_S$ is the (projective) Hilbert function of the
389: $k$-algebra $k[S] / J$:
390: \begin{equation} \label{eq:hilbert}
391: M_S(n) = H_{k[S] / J}(n).
392: \end{equation}
393: \end{lem}
394: {\em Example:}
395: If $S=[4]$, then
396: $I = \langle
397: q_1 - q, \,
398: q_2 - q^2 x_2, \,
399: q_3 - q^3 x_2 x_3, \,
400: q_4 - q^4 x_2^3 x_3 \rangle$
401: and $J = \langle q_4 q_1^3 - q_3 q_2^2 \rangle$.
402: For $M_{[4]}(n)$, see (\ref{eq:M4n}) on page~\pageref{eq:M4n}.
403:
404:
405: \section{Explicit Answers}\label{sec:finite}
406:
407: Let $S$ be a given finite set throughout this section.
408: Lemma~\ref{lem:hilbert} allows to compute a closed form for the
409: sequence $M_S(n)$ by well-known methods from computational
410: commutative algebra. For the sake of completeness, let us briefly
411: review them:
412:
413: \begin{enumerate}
414: \item Fix a term order $\preceq$ on $k[S,q,x]$ that
415: allows the elimination of the variable $q$ and the variables
416: $x_p$ in step 2 below. Compute a Gr\"{o}bner basis $F$ for the
417: (toric) ideal $I = \langle q_j - \h(q_j) \,:\, j\in S
418: \rangle$ with respect to this term order using Buchberger's
419: algorithm \cite{BB70,BB85}. \item Let $G := F \cap k[S]$. By
420: the elimination property of Gr\"{o}bner bases with respect to a
421: suitable elimination order $\preceq$, the set $G$ is a Gr\"{o}bner
422: basis for the elimination ideal $ J = I \cap k[S]$. \item Let $L
423: := I_\preceq (G)$ be the set of leading terms of
424: polynomials in $G$. \item Compute $\mathcal{M}_S(q)$ using
425: \begin{equation}
426: \mathcal{M}_S(q)
427: = \mathcal{H}_{k[S] / J}(q)
428: = \mathcal{H}_{k[S] / I_\preceq (J)}(q)
429: = \mathcal{H}_{k[S] / \langle L \rangle}(q). \end{equation}
430: The first equality holds by Lemma~\ref{lem:hilbert}. The second
431: equality is an identity of Macaulay~\cite{Mac27}.
432: Since $G$ is a Gr\"{o}bner basis, its initial terms $L$ generate
433: the initial term ideal of $\langle G \rangle$ with respect to $\preceq$,
434: which explains the third equation sign.
435: A naive method
436: for computing the Hilbert-Poincar\'{e} series of
437: $k[S] /\langle L \rangle$
438: is to apply the inclusion-exclusion relation
439: \begin{equation}\label{eq:HtL}
440: \mathcal{H}_{k[S] / \langle \{t\} \cup L \rangle}(q)
441: =
442: \mathcal{H}_{k[S] / \langle L \rangle}(q)
443: - q^{\deg t} \mathcal{H}_{k[S] / \langle L \rangle : t}(q),
444: \end{equation}
445: recursively until the base case
446: \begin{equation}\label{eq:Hbase}
447: \mathcal{H}_{k[S] / \langle \rangle}(q)
448: = \mathcal{H}_{k[S]}(q) = \frac{1}{\prod_{j \in S} (1 - q_j)}
449: \end{equation}
450: is reached. For better (faster) algorithms, see~\cite{B97}.
451: \item Extract a closed form expression for
452: $H_{k[S] / \langle L \rangle}(n)$ from its
453: generating function $ \mathcal{H}_{k[S] / \langle L \rangle}(q)$.
454: (Use partial fraction decomposition and the binomial series).
455: It is the desired answer $M_S(n)$.
456: \end{enumerate}
457: One of the authors computed 1 -- 4 for several finite $S$ using
458: different computer algebra systems. It turned out that \cocoa
459: \cite{CocoaSystem} was fastest for that purpose.
460: \begin{thm}\label{thm:form}
461: Let $S$ be a finite subset of the positive natural numbers.
462: Then
463: \begin{enumerate}
464: \item $\mathcal{M}_S(q)$ can be written as
465: \begin{equation}\label{eq:form}\mathcal{M}_S(q)
466: = \frac{f_S(q)}{\prod_{j\in S} (1-q^j)}\end{equation}
467: where $f_S(q)$ is a polynomial with integer coefficients.
468: \item
469: There exists $n_0$ such that
470: $M_S(n)$ can be written as a quasipolynomial~\cite{St}
471: for $n \geq n_0$.
472: Moreover, it suffices to use periods which are divisors of elements of $S$.
473: \end{enumerate}
474: \end{thm}
475: \begin{pf}
476: Relations (\ref{eq:HtL}) and (\ref{eq:Hbase}) prove the first statement.
477: The second statement follows from the first easily.
478: \qed\end{pf}
479:
480: Let us follow the algorithm in the case $S = [4]$, which
481: is the simplest nontrivial case. We have $ I = \langle q_1 - q,\,
482: q_2 - q^2 x_2,\, q_3 - q^3 x_2 x_3,\, q_4 - q^4 x_2^3 x_3
483: \rangle$. To
484: eliminate the variables $x_3$, $x_2$ and $q$ we
485: choose the lexical term order where $x_3 \succ x_2 \succ q \succ
486: q_4 \succ q_3 \succ q_2 \succ q_1$. The corresponding reduced
487: Gr\"{o}bner basis of $I$ is $F = \{ q_1^3 q_4 - q_2^2 q_3,\, q -
488: q_1,\, q_1^2 x_2 - q_2,\, q_2 q_3 x_2 - q_1 q_4 ,\, q_1 q_3
489: x_2^2-q_4,\, q_1 q_2 x_3-q_3,\, q_2^2 x_3- q_1 q_3 x_2,\, q_1^2
490: q_4 x_3- q_3^2 x_2,\, q_2 q_4 x_3- q_3^2 x_2^2,\, q_1 q_4^2 x_3-
491: q_3^3 x_2^3,\, q_4^3 x_3- q_3^4 x_2^5 \}$.
492: By the elimination
493: property of Gr\"{o}bner bases $G := F \cap k[q_1,
494: q_2, q_3, q_4] = \{ q_1^3 q_4 - q_2^2 q_3 \}$ is
495: a Gr\"{o}bner basis for the elimination ideal $J = I \cap k[q_1,
496: q_2, q_3, q_4]$.
497: Collecting leading terms of $G$
498: gives $L = \{ q_1^3 q_4 \}$.
499: Since $G$ is a Gr\"obner basis of $J$ we know that $I_\preceq(J) =
500: \langle q_1^3 q_4 \rangle$. The Hilbert-Poincar\'{e} series
501: of $k[q_1, q_2, q_3, q_4] / \langle q_1^3 q_4 \rangle$ gives
502: \begin{equation}\label{eq:4}
503: \mathcal{M}_{[4]}(q) = \frac{1 - q^7}{(1-q) (1-q^2) (1-q^3)
504: (1-q^4)}.\end{equation}
505: It is clear that we may replace
506: any occurrence of the partition $4+1+1+1$ in
507: a multinomial coefficient
508: by $3+2+2$ without changing
509: the value of the multinomial coefficient.
510: Therefore, there are at most
511: as many multinomial coefficients
512: as there are partitions
513: avoiding $4 + 1 + 1 + 1$.
514: Equation~(\ref{eq:4}) states that
515: this upper bound gives in fact the exact number in the case
516: of $S=\{1,2,3,4\}$.
517:
518: Note that all denominators in the partial fraction
519: decomposition
520: \begin{multline}\label{eq:M4pfd}
521: \mathcal{M}_{[4]}(q) =
522: % edited version:
523: -\frac{7}{24} \frac{1}{(q-1)^3}
524: - \frac{77}{288} \frac{1}{(q-1)}
525: + \frac{1}{16} \frac{1}{(q+1)^2} + \\
526: + \frac{1}{32} \frac{1}{(q+1)}
527: + \frac{1}{9} \frac{(q+2)}{(q^2 + q + 1)}
528: + \frac{1}{8} \frac{(q + 1)}{(q^2 + 1)}
529: % unedited Mathematica output:
530: %\frac{-7}{24\,{\left( -1 + q \right) }^3} - \frac{77}{288\,\left( -1 + q \right) } +
531: % \frac{1}{16\,{\left( 1 + q \right) }^2} + \frac{1}{32\,\left( 1 + q \right) } +
532: % \\ +
533: % \frac{1 + q}{8\,\left( 1 + q^2 \right) } + \frac{2 + q}{9\,\left( 1 + q + q^2 \right) }
534: \end{multline}
535: of (\ref{eq:4}) are powers of cyclotomic polynomials
536: $C_j(q)$ where $j$ divides an element of $S = \{1,2,3,4\}$.
537: We rewrite this as
538: \begin{multline}\label{eq:M4pfderweitert}
539: \mathcal{M}_{[4]}(q) =
540: \frac{7}{24} \frac{1}{(1 - q)^3}
541: + \frac{77}{288} \frac{1}{(1 - q)}
542: + \frac{1}{16} \frac{(1-q)^2}{( 1 - q^2 )^2} + \\
543: + \frac{1}{32} \frac{(1-q)}{(1 - q^2)}
544: + \frac{1}{9} \frac{(2 - q - q^2)}{(1 - q^3)}
545: + \frac{1}{8} \frac{(1 + q - q^2 - q^3)}{(1 - q^4)}.
546: \end{multline}
547: in order to use the binomial series
548: $(1-z)^{-a-1} = \sum_{n=0}^\infty \binom{a+n}{a} z^n$.
549: The result is
550: \begin{multline}\label{eq:M4n}
551: M_{[4]}(n) =
552: \frac{7}{48} n^2
553: + \left(\frac{1}{16} [1,-1](n) + \frac{7}{16} \right) n + \\
554: + \frac{1}{8} [1,1,-1,-1](n)
555: + \frac{1}{9} [2,-1,-1](n)
556: + \frac{3}{32}[1,-1](n)
557: + \frac{161}{288}
558: \end{multline}
559: where $[a_0, a_1, \dots, a_m](n) := a_j$ for $n \equiv j (m)$.
560: Similar computations show that
561: \begin{equation}\label{eq:5}
562: \mathcal{M}_{[5]}(q) =
563: \frac{1 - q^7}{(1-q) (1-q^2) \dots (1-q^5)},\end{equation}
564: \begin{equation}\label{eq:6}
565: \mathcal{M}_{[6]}(q) = \frac{1 - q^7 - q^8 - q^{10} +
566: q^{12} + q^{13}} {(1-q) (1-q^2) \dots
567: (1-q^6)},\end{equation} and
568: \begin{equation}\label{eq:7}
569: \mathcal{M}_{[7]}(q) = \frac{1 - q^7 - q^8 - q^{10} + q^{12} +
570: q^{13}} {(1-q) (1-q^2) \dots
571: (1-q^7)}.\end{equation}
572: It is no coincidence that the numerators of (\ref{eq:6}) and (\ref{eq:7})
573: agree (Theorem~\ref{thm:SuP'}).
574:
575:
576: \section{Upper Bounds}
577: \label{sec:uppers} Trivially, $M(n) \leq p(n)$. Our goal is to
578: find sharper upper bounds.
579: \begin{lem}\label{lem:upper}
580: Assume $S'\subseteq S$.
581:
582: Let $\tilde{I}$ be the ideal of $k[S,q,x]$ generated by the set of
583: polynomials $\{ q_j - \h(q_j) \,:\, j\in S'\}$. Let $\tilde{J}$ be
584: the ideal generated by $\tilde{I} \cap k[S']$ in the ring $k[S]$.
585: Let $U_{S,\,S'}(n) := H_{k[S] / \tilde{J}}(n).$ Then
586: \begin{enumerate}
587: \item
588: $M_S(n) \leq U_{S,\,S'}(n)$.
589: \item
590: We have
591: \begin{equation}\label{eq:h1}
592: \sum_n U_{S,\,S'}(n) q^n
593: = \frac{f_{S'}(q)}{\prod_{j\in S} (1-q^j)}
594: \end{equation}
595: where $f_{S'}(q)$ is defined by
596: \begin{equation}\label{eq:h2}
597: \sum_n M_{S'}(n) q^n
598: = \frac{f_{S'}(q)}{\prod_{j\in S'} (1-q^j)}.
599: \end{equation}
600: \end{enumerate}
601: \end{lem}
602: \begin{pf}
603: We prove the first statement. Let $I$ be the ideal of $k[S,q,x]$
604: generated by the set of polynomials $\{ q_j - \h(q_j) \,:\, j\in
605: S\}$ and let $J = I \cap k[S]$. Since $\tilde{J}$ is a $k$-vector
606: subspace of $J$ we have
607: \begin{equation}
608: \dim_k k[S]_n \cap J \geq \dim_k k[S]_n \cap \tilde{J} \end{equation}
609: and therefore
610: \begin{equation} \dim_k (k[S] / J)_n
611: \leq \dim_k (k[S] /\tilde{J})_n\end{equation}
612: i.e.
613: \begin{equation} M_S(n) \leq U_{S,\,S'}(n).\end{equation}
614:
615: To prove the second statement, let $I'$ be the ideal generated by
616: $\{q_j - \h(q_j) \,:\, j\in S'\}$ in the ring $k[S',q,x]$ and let
617: $J' := I' \cap k[S']$. Since the ideals $\tilde{J}$ and $J'$ are
618: generated by the same set of polynomials (albeit in different
619: rings), the Hilbert functions $U_{S,S'}(n) = H_{k[S] /
620: \tilde{J}}(n)$ and $M_S'(n) = H_{k[S']/ J'}(n)$ correspond in the
621: way claimed by (\ref{eq:h1}) and (\ref{eq:h2}). \qed\end{pf} To
622: get upper bounds for $M(n)$, we use the preceding Lemma in the
623: special case $S=\mathbb{N}$ getting:
624: \begin{thm}\label{thm:Mleq}\label{thm:Mpp7}
625: For any $S'$ we have
626: \begin{equation}
627: M(n) \leq [q^n] \frac{f_{S'}(q)}{\prod_{j=1}^{\infty} (1-q^j)}
628: \end{equation}
629: (where $[q^n]\mathcal{A}(q)$ denotes the the coefficient of $q^n$ in the
630: power series expansion of $\mathcal{A}(q)$).
631: For instance, the cases $S'=[4]$ and $S'=[6]$ yield the bounds
632: \begin{equation}
633: M(n) \leq p(n) - p(n-7),
634: \end{equation}
635: and
636: \begin{equation}
637: M(n) \leq p(n) - p(n-7) - p(n-8) - p(n-10) + p(n-12) + p(n-13).
638: \end{equation}
639: \end{thm}
640: Note that a direct proof of $M(n) \leq p(n) - p(n-7)$ could
641: be given by exploiting the equivalence of the partitions $4+1+1+1$
642: and $3+2+2$ in the sense of Equation~(\ref{eq:3224111}).
643:
644: The bound $M(n) \leq p(n) - p(n-7)$ is good enough to imply:
645: \begin{thm}\label{thm:Mp0}
646: $M(n)=o(p(n))$, i.e. $\lim_{n\rightarrow\infty} M(n)/p(n) = 0$.
647: \end{thm}
648:
649: \begin{pf}
650: Due to the monotonicity of $p(n)$ and the fact that the unit
651: circle is the natural boundary for
652: \begin{equation}
653: \sum_{n=0}^{\infty} p(n) q^n
654: = \prod_{n=1}^{\infty} \frac{1}{1 - q^n}\;, \end{equation}
655: we see that
656: \begin{equation}
657: \lim_{n\rightarrow \infty} \frac{p(n - 7)}{p(n)} = 1\,.
658: \end{equation} Hence
659: \begin{equation}
660: 0 \leq \lim_{n\rightarrow \infty} \frac{M(n)}{p(n)} \leq
661: \lim_{n\rightarrow \infty} \frac{p(n) - p(n - 7)}{p(n)}
662: = 1 - 1 = 0\,,
663: \end{equation}
664: which proves Theorem \ref{thm:Mp0}.
665: \qed\end{pf}
666:
667:
668:
669:
670: \section{Lower Bounds} \label{sec:lowers}
671: Recall that $M(n) \geq p_{\Primes}(n)$
672: (Theorem~\ref{lower-bound-thm}).
673: The numbers given on page \pageref{numbers} suggest
674: that $M(n)$ grows much faster than $p_{\Primes}(n)$.
675: We will prove that this is indeed the case:
676: $\lim_{n\rightarrow\infty} p_{\Primes}(n)/M(n) = 0$
677: (Theorem \ref{thm:pM0})
678: and we will give better lower bounds for $M(n)$.
679:
680: Let us write $S < P$ if each element of $S$ is less than each element
681: of $P$.
682: We need the following generalization of
683: Lemma~\ref{NeverEquivalent}:
684: \begin{lem}\label{lem:pp}
685: Assume $S < P$ where $P$ is a set of primes. Let $s$ and $s'$ be
686: any two power products in $k[S]$ and let $p$ and $p'$ be distinct
687: power products in $k[P]$. Then $\h(s p) \neq \h(s' p')$.
688: \end{lem}
689:
690: In the case $S=\emptyset$, Lemma~\ref{lem:pp} states that distinct
691: partitions $p$ and $p'$ into primes yield different multinomial
692: coefficients: $\h(p) \neq \h(p')$. Lemma~\ref{lem:pp} can be
693: proved by the same induction argument as
694: Lemma~\ref{NeverEquivalent}.
695:
696: \begin{lem}\label{lem:SuP}
697: Assume $S < P$ where $P$ is a set of primes. Define $\h$ on $k[S
698: \cup P]$ by (\ref{eq:phidef}). Then $\ker \h$ is generated, as an
699: ideal of $k[S \cup P]$, by $\ker \h \, \cap \, k[S]$.
700: \end{lem}
701: \begin{pf}
702: Let $f \in \ker \h$. Since $k[S \cup P] = k[S][P]$, we can $f$ as
703: a finite sum $f = \sum_s \sum_p c_{s,p}\, s p$ indexed by power
704: products $s$ and $p$ from $k[S]$ and $k[P]$ respectively, with
705: coefficients $c_{s,p} \in k$. As $f \in \ker \h$, $\sum_s \sum_p
706: c_{s,p}\, \h(s p) = 0.$ By Lemma~\ref{lem:pp}, this implies
707: $\sum_s c_{s,p}\, \h(s p) = 0 $ for arbitrary but fixed $p$.
708: Cancelling $\h(p)$ from this equation shows that $\h(f_p)=0$ where
709: $f_p := \sum_s c_{s,p}\, s$. In this way we succeed in writing $f$
710: as $f = \sum_p f_p p$ where each $f_p$ is in $\ker \h \cap k[S]$.
711: \qed\end{pf} As an immediate consequence of Lemma~\ref{lem:SuP} we
712: get:
713: \begin{thm}\label{thm:SuP'}
714: Assume $S < P$ where $P$ is a set of primes.
715: Then
716: \begin{equation}
717: \mathcal{M}_{S \cup P}(q) = \mathcal{M}_{S}(q) / \prod_{j\in P} (1-q^j).
718: \end{equation}
719: \end{thm}
720: As a first application of Theorem~\ref{thm:SuP'}, we
721: count multinomial coefficients with lower entries which
722: are either prime or equal to $1$:
723: \begin{equation}\label{eq:M1P}
724: \mathcal{M}_{\{1\} \cup \Primes}(q) =
725: \frac{1}{(1-q) \prod_{j \in \Primes}(1-q^j) },
726: \end{equation}
727: which
728: allows for improving
729: Theorem~\ref{lower-bound-thm}:
730: \begin{thm}\label{thm:pM0}
731: We have
732: \begin{equation}
733: \lim_{n\rightarrow\infty}
734: p_{\Primes}(n) / M_{\{1\} \cup \Primes}(n) = 0
735: \end{equation}
736: and therefore
737: $\lim_{n\rightarrow\infty} p_{\Primes}(n)/M(n) = 0.$
738: \end{thm}
739: \begin{pf}
740: Let $A(n) := M_{\{1\} \cup \Primes}(n)$.
741: Due to the monotonicity of $A(n)$
742: and the fact that the unit
743: circle is the natural boundary for we see that
744: \begin{equation}\lim_{n\rightarrow \infty} A(n-1) / A(n)
745: = 1.\end{equation}
746: By (\ref{eq:M1P}),
747: \begin{equation}
748: p_{\Primes} (n) =
749: A(n) - A(n-1).
750: \end{equation}
751: Therefore,
752: \begin{equation}
753: 0 \leq \lim_{n\rightarrow \infty}
754: \frac{p_\Primes(n)}{A(n)} \leq
755: \lim_{n\rightarrow \infty} \frac{A(n) - A(n-1)}{A(n)}
756: = 1 - 1 = 0\,,
757: \end{equation}
758: which proves Theorem \ref{thm:pM0}.
759: \qed\end{pf}
760:
761: Let $L_S(n) := M_{S \cup \Primes}(n)$; clearly, $L_S(n)$ is a
762: lower bound for $M(n)$.
763: Theorem~\ref{thm:SuP'} allows us deduce
764: \begin{equation}\label{eq:L45}
765: \mathcal{L}_{[4]}(q) = \mathcal{L}_{[5]}(q) = \frac{1 - q^7}
766: {
767: \prod_{j \in [4] \cup \Primes} (1-q^j)
768: }
769: \end{equation}
770: and
771: \begin{equation}\label{eq:L67}
772: \mathcal{L}_{[6]}(q) = \mathcal{L}_{[7]}(q) =
773: \frac{1 - q^7 - q^8 - q^{10} +
774: q^{12} + q^{13}}
775: {
776: \prod_{j \in [6] \cup \Primes} (1-q^j)
777: }\end{equation}
778: from the Equations (\ref{eq:4}) -- (\ref{eq:7}); some values
779: of $L_{[4]}(n)$ are listed on page \pageref{numbers}.
780:
781:
782: \section{The Irreducible Pairs} \label{sec:in}
783:
784: An {\em irreducible pair} is a pair of partitions of $n$ that
785: yield the same multinomial coefficient but have no parts in
786: common. For example,
787: \begin{equation}\label{eq:pair4111}
788: (4,1,1,1)
789: \text{ and }
790: (3,2,2)
791: \end{equation}
792: is an irreducible pair.
793:
794: It turns out that there are infinitely many irreducible
795: pairs of partitions. The following is a partial list:
796: Generalizing (\ref{eq:pair4111}) we see that
797: \begin{equation}
798: (2^m,\underbrace{1,1,\dots,1}_{2m-1}\,) \text{ and }
799: (2^m - 1, \underbrace{2,2,\dots,2}_{m}\,)
800: \end{equation} form an irreducible pair of partitions of $2^m + 2m - 1$.
801: More generally, for any integers $a \geq 2$ and $m \geq 1$
802: the partitions
803: \begin{equation}
804: \label{eq:pn}
805: (a^m, \underbrace{a-1,a-1,\dots,a-1}_{m},
806: \underbrace{1,1,\dots,1}_{m-1}\,) \text{ and }
807: (a^m - 1, \underbrace{a,a,\dots,a}_m \,)
808: \end{equation} form an irreducible pair of partitions of $a^m + am - 1$.
809:
810: The pair
811: \begin{equation} \label{eq:pair611}
812: (6,1,1) \text{ and } (5,3)
813: \end{equation}
814: can be generalized to irreducible pairs
815: \begin{equation}\label{eq:jfactorial}
816: (j!,\, \underbrace{1,\dots,1}_{(j-1)}\,)
817: \text{ and }
818: (j!-1,\,j)
819: \end{equation}
820: of partitions of $(j! + j - 1)$ for $j \geq 3$.
821:
822: From any two irreducible pairs we can get a third one
823: by combining them in a natural way.
824: For instance, combining $a$ copies of (\ref{eq:pair611})
825: with $b$ copies of (\ref{eq:pair4111}) gives
826: the pair (\ref{eq:ab}) which is used in the proof below.
827:
828: The above examples show that $i(n)$ is positive infinitely often.
829: Indeed we have:
830: \begin{thm} \label{thm:n56}
831: $i(n) \geq \frac{n}{56} - 1$.
832: \end{thm}
833: \begin{pf} For each pair of non-negative integers $a$ and $b$
834: satisfying \begin{equation} \label{eq:37}
835: 8a + 7b = n\,,
836: \end{equation} we see that \begin{equation} \label{eq:ab}
837: (\underbrace{6,\dots,6}_a,\underbrace{4,\dots,4}_b,
838: \underbrace{1,\dots,1}_{2a+3b}) \text{ and }
839: (\underbrace{5,\dots,5}_a,\underbrace{3,\dots,3}_{a+b},
840: \underbrace{2,\dots,2}_{2b})
841: \end{equation} forms a new irreducible pair of partitions of $n$. Consequently
842: $i(n)$ is at least as large as the number of non-negative
843: solutions of the linear Diophantine equation (\ref{eq:37}).
844:
845: Now the segment of the line $8a + 7b = n$ in the first quadrant is
846: of length $n\sqrt{113}\big/56$. Furthermore from the full
847: solution of the linear Diophantine equation we note that the
848: integral solutions of (3.7) are points on this spaced a distance
849: $\sqrt{113}$ apart. Hence in the first quadrant there must be at
850: least \begin{equation}
851: \left\lfloor \frac{n\sqrt{113}/56}{\sqrt{113}}\right\rfloor
852: = \left\lfloor \frac{n}{56} \right\rfloor > \frac{n}{56} - 1
853: \end{equation} such points. Therefore \begin{equation}
854: i(n) > \frac{n}{56} - 1\,.
855: \end{equation}
856: \qed\end{pf}
857:
858: Theorem~\ref{thm:n56} shows that $i(n) > 0$ for all $n \geq 56$.
859: Direct computation shows that $i(n) > 0$ for all $n > 7$ with the
860: exception of $n = 9,11$ and $12$.
861:
862:
863: \section{Further Problems}
864: Clearly we have only scratched the surface concerning the order of
865: magnitude of $M_k(n)$, $M(n)$ and $i(n)$. We have computed tables
866: of the functions, and based on that evidence we make the following
867: conjectures.
868: \begin{conj}\label{con:p*}
869: $M(n) \geq p^* (n)$ for $n \geq 0$, where $p^*(n)$ is the total
870: number of partitions of $n$ into parts that are either $\leq 6$ or
871: multiples of $3$ or both.
872: \end{conj}
873: \begin{conj}\label{con:C}
874: There exists a positive constant $C$ so that
875: \begin{equation}
876: \lim_{n\rightarrow\infty} \frac{\log M(n)}{\sqrt{n}} = C.
877: \end{equation}
878: \end{conj}
879: If $C$ exists and if Conjecture \ref{con:p*} is true, then
880: \cite[Th. 6.2, p.89]{A76}
881: \begin{equation}
882: \frac{\pi}{3} \sqrt{2} \leq C \leq \pi \sqrt{\frac23}\;.
883: \end{equation}
884: \begin{conj}\label{con:Ck}
885: Let $C_k$ be the infimum of the quotients $M_k(n) / p_k(n)$ where $n$ ranges
886: over the natural numbers.
887: Then $C_k > 0$ for all natural numbers $k$. Moreover,
888: $C_k$ is a strictly decreasing function of $k$ for $k
889: \geq 3$ and $C_k \to 0$ as $k \to \infty$.
890: \end{conj}
891: \begin{conj}\label{con:phash}$M(n) \leq p^{\#}(n)$ for $n \geq 0$ where
892: $p^{\#}(n)$ is the total number of partitions of $n$ into parts
893: that are either $\leq 7$ or multiples of $3$ or both.
894: \end{conj}
895: Conjecture \ref{con:phash} together with Conjecture \ref{con:p*}
896: allows us to replace Conjecture \ref{con:C} with
897: \begin{conj}
898: \begin{equation}
899: \lim_{n\rightarrow \infty} \frac{\log M(n)}{\sqrt{n}} =
900: \frac{\pi}{3} \sqrt{2}\,.
901: \end{equation}
902: \end{conj}
903:
904: {\em Acknowledgements:} {\small We thank Anna Bigatti for expert
905: advice on \cocoa \cite{CocoaSystem} and Bogdan Matasaru for
906: rewriting a program for computing $M(n)$ in the programming language
907: {\tt C}.}
908:
909: \begin{table}
910: \begin{tabular} {rrrrrrrr}
911: $n$ & $p_{\Primes}(n)$ & $L_{[4]}(n)$ & $p^*(n)$ & $M(n)$
912: &
913: $p^{\#}(n)$ & $U_{\mathbb{N}^+,[4]}(n)$ & $p(n)$\\
914: \hline
915: $0$ & $1$ & $1$ & $1$ & $1$ & $1$ & $1$ & $1$ \\
916: % $1$ & $0$ & $1$ & $1$ & $1$ & $1$ & $1$ & $1$ \\
917: % $2$ & $1$ & $2$ & $2$ & $2$ & $2$ & $2$ & $2$ \\
918: % $3$ & $1$ & $3$ & $3$ & $3$ & $3$ & $3$ & $3$ \\
919: % $4$ & $1$ & $5$ & $5$ & $5$ & $5$ & $5$ & $5$ \\
920: % $5$ & $2$ & $7$ & $7$ & $7$ & $7$ & $7$ & $7$ \\
921: % $6$ & $2$ & $10$ & $11$ & $11$ & $11$ & $11$ & $11$ \\
922: % $7$ & $3$ & $13$ & $14$ & $14$ & $15$ & $14$ & $15$ \\
923: % $8$ & $3$ & $18$ & $20$ & $20$ & $21$ & $21$ & $22$ \\
924: % $9$ & $4$ & $23$ & $27$ & $27$ & $29$ & $28$ & $30$ \\
925: $10$ & $5$ & $30$ & $36$ & $36$ & $39$ & $39$ & $42$ \\
926: $20$ & $26$ & $232$ & $357$ & $366$ & $445$ & $526$ & $627$ \\
927: $30$ & $98$ & $1102$ & $2064$ & $2131$ &
928: $2875$ & $4349$ & $5604$ \\
929: $40$ & $302$ & $4020$ & $8853$ & $9292$ &
930: $13549$ & $27195$ & $37338$ \\
931: $50$ & $819$ & $12405$ & $31639$ & $33799$ &
932: $52321$ & $140965$ & $204226$ \\
933: $60$ & $2018$ & $34016$ & $99245$ & $107726$ &
934: $175426$ & $636536$ & $966467$ \\
935: $70$ & $4624$ & $85333$ & $281307$ & $310226$ &
936: $527909$ & $2582469$ & $4087968$ \\
937: \hline
938: \end{tabular}
939: %\caption{\label{numbers}Some values of $M(n)$.}
940: \label{numbers}
941: \end{table}
942:
943: \nocite{A94}
944: \bibliographystyle{elsart-num}
945: \bibliography{mn}
946: \end{document}
947: