2
$\begingroup$

Fix three integers $a, b, c$ and consider a sequence of integers $a_{i,j}$ defined, for $i \ge 0, j \ge 0$, recursively as follows:

$a_{i,0}=1$ for every $i$, $a_{0,j}=a+bj+cj^2$ and, for $i \ge 1, j \ge 1$, $$a_{i,j}=a_{i,j-1}+a_{i-1,j}.$$ Is there a closed formula for the $a_{i,j}$'s?

$\endgroup$
5
  • $\begingroup$ Working out some examples, it looks like the coefficients of $a$ and $b$, along with the constant terms, are binomial coefficients. The coefficients of $c$ seem trickier. $\endgroup$ Commented Feb 13, 2023 at 17:40
  • 1
    $\begingroup$ Is $\ a_{0,0}=1\ $ or $\ a_{0,0}=a$? $\endgroup$
    – Wlod AA
    Commented Feb 13, 2023 at 17:42
  • 1
    $\begingroup$ I presume you’re asking for a closed formula $\endgroup$ Commented Feb 14, 2023 at 1:29
  • 1
    $\begingroup$ @J.W.Tanner: Yes, thanks (it was a typo) $\endgroup$
    – Cob
    Commented Feb 14, 2023 at 16:39
  • 1
    $\begingroup$ @Wlod AA: Thanks, better with $a=1$ $\endgroup$
    – Cob
    Commented Feb 14, 2023 at 17:17

2 Answers 2

2
$\begingroup$

As for the Fibonacci sequence, let us find a basis.

  1. If $a_{0,j} = 0$, then $a_{ij} = \binom{i+j-1}{j}$.
  2. If $a_{0,j} = 1$, then $a_{ij} = \binom{i+j}{j}$.
  3. If $a_{0,j} = j+1$, then $a_{ij} = \binom{i+j+1}{j}$.
  4. If $a_{0,j} = \frac{(j+1)(j+2)}2$, then $a_{ij} = \binom{i+j+2}{j}$.

Any quadratic polynomial $a+bj+cj^2$ is an affine combination of $0$, $1$, $j+1$, $\frac{(j+1)(j+2)}2$. The answer will be the combination of $\binom{i+j-1}{j}$, $\binom{i+j}{j}$, $\binom{i+j+1}{j}$, $\binom{i+j+2}{j}$ with the same coefficients.

If I have solved the linear system correctly, one gets $$a_{ij} = (1-a)\binom{i+j-1}{j} + (a-b+c) \binom{i+j}{j} + (b-3c) \binom{i+j+1}{j} + 2c \binom{i+j+2}{j}$$

Note that for $i=j=0$ this formula yields $a_{00} = a$, which was not well-defined in the question (but does not participate in the recurrence and is therefore irrelevant).

$\endgroup$
1
  • $\begingroup$ very nice, I got it, thanks $\endgroup$
    – Cob
    Commented Feb 14, 2023 at 17:17
1
$\begingroup$

I believe the pattern simplifies to \begin{align} a_{ij} = \binom{i+j-1}{j} & + \binom{i+j-1}{j-1}a + \binom{i+j}{j-1}b \\ & + \left[ \binom{i+j-1}{j-1} + 3\binom{i+j-1}{j-2} + 2\binom{i+j-1}{j-3} \right] c. \end{align}

The triangle of coefficients of $c$ is in the OEIS if you reverse the rows; it's A125165 which has a nice expression as a Riordan array.

Note: I had started on this before I saw Ivan Izmestiev's answer; I believe our claims are compatible.

$\endgroup$
2
  • 1
    $\begingroup$ Nice! Note that the coefficient at c can be simplified to $\binom{i+j+1}{j-1} + \binom{i+j}{j-2}$. $\endgroup$ Commented Feb 13, 2023 at 18:52
  • $\begingroup$ @Brian Hopkins: very nice, I got it, thanks $\endgroup$
    – Cob
    Commented Feb 14, 2023 at 17:18

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .