In an online lecture (link to Youtube), the professor proves the Chain Rule using the following statement:

$$\frac{\Delta y}{\Delta x} = \frac{dy}{dx}\biggr|_{x=x_1} + k$$

$$\Delta y = \frac{dy}{dx}\biggr|_{x=x_1} \Delta x+ k\Delta x$$ where $$\lim_{\Delta x \to0}k=0.\\[0.4in]$$

Then he says that $x$ and $y$ are functions of a third variable $t$, so:

$$\begin{align*} \frac{\Delta y}{\Delta t} &= \frac{dy}{dx}\biggr|_{x=x_1} \frac{\Delta x}{\Delta t}+ k\frac{\Delta x}{\Delta t}\\[0.2in] \lim_{\Delta t \to 0}\frac{\Delta y}{\Delta t} &=\lim_{\Delta t \to 0} \frac{dy}{dx}\biggr|_{x=x_1} \lim_{\Delta t \to 0}\frac{\Delta x}{\Delta t}+ \lim_{\Delta t \to 0}k \lim_{\Delta t \to 0}\frac{\Delta x}{\Delta t}\\[0.2in] \frac{dy}{dt} &= \frac{dy}{dx} \frac{dx}{dt}+ 0\cdot\frac{dx}{dt} \end{align*}$$

The part I don't understand is, when he says that the last term in the last line, is not $0$ because of $k$, but also because as $\Delta t \to 0$, $dx$ approaches $0$ as well.

But why? I don't understand how $dx$ approaches zero as $dt$ approaches $0$? Shouldn't it be the other way around? That $dx/dt$ should increase without bound as $dt$ approaches $0$?

Also, another question... how are we allowed to simply define $\lim_{\Delta x \to0}k=0$? Although it makes sense graphically, but since $k$ is a number, doesn't that violate the fact that $\lim_{\Delta x \to0}c=c$? is it something to do with $\Delta x$?


Ultimately, I think the reason is that $x$ is a differentiable function of $t$ which particularly means that as $\Delta t\rightarrow 0$, $\Delta x\rightarrow 0$. That said I really don't like this approach to the chain rule. It is a bit cobbled together.

I feel that this is the better way to do chain rule (which is much clearer from the get go, I think). Let $f$ and $g$ be differentiable and $g$ non-constant. If $g$ is the constant function, then clearly $(f\circ g)'$ is zero since $f\circ g$ would be constant. So trivially it is true that $(f\circ g)' = f'(g(x))g'(x)$ since $g' = 0$. I'll assume they're both defined on all of $\mathbb{R}$ to avoid annoyances with domains and ranges.

Then we wish to evaluate $(f\circ g)'(x)$. From the definition of the derivative, we have

$$ (f\circ g)'(x) = \frac{d}{dx}f(g(x)) = \lim_{\Delta x\rightarrow 0} \frac{f(g(x+\Delta x))-f(g(x))}{\Delta x}.$$

Let's multiply by a clever form of $1$ to make things easier on ourselves, particularly we'll multiply by

$$\frac{g(x+\Delta x)-g(x)}{g(x+\Delta x)-g(x)}.$$

This is where I required that $g$ be non-constant. If it were constant, the above expression would make no sense since we would be dividing $0$ by $0$. Note that this resembles the terms inside of $f$. This is not by accident. Additionally we have that $g(x+\Delta x)\approx g(x)+g'(x)\Delta x$ for small $\Delta x$ (this is what the derivative is for - linear approximations). If $g'(x) = 0$, then $g(x+\Delta x)\approx g(x)$ and in this case, we have

$$\lim_{\Delta x\rightarrow 0}\frac{f(g(x+\Delta x))-f(g(x))}{\Delta x} = 0.$$

Again, this is clearly equal to $f'(g(x))g'(x)$ since $g' = 0$. If $g'(x)\neq 0$, we can make use of our clever form of $1$ to get

$$(f\circ g)'(x) = \lim_{\Delta x\rightarrow 0}\frac{f(g(x+\Delta x))-f(g(x))}{g(x+\Delta x)-g(x)}\frac{g(x+\Delta x)-g(x)}{\Delta x}.$$

Now both pieces look eerily like a derivative (which is what we want), except that the first piece has $g$ in it. However if $\Delta x\rightarrow 0$, we know that $g(x+\Delta x)\rightarrow g(x)$ since $g$ is differentiable (and therefore continuous). Clearly $\lim_{\Delta x\rightarrow 0}\frac{g(x+\Delta x)-g(x)}{\Delta x}$ exists since $g$ is differentiable so we need only to argue that $\lim_{\Delta x\rightarrow 0}\frac{f(g(x+\Delta x)-f(g(x))}{g(x+\Delta x)-g(x)}$ is well-defined. By limit theorems we know that if both limits exists, we can distribute the limit to each piece and evaluate.

So we want to argue that

$$\lim_{\Delta x\rightarrow 0}\frac{f(g(x+\Delta x))-f(g(x))}{g(x+\Delta x)-g(x)}$$

is well-defined. Using our approximation for $g(x+\Delta x)$ from above, we have

$$\lim_{\Delta x\rightarrow 0}\frac{f(g(x)+g'(x)\Delta x)-f(g(x))}{g(x)+g'(x)\Delta x-g(x)}.$$

Cancelling appropriate terms we have

$$\lim_{\Delta x\rightarrow 0}\frac{f(g(x)+g'(x)\Delta x)-f(g(x))}{g'(x)\Delta x}.$$

Repeating the same logic as above with $f(g(x)+g'(x)\Delta x)$, we have that $f(g(x)+g'(x)\Delta x)\approx f(g(x))+f'(g(x))g'(x)\Delta x$. And so we get

$$\lim_{\Delta x\rightarrow 0}\frac{f(g(x))+f'(g(x))g'(x)\Delta x-f(g(x))}{g'(x)\Delta x} = f'(g(x)).$$

Since the limit of the first piece makes sense and the limit of the second piece makes sense, we can distribute the limits to get that

$$(f\circ g)'(x) = \lim_{\Delta x\rightarrow 0}\frac{f(g(x+\Delta x))-f(g(x))}{g(x+\Delta x)-g(x)}\lim_{\Delta x\rightarrow 0}\frac{g(x+\Delta x)-g(x)}{\Delta x} = f'(g(x))g'(x)$$

by our above calculations. So in each case that emerged we had that $(f\circ g)'(x) = f'(g(x))g'(x)$ and so we conclude that the chain rule holds.


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.