0

Let $\vec k=(k_1,...,k_d)$ be a $d$-tuple with $k_i$ being non-negative integers satisfying $\sum_i k_i=n$; $\vec k$ is then referred to as a weak $d$-composition of $n$. It is clear that if $\vec a$ is a weak $d$-composition of $l$, and $\vec b$ is a weak $d$-composition of $m$, then $\vec k\equiv\vec a +\vec b=(a_1+b_1,...,a_d+b_d)$ is a weak $d$-composition of $l+m\equiv n$.

I want to know, for a given $\vec k$ and $l$, how many pairs $\vec a$ and $\vec b$ exist? Alternatively, we are looking for all $\vec a$ which are weak $d$-composition of $l$ satisfying $a_i\leq k_i$ for all $i$. Let $\mathbb{A}^l(\vec k)$ denote the set of all such $\vec a$, for a fixed $l$ and $\vec k$.

Q1: Is this concept and notation already in the literature?

Q2: What is the cardinality of this set?

It is clear that the cardinality is symmetric under $l\to n-l$. Please provide references, as I am quite new to combinatorics. This problem emerges when writing Vandermonde's identity for multinomials (e.g. Eq. 4 here, different notation), where we implicitly sum over all $\vec a \in \mathbb{A}^l(\vec k)$. I wrote code on Mathematica that counts the elements for individual cases of $\vec k$, but I didn't succeed in conjecturing a formula.

David Raveh
  • 1,795
  • The concept you want is "stars and bars." There are $\binom{n+d-1}{d-1}$ weak $d$-compositions of $n$. The number of such "cuts" is $\binom{l+d-1}{d-1} \cdot \binom{n-l+d-1}{d-1}$. – Robert Shore Jul 10 '23 at 21:09
  • @RobertShore "Alternatively, $\vec a$ is a weak $d$-composition of $l$ satisfying $a_i\leq k_i$ for all $i$." This is a subset of the set of all weak $d$-compositions of $l$. – David Raveh Jul 10 '23 at 21:12
  • So given a specific weak $d$-composition of $n$, you want to know how many ways that specific composition can be "cut"? – Robert Shore Jul 10 '23 at 21:13
  • @RobertShore how many ways it can be cut into pieces $l$ and $n-l$. – David Raveh Jul 10 '23 at 21:16
  • 1
    To confirm, can you phrase Q2 like this? You are given $\vec k$ and $l$, and you are asking for the number of ways to choose $a_1, \dots,a_d$, such that $0\le a_i\le k_i$ for each $i∈ {1,\dots, d}$, and such that $a_1+\dots+a_d=l$? – Mike Earnest Jul 11 '23 at 14:09
  • @MikeEarnest precisely! – David Raveh Jul 11 '23 at 16:11

0 Answers0