Last time | Next time |
We're on to Vol 1, Sec 3.7: Derivatives of Inverse functions
We'll approach this topic via several examples, making good use of the chain rule:
Notice that the equation \(F=T(C)\) is written in slope-intercept form, whereas the equation \(C=T^{-1}(F)\) is written in point-slope form.
Why does this make sense in this particular case, of these two temperature scales?
We might look at how this plays out in problem 266, for example.
(WARNING: don't be fooled by the notation: $f^{-1}(x)$ is the inverse function; $f(x)^{-1}$is the multiplicative inverse, $\frac{1}{f(x)}$. They are completely different animals!)
\[ f(x)=x^2 \] and \[ f^{-1}(x)=\sqrt{x} = x^{\frac{1}{2}} \]
This is an interesting place to start, because the square function isn't actually invertible!
By convention, we restrict the domain, and choose to consider the square function on only half of its domain (positive values of $x$).
And you can see what goes wrong if we try to reflect the part to the left of the $y$-axis: the inverse "function" would fail the vertical line test -- very naughty "function".
It's unfortunate that we, being lazy mathematicians, will typically use \(x\) as the variable in both of these formulas (as we see in this picture):
But we need to remember that \(f^{-1}\) operates on the range of f, \(R_f\), that is on the values of \(y=f(x)\), and takes them back to the domain of \(f\).
We begin by observing that \[ f(x)=x^2 \] \[ f^{-1}(y)=\sqrt{y}. \] and \[ f(f^{-1}(y))=y. \]
We assert that if the two functions are equal, then their derivatives must be equal as well (does that make sense?). So let's differentiate with respect to \(y\):
\[ \frac{d}{dy}\left(f(f^{-1}(y))\right) = \frac{d}{dy}(y) \] Then, using the chain rule on the left, \[ f'(f^{-1}(y)) \frac{d}{dy}\left(f^{-1}(y)\right) = 1 \] So \[ \frac{d}{dy}\left(f^{-1}(y)\right) = \frac{1}{f'(f^{-1}(y))} \] This is the key result, summarized in Theorem 3.11, where the same result is written in terms of the variable \(x\):
One other thing that they do in their theorem is give the name of \(g\) to \(f^{-1}\) -- not a bad idea, since that notation is so egregious.
Let's do an example: for $f(x)=x^2$, $f'(x)=2x$: so \[ f'(f^{-1}(y))=2f^{-1}(y) \] and \[ \frac{d}{dy}\left(f^{-1}(y)\right) = \frac{1}{2f^{-1}(y)} = \frac{1}{2\sqrt{y}} = \frac{1}{2}y^{-\frac{1}{2}} \] Re-written in terms of \(x\), where we're more familiar, we say that
\[ \frac{d}{dx} \left( f^{-1}(x) \right) = \frac{d}{dx} \left( \sqrt{x} \right) = \frac{d}{dx} \left(x^{\frac{1}{2}}\right) = \frac{1}{2}x^{-\frac{1}{2}} \] Just as we suspected!
This example I'll switch back to using \(x\) as the independent variable, just because we're a little more comfortable with that.
There will be enough other stuff to make us uncomfortable....:)
For $f(x)=\sin(x)$, $f'(x)=\cos(x)$ -- easy! But this time it will be a little more unpleasant to compute \[ f'(f^{-1}(x)) = \cos(\arcsin(x)). \]
There's a trick you know, however: we can rewrite cosine in terms of sine, using our most important trig identity: \[ \sin(x)^2+\cos(x)^2=1, \] so \[ \cos(x)=\pm\sqrt{1-\sin(x)^2}, \] Hence \[ \frac{d}{dx}\left(f^{-1}(x)\right) = \frac{1}{\pm\sqrt{1-\sin(\arcsin(x))^2}} \] becomes \[ \frac{d}{dx}\left(\arcsin(x)\right) = \frac{1}{\sqrt{1-x^2}} \] where you will notice that I chose the positive "branch". Why?
Look at sine on this interval. If it is increasing everywhere, its inverse must be increasing everywhere: and what does that say about its derivative?
There are two other functions which deserve our attention. We call their inverses "arctan" and "arccos".
Do exactly the same thing that we've just done in these three examples, but with two other important functions:
Observe that we could compute this derivative using a transformation: we've already observed that \[ \arccos(x)=\frac{\pi}{2}-\arcsin(x) \]
\[ \arccos'(x)=-\arcsin'(x) \]