Proof of the Chain Rule


By Louise Johnson on Wednesday, September 19, 2001 - 09:00 pm:

Where can I find the proof that dy/dx can be treated as a fraction?


By David Loeffler on Monday, October 01, 2001 - 09:56 am:

It can't. You have to prove the chain rule by entirely different methods; although the initial premises and the end result may look like you've treated it as a fraction, in fact you haven't. If that makes sense.

David


By Louise Johnson on Friday, October 05, 2001 - 07:21 pm:

It does. Thankyou. Still where can I find the proof?


By Tristan Marshall on Thursday, October 11, 2001 - 03:34 pm:
Ok, here goes:

If u=g(x) define du/dx as: the limit as δx0 of:

g(x+δx)-g(x))/δx

You should have seen this definition before. The key is the precise meaning of 'limit'. Here it means that, for a given value of x, the above expression can be made as close as we like to x. (This is, roughly, the precise definition of 'limit')

More precisely, given ε>0, we can find δx such that the above expression is within ε of du/dx.

From now on I'll write g'(x) for du/dx etc.

Ok, now for the proof:

Let u=g(x), y=f(u). We want to prove that

dy/dx=(dy/du).(du/dx)

Let δx be a change in x, and δu and δy be the corresponding changes in u and y.

By playing around with the limit expression above, we get:

δu=(g'(x)+ε)δx

where ε depends on x and δx and 0 as δx0.

Similarly for y:

δy=(f'(u)+ε')δu

where, as before, ε'0 as δu0.

Combining this with the expression for δu, we have:

δy=(f'(u)+ε')(g'(x)+ε)δx

NOTE: This expression is EXACT, once ε and ε' have been found, and these are uniquely defined by the value of δx.

So now divide both sides by δx and let δx0. We get:

dy/dx=(dy/du).(du/dx)

I hope that helps.


By Louise Johnson on Tuesday, October 30, 2001 - 10:53 pm:

wonderful