# Limit Definition Derivative: e^(-1/x)

1. Oct 13, 2013

### Vale132

1. The problem statement, all variables and given/known data

Let

f(x) =\begin{cases}
0 & \text{ if } x\leq 0 \\
e^\left ( -1/x \right ) & \text{ if } x> 0
\end{cases}
Compute f'(x) for x < 0 and x > 0.

2. Relevant equations

$$f'(x) = \lim \ \ \ \ \ \ \displaystyle{\frac{e^{(-1/(x+h)} - e^{-1/x}}{h}} \\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ h\rightarrow0$$

3. The attempt at a solution

Did the x < 0 part already, seemed straightforward (got 0/h = 0).

For x > 0:

$$f'(x) = \lim \ \ \ \ \ \ \frac{1}{he^{1/(x+h)}} - \ \ \ \ \lim \ \ \ \ \ \frac{1}{he^{1/x}} \\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ h\rightarrow0 \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ h\rightarrow0$$

It is it true that because, as h -> 0 the h in the exponential's denominator (in the first limit) goes to zero, the two limits are approximately the same (large) quantity, so their difference is zero?

EDIT: Oops, looks like I broke up the limit incorrectly, since it's not a limit of two separate functions of x. Should I instead take the limit of the difference of two quotients?

Last edited: Oct 13, 2013
2. Oct 13, 2013

### joshmccraney

how have you defined e^x?? sometimes this definition is very helpful to solve

3. Oct 13, 2013

### Vale132

Do you mean I could use the Taylor series definition for e^x, one centered around 0 and one around h, to turn the exponentials into polynomials? Then the terms without h in the numerator would cancel, and I could use the remaining ones to cancel h in the denominator?

4. Oct 13, 2013

### Vale132

Or

$$e^{x} = \lim \ \ \ \ \ \ \ (1 + \frac{x}{n})^n \ \ \ ?\\ \ \ \ \ \ \ \ \ n\rightarrow∞$$

EDIT: Then inserting -1/(x+h) for x, the exponential would go to 1, as would e^(-1/x)?

Last edited: Oct 13, 2013
5. Oct 13, 2013

### Dick

Do you really have to do this using difference quotients? Why not just apply the chain rule to get the derivative for x>0? The only case where you might need to resort to the difference quotient is x=0.

6. Oct 13, 2013

### Vale132

You may be right; I had gotten in the habit from using quotients in earlier parts of the question. I guess "compute" implies that using the chain rule is okay. Thanks for the suggestion. Out of curiosity, would one of the above methods be useful if I had to do it this way?

7. Oct 13, 2013

### joshmccraney

sorry, judging by the time of the current academic term i was under the impression you needed to use the difference quotient. if you know the series, then why not use lopital's rule?

8. Oct 13, 2013

### Dick

I suppose you can tough it out with the power series expansion. But I don't think you have to. Just use the chain rule for x>0. Your function is well behaved for x>0 and x<0. At x=0 it's different. It's the interesting point. 1/0 is undefined. There I think you need to apply the difference quotient. And as joshmccraney said, there you might find l'Hopitals rule helpful.

9. Oct 13, 2013

### ibysaiyan

This problem has caught my attention.. I was thinking of using binomial expansion for e^(x+h)^-1 term and then to apply e^k -1 / k = 1 as lim k--> 0

10. Oct 13, 2013

### Dick

Sorry, but I'm not sure what part of the problem you trying to deal with here.

11. Oct 13, 2013

### ibysaiyan

I was referring to the second part of the question. Suppose f(x) = e^(-1/x)

$$f'(x) = \lim_{h \to 0} e^{(1/x+h)} - e^{(1/x)} / h$$

A taylor expansion about $$e^-{(1/x)} = \sum_{n=0} ^ \infty - (1/x^n) /n!$$

$$f'(x) = \lim_{h \to 0} (-1 - 1/{(x+h)}) - (-1 - {1/x}) / h$$

$$f'(x) = \lim_{h \to 0}{ {-(x+h)^{-1}} + (1/x) / h }$$

Is this a reasonable approach ?

12. Oct 13, 2013

### Dick

There are two problems here. The first is that it isn't right. There is no reason to ignore the higher power terms. And the second is that this whole thing is completely unnecessary. You know how to differentiate e^x and you know how to differentiate -1/x. Use difference quotients on those if you have to. Then use the chain rule. The only place you even need to think about using a difference quotient is at x=0 where the whole chain rule thing is undefined. How many times do I have to say this?

13. Oct 14, 2013

### ibysaiyan

Thanks for the explanation, apologies if I have annoyed you.

14. Oct 14, 2013

### Dick

That's ok, but the only real issue in this problem that takes thinking about is whether the function is differentiable at x=0. And everybody seems to be missing that. That's what's annoying.

Last edited: Oct 14, 2013
15. Oct 14, 2013

### D H

Staff Emeritus
Why is that so annoying? This is an introductory calculus problem, and the problem statement specifically avoids the issue of behavior t x=0. It asks the student to "Compute f'(x) for x < 0 and x > 0."

That said, this function is differentiable at x=0 in the sense that $\lim_{x\to 0^-} f'(x) = \lim_{x\to 0^+} f'(x) = 0$.

16. Oct 14, 2013

### Dick

Ok, so now I'm annoyed with myself for seeing a different problem than was intended. Sorry! And the function isn't just differentiable in the sense of that limit. The difference quotient also shows f'(0)=0. I'm guessing that might be in another part of the same question.