6
$\begingroup$

This is a problem from my calculus of variations class:

Let $X=\{v:[0,1]\rightarrow\mathbb{R}$ of class $C^1$, $v(0)=1$,$\ v(1)=0\}$, let $F:X\rightarrow\mathbb{R}$ be the functional $$F(v):=\int_0^1 (e^{v'(x)}+v^2(x))dx.$$ Integrate the Euler-Lagrange equation with conditions $v(0)=1$, $\ v'(0)=a$, $a\in\mathbb{R}$ and show that $F$ has no minimum on $X$.

The professor also said that we could restrict ourselves to the case $a<0$.
Ok, I did my homework: first we define $X_a:=\{v:[0,1]\rightarrow\mathbb{R}$ of class $C^1$, $v(0)=1$,$\ v'(0)=a\}$. The Euler-Lagrange equation for $F$ is $v''e^{v'}=2v$. Since the functional $F$ doesn't depend explicitely on $x$ the quantity $v'e^{v'}-e^{v'}-v^2$ is some constant, say $k$, thus imposing the boundary conditions at $x=0$ we get $e^{v'}(v'-1)-v^2=e^{a}(a-1)-1$ (which is the du Bois-Reymond equation and it is exactly what we would have got integrating $v'(v''e^{v'}-2v)=0$). Now, if a minumum exists in $X$ it must have bounded derivative in $x=0$, so it must be in $X_a$ for some $a<0$. Let's see if the boundary conditions of $X$ are compatible with the boundary conditions of $X_a$. In $x=1$ we have $e^{v'(1)}(v'(1)-1)=e^{a}(a-1)-1$. Since $v'(1)$ is something in $\mathbb{R}$, we reduced the problem to the study of the zeros of the function $g(y):=e^{y}(y-1)-e^{a}(a-1)+1$. It turns out that $g$ has no zeros for $a<0$, so we are done.
Let's get to the question: why can we restrict ourselves to $a<0$?
For $a\geq0$ g has one zero, so there could be a candidate minimum in $X\cap X_a$. Graphically, I can believe that the function $v$ has to be decreasing to minimize the functional, but this is not a proof, nor a satisfactory justification.
I thought that if we can bound from below the values that $F$ assumes on $X\cap X_a$, $a\geq0$ with $F(w)$ for some $w$ not in $X\cap X_a$, this will be enough to show that $F$ has no minimum in $X$. Nevertheless this seems not so simple, since we don't have an explicit formula for the candidate minimum.

$\boldsymbol{Edit}$: A colleague of mine told me that for the case $a>0$ I have to look at the sign of the second derivative. From the Euler-Lagrange equation we get $sign(v''(x))=sign(v(x))$ $\forall x\in(0,1)$, which implies that the first derivative is increasing in $(0,1)$. By continuity and positivity at 0 of the first derivative we get that $v$ is increasing in $[0,1]$, thus it cannot reach the point 0 at $x=1$.
My question is: does this argument cover also the case $a=0$?

$\endgroup$
0

1 Answer 1

1
$\begingroup$
  1. Summarizing OP's analysis: Define function $$f(a)~:=~(a-1)e^a.\tag{1}$$ Now it is a fact that $$ \exists ! a_0 >0:~~ f(a_0)~=~0. \tag{2}$$ $$ \forall a< a_0: ~~ -1~\leq~f(a)~<~0.\tag{3}$$ $$ \forall a> a_0: ~~ f(a)~>~0.\tag{4}$$ Here $a_0=1$. There is a first integral to OP's EL equation. Together with OP's boundary conditions, it becomes $$ f(v^{\prime}(0))-1~=~ f(v^{\prime}) - v^2 ~=~f(v^{\prime}(1)) .\tag{5} $$ In light of eqs. (2)-(4) we see that eq. (5) is impossible to satisfy for $v^{\prime}(0)<a_0$ because then $f(v^{\prime}(0))<0$ while $f(v^{\prime}(1))\geq -1$.

  2. Next consider the case $$v^{\prime}(0)~\geq~ 0.\tag{6}$$ The EL equation reads $$ \left(e^{v^{\prime}}\right)^{\prime}~=~2v.\tag{7} $$ If in the interval $[0,x_0]$ the function $v > 0$ is positive, then $e^{v^{\prime}}$ (and thereby $v^{\prime}$) is increasing, and hence $v^{\prime}\geq 0$ is non-negative by assumption (6). Therefore the function $v > 0$ is weakly increasing. Therefore we may assume $x_0=1$. Contradiction with the boundary condition $v(1)=0$. $\Box$

$\endgroup$
5
  • $\begingroup$ Actually I think I made a mistake studying the zeros of the function $f$: $f$ is strictly larger that zero even for $a=0$. Thus the case $a=0$ was included in the first case. Anyway, thanks for your answer. $\endgroup$
    – Uskebasi
    Apr 3, 2017 at 17:52
  • $\begingroup$ How did you get this expression $v'e^{v'}-e^{v'}-v^2$? How did you solve the E-L eqn? $\endgroup$
    – A witch
    Feb 6, 2021 at 22:48
  • $\begingroup$ I updated the answer. $\endgroup$
    – Qmechanic
    Feb 6, 2021 at 23:13
  • $\begingroup$ Can you check this question? It is similar to this one but I am stuck st a step math.stackexchange.com/questions/4015486/… $\endgroup$
    – Véronique
    Feb 7, 2021 at 1:38
  • $\begingroup$ I wonder why there were no upvotes so far ;) $\endgroup$
    – metamorphy
    Feb 7, 2021 at 12:26

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.