logo

Cauchy Product: The Product of Two Convergent Power Series 📂Analysis

Cauchy Product: The Product of Two Convergent Power Series

Theorem 1

If the convergence interval for f(x):=k=0akxkf(x) : = \sum _{k=0}^{\infty} a_{k} x^{k} and g(x):=k=0bkxkg(x) : = \sum_{k=0}^{\infty} b_{k} x^{k} is (r,r)(-r,r) and assuming ck:=j=0kajbkjc_{k} := \sum_{j=0}^{k} a_{j} b_{k-j}, then k=0ckxk\sum_{k=0}^{\infty} c_{k} x^{k} converges to f(x)g(x)f(x)g(x) within the convergence interval (r,r)(-r,r).

Description

The fact that the products of coefficients converge to the product of the coefficients of the two functions on their own is quite fascinating. It would be taken for granted if it were just a polynome, but power series have infinitely many terms, after all.

Proof

Let’s fix x(r,r)x \in (-r,r) and nNn \in \mathbb{N} one at a time and define sequences of functions as follows.

fn(x):=k=0nakxkgn(x):=k=0nbkxkhn(x):=k=0nckxk \begin{align*} f_{n} (x) : =& \sum_{k=0}^{n} a_{k} x^{k} \\ g_{n} (x) : =& \sum_{k=0}^{n} b_{k} x^{k} \\ h_{n} (x) : =& \sum_{k=0}^{n} c_{k} x^{k} \end{align*}

Since commutativity of addition holds for a finite number of terms, hn(x)=k=0nckxk=k=0nj=0kajbkjxjxkj=j=00ajb0jxjx0j+j=01ajb1jxjx1j++j=0najbnjxjxnj=+a0b0x0x0+a0b1x0x1+a1b0x1x0+a0bnx0xn+a1bn1x1xn1++anbnxnx0(sum by column)=a0x0k=0nbkxk+a1x1k=1nbkxk1++anxnk=nnbkxkk=j=0najxjk=jnbkjxkj=j=0najxjgnj(x)=j=0najxj[gnj(x)+g(x)g(x)]=g(x)j=0najxj+j=0najxj[gnj(x)g(x)]=g(x)fn(x)+j=0najxj[gnj(x)g(x)] \begin{align*} h_{n} (x) =& \sum_{k=0}^{n} c_{k} x^{k} \\ =& \sum_{k=0}^{n} \sum_{j=0}^{k} a_{j} b_{k-j} x^{j} x^{k-j} % \\ =& \sum_{k=0}^{n} \left[ a_{0} b_{k} x^{0} x^{k} + a_{1} b_{k-1} x^{1} x^{k-1} + \cdots + a_{k-1} b_{1} x^{1} x^{k-1} + a_{k} b_{0} x^{0} x^{k} \right] \\ =& \sum_{j=0}^{0} a_{j} b_{0-j} x^{j} x^{0-j} + \sum_{j=0}^{1} a_{j} b_{1-j} x^{j} x^{1-j} + \cdots + \sum_{j=0}^{n} a_{j} b_{n-j} x^{j} x^{n-j} \\ =& + a_{0} b_{0} x^{0} x^{0} \\ & + a_{0} b_{1} x^{0} x^{1} + a_{1} b_{0} x^{1} x^{0} \\ & \vdots \\ & + a_{0} b_{n} x^{0} x^{n} + a_{1} b_{n-1} x^{1} x^{n-1} + \cdots + a_{n} b_{n} x^{n} x^{0} \\ (\text{sum by column}) =& a_{0} x^{0} \sum_{k=0}^{n} b_{k} x^{k} + a_{1} x^{1} \sum_{k=1}^{n} b_{k} x^{k-1} + \cdots + a_{n} x^{n} \sum_{k=n}^{n} b_{k} x^{k-k} \\ =& \sum_{j=0}^{n} a_{j} x^{j} \sum_{k=j}^{n} b_{k-j} x^{k-j} \\ =& \sum_{j=0}^{n} a_{j} x^{j} g_{n-j} (x) \\ =& \sum_{j=0}^{n} a_{j} x^{j} \left[ g_{n-j} (x) + g(x) - g(x) \right] \\ =& g(x) \sum_{j=0}^{n} a_{j} x^{j} + \sum_{j=0}^{n} a_{j} x^{j} \left[ g_{n-j} (x) - g(x) \right] \\ =& g(x) f_{n} (x) + \sum_{j=0}^{n} a_{j} x^{j} \left[ g_{n-j} (x) - g(x) \right] \end{align*}

Thus, it only remains to show that limnfn(x)=f(x)\lim _{n \to \infty} f_{n} (x) = f(x) implies limnj=0najxj[gnj(x)g(x)]=0\lim _{n \to \infty} \sum_{j=0}^{n} a_{j} x^{j} \left[ g_{n-j} (x) - g(x) \right] = 0.

Given any positive number ε>0\varepsilon > 0, assuming that within the convergence interval limngn(x)=g(x)\lim _{n \to \infty} g_{n} (x) = g(x) and f(x)f(x) converges absolutely, for all natural numbers n>jn > j

gnj(x)g(x)M | g_{n- j } (x) - g (x) | \le M

k=0akxk<M \sum_{k=0}^{\infty} \left| a_{k} x^{k} \right| < M

there exists a M>0M > 0 that satisfies the above. For the same reason, for this MM

lN    gl(x)g(x)<ε2M l \ge N \implies | g_{ l } (x) - g (x) | < {{\varepsilon} \over {2M}}

k=N+1akxk<ε2M \sum_{k=N+1}^{\infty} \left| a_{k} x^{k} \right| < {{\varepsilon} \over {2M}}

there can be chosen a NNN \in \mathbb{N}.

Now, setting n>2Nn > 2N

j=0najxj[gnj(x)g(x)]=j=0Najxj[gnj(x)g(x)]+j=N+1najxj[gnj(x)g(x)]j=0Najxj[gnj(x)g(x)]+j=N+1najxj[gnj(x)g(x)]j=0Najxjgnj(x)g(x)+j=N+1najxjgnj(x)g(x)ε2Mj=0Najxj+Mj=N+1najxjε2MM+Mε2Mε2+ε2=ε \begin{align*} \left| \sum_{j=0}^{n} a_{j} x^{j} \left[ g_{n-j} (x) - g(x) \right] \right| =& \left| \sum_{j=0}^{N} a_{j} x^{j} \left[ g_{n-j} (x) - g(x) \right] + \sum_{j=N+1}^{n} a_{j} x^{j} \left[ g_{n-j} (x) - g(x) \right] \right| \\ \le & \left| \sum_{j=0}^{N} a_{j} x^{j} \left[ g_{n-j} (x) - g(x) \right] \right| + \left| \sum_{j=N+1}^{n} a_{j} x^{j} \left[ g_{n-j} (x) - g(x) \right] \right| \\ \le & \sum_{j=0}^{N} \left| a_{j} x^{j} \right| \left| g_{n-j} (x) - g(x) \right|+ \sum_{j=N+1}^{n} \left| a_{j} x^{j} \right| \left| g_{n-j} (x) - g(x) \right| \\ \le & {{\varepsilon} \over {2M}} \sum_{j=0}^{N} \left| a_{j} x^{j} \right| + M \sum_{j=N+1}^{n} \left| a_{j} x^{j} \right| \\ \le & {{\varepsilon} \over {2M}} \cdot M + M \cdot {{\varepsilon} \over {2M}} \\ \le & {{\varepsilon} \over {2}} + {{\varepsilon} \over {2}} \\ =& \varepsilon \end{align*}


  1. Wade. (2013). An Introduction to Analysis(4th Edition): p244. ↩︎