3

I was working with this recurrence relation :

$$\begin{cases}A(n,k) = A(n-1, k-1)+A(n-2, k-1)+A(n-1, k)\\ A(n, 0) = 1\\ A(n, 1) = 2n \end{cases}$$ Generating function : $(1+x)/(1-x-x*y-x^2*y)$

Now since this involves two parameters, I tried changing it to a single parameter by fixing $k$ and iterating over $n$ for $n=k$, $n=k+1$, $n=k+2$.. and so on.

What I found was that $A(k, k)=2$, $A(k+1, k)=4k$, $A(k+2, k)=4k^2+2$ and the later results were too haphazard to write i.e. they did not form a pattern to the previous formulas. I tried to use the above equation to find characteristic roots and use $A(n) = a(x^n)+b(y^n)$ where $x$ and $y$ are roots. However, I wasn't able to compute it correctly.

What am I missing here? Am I tackling the problem in a complete wrong way? Is there any other method to solve such recurrences? Any help is appreciated. I assume the formula for $A(n)$ could be pretty cumbersome but I still wish to find it.

resound
  • 128
  • You can create a spreadsheet with $n$ in the rows and $k$ in the columns. You need a starting rule for $n=0$ as well. Maybe the $A(n,0)=1$ was supposed to be $A(0,k)=1$. Once you have the starting rule you can just put the first line in the first place to calculate and copy right and down. That will give you values, not an expression. – Ross Millikan Oct 13 '18 at 21:31
  • $A(n,k)$ depends on $A(n-1, k)$, and that depends on $A(n-2, k)$, and so on. Are the boundary conditions missing? – peterwhy Oct 13 '18 at 21:31
  • Try generating functions. That is multiply both sides by $z_1^n z_2^k$ then sum both sides over all $n$ and $k$. You get some equation about a function of two variables. You solve for that function and then extract the coefficent of $z_1^n z_2^k$ at the end. – AHusain Oct 13 '18 at 21:38
  • @peterwhy The boundary conditions mentioned are the only ones I have, and to the best of my knowledge, they should be correct. I looked at these relations a bit on Oeis and they can also be represented as nB(n)=2(n-k+1)B(n-1) + (n-2)B(n-2) ; B(0)=1 and B(1) = 2*(n-k+1) – resound Oct 13 '18 at 21:38
  • G.f. : (1+x)/(1-x-xy-x^2y) - What I found on OEIS. I don't know how we could use this to solve it. Over my head! @AHusain – resound Oct 13 '18 at 21:42
  • @RossMillikan It surely is possible to generate a dp table but I wished for a closed formula and which I can say is obtainable for this relation. Thanks for the suggestion there! :) – resound Oct 13 '18 at 21:48
  • Please change the title to something which better describes the question, and without the exclamation mark – Yuriy S Oct 17 '18 at 11:02
  • @YuriyS I'll keep this in mind for my future questions. – resound Oct 18 '18 at 12:09

2 Answers2

2

We look at the generating function $\frac{1+x}{1-x-xy-x^2y}$ and derive the coefficients $A(n,k)$. It is convenient to use the coefficient of operator $[x^n]$ to denote the coefficient of $x^n$ of a series.

We obtain \begin{align*} \color{blue}{A(n,k)}&=[x^ny^k]\frac{1+x}{1-x-xy-x^2y}\\ &=[x^ny^k]\frac{1+x}{1-x-x(1+x)y}\\ &=[x^ny^k]\frac{1+x}{1-x}\cdot\frac{1}{1-\frac{x(1+x)}{1-x}y}\\ &=[x^n]\frac{1+x}{1-x}[y^k]\sum_{j=0}^\infty\left(\frac{x(1+x)}{1-x}\right)^jy^j\tag{1}\\ &=[x^n]\frac{1+x}{1-x}\left(\frac{x(1+x)}{1-x}\right)^k\tag{2}\\ &=[x^{n-k}]\frac{(1+x)^{k+1}}{(1-x)^{k+1}}\tag{3}\\ &=[x^{n-k}](1+x)^{k+1}\sum_{j=0}^\infty\binom{-k-1}{j}(-x)^j\tag{4}\\ &=[x^{n-k}](1+x)^{k+1}\sum_{j=0}^\infty\binom{k+j}{j}x^j\tag{5}\\ &=\sum_{j=0}^{n-k}\binom{k+j}{j}[x^{n-k-j}](1+x)^{k+1}\tag{6}\\ &=\sum_{j=0}^{n-k}\binom{k+j}{j}\binom{k+1}{n-k-j}\tag{7}\\ &\,\,\color{blue}{=\sum_{j=0}^{n-k}\binom{n-j}{k}\binom{k+1}{j}}\tag{8} \end{align*}

Comment:

  • In (1) we do a geometric series expansion with respect to $y$.

  • In (2) we select the coefficient of $y^k$.

  • In (3) we do some simplifications and apply the rule $[x^p]x^qA(x)=[x^{p-q}]A(x)$.

  • In (4) we do a binomial series expansion.

  • In (5) we apply the binomial identity $\binom{-p}{q}=\binom{p+q-1}{q}(-1)^q$.

  • In (6) we select the coefficient of $x^{n-k}$ and restrict the upper bound of the sum by $n-k$ since other terms do not contribute to $[x^{n-k}]$.

  • In (7) we select the coefficient of $x^{n-k-j}$.

  • In (8) we do a final rearrangement by changing the order of summation $j\to n-k-j$.

We use the formula (8) to check the boundary conditions. We obtain \begin{align*} A(n,0)&=\sum_{j=0}^n\binom{n-j}{0}\binom{1}{j}=\sum_{j=0}^n\binom{1}{j}=\begin{cases} 1&\qquad\qquad\qquad\qquad n=0\\ 2&\qquad\qquad\qquad\qquad n>0 \end{cases}\\ A(n,1)&=\sum_{j=0}^{n-1}\binom{n-j}{1}\binom{2}{j}=\sum_{j=0}^{n-1}(n-j)\binom{2}{j}=\begin{cases} 1&\qquad n=1\\ 4(n-1)&\qquad n>1 \end{cases} \end{align*} We observe, the generating function $\frac{1+x}{1-x-xy-x^2y}$ does not follow OPs stated boundary conditions $A(n,0)=1\ (n\geq 0)$ and $A(n,1)=2n\ (n\geq 1)$ and this is a reason for the difficulties OP has to cope with.

Note: We find with some help of Wolfram Alpha a series expansion \begin{align*} \frac{1+x}{1-x-xy-x^2y}&=1+(2+y)x+(2+4y+y^2)x^2+(2+8y+6y^2+y^3)x^3\\ &\qquad+(2+12y+18y^2+8y^3+y^4)x^4+\cdots \end{align*} The corresponding sequence of the coefficients $A(n,k)$ starting with \begin{align*} &1;\\ &\color{blue}{2},\color{blue}{1};\\ &\color{blue}{2},\color{blue}{4},1;\\ &\color{blue}{2},\color{blue}{8},6,1;\\ &\color{blue}{2},\color{blue}{12},18,8,1;\ldots \end{align*} can be found as A113413 in OEIS. We can find there OPs stated generating function $\frac{1+x}{1-x-xy-x^2y}$ as well as the recurrence relation $A(n,k) = A(n-1, k-1)+A(n-2, k-1)+A(n-1, k)$, but we have different boundary conditions (marked above in $\color{blue}{\mathrm{blue}}$).

Markus Scheuer
  • 108,315
  • I think at the max it can be reduced to the single dp I mentioned in the above comments. A closed simple formula is actually very highly unlikely. I'll mark this as a correct answer as it does define the intricacies of this question. – resound Oct 18 '18 at 12:07
  • @shimak: Thanks for accepting the answer. – Markus Scheuer Oct 18 '18 at 13:11
0

Here we solve OP's recurrence relation by a generating function approach. We will see the recurrence relation leads to a slightly different generating function than stated by OP due the specific boundary conditions. Then we use the generating function to show \begin{align*} \color{blue}{A(n,k)=\sum_{j=0}^{n-k}\binom{k+j}{j}\binom{k}{n-k-j}+\sum_{j=0}^{n-k}\binom{k+j-1}{j}\binom{k-1}{n-k-j}[[k\geq 1]]}\tag{1} \end{align*}

We use in (1) Iverson brackets $[[k\geq 1]]$ which is $1$ iff $k\geq 1$ and $0$ otherwise.

Note: OP's recurrence relation is not completely specified. We add a boundary condition $A(0,k), k\geq 1$ which do not conflict with anything else and consider

\begin{align*} A(n,k) &= A(n-1, k-1)+A(n-2, k-1)+A(n-1, k)\tag{2}\\ A(n, 0) &= 1\\ A(n, 1) &= 2n\\ \color{blue}{A(0, k) }&\color{blue}{= 0\qquad k\geq 1} \end{align*}

We define the generating function $a(x,y)=\sum_{n,k\geq 0}A(n,k)x^ny^k$ and obtain from the recurrence relation (2) \begin{align*} \sum_{{n\geq 2}\atop{k\geq 1}}&A(n,k)x^ny^k\\ &=\sum_{{n\geq 2}\atop{k\geq 1}}A(n-1,k-1)x^ny^k+\sum_{{n\geq 2}\atop{k\geq 1}}A(n-2,k-1)x^ny^k +\sum_{{n\geq 2}\atop{k\geq 1}}A(n-1,k)x^ny^k\\ &=xy\sum_{{n\geq 1}\atop{k\geq 0}}A(n,k)x^ny^k+x^2y\sum_{{n\geq 0}\atop{k\geq 0}}A(n,k)x^ny^k +x\sum_{{n\geq 1}\atop{k\geq 1}}A(n,k)x^ny^k\\ &=xy\left(a(x,y)-\sum_{k\geq 0}A(0,k)y^k\right)+x^2ya(x,y)+x\left(a(x,y)-\sum_{n\geq 0}A(n,0)x^n-\sum_{k\geq 0}A(0,k)y^k+1\right)\\ &=xy\left(a(x,y)-1\right)+x^2ya(x,y)+x\left(a(x,y)-\sum_{n\geq 0}x^n-1+1\right)\\ &=xy\left(a(x,y)-1\right)+x^2ya(x,y)+x\left(a(x,y)-\frac{1}{1-x}\right)\tag{3} \end{align*}

Since \begin{align*} \sum_{{n\geq 2}\atop{k\geq 1}}A(n,k)x^ny^k&=a(x,y)-\sum_{n\geq 0}A(n,0)x^n-\sum_{k\geq 1}A(k,0)y^k-x\sum_{k\geq 1}A(k,1)y^k\\ &=a(x,y)-\frac{1}{1-x}-2xy\tag{4} \end{align*}

Putting (3) and (4) together we obtain \begin{align*} \color{blue}{a(x,y)}&\color{blue}{=\frac{1+xy}{1-x-xy-x^2y}}\\ &=\color{blue}{1}+(\color{blue}{1}+\color{blue}{2}y)x+(\color{blue}{1}+\color{blue}{4}y+2 y^2)x^2+(\color{blue}{1}+\color{blue}{6}y+8y^2+2 y^3)x^3\\ &\qquad + (\color{blue}{1}+\color{blue}{8}y+18y^2+12y^3+2y^4)x^4+\cdots \end{align*} where the expansion was done with some help of Wolfram Alpha. The coefficients specified by the boundary conditions are marked in blue.

Note, the numerator $1+xy$ in $a(x,y)$ which is different to the numerator $1+x$ stated in OP's generating function.

Now we derive the formula (1) for $A(n,k)$. Since the derivation is similar to that of my other answer the comments and a few details are omitted. We obtain \begin{align*} \color{blue}{A(n,k)}&\color{blue}{=[x^ny^k]\frac{1+xy}{1-x-xy-x^2y}}\\ &=[x^ny^k]\frac{1+xy}{1-x-x(1+x)y}\\ &=[x^ny^k]\frac{1+xy}{1-x}\sum_{j=0}^\infty\left(\frac{x(1+x)}{1-x}\right)^jy^j\\ &=[x^{n-k}]\frac{(1+x)^k}{(1-x)^{k+1}}+[x^{n-k}]\frac{(1+x)^{k-1}}{(1-x)^k}[[k\geq 1]]\\ &=[x^{n-k}](1+x)^k\sum_{j=0}^\infty\binom{k+j}{j}x^j+[x^{n-k}](1+x)^{k-1}\sum_{j=0}^\infty\binom{k+j-1}{j}x^j[[k\geq 1]]\\ &\,\,\color{blue}{=\sum_{j=0}^{n-k}\binom{k+j}{j}\binom{k}{n-k-j}+\sum_{j=0}^{n-k}\binom{k+j-1}{j}\binom{k-1}{n-k-j}[[k\geq 1]]} \end{align*} and the claim (1) follows.

Markus Scheuer
  • 108,315