# Find the first derivative of this functional

1. Nov 28, 2006

### Dragonfall

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

For $$f\in C([0,1])$$, let $$N(f)=\int_0^1(f(t))^2dt$$. Show that $$N\in C^{\infty}$$ and calculate the first derivative.

2. Relevant equations

Can I use Leibniz's integral rule for this?

3. The attempt at a solution

If I just blindly plug in the formula, I get

$$dN/df=\int_0^12f(t)dt$$

2. Nov 28, 2006

### AKG

You have some very strange notation. I assume:

$$C([0,1]) = \{ f : [0,1] \to \mathbb{R} | f\mbox{ is continuous}\}$$

That would make

$$C^{\infty} = \{ F : C([0,1]) \to \mathbb{R} | F\mbox{ is infinitely differentiable}\}$$

Is this what you mean? If so, how is the derivative of an elment of $C^{\infty}$ defined?

3. Nov 28, 2006

### Dragonfall

Yes, that's what I mean. I don't understand your question. I suppose they can be defined in terms of its power series.

4. Nov 28, 2006

### AKG

It was a pretty straightforward question. What is the definition of the derivative of an element of $C^{\infty}$?

5. Nov 29, 2006

### HallsofIvy

Staff Emeritus
AKG's point is that N is NOT a function from the real numbers to the real numbers. It is a function whose domain is a set of functions, and whose range is the real numbers. How are you defining the derivative of such a function?

6. Nov 29, 2006

### Dragonfall

I actually don't know. I just know that in a general metric space, a function f is differentiable at one point y if it can be written as f(x)=f(y)+df(x-y)+p(x) where p is little-o norm and df(x-y) is a linear continous function. I suspect df is the derivative, but I don't know how to compute it.

There is no mentioning of differentiation in my book (Elements of the Theory of Functions and Functional Analysis) and I might have made a mistake in my notes.

Last edited: Nov 29, 2006
7. Nov 29, 2006

### HallsofIvy

Staff Emeritus
In a "general metric space", addition is not defined. What you are talking about is a topological vector space (the most general situation in which the derivative can be defined). Sets of functions do form topological vector spaces in which the topology is based on a metric but there are several different ones that are not equivalent. You haven't said which metric you are using.

8. Nov 29, 2006

### Dragonfall

Neither did the question.

Let's assume the supremum metric.

Last edited: Nov 29, 2006
9. Nov 29, 2006

### AKG

How is the little-o norm defined?

10. Nov 29, 2006

### Dragonfall

Let V, W be vector spaces, let $$M\subset V$$ be open and let $$x_0\in M$$. Then $$f:M\rightarrow W$$ is little-o norm if $$\forall\epsilon>0\exists\delta>0\forall x(||x-x_0||_V<\delta\Rightarrow||f(x)-f(x_0)||_W<\epsilon||x-x_0||_V)$$.

11. Nov 29, 2006

### AKG

Your definition of little-o norm depends on x0, i.e. it could only be read to say that f is little-o norm at x0. In your definition of differentiability, then, do you mean p to be little-o norm at some point, or at all points?

Also, the only way your definition makes sense is if you mean df to be linear continuous, not df(x-y).

Last edited: Nov 29, 2006
12. Nov 30, 2006

### Dragonfall

Is df(x-y) df evaluated at x-y or df times (x-y)? Also, the little-o norm is probably defined at the point which is differentiable, but I'm not sure. Is there a book where I might find the definition?

13. Nov 30, 2006

### AKG

The definition that would make the most sense to me is that given f : V -> W, f is differentiable at some y in V iff there exists a function p : V -> W that is little-o norm at y and a function df(y) : V -> W that is continuous and linear such that for all x in V, f(x) = f(y) + (df(y))(x-y) + p(x). Note that if V = W = R, then a continuous linear function df(y) : V -> W just multiplies its argument by a constant. So if f'(y) is what we think of as being the derivative of f at y in the normal sense (i.e. it is just a number), then df(y)(z) = f'(y)z, where on the left we have the function df(y) acting on z, and on the right we have a number f'(y) multiplying with z.

I think N is differentiable, so I would proceed as follows: Pick an arbirary function f. Guess a function $dN(f) : C^{\infty} \to \mathbb{R}$ that's continuous and linear. Pick an arbitrary function g, and define $p : C^{\infty} \to \mathbb{R}$:

p(g) = N(g) - N(f) - (dN(f))(g-f)

Check that p so-defined is little-o norm at f. Guessing dN(f) is the tricky part, because as far as I know, you just have to make a reasonable guess, there's no smart way I know of of picking the right dN(f). This might be because this is the first time I'm seeing differentiation on an arbitrary normed vector space. If you pick the right dN(f), then showing p is little-o norm at f is just a first-year analysis sort of $\delta -\epsilon$ problem.

14. Nov 30, 2006

### Dragonfall

That makes sense, but to be sure, can you recommend a book that would contain the definition? I'm not even sure which field this falls into.