8
$\begingroup$

So, I've done most of the problem to this point, but just cannot figure out the last piece. I may just be missing the math skills needed to complete the proof (differential equations).

Problem (from Rudin): If $\alpha$ is real and $-1 < x < 1$, prove Newton's binomial theorem;

$(1+x)^\alpha$ = $1 + \sum{\frac{\alpha (\alpha-1) \dots(\alpha - n + 1)}{n!}}$$x^n$

Where the sum goes from $n=1$ to infinity. The book suggests calling the right side $f(x)$, proving the series converges, proving $(1+x)f'(x) = \alpha f(x)$. I've done all this. All that is left is solving this differential equation, which I simply cannot figure out. Could anyone help with this last step? It would be appreciated!

$\endgroup$
2
  • 1
    $\begingroup$ Do you know the identity $\frac{\mathrm d}{\mathrm dx}\log(f(x))=\frac{f^\prime(x)}{f(x)}$? $\endgroup$ Apr 7 '13 at 16:25
  • $\begingroup$ That was the one I needed! Totally slipped my mind. Thank you! $\endgroup$
    – user23658
    Apr 7 '13 at 16:35
3
$\begingroup$

To settle this: there is the formula for logarithmic differentiation,

$$\frac{\mathrm d}{\mathrm dx}\log(f(x))=\frac{f^\prime(x)}{f(x)}$$

which means

$$\frac{\mathrm d}{\mathrm dx}\log(f(x))=\frac{\alpha}{1+x}$$

Integrate both sides, mind the boundary condition, and you should be able to recover your needed result.

$\endgroup$
6
  • $\begingroup$ Can showing $(1+x)f'(x)=\alpha f(x)$ be proven a quicker way? I essentially used a long chain of algebra. $\endgroup$
    – user23658
    Apr 7 '13 at 17:45
  • $\begingroup$ Can you edit your question to summarize this "long chain" you speak of? $\endgroup$ Apr 7 '13 at 17:51
  • $\begingroup$ I started by using the limit definition of f'(x) with lim of x+1 going to zero but this feels like an odd way to approach it. I actually think it may be completely wrong. I'm a non-math student trying some Rudin problems so you'll have to forgive me if I'm missing something elementary. I appreciate the help! $\endgroup$
    – user23658
    Apr 7 '13 at 18:01
  • $\begingroup$ Ack. Are you already at the stage where you can justify when an infinite series can be differentiated term-by-term? (P.S. I'm a non-mathematician, too.) $\endgroup$ Apr 7 '13 at 18:06
  • $\begingroup$ I've used that if f(x) = $\sum{cn x^n}$ (n=0 to inf) converges for x $\in (-1,1)$, (which I have shown), then $f'(x) = \sum{n cn x^{n-1}}$ (n=1 to inf), if that's what you're referring to. $\endgroup$
    – user23658
    Apr 7 '13 at 18:16
1
$\begingroup$

For compactness, use Knuth's notation for falling factorial powers: $$ \alpha^{\underline{n}} = \alpha \cdot (\alpha - 1) \cdot \ldots \cdot (\alpha - n + 1) = \prod_{0 \le k \le n - 1} (\alpha - k) $$ Now use Maclaurin's formula: $$ \begin{align*} f(z) &= (1 + z)^\alpha \\ f^{(n)}(x) &= \alpha^{\underline{n}} (1 + z)^{\alpha - n} \end{align*} $$ This gives directly: $$ f(z) = \sum_{n \ge 0} \frac{f^{(n)}(0)}{n!} z^n = \sum_{n \ge 0} \frac{\alpha^{\underline{n}}}{n!} z^n $$ We can define: $$ \binom{\alpha}{n} = \frac{\alpha^{\underline{n}}}{n!} $$ to get the familiar: $$ (1 + z)^\alpha = \sum_{n \ge 0} \binom{\alpha}{n} z^n $$

$\endgroup$
1
$\begingroup$

Note that the differential equation $$f'-\frac{\alpha}{1+x}f=0$$ with initial conditions $f(0)=1$ has a unique solution.

This can be seen by taking $f,f_0$ two solutions and looking at $g=\frac{f}{f_0}$. You'll get that $$g'=\frac{ff_0^\prime-f'f_0}{f_0^2}$$

But then $$g'=\frac{1}{f_0^2} \left(f \frac{\alpha}{1+x}f_0-\frac{\alpha}{1+x}f f_0\right)=0$$

It follows $g$ is contant, and since $f(0)=f_0(0)=1$, $g=1$ so $f_0=f$. Thus, all you need to show is that $$f=\sum_{n=0}^\infty {\alpha\choose n}x^n$$ converges over $|x|<1$,that so does its derivative and that $(1+x)f'=\alpha f$. This last equality follows from $$n{\alpha\choose n}=\alpha{{\alpha-1}\choose n-1}$$

A much more tedious alternative is possible: you can show that the remainder goes to $0$ for $|x|<1$. To this end, first consdier $0\leq x <1$. Then $$R_{m,0}(x)=\frac{f^{m+1}(t)}{(m+1)!}x^{m+1}$$ for some $t\in(0,1)$. But if $f=(1+x)^{\alpha}$ then $$f^{(m+1)}=\prod_{k=0}^m (\alpha-k)(1+x)^{\alpha-m-1}$$

so that $$R_{m,0}(x)={\alpha\choose m+1}(1+t)^{\alpha-m-1}x^{m+1}$$

But $$0<(1+t)^{\alpha-m-1}\leq 1$$ for $\alpha<m+1$ It follows $\lim_{m\to\infty} R_{m,0}=0$ for any $x\in[0,1)$. Now suppose $-1<x<0$. By Cauchy's remainder formula. $$R_{m,0}=\frac{f^{(m+1)}(t)}{m!}(x-t)^m x$$ for some $x<t\leq 0$. Yet again $$f^{(m+1)}=\prod_{k=0}^m (\alpha-k)(1+x)^{\alpha-m-1}$$ so that $$R_{m,0}(x)=(m+1){\alpha\choose m+1}\left(\frac{x-t}{1+t}\right)^m(1+t)^{\alpha-1}x$$

But $$\left| {\frac{{x - t}}{{1 + t}}} \right| = \left| x \right|\left| {\frac{{1 - t/x}}{{1 + t}}} \right|$$ and since $-1<x<t< 0$, $0\leq t/x$ and $0\leq t+1< 1$ so $$\frac{1}{1+t}\geq 1$$ and $$\left| x \right|\left| {\frac{{1 - t/x}}{{1 + t}}} \right| \leqslant \left| x \right|$$

On the other hand, $$|x(1+t)^{\alpha-1}|\leq M_{\alpha,x}$$ where $M_{\alpha,x}=\max(1,(1+x)^{\alpha-1})$. Indeed, if $\alpha \geq 1$ then $0<1+x<1+t<1$ so $0<(1+x)^{\alpha-1}<(1+t)^{\alpha-1}<1$ and if $\alpha<1$ then $1< (1+t)^{\alpha-1}<(1+x)^{\alpha-1}$

It follows that $$\displaylines{ \left| {{R_{m,0}}\left( x \right)} \right| = \left( {m + 1} \right){\alpha\choose m+1}{\left| {\frac{{x - t}}{{1 + t}}} \right|^m}\left| {{{\left( {1 + t} \right)}^{\alpha - 1}}x} \right| \cr \leqslant \left( {m + 1} \right){\alpha\choose m+1}{\left| x \right|^{m + 1}}{M_{\alpha ,x}} \cr = \alpha {\alpha-1\choose m}{\left| x \right|^m}{M_{\alpha ,x}} \cr} $$

and this last quantity goes to zero when $m\to\infty$.

All in all $${\left( {1 + x} \right)^\alpha } = \sum\limits_{n = 0}^\infty{\alpha\choose m} {{x^n}} $$ over $|x|<1$.

$\endgroup$
2
  • $\begingroup$ Do you have any suggestions for proving the original equality? $(1+x)f'(x) = \alpha f(x)$. I think my original thinking danced around a clear proof $\endgroup$
    – user23658
    Apr 7 '13 at 17:48
  • $\begingroup$ @user23658 Yes. Use Pascal's Rule and the identity I remarked in the ODE based solution. $\endgroup$
    – Pedro Tamaroff
    Apr 7 '13 at 17:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.