Suppose f(m,n) is a function that is defined for non-negative integers m,n where:
f(m,n) = 0 if m < n
f(m,0) = 1 \forall m
\displaystyle f(m,n) = \sum_{k=1}^{n} f(k-1,k-1) f(m-k,n-k) + f(m-1,n) \forall m \geq n
Prove that f(m,n) = \frac{m-n+1}{m+1} \binom{m+n}{m}
Solution
First, if m=n, then the recursion becomes the recursion of Catalan numbers, and the initial condition f(0,0) = 1 matches. So f(n,n) = C_n = \frac{1}{n+1}\binom{2n}{n} is the n-th Catalan number.
Our recursion now becomes:
\displaystyle f(m,n) = \sum_{k=1}^{n} C_{k-1} f(m-k,n-k) + f(m-1,n)
Let d = m-n, we know that the formula holds for d=0. For d>0, we introduce g(n,d) = f(n+d,n) that is defined over all non-negative n,d, with g(n,0) = C_n \forall n.
Then our recursion formula becomes:
\displaystyle g(n,d) = \sum_{k=1}^{n} C_{k-1} g(n-k,d) + g(n,d-1)
Due to the last term, we also define that g(n,-1) = 0 \forall n to allow that recursion formula to work generally. Note that this convention does not violate the original definition of f.
Suppose P(x) = C_0 + C_1x + C_2x^2 + \cdots + C_nx^n + \cdots be an infinite polynomial defined on a region near x=0 (within its converging radius). We know that P(x) is the generating function for Catalan numbers, and it's known to be
\displaystyle P(x) = \frac{1-\sqrt{1-4x}}{2x}
Now let Q(x,y) = \sum_{n,d \geq 0} g(n,d)x^ny^d also defined near zero.
Consider the coefficient of x^n y^d in P(x)Q(x,y). It is the sum C_0 g(n,d) + C_1g(n-1,d) + \cdots + C_ng(0,d) = g(n+1,d) - g(n+1,d-1)
But \sum_{n \geq 1, d \geq 0}g(n,d)x^ny^d = Q(x,y) - (1+y+y^2 + \cdots) = Q - \frac{1}{1-y}
Which means the coefficients of P(x) Q(x,y) is identical to \frac{1}{x}(Q - \frac{1}{1-y}}) - \frac{y}{x}(Q - \frac{1}{1-y})
PQ = \frac{1}{x}(Q - \frac{1}{1-y}} )- \frac{y}{x}(Q - \frac{1}{1-y}) solving for Q yields
Q(x,y) = \frac{1}{1-y-xP(x)}
As a sanity check, when y =0, all the y terms vanish and Q should be identical to P. But since P is the Catalan generating function, it satisfies P(x) = 1+xP(x)^2, so Q = \frac{1}{1-xP} = P.
One possible approach here is to solve explicitly for Q(x,y) and find its Taylor series to get an explicit formula for g and thus f, but it would make you want to gouge your eyes out.
Instead, we note that P = 1 + xP^2, so 1-xP = 1-\frac{P-1}{P} = \frac{1}{P}, thus
Q=\frac{1}{1-y-xP} = \frac{1}{\frac{1}{P}-y} = \frac{P}{1-Py}
Which means that Q-P = yPQ is an identity near zero. Note that
\displaystyle Q-P = \sum_{n \geq 0, d \geq 1}g(n,d)x^ny^d
And that
\displaystyle yPQ = \sum_{n \geq 0, d \geq 1}x^ny^d(g(n+1,d-11) - g(n+1,d-2))
From there we get another identity: g(n,d) = g(n+1,d-1) - g(n+1,d-2) for all n \geq 0, d \geq 1
In other words, g(n,d) = g(n-1,d+1) + g(n,d-1) \forall n\geq 1, d \geq 0
Translating back, we have f(m,n) = f(m,n-1) + f(m-1,n)
Suppose s = m+n, we can prove the desired formula by induction on s. First, for m+n=1, it must be that m=1,n=0, and the formula is satisfied trivially. One can also check that the formula still holds for low values of s. The induction step holds because:
f(m,n-1) + f(m-1,n) = \frac{m-n+2}{m+1} \binom{m+n-1}{m} - \frac{m-n}{m} \binom{m+n-1}{m-1}
=\frac{n(m-n+2)}{(m+n)(m+1)} \binom{m+n}{m} + \frac{m-n}{m+n} \binom{m+n}{m} = \frac{m-n+1}{m+1} \binom{m+n}{m+1}
[...] The solution for this recursive formula is described here: http://dharmath.thehendrata.com/2009/10/23/2-variable-recursion/ [...]
ReplyDelete