Show that each term of a composition of formal power series $A(B(q))$ is computed in finitely many steps if $b_0 = 0$

75 Views Asked by At

I'm doing an online course on Enumerative Combinatorics. In the chapter on formal power series and nonlinear recurrence relations, we start off by looking at compositions of formal power series $A(B(q))$ expressed as:

$$\sum_{n\geq0} a_{n}(B(q))^n$$

with $a_n \in A(q)$.

An exercise proposed by the instructor is to determine why computing each term of the above sum takes finitely many steps if $b_{0} \in B(Q) = 0$.

I had a little think and it doesn't seem to me why this would be possible if $b_0 = 0$. Let's take the second term of $A(B(q))$, $a_1B(q)$, and assume $b_0 \in B(q)$ is $0$.

We have $(a_1b_0=0) + a_1b_1q + a_1b_2q^2 + \dots$.

Is this last expression finite somehow?

1

There are 1 best solutions below

4
On BEST ANSWER

Let $A(B(q))=\sum_{n\ge 0}c_nq^n$, and say you want to compute $c_n$. The key idea is that the only contributions to $c_n$ come from the terms $$a_0+a_1 (B(q))^1+a_2(B(q))^2+\dots a_n (B(q))^n.$$ Indeed, for any $m>n$, the term $(B(q))^m$ will only have powers of $q$ which are greater than $n$. Therefore, you just need to extract the coefficient of $q^n$ in each of the above terms, which it should be clear can be done in finitely many steps.

We require $b_0=0$, because otherwise the composition is not a well defined power series. For example, let $A(q)=1/(1-q)=1+q+q^2+\dots$, and let $B(q)=1$. Then we would have $A(B(q))=\frac10=1+1+1+\dots$, which is impossible.