29

While solving a problem, I reduced it in the form of the following recurrence relation.

$ C_{0} = 1, C_{n} = \displaystyle\sum_{i=0}^{n - 1} C_{i}C_{n - i - 1} $

However https://en.wikipedia.org/wiki/Catalan_number tells me, this is the recurrence relation for catalan numbers and it can be further simplified as,

$ C_{0} = 1, C_{n} = \displaystyle\frac {2(2n - 1)}{n + 1} C_{n - 1}$

How can I derive the second relationship from the first one ? One way is to prove it is by induction, but we don't know the simplified recurrence so far.

Shashwat
  • 395
  • 1
    The last sentence is ungrammatical and unclear. – joriki Mar 22 '13 at 13:05
  • Getting the last recurrence from the explicit formula is considered cheating? – vonbrand Mar 22 '13 at 13:45
  • @joriki: I apologize. What I meant was, can we get to the second recurrence relation using only the first recurrence relation ? Assuming we already know the second recurrence relation, we can prove its correctness or we can derive it using closed form as done by MhenniBenghorbal, but what if we only have the first recurrence relation ? – Shashwat Mar 22 '13 at 15:44
  • @vonbrand: If you do not have the explicit formula and you are unaware of the second recurrence, what to do in that case ? The same happened in my case. – Shashwat Mar 22 '13 at 15:47
  • There is no simple answer to that. The recurrences you can come up with are infinite. The best that can be done is to come up with guesses that cover a large(ish) collection. Those are then called "methods to solve ..." ;-) – vonbrand Mar 22 '13 at 15:51
  • @vonbrand: True that. However, can't we attempt to simplify the recurrence relation to a linear recurrence relation ? – Shashwat Mar 22 '13 at 16:56
  • 2
    If you don't mind a slight detour, one elegant proof is to show that the number of right-up lattice paths from (0,0) to (n,n) that do not cross over the diagonal satisfies the first recurrence. This is almost natural because we want to find something that decomposes in exactly one way into two parts, which in general are objects of this sort. If we didn't see the relation to lattice paths, we could have thought of valid bracket sequences because the first part can be everything in the first matching pair of brackets and the second part can be everything outside. – user21820 Jul 17 '14 at 02:44
  • 1
    After that the reflection proof is also natural because when we consider the paths that cross the diagonal they must cross the diagonal for the first time at a unique point that is solely determined by the part of the path before that, and so reflecting the rest of the path will be a bijection to a more easily countable set. – user21820 Jul 17 '14 at 02:48
  • Catalon number derivation is given here http://www.cs.nthu.edu.tw/~wkhon/algo08-tutorials/tutorial-catalan.pdf – Rajesh Sep 17 '14 at 06:55

3 Answers3

35

You can probably find it somewhere online, but for completeness here’s a derivation of the familiar closed form for $C_n$ from the recurrence $$C_n=\sum_{k=0}^{n-1}C_kC_{n-1-k}\tag{0}$$ and the initial value $C_0$, via the ordinary generating function. Then, as in Mhenni Benghorbal’s answer, you can easily (discover and) verify the first-order recurrence. I don’t see any nice way to get it directly from $(0)$.

Let the ordinary generating function for the Catalan numbers be

$$c(x)=\sum_{n\ge 0}C_nx^n=\sum_{n\ge 0}\sum_{k=0}^{n-1}C_kC_{n-1-k}x^n\;.$$

Then since $C_0=1$, we have

$$\begin{align*} c(x)&=\sum_{n\ge 0}\sum_{k=0}^{n-1}C_kC_{n-1-k}x^n\\ &=1+\sum_{n\ge 1}\sum_{k=0}^{n-1}C_kC_{n-1-k}x^n\\ &=1+x\sum_{n\ge 0}\sum_{k=0}^nC_kC_{n-k}x^n\\ &=1+x\left(\sum_{n\ge 0}C_nx^n\right)^2\\ &=1+xc(x)^2\;, \end{align*}$$

or $xc(x)^2-c(x)+1=0$. The quadratic formula then yields

$$c(x)=\frac{1\pm\sqrt{1-4x}}{2x}\;,\tag{1}$$

and since

$$\lim_{x\to 0}c(x)=\lim_{x\to 0}\sum_{n\ge 0}C_nx^n=C_0=1\;,$$

it’s clear that we must choose the negative square root in $(1)$, so that

$$c(x)=\frac{1-\sqrt{1-4x}}{2x}\;.$$

Now apply the binomial theorem to $\sqrt{1-4x}$:

$$\begin{align*} \left(1-4x\right)^{1/2}&=1+\sum_{n\ge 1}\binom{1/2}n(-4x)^n\\ &=1+\sum_{n\ge 1}\frac{\left(\frac12\right)\left(-\frac12\right)\left(-\frac32\right)\dots\left(-\frac{2n-3}2\right)}{n!}(-4x)^n\\ &=1+\sum_{n\ge 1}(-1)^{n-1}\frac{(2n-3)!!}{2^nn!}(-4x)^n\\ &=1-\sum_{n\ge 1}\frac{2^n(2n-3)!!}{n!}x^n\\ &=1-2\sum_{n\ge 1}\frac{2^{n-1}\prod_{k=1}^{n-1}(2k-1)}{n(n-1)!}x^n\\ &=1-2\sum_{n\ge 1}\frac{2^{n-1}(n-1)!\prod_{k=1}^{n-1}(2k-1)}{n(n-1)!^2}x^n\\ &=1-2\sum_{n\ge 1}\frac{\left(\prod_{k=1}^{n-1}(2k)\right)\left(\prod_{k=1}^{n-1}(2k-1)\right)}{n(n-1)!^2}x^n\\ &=1-2\sum_{n\ge 1}\frac{(2n-2)!}{n(n-1)!^2}x^n\\ &=1-2\sum_{n\ge 1}\frac1n\binom{2(n-1)}{n-1}x^n\;, \end{align*}$$

so

$$\begin{align*} c(x)&=\frac1{2x}\cdot2\sum_{n\ge 1}\frac1{n}\binom{2(n-1)}{n-1}x^n\\ &=\sum_{n\ge 1}\frac1n\binom{2(n-1)}{n-1}x^{n-1}\\ &=\sum_{n\ge 0}\frac1{n+1}\binom{2n}nx^n\;, \end{align*}$$

and we have the familiar closed form $C_n=\dfrac1{n+1}\dbinom{2n}n$.

Brian M. Scott
  • 616,228
  • Amazing. If I understood good, it means that $C_n$ is the Catalan sequence if and only if it satisfies this: $C_{0} = 1, C_{n} = \displaystyle\sum_{i=0}^{n - 1} C_{i}C_{n - i - 1}$. –  Mar 22 '13 at 23:57
  • @Thus: That’s right: that initial value and recurrence completely determine the entire sequence of Catalan numbers. – Brian M. Scott Mar 23 '13 at 00:02
  • Is that true for every sequence generated by this "cauchyproductlike" recurrence if we change the initial value to some other number? –  Mar 23 '13 at 00:08
  • @Thus: Yes, each choice of initial value will generate a unique sequence. With $a_0=2$, for instance, we have $a_1=2\cdot2=4$, $a_2=2\cdot4+4\cdot2=16$, $a_3=2\cdot16+4\cdot4+16\cdot2=80$, and so on. – Brian M. Scott Mar 23 '13 at 00:12
  • You awakened my interest, what is your opinion about this: $C_{0} = k, C_{n} = \displaystyle\sum_{i=0}^{n - 1}\dfrac {C_{i}}{C_{n - i - 1}}$

    Will it also generate an unique sequnce?

    –  Mar 23 '13 at 00:19
  • @Thus: Provided that it makes sense at all: you have to be sure that it never produces a $0$ term. With $a_0=1$ I get $a_1=\frac11=1$, $a_2=\frac11+\frac11=2$, $a_3=\frac12+\frac11+\frac21=\frac72$, $a_4=\frac{7/2}1+\frac21+\frac12+\frac1{7/2}=6+\frac27=\frac{44}7$, and so on. As long as I never get a $0$ term $-$ and I won’t if $a_0>0$ $-$ the recurrence completely determines $a_n$ given the earlier values. – Brian M. Scott Mar 23 '13 at 00:24
  • So then, apart from these recurrence relations, there seems that it is reasonable to believe that there is a large class of recurrence relations such that they define unique sequences? –  Mar 23 '13 at 00:35
  • @Thus: If a recurrence together with its initial value(s) defines a sequence at all, that sequence is unique; this can be proved by induction. – Brian M. Scott Mar 23 '13 at 00:37
  • @BrianM.Scott: Thank you. This answers my question. – Shashwat Mar 23 '13 at 10:54
  • 3
    @BrianM.Scott In the first half of your answer, how do you go from $1+x\sum_{n\ge 0}\sum_{k=0}^nC_kC_{n-k}x^n$ to $1+x\left(\sum_{n\ge 0}C_nx^n\right)^2$ – TheRealFakeNews Apr 11 '13 at 02:32
  • 6
    @Alan: Just multiply out the square: the coefficient of $x^n$ in $$\left(\sum_{n\ge 0}a_nx^n\right)\left(\sum_{n\ge 0}b_nx^n\right)$$ is $$\sum_{k=0}^na_kb_{n-k};.$$ – Brian M. Scott Apr 11 '13 at 03:36
  • I think that the split-up of the sum into $n = 0$ and $n > 0$ has to be done earlier (before splitting up $n! = n(n-1)!$. Otherwise $\frac{1}{n}$ is not defined for $n = 0$. – user7427029 Mar 07 '21 at 12:22
  • @user7427029: You’re right; thanks for catching that. I’m not quite awake yet, but I think that I’ve taken care of it now. – Brian M. Scott Mar 07 '21 at 19:57
  • 1
    Looks. wonderful. :-) – user7427029 Mar 07 '21 at 20:30
9

A related problem. It is easier to prove it using the identity

$$ C_n = \frac{1}{n+1}{2n\choose n} = \frac{(2n)!}{(n+1)!\,n!} \implies C_{n-1}=\frac{(2(n-1))!}{(n)!\,(n-1)!} $$

$$ \frac{C_n}{C_{n-1}}= \frac{ (2n)(2n-1)(2n-2)!(n-1)! }{(n+1)n(n-1)!(2n-2)!}=\frac{2(2n-1)}{n+1} $$

$$ \implies C_n = \frac{2(2n-1)}{n+1} C_{n-1}. $$

Added: We will find the ordinary generating function. Let $g(x)=\sum_{n=0}^{\infty}C_{n}x^{n} $

$$ C_{n+1} = \displaystyle\sum_{i=0}^{n } C_{i}C_{n - i } \implies \sum_{n=0}^{\infty}C_{n+1}x^n = \sum_{n=0}^{\infty} \sum_{i=0}^{n } C_{i}C_{n - i } x^n $$

$$ \implies \sum_{n=1}^{\infty}C_{n}x^{n-1} = \sum_{i=0}^{\infty}C_i\sum_{n=i}^{\infty}C_{n-i}x^n= \sum_{i=0}^{\infty}C_i\sum_{n=0}^{\infty}C_{n}x^{n+i}$$

$$\implies \frac{1}{x}\sum_{n=0}^{\infty}C_{n}x^{n}-\frac{C_0}{x}= \sum_{i=0}^{\infty}C_ix^i\sum_{n=0}^{\infty}C_{n}x^{n} $$

$$ \implies \frac{g(x)}{x}-\frac{1}{x} = g(x)g(x) = g(x)^2 $$

$$ \implies g(x)^2-\frac{g(x)}{x}+\frac{1}{x} = 0. $$

6

Note: This answer is merely a slight variation of the answers already given. The derivation of the generating function of the Catalan Numbers is somewhat different which may be convenient for the reader.

The following is valid: The recurrence relation \begin{align*} C_{0} = 1, C_{n} = \displaystyle\sum_{i=0}^{n - 1} C_{i}C_{n - i - 1}\qquad(n\geq 1)\tag{1} \end{align*} specifies the Catalan Numbers \begin{align*} \qquad\qquad\frac{1}{n+1}\binom{2n}{n}\qquad\qquad(n\geq 0)\tag{2} \end{align*}

Note: The connection between the closed formula (2) and the formula stated in the question is given at the beginning of the answer from @MhenniBenghorbal.

Generating function for $C_n$:

When looking at the recurrence relation we see that $\sum_{i=0}^{n - 1} C_{i}C_{n - i - 1}$ is a Cauchy-Product. Since Cauchy products occur when multiplying series it seems natural to work with the following generating functions:

\begin{align*} C(z) = \sum_{n\geq 0}C_nz^n\qquad\text{and}\qquad C^2(z)=\sum_{n\geq 0}\left(\sum_{i=0}^{n}C_iC_{n-i}\right)z^n\tag{3} \end{align*}

Let $[z^n]$ denote the coefficient operator. We observe with the help of (1) and (3):

\begin{align*} [z^n]C(z)&=C_n\\ &=\sum_{i=0}^{n-1}C_iC_{(n-1)-i}\\ &=[z^{n-1}]C^2(z)\\ &=[z^n]zC^2(z)\qquad\qquad\qquad(n\geq 1)\\ \\ [z^0]C(z)&=C_0=1\\ \end{align*}

Therefore we get

\begin{align*} C(z)=zC^2(z)+C_0=zC^2(z)+1 \end{align*}

and solving the quadratic equation gives

\begin{align*} C_{1,2}(z)=\frac{1}{2z}\left(1\pm\sqrt{1-4z}\right) \end{align*} Since the expansion of $\sqrt{1-4z}=1-2z-\ldots$ and $C(z)= \sum_{n\geq 0}C_nz^n$ is a power series in $z$ we conclude that following solution is valid:

\begin{align*} C(z)=\frac{1}{2z}\left(1-\sqrt{1-4z}\right) \end{align*}

Now:

Calculation of $C_n$:

With the help of the well known binomial identity $$\binom{\frac{1}{2}}{n}=\frac{(-1)^{n+1}}{4^n(2n-1)}\binom{2n}{n}$$ we get \begin{align*} C_n&=[z^n]\frac{1}{2z}\left(1-\sqrt{1-4z}\right)\\ &=-\frac{1}{2}[z^{n+1}]\sqrt{1-4z}\\ &=-\frac{1}{2}[z^{n+1}]\sum_{n\geq 0}\binom{\frac{1}{2}}{n}\left(-4z\right)^n\\ &=-\frac{1}{2}\binom{\frac{1}{2}}{n+1}\left(-4\right)^{n+1}\\ &=\frac{1}{2}\frac{1}{2n+1}\binom{2n+2}{n+1}\\ &=\frac{1}{2}\frac{1}{2n+1}\frac{(2n+2)(2n+1)}{(n+1)^2}\binom{2n}{n}\\ &=\frac{1}{n+1}\binom{2n}{n} \end{align*}

Markus Scheuer
  • 108,315