How can we find the derivative of xqp, where p, q are positive integers?
We let p=qr to keep things simple.
Intuitively, the answer should be rxr−1, because this is the answer when r is an integer.
Do you remember the proof that the derivative of xn is nxn−1, where n is an integer?
We can use the binomial theorem here, too!
** Generalized Binomial Theorem **
This is the binomial theorem generalized when the exponent isn't an integer.
Basically we redefine (kn) when n is not an integer to mean k!n(n−1)...(n−k+1).
By redefining binomial coefficients like this, we obtain
(x+y)r=xr+rxr−1y+2!r(r−1)xr−2y2+3!r(r−1)(r−2)xr−3y3+....
Notice that this is an **infinite series **. When r is an integer, we will eventually have a r−r=0 factor, so the series isn't infinite. However, when r is not an integer, this doesn't happen.
Do you get the generalized binomial theorem?
Anyway, applying the generalized binomial theorem, we find
(x+h)r=xr+rxr−1h+2!r(r−1)xr−2h2+..
Then h(x+h)r−xr becomes
rxr−1+2!r(r−1)xr−2h+3!r(r−1)(r−2)xr−3h2 + ...
We want the limit of this as h approaches 0, so we are left with
rxr−1.
So we have shown the derivative of xr is rxr−1, not just for integers but for all rationals.
Does this make sense? You don't need to understand the proof of the generalized binomial theorem.
Now we know that dxdxn=nxn−1 when n is a positive integer. Is the same thing true for other numbers?
With some work, we can extend this rule to the case when n is a rational number: n=qp where p and q are integers. Here are three ways to do it.
Using implicit differentiation
If y=xp/q, then yq=xp. We can differentiate both sides to get
qyq−1dxdy=pxp−1.
Solving for dxdy and then substituting back y=xp/q yields:
Using the chain rule
Let f(x)=xp/q, g(x)=xq, and h(x)=xp. We've set this up so that g(f(x))=h(x). We also know that g′(x)=qxq−1 and h′(x)=pxp−1, but we don't (yet) know f′(x).
By using the chain rule, we get:
Note that we're using the chain rule on the composition g(f(x)) when we don't yet know that f has a derivative. So technically, all this approach tells us is that if f(x) is differentiable, this must be its derivative.
Using the limit definition
This approach is exceptionally painful, but it can be done. To make things slightly simpler, here is a way to do it for f(x)=x=x1/2.
f′(x)=h→0limhx+h−x=h→0limh⋅(x+h+x)(x+h−x)⋅(x+h+x)=h→0limh⋅(x+h+x)(x+h)−x=h→0limx+h+x1=2x1.
So f′(x)=2x1=21⋅x−1/2.
This approach used the "difference of squares" factorization a2−b2=(a+b)(a−b), with a=x+h and b=x, to convert the numerator into the form (x+h)−x=h. To differentiate x1/q, we could use the same approach with the factorization
aq−bq=(a−b)(aq−1+aq−2b+⋯+abq−2+bq−1).
Then the chain rule (applied in the usual direction this time) tells us the derivative of xp/q=(x1/q)p.
For f(x)=xqp, we can use logarithmic differentiation to find dxd(f(x)):
dxd(ln(f(x)))=dxd(qpln(x))
Using the chain rule, we get
f(x)1⋅dxd(f(x))=qp⋅x1