I learned from a textbook that if $f(x)$ can be expressed as a function of $u(x)$, for example $$f(x) = u(x)^3,$$ and if $\delta f$, $\delta u$, $\delta x$ are small finite quantities, then $${\delta f \over \delta x} = {\delta f \over \delta u} {\delta u \over \delta x}.$$ As the quantities become infinitesimally small, we get $$\frac{df}{dx} = \frac{df}{du} \frac{du}{dx}.$$ The above definitions seems to suggest the the quantities ${\delta u}$ and $du$ have to cancel in the expressions.

I expanded the above definition using the gradient definition of a derivative into $${df \over dx}=\lim_{\delta u \to 0} {f(u+\delta u)-f(u) \over \delta u} \times \lim_{\delta x \to 0} {u(x+\delta x)-u(x) \over \delta x}$$

Does this imply that the $\delta u$ and $u(x+\delta x)-u(x)$ cancel each other?If that is so, why are they equal? It does not seem intuitive to me that they equal.

  • $\begingroup$ How do you define $\delta u$ ? $\endgroup$ – Yves Daoust Jun 13 '18 at 7:28
  • $\begingroup$ @TaeNyFan Please recall that if the OP is solved you can evaluate to accept an answer among the given, more details here meta.stackexchange.com/questions/5234/… $\endgroup$ – user Aug 4 '18 at 21:03

We can give such kind of not rigorous but intuitive explanation of chain rule as follow for

$$f(x) = f(u(x))$$


$$\Delta f\approx f'(u(x))\Delta u$$

$$\Delta u\approx u'(x)\Delta x$$


$$\Delta f\approx f'(u(x))u'(x)\Delta x\implies\frac {\Delta f}{\Delta x}\approx f'(u(x))u'(x)\implies \frac{d f}{dx}=f'(u(x))u'(x)=\frac{df}{du}\frac{du}{dx}$$


$${\delta f \over \delta x} = {\delta f \over \delta u} {\delta u \over \delta x}$$ is an ordinary relation between fractions and the $\delta u$ truly cancel each other as they are the same number.

When going to the limit, by the definition of the derivatives and the product rule,

$$\frac{df}{dx}=\lim_{\delta x\to0}{\delta f \over \delta x} = \lim_{\delta x\to0}{\delta f \over \delta u} {\delta u \over \delta x}=\lim_{\color{blue}{\delta x\to0}}{\delta f \over \delta u}\lim_{\delta x\to0} {\delta u \over \delta x}=\lim_{\color{blue}{\delta u\to0}}{\delta f \over \delta u}\lim_{\delta x\to0} {\delta u \over \delta x}=\frac{df}{du}\frac{du}{dx}.$$

The subsitution of $\delta x$ for $\delta u$ (in blue) is possible because $\delta u$ and $\delta x$ go to zero simultaneously.

  • $\begingroup$ This argument is fine as long as $\delta u$ doesn't vanish. If it does (e.g. $u$ is constant), this simple proof of the chain rule fails although the rule remains true. Is there a neat way to handle this? $\endgroup$ – Joonas Ilmavirta Jun 13 '18 at 17:13
  • $\begingroup$ @JoonasIlmavirta: the rule remains true, but the rule $df/dx=(-df/du+\pi r^2\cos x)\,du/dx$ is true as well. There is no real point keeping the theorem in this case. $\endgroup$ – Yves Daoust Jun 13 '18 at 18:49

By definition, you have

\begin{align} \delta f(u) &= f(u + \delta u) - f(u) \\ \delta u(x) &= u(x + \delta x) - u(x) \end{align}

The fact that the ratio cancels is immediately obvious here

$$ \frac{df}{du}\frac{du}{dx} = \lim_{\delta u \to 0}\frac{f(u + \delta u) - f(u)}{u(x+\delta x) - u(x)}\cdot \lim_{\delta x \to 0} \frac{u(x + \delta x) - u(x)}{\delta x} = \lim_{\delta x\to 0} \frac{f(u+\delta u)-f(u)}{\delta x} $$

The part you should care more about is going back from $u$ to $x$

\begin{align} f(u(x)) &= f(x) \\ f(u(x) + \delta u(x)) &= f\big(u(x+\delta x)\big) = f(x + \delta x) \end{align}


$$ \lim_{\delta x\to 0} \frac{f(u+\delta u)-f(u)}{\delta x} = \lim_{\delta x \to 0} \frac{f(x+\delta x) - f(x)}{\delta x} = \frac{df}{dx} $$


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.