I have just started studying Calculus of Variations, and need some help about deriving the Euler-Lagrange equation.

In the book I'm reading, the writer starts by imposing the following inner product on the infinite dimensional function space: $$\langle f;g\rangle=\int_a^bf(x).g(x)dx $$Then he states what he called "the basic directional derivative formula" $$\langle\nabla J[u];v\rangle=\frac{d}{d\varepsilon}J[u+\varepsilon v]$$ at $\varepsilon=0$ and where $J$ is the objective functional.

The minimizing function of $J$ is a one satisfying $\langle \nabla J[u];v\rangle=0$

The writer then works for a formula of $\langle\nabla J[u];v\rangle$ and finally reaches :$$\langle\nabla J[u];v\rangle=\int_a^bv.[\dfrac{\partial L}{\partial u}-\dfrac{d}{dx}(\dfrac{\partial L}{\partial u'})]dx +\dfrac{\partial L}{\partial u'}(b).v(b)-\dfrac{\partial L}{\partial u'}(a)v(a)$$Then states that, for suspect functions (those verifing boundary conditions), $v(a)=v(b)=0$, and finally deduces Euler-Lagrange equation for the minmizing function :$$\dfrac{\partial L}{\partial u}-\dfrac{d}{dx}(\dfrac{\partial L}{\partial u'})=0$$

I have two questions now:

  1. Shouldn't the norm of $v$ be equal to 1 in order to satisfy the formula of directional derivative?

  2. Shouldn't we prove that the region of suspect functions is connected so that we can talk about a directional derivative at a function in the direction of another?

Please help.


1) In principle you are right, and in the definition of $\nabla J$ there should be a division by $\| v \|$ in order to mimick the usual formula from calculus on $\Bbb R ^n$. Nevertheless, keep in mind that you are interested in the solutions of $\nabla J = 0$, and these do not care about the equation being divided by non-zero numbers.

There is a deeper reason, though, for $\| v \|$ not being present in the denominator: in a normed space, you may speak about the Fréchet derivative. In a space without norm, though, what can you do? The concept to use is the Gâteaux derivative, which is exactly the one used in defining $\nabla J$. It happens that you have a norm on your space, but the definition given works on spaces without norm, too; it wouldn't have worked if there had been that $\| v \|$ in the denominator (what would you have replaced it with in the absence of a norm?).

2) I don't see why you require connectedness, but if you really want it then note that the set of functions that vanish on the boundary is even better - it is pathwise connected: for every $f, g$ in it, the segment $tf + (1-t)g$ vanishes on the boundary too. Anyway, you don't need this (think about derivability in $\Bbb R ^n$: where do you use connectedness? nowhere).

  • $\begingroup$ Thanks for answering my questions. In the second question I wasn't asking about connectedness of variations $v$, but rather the functions $u$ from which we have to pick the minimizing function, those need to satisfy some boundary conditions but not necessarily vanish at the boundaries. I thought that if these functions weren't connected (were singular distinct points for example), then it wouldn't be possible to have a directional derivative at one function in the direction of another. $\endgroup$ – Tofi Mar 31 '16 at 16:06
  • $\begingroup$ @H.Tofaili: Wait, you're confusing me: what do you mean by "connectedness of a function"? Such a concept doesn't even exist! $\endgroup$ – Alex M. Mar 31 '16 at 16:17
  • $\begingroup$ Connectedness doesn't matter for differentiation, you only need your underlying set to be open. $\endgroup$ – Alex M. Mar 31 '16 at 17:04
  • $\begingroup$ I had to say : connectedness of the domain {$u$/$u$ satisfies boundary conditions}. $\endgroup$ – Tofi Mar 31 '16 at 17:06
  • $\begingroup$ Connectedness DOES NOT MATTER! $\endgroup$ – Alex M. Mar 31 '16 at 17:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.