4
$\begingroup$

Suppose we have two independent functions $f(u)$ and $g(v)$ then

(i) :- $$\frac {d \;(f(u)\cdot g(v))}{dx} = \frac {d (f(u)\cdot g(v))}{du}\cdot\frac{du}{dx} $$

Since $g(v)$ doesn't change with changing $u$ $$\frac {d (g(v))}{du} = 0$$

So the first equation becomes

(ii) :- $$\frac {d \;(f(u)\cdot g(v))}{dx} = g(v) \cdot f'(u) \cdot\frac{du}{dx} $$

Similarly we can get this equation

(iii) :- $$\frac {d \;(f(u)\cdot g(v))}{dx} = f(u)\cdot g'(v)\cdot \frac{dv}{dx} $$

Now if I add both the equations I will get

$$ \frac {d \;(f(u)\cdot g(v))}{dx} = \left( \frac{1}{2} \right) \left(g(v)\cdot f'(u)\cdot\frac{du}{dx} + f(u)\cdot g'(v)\cdot\frac{dv}{dx}\right)$$

But this can't be right since it is half of what we get from the product rule.

So where am I wrong ?

$\endgroup$
4
  • $\begingroup$ The equations aren't organised with the equality sign . So if it's possible please edit it so that the equations fit in a single line. $\endgroup$
    – Ankit
    Jul 20, 2021 at 4:18
  • $\begingroup$ What do you mean by independent? There seems to be a misunderstanding in differentiating a composition of two functions (a.k.a. chain rule), so this might be the problem. $\endgroup$ Jul 22, 2021 at 8:13
  • $\begingroup$ There shoundn't be the $\dfrac12$ in the final equation. You should just add those two instead of averaging out. Think about differentiating $x^2=x\cdot x$ with the multiplication rule, i.e. $(x\cdot x)'=x\cdot x'+x'\cdot x=2x$. $\endgroup$
    – Kay K.
    Jul 22, 2021 at 9:55
  • $\begingroup$ What you really need is the mutlivariable chain rule, trying to work in MV w/o the proper MV tiools will lead to a lot of confusion $\endgroup$ Jul 23, 2021 at 5:25

4 Answers 4

3
$\begingroup$

Your derivation is based on the assumption that, $$\dfrac{d\left(f(u)g(v)\right)}{dx}=\dfrac{d\left(f(u)g(v)\right)}{du}\dfrac{du}{dx}=\dfrac{d\left(f(u)g(v)\right)}{dv}\dfrac{dv}{dx}$$
But, $$ \dfrac{d\left(f(u)g(v)\right)}{du}\dfrac{du}{dx}\;\text{ doesn't necessarily be equal to }\;\dfrac{d\left(f(u)g(v)\right)}{dv}\dfrac{dv}{dx}.$$

For example, let $f(u) = u^2, g(v) = v^3+v, u(x) = \sin (x), v(x) = x+1$.
Here,
$$ \dfrac{d\left(f(u)g(v)\right)}{du}\dfrac{du}{dx} =g(v)\dfrac{d\left(f(u)\right)}{du}\dfrac{du}{dx}= (v^3+v)(2u)\cos(x) $$ but, $$ \dfrac{d\left(f(u)g(v)\right)}{dv}\dfrac{dv}{dx} =f(u)\dfrac{d\left(g(v)\right)}{du}\dfrac{dv}{dx}=u^2(3v^2+1)(1) $$
And both aren't the same.


What you have to do is,
$$ \dfrac{d\left(f(u)g(v)\right)}{dx} = g(v)\dfrac{d\left(f(u)\right)}{dx} + f(u)\dfrac{d\left(g(v)\right)}{dx}=g(v)\dfrac{d\left(f(u)\right)}{du}\dfrac{du}{dx} + f(u)\dfrac{d\left(g(v)\right)}{dv}\dfrac{dv}{dx}$$


Moreover, if $$ \begin{split}& g(v)\dfrac{d\left(f(u)\right)}{du}\dfrac{du}{dx}=f(u)\dfrac{d\left(g(v)\right)}{dv}\dfrac{dv}{dx}\\ \implies&\dfrac{f'(u)du}{f(u)} =\dfrac{g'(v)dv}{g(v)}\\ \implies& \ln|f(u)| = \ln|kg(v)| \\ \implies& \boxed{f(u) = kg(v)} \end{split} $$

But $f(u)$ and $g(v)$ are dependent only on $u$ and $v$ respectively.

Both $f(u)$ and $g(v)$ have to be constant functions for this condition to occur and in this case the total derivative will be zero, as $f'(u)$ and $g'(v)$ are zeroes. So in this particular condition, it doesn't matter whether you have half the formula or the exact formula.

$\endgroup$
2
  • $\begingroup$ But why can't I differentiate them separately with respect to two different variables ?? $\endgroup$
    – Ankit
    Jul 20, 2021 at 6:58
  • $\begingroup$ You can do, but that won't be equal to $\frac{d(fg)}{dx}$ $\endgroup$
    – 19aksh
    Jul 20, 2021 at 7:07
1
+25
$\begingroup$

Let us make the implicit assumptions explicit, you are assuming that $u(x)$ and $v(x)$ which means that as you change a parameter $x$, there are two other functions which change according to it.

Physical example: You pump in air (volume) into a balloon, then as you increase it, both the mass of the balloon and density of balloon changes. Now, you may have other two function dependent on mass and density of balloon.

Now, we need to compute $\frac{d}{dx} \left[ g(v) f(u) \right]$ , by the multivariable chain rule:

$$ \frac{d}{dx} g(v) f(u) = \frac{\partial g}{\partial v} \frac{dv}{dx} f(u) + \frac{\partial f}{\partial u} \frac{du}{dx} g(v)$$

To reduce this to the single variable product rule we consider the fact that $g$ is only a function of $v$ and $f$ is only a function of $u$:

$$ \frac{d}{dx} g(v) f(u) = g'(v) f(u) \frac{dv}{dx} + f'(u) \frac{du}{dx} g(v)$$

The general MV chain rule for two variable is given as:

$$ \frac{d}{dt} F(u(t) , v(t) ) = \frac{\partial F}{\partial u} \frac{du}{dt} + \frac{\partial F}{\partial v} \frac{dv}{dt}$$

$\endgroup$
1
  • $\begingroup$ Khan Academy has a good article on the intuition behind this. $\endgroup$ Jul 23, 2021 at 5:25
1
$\begingroup$

I think that this statement is wrong:

Since $g(v)$ doesn't change with changing $u$

if $u=\phi(x)$, then we can probably partition its domain in such a way that we can always find $x=\phi^{-1}(u)$ for a given partition.

since $v=\psi(x)$ and $x=\phi^{-1}(u)$, we get $v=\psi(\phi^{-1}(u))=\xi(u)$.

$\endgroup$
1
$\begingroup$

If your method were right, you could also subtract your partial result to arrive at $$g(v) \cdot f'(u) \cdot\frac{du}{dx}-f(u)\cdot g'(v)\cdot \frac{dv}{dx}=0$$ which is certainly not generally true. What your partial results actually compute, is in fact not $\frac {d \;(f(u)\cdot g(v))}{dx}$ in both cases, but only a partial contribution to$\frac {d \;(f(u)\cdot g(v))}{dx}$, and you have to really add these contributions, not take their average.


By the end of the day, this (or the product rule per se) is just the (multidimensional) chain rule, which circumvents the magic of keeping partial results and combining them by having several dimensions available; in short, the derivative of $\pmatrix{x_1\\\vdots\\x_n}\mapsto \pmatrix{f_1(x_1,\ldots,x_n)\\\vdots\\f_m(x_1,\ldots,x_n)}$ is $\pmatrix{\frac{\partial}{\partial x_1}f_1&\cdots&\frac{\partial}{\partial x_n}f_1\\\vdots&\ddots&\vdots\\\frac{\partial}{\partial x_1}f_m&\cdots&\frac{\partial}{\partial x_n}f_m}$ and we conveniently identify $1\times 1$ matrices with scalars:

We have the map sequence $$ x\mapsto \pmatrix{u(x)\\v(x)}\mapsto \pmatrix{f(u(x))\\g(v(x))}\mapsto f(u(x))g(v(x)).$$

  • The derivative of the first step $x\mapsto\pmatrix{u(x)\\v(x)}$ is $\pmatrix{u'(x)\\v'(x)}$,
  • the derivative of the second step $\pmatrix{u\\v}\mapsto \pmatrix{f(u)\\g(v)}$ is $\pmatrix{f'(u)&0\\0&g'(v)}$ (where the zeroes indicate the "independent functions" property), or evaluated at $\pmatrix{u(x)\\v(x)}$: $\pmatrix{f'(u(x))&0\\0&g'(v(x))}$
  • the derivative of the last step $\pmatrix{p\\q}\to pq$ is $(q\;p)$, or evaluated at $\pmatrix{f(u(x))\\g(v(x))}$: $(g(v(x))\;f(u(x)))$.

Multiplying all out gives $$ (g(v(x))\;f(u(x)))\pmatrix{f'(u(x))&0\\0&g'(v(x))}\pmatrix{u'(x)\\v'(x)}= g(v(x))f'(u(x))u'(x)+f(u(x))g'(v(x))v'(x)$$

$\endgroup$

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.