5

Firstly, yes, I know how to expand expressions like $(a+b)^{({1}/{2})}$ but I want a rigorous proof of why it is okay to replace binomial coefficients like ${1/2}\choose{2} $ with $({1/2})({1/2-1}))/2$

(Of course, I already understand the proof of binomial theorem for integers using combinatorics or induction)

A similar question was posed on this site earlier, but I don't feel the answers really get to the bottom of my doubt:

Binomial theorem for non integers ? O_o ??

For example, one of the answers cited Taylor's theorem from calculus. My problem with that is that my understanding of derivatives of monic polynomials mainly comes from using the binomial theorem to expand $(x+\epsilon)^n$ for arbitrarily small $\epsilon$

Thus, this felt like a circular proof to me.

So I would either like:

A basic proof of the binomial theorem without derivatives

OR

An alternative understanding of derivatives of $x^n$ without binomial expansions

Daniel Smania
  • 10,152
aman
  • 1,414
  • The rule $\dfrac{d}{dx}x^{\alpha} = \alpha x^{\alpha-1}$ works for all real numbers $\alpha$ (and whenever $x^{\alpha}$ is defined). Note that if the exponent $\alpha$ is not an integer, then one of the ways to define it is $x^{\alpha} := e^{\alpha \ln(x)}$ (so we require $x > 0$). So, applying Taylor's theorem is definitely not circular, once you have properly established a definition for the symbol $x^{\alpha}$ for positive $x$, and any real number $\alpha$. – peek-a-boo May 02 '20 at 06:05
  • I don't understand how the $nx^{n-1}$ works for non-integers. The proof for this expression relies on binomial theorem right? – aman May 02 '20 at 06:08
  • Yes, I guess I don't really understand what raising a number to any real power means. Can you help me with this? – aman May 02 '20 at 06:10
  • 1
    That's the proof of the baby version, which is usually presented first in elementary classes. But it is true in general. The first question you should be asking isn't "why is $\dfrac{d}{dx}(x^n) = nx^{n-1}$". It should be "what's the meaning of $x^n$ if $n$ is not an integer or rational number". And for this, refer to my previous comment. One has to first define the exponential function $\exp: \Bbb{R} \to [0, \infty)$ (either using a power series definition, or for example, by first defining the logarithm, $\ln(x):= \int_1^x \dfrac{1}{t} , dt$, then defining $\exp$ to be the inverse of $\ln$) – peek-a-boo May 02 '20 at 06:12
  • in any case, once you define $\exp$ and $\ln$, you can then define the symbol $x^{\alpha}$, for any positive number $x$, and any real number $\alpha$. Then, you can apply the chain rule to prove that differentiation rule. Anyway, the questions you're asking are very good, but they're not easy to answer if you don't have the proper calculus background to understand the various definitions. But if you do, then you should take a look at any good calculus text, for example, Michael Spivak's Calculus. – peek-a-boo May 02 '20 at 06:16
  • you define $x^{\alpha} := e^{\alpha \ln(x)}$ (i.e the RHS is a known quantity, and you're using it to define the LHS). And in the case of integer $\alpha$, this will coincide with the usual definition – peek-a-boo May 02 '20 at 06:21
  • Here is my understanding so far: You first define $ln$ as the function with an instantaneous rate of change as $1/x$ Then, you define $e^x$ to be the inverse of that function. Now, you know nothing whatsoever about the properties of this function (why is it its own derivative, etc) So how do you move on to differentiating $x^n$ – aman May 02 '20 at 06:23
  • In fact, how do you even know that $x^n=e^{nln(x)}$ from this definition... – aman May 02 '20 at 06:25
  • you have to establish all of those results along the way, in a systematic fashion of course. I'm not claiming these results are obvious by any means. There is an entire chapter of a textbook devoted to this, so there's far too much information for me to summarize in a short stackexchange discussion. I really suggest you to take a look at the textbook I referenced. All your questions will be answered there. – peek-a-boo May 02 '20 at 06:27
  • 1
    Ok I don't know if I'm ready for it though. Wouldn't it be easier to define $x^k$ as the limit of the products of the exponents in decimal notation like $x^\pi=x^3*x^{0.1}...$ – aman May 02 '20 at 06:59
  • 1
    Then one can somehow figure out derivative of $x^{1/n}$ and using chain rule $x^{a/b}$ From this we can work for some kind of limit? Maybe I am being too simplistic? – aman May 02 '20 at 07:02
  • There's no easy way to show this for fractional exponents without calculus. – Allawonder May 02 '20 at 07:10
  • You can read various approaches to general binomial theorem here and here. A typical proof (first link) uses the derivative formula for $x^n$ which holds for all real $n$. Second link is based on more intuitive idea. – Paramanand Singh May 02 '20 at 16:05
  • For derivative of $x^n$ with rational $n$ see https://math.stackexchange.com/a/1782225/72031 – Paramanand Singh May 02 '20 at 16:10
  • Your approach to definition of $x^n$ when $n$ is irrational is handled in this post. – Paramanand Singh May 02 '20 at 16:14
  • @peek-a-boo: For what it's worth, the strategy of defining $\ln$ first is fleshed out in this answer. – Gavin R. Putland Aug 09 '25 at 07:36

4 Answers4

2

Proof of derivative of xn for real n without binomial expansion?

From the definition of the derivative, we readily establish that $$\tfrac{d}{dx}(uv) = \tfrac{du}{dx}\;\!v + u\;\!\tfrac{dv}{dx} \qquad \text{[product rule]},$$ $$\tfrac{d}{dx}\Big(y\big(u(x)\big)\Big) = \tfrac{dy}{du}\tfrac{du}{dx} \qquad \text{[chain rule]},$$ $$\tfrac{d}{dx}\big(x^0\big) = 0$$ and $$\tfrac{d}{dx}\big(x^1\big) = 1 \,.$$

Higher integer powers can then be handled by repeated application of the product rule: $$\tfrac{d}{dx}\big(x^2\big) = \tfrac{d}{dx}(xx) = \tfrac{d}{dx}(x)\;\!x + x\;\!\tfrac{d}{dx}(x) = 1\;\!x + x\;\!1 = 2x \,;$$ $$\tfrac{d}{dx}\big(x^3\big) = \tfrac{d}{dx}(x^2x) = \tfrac{d}{dx}\big(x^2\big)\;\!x + x^2\tfrac{d}{dx}(x) = 2x\;\!x + x^2\;\!1 = 3x^2 \,;$$ $$\tfrac{d}{dx}\big(x^4\big) = \tfrac{d}{dx}(x^3x) = \tfrac{d}{dx}\big(x^3\big)\;\!x + x^3\tfrac{d}{dx}(x) = 3x^2x + x^3\;\!1 = 4x^3 \,;$$ and in general, for non-negative integers $n$, \begin{equation}\label{grpe1}\tag{1} \tfrac{d}{dx}\big(x^n\big) = nx^{n-1} . \end{equation} Confirmation of (\ref{grpe1}) for integers ${n\!\gt\!4}$: \begin{align} \frac{\xi^n - x^n}{\xi-x} &= \frac{(\xi-x)\big(\xi^{n-1} + \xi^{n-2}x + \cdots + \xi\:\!x^{n-2} + x^{n-1}\big)}{\xi-x}\\[1ex] &= \xi^{n-1} + \xi^{n-2}x + \cdots + \xi\:\!x^{n-2} + x^{n-1} . \end{align} As this has $n$ terms, taking limits as ${\xi\!\to\!x}$  gives (\ref{grpe1}).

Negative integer powers: If $n$ is a positive integer (and ${x\!\ne\!0}$), $$x^n x^{-n} = 1 \,.$$ Differentiating both sides w.r.t. $x$, using the product rule and rule (\ref{grpe1}) on the left, we get $$nx^{n-1}x^{-n} + x^n\tfrac{d}{dx}\big(x^{-n}\big) = 0 \,,$$ whence $$\tfrac{d}{dx}\big(x^{-n}\big) = -nx^{-n-1},$$ which is rule (\ref{grpe1}) with $-n$ in place of $n$.

Rational powers:  If  $p$ and $q$ are integers with ${q\!\ne\!0}$  (and ${x\!>\!0}$), $$\big(x^{p/q}\big)^q = x^p .$$ Differentiating both sides w.r.t. $x$, using the chain rule and rule (\ref{grpe1}) on the left, we get $$q\big(x^{p/q}\big)^{q-1} \tfrac{d}{dx}\big(x^{p/q}\big) = px^{p-1} ,$$ whence $$\tfrac{d}{dx}\big(x^{p/q}\big) = \tfrac{\,p\,}{q}\;\!x^{p/q\,-1},$$ which is rule (\ref{grpe1}) with $p/q$ in place of $n$.

Real powers: An irrational value of $n$ is the limit of a sequence of rational values. Hence $x^n$ for the irrational $n$ is the corresponding limit of $x^n$ for the rational values. If we accept the premise that the derivative of this limit is the limit of the derivatives, the rest of the argument leading to (\ref{grpe1}) is straightforward. The "premise" is discussed under the question "What is the derivative of xn ?"

Having obtained rule (\ref{grpe1}) with a certain generality without using any form of the binomial theorem, we can then use it to prove the binomial theorem with the same generality without tacitly arguing in a circle.

0

Enough is to prove that for $|x|<1$ $$\frac{1}{(1-x)^a}\stackrel{?}{=}\sum_{n=0}^{\infty}\frac{(a)_n}{n!}x^n=f_a(x)\ \ \ \ (*)$$ where $(a)_n=a(a+1)(a+2)\ldots (a+n-1)$ -with $(a)_0=1.$ Now since $$(a+1)_n-n(a+1)_{n-1}=(a)_n$$ we get $f'_a(x)=af_{a+1}(x),$ implying the differential equation $(1-x)f'_a(x)=af_a(x).$ Solving it and using $f_a(0)=1$ gives (*).

0

Most Calculus books do not rigorously define and prove the main properties of basic functions such as $\sqrt{x}$, $\sin(x)$, and $e^x$. The reason is that this is not simple. For instance, one may ask how a computer or calculator actually evaluates these functions. One way is by using power series, which usually appear later in Calculus. There are, however, many other approaches, some of them even more elementary.

I will give a deduction of the Binomial formula for $(1+x)^{1/2}$ using only basic facts about power series (we will not use Taylor series or the logarithm explicitly). The idea is to invert the argument: we first define a function $f_{\alpha}$ through the generalized binomial coefficients for $\alpha\in \mathbb{R}$, and then we prove at the end that $f_{1/n}(x) = (1+x)^{1/n}$ for $|x|<1$ and $n \in \mathbb{N}^\star$.

Define

$$ \binom{\alpha}{k} = \frac{\alpha(\alpha-1)(\alpha-2)\cdots(\alpha-(k-1))}{k!}, \qquad k \in \mathbb{N}, \ \alpha \in \mathbb{R}. $$

Note that if $\alpha \in \mathbb{N}$ and $\alpha \geq k$ this coincides with the usual definition of the binomial coefficients.

Define

$$ f_\alpha(z) = \sum_{k=0}^\infty \binom{\alpha}{k} z^k. $$

Our goal is to show that

(1) This function is well-defined for $|z|<1$ (the power series converges) and $f_\alpha(z) > 0$ for $|z|<1$. In particular derivatives of arbitrary order exist on this interval.

(2) If $n \in \mathbb{N}^\star$ then $f_n(z) = (1+z)^n$.

(3) If $n \in \mathbb{N}^\star$ then $(f_{1/n}(z))^n = 1+z$.

This allows us to define for $x \in (0,2)$

$$ x^{1/2} := f_{1/2}(x-1) $$

and to prove that $x^{1/2}$ is differentiable of arbitrary order, and to use the chain rule to calculate all its derivatives at $x=1$ without circular arguments.

Step 1. Radius of convergence.

  • If $\alpha \notin \mathbb{N}$ apply the ratio test to conclude that the radius of convergence is $R=1$.

  • If $\alpha \in \mathbb{N}$ then the product $\alpha(\alpha-1)(\alpha-2)\cdots(\alpha-(k-1))$ vanishes once $k>\alpha$. Therefore $f_\alpha$ is a polynomial. In this case the radius of convergence is $+\infty$.

Step 2. Functional relation.

$$ (1+z) f_\alpha(z) = f_{\alpha+1}(z), \qquad |z|<1. $$

Hint: expand both sides as power series and check that the coefficients match.

Step 3. For $n \in \mathbb{N}^\star$ we have

$$ f_n(z) = (1+z)^n. $$

Hint: use induction on $n$, Step 2, and the fact that $f_1(z) = 1+z$. This gives the Binomial Theorem for $(1+z)^n$ with $n \in \mathbb{N}$.

Step 4. Derivative relation.

$$ f_\alpha'(z) = \alpha f_{\alpha-1}(z), \qquad |z|<1, \ \alpha \neq 0. $$

Hint: if $\alpha \notin \mathbb{N}$ differentiate the series term by term and compare coefficients. If $\alpha \in \mathbb{N}$ use Step 3.

Step 5. Differential equation.

$$ (1+z) f_\alpha'(z) = \alpha f_\alpha(z), \qquad |z|<1. $$

This is a first-order differential equation satisfied by $f_\alpha$.

Hint: combine Steps 2 and 4.

Step 6. If $\alpha \in \mathbb{N}^\star$ then

$$ (f_{1/\alpha}(z))^\alpha = 1+z, \qquad |z|<1. $$

In particular, taking $\alpha=2$ we obtain $f_{1/2}(z) = \sqrt{1+z}$.

Proof: define

$$ h(z) = (f_{1/\alpha}(z))^\alpha. $$

By Step 5 we get

$$ h'(z) = \alpha (f_{1/\alpha}(z))^{\alpha-1} \cdot \frac{1}{\alpha} \frac{\left(f_{1/\alpha}(z)\right)^\alpha}{1+z} = \frac{h(z)}{1+z}. $$

Thus

$$ (1+z) h'(z) - h(z)=0, $$

so since $h(0)=1$ and

$$\left(\frac{h(z)}{1+z} \right)'= \frac{(1+z) h'(z) - h(z)}{(1+z)^2} =0$$

we conclude that $h(z)=1+z$.

Step 7. For all $\alpha\in \mathbb{R}^\star$ we have $$f_\alpha(z)f_{-\alpha}(z)=1, \qquad |z|<1.$$

Hint: Differentiate the left-hand side using the product rule and Step 5, and conclude that its derivative is zero.

In particular $f_\alpha(z)\neq 0$ for $|z|< 1$ and since $f_\alpha$ is continuous on $(-1,1)$ and $f_\alpha(0)=1> 0$ it follows that $f_\alpha(z) > 0$ for $|z|< 1$.

Daniel Smania
  • 10,152
0

As I recall, Newton proved it without calculus. It actually helped develop it.

$(x+a)=(c_0+c_1x+c_2x^2+c_3x^3+...)(c_0+c_1x+c_2x^2+c_3x^3+...)$

$(c_0^2+c_1^2x^2+c_2^2x^4+...)+(2c_0c_1x+2c_0c_2x^2+2c_0c_3x^3+...+ 2c_1c_2x^3+2c_1c_3x^4+...)$

$(x+a)=\sum_{k=0}c_k^2x^{2k}+2\sum_{i\ne j}c_ic_jx^{i+j}$

$c_0^2=a $

$2c_1\sqrt{a}=1\implies c_1=\frac{1}{2\sqrt{a}}$

$c_1^2+2c_0c_2=0\implies \frac{1}{4a}+2\sqrt{a}c_2=0\implies c_2=\frac{-1}{8a\sqrt{a}}$

$2(c_0c_3+c_1c_2)=0\implies c_3\sqrt{a}-\frac{1}{16a^2}\implies c_3 = \frac{1}{16a^{5/2}}$

$c_2^2+2(c_0c_4+c_1c_3)=0 \implies \frac{1}{64a^3}+2c_4\sqrt{a}+\frac{2}{2\sqrt{a} \cdot 16\cdot a^{5/2}}\implies c_4=\frac{-5}{128a^{7/2}}$

So for every higher coefficient, we have a linear equation expressing the new coefficient in terms of ones that came before, allowing an iterative process to determine the coefficients.

Induction can be proven to show the general pattern, then that in turn can be expressed in terms of the generalized binomial coefficient.

TurlocTheRed
  • 6,679