Fundamental Theorem of the Calculus of Variations. Let $u \in L^1_{\text{loc}}(a,b)$ and $$ \int_{a}^{b} u(x) \varphi(x) dx = 0 \quad \forall \varphi \in \mathcal{C}^{\infty}_{\text{c}}(a,b). $$ Then, we have $u|_{(a,b)} \equiv 0$ almost everywhere.

I am going to present the proof our professor did and the one I found in the lecture notes from the same course a few years back. I would like to know if both approaches are correct or the first is not because it ignores some details or if the second proof is just unnecessarily complicated.

Proof 1: Let $[c,d] \subset (a,b)$ a compact interval with $c < d$. We aim to show $\int_{c}^{d} | u(x) | dx = 0$, which yields the claim since $[c,d]$ was chosen arbitrarily. Define $\omega := \text{sgn}(u) \cdot \chi_{[c,d]}$. Furthermore, define $\omega_{\varepsilon} := \omega \star J_{\varepsilon}$ ($\star$ = convolution), where $$ J_{\varepsilon}(x) := \begin{cases} c_{\varepsilon}\exp\left(\frac{1}{x^2 - 1}\right), & | x | < 1, \\ 0, & \text{else.} \end{cases} $$ and $c_{\varepsilon}$ is chosen such that $\int_{\mathbb{R}} J_{\varepsilon}(x) dx = 1$.

In a previous lemma we have shown that

  1. for sufficiently small $\varepsilon > 0$ we have $\omega_{\varepsilon} \in \mathcal{C}^{\infty}_{\text{c}}(a,b)$ and
  2. $\omega_{\varepsilon} \xrightarrow{\varepsilon \searrow 0} \omega$ almost everywhere on $(a,b)$.

We now want to use Lebesgue's theorem to show \begin{equation*} \int_{a}^{b} \omega_{\varepsilon}(x) u(x) = \int_{c}^{d} | u(x) | dx = 0. \end{equation*} To find a integrable majorant for $\omega_{\varepsilon}$ we observe that \begin{equation*} | \omega_{\varepsilon}(x) | = \left| \int_{\mathbb{R}} J_{\varepsilon}(x - \xi) \omega_{\varepsilon}(\xi) d\xi \right| \le \max_{\xi \in \mathbb{R}} | \omega_{\varepsilon}(\xi) | \cdot \int_{\mathbb{R}} J_{\varepsilon}(x - \xi) d \xi = 1 \cdot 1 = 1. \end{equation*} Therefore, $| u(x) \omega_{\varepsilon}(x) | \le | u(x) |$.

Because $u \in L^1_{\text{loc}}(a,b)$ we have $u \in L^1(c,d)$ and therefore, $| u |$ is a integrable majorant for $u \omega_{\varepsilon}$. $\square$

Proof 2 Let $u \in L^1_{\text{loc}}(a,b)$ and $[c,d] \subset (a,b)$. Define $\omega = \text{sgn}(u) \chi_{[c,d]}$. Then we have $\omega \in L^1_{\text{loc}}(a,b)$ and $\text{supp}(\omega) \subset [c,d]$.

!The $\tilde{J}_{\varepsilon}$ is the $J_{\varepsilon}$ from above!

Now define $\omega_{\varepsilon} := \tilde{J}_{\varepsilon} \ast \omega$. Then, $\omega_{\varepsilon} \to \omega$ almost everywhere on $(a,b)$ and $\text{supp}(\omega_{\varepsilon}) \subset [c - \varepsilon, d + \varepsilon]$, hence $\omega_{\varepsilon} \in \mathcal{C}^{\infty}_{\text{c}}(a,b)$ if $\varepsilon$ is small enough.

We obtain \begin{align*} 0 = \int_{a}^{b} \underbrace{u(x) \omega_{\varepsilon}(x)}_{\xrightarrow{\textrm{a.e.}} u(x) \omega(x)} dx & = \int_{c - \varepsilon}^{d + \varepsilon} u(x) \omega_{\varepsilon}(x) dx \\ & = \int_{a}^{b} u(x) \chi_{[c - \varepsilon, d + \varepsilon]}(x) \omega_{\varepsilon}(x) dx. \end{align*} As \begin{equation*} | \omega_{\varepsilon}(x) | \le \int \tilde{J}_{\varepsilon}(x - y) \underbrace{| \omega(y) |}_{\le 1} dy \le 1, \end{equation*} For $\varepsilon_0 < \min(c - a, b - d)$ and all $\varepsilon < \varepsilon_0$ we get \begin{equation*} | u(x) \omega_{\varepsilon}(x) | \le | u(x) | \chi_{[c - \varepsilon_0, d + \varepsilon_0]}(x) \end{equation*} This function is integrable on $(a,b)$. Therefore, Lebesgues theorem shows \begin{equation*} 0 = \int_{a}^{b} u(x) \omega(x) dx = \int_{c}^{d} | u(x) | dx, \end{equation*} hence $u \equiv 0$ almost everywhere on $[c,d]$. As $[c,d] \subset (a,b)$ was chosen arbitrarily, this yields the claim. $\square$

  • $\begingroup$ I somehow without noticing down voted this question: If you edit it, I can up vote it. $\endgroup$ – onurcanbektas May 7 at 7:52
  • $\begingroup$ @onurcanbektas done! $\endgroup$ – Viktor Glombik May 7 at 9:07

Both proofs are fine, although there is a limit missing in the first one, it should say \begin{equation*} \lim_{\epsilon \to 0}\int_{a}^{b} \omega_{\varepsilon}(x) u(x) = \int_{c}^{d} | u(x) | dx = 0. \end{equation*} and the definition of $\omega$ is wrong, it should be $$\omega(x) := \text{sgn}(u(x))\chi_{[c,d]}(x).$$ The main difference between the proofs is this: the second proof contains an argument why $\text{supp}(\omega_{\epsilon}) \subseteq (a,b)$ for $\epsilon$ small enough, while in the first proof they argue this point by referring to a "previous lemma".

  • $\begingroup$ Thanks for noticing the typo in the definition of $\omega$ in the first proof! $\endgroup$ – Viktor Glombik May 5 at 15:15

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.