Intentamos probar que el valor más pequeño de la función $F$ definida por \begin{equation*} F(\alpha) = \int_0^2f(x)f(x+\alpha) \, dx \tag{1} \end{equation*> es $F(1)$. Al principio le otorgamos a $f$ la propiedad adicional de ser diferenciable con una derivada continua. Al final llenaremos ese vacío.
La función $F$ será continua y tendrá el período $T = 2$. Por lo tanto, existe un valor más pequeño de $F(\alpha)$.
Si $f$ es diferenciable, entonces $F$ también será diferenciable.
Intentamos probar que \begin{equation*} F'(\alpha) \le 0 \text{ para } 0 \le \alpha \le 1. \tag{2} \end{equation*> Ya que \begin{equation*} f(x) = f(2-x) = f(-x) \tag{3} \end{equation*> entonces $F(\alpha) = F(2-\alpha)$. Entonces (2) implica que \begin{equation*} F'(\alpha) \ge 0 \text{ para } 1 \le \alpha \le 2. \end{equation*> Por lo tanto, sabemos que $F(1)$ es el valor más pequeño.
Comenzamos estudiando (1). Si cambiamos $x$ a $x+1$ y luego $x$ a $1-x$ obtenemos \begin{gather*> \int_1^2f(x)f(x+\alpha) \, dx = \int_0^1f(x+1)f(x+1+\alpha) \, dx = [(3)] \\ = \int_0^1f(1-x)f(1-x-\alpha) \, dx = \int_0^1f(x)f(x-\alpha) \, dx. \end{gather*> Ahora podemos reescribir $F(\alpha)$ como \begin{equation*} F(\alpha) = \int_0^1f(x)(f(x+\alpha) + f(x-\alpha)) \, dx. \end{equation*> A través de la diferenciación bajo el signo de integral (ahora usamos que $f$ es diferenciable), seguida de integración por partes, obtenemos \begin{gather*> F'(\alpha) = \int_0^1f(x)(f'(x+\alpha) - f'(x-\alpha)) \, dx = \left[f(x)(f(x+\alpha) - f(x-\alpha))\right]_0^1 \\ - \int_0^1f'(x)(f(x+\alpha) - f(x-\alpha)) \, dx = f(1)\underbrace{(f(1+\alpha) - f(1-\alpha))}_{=0} - f(0) \underbrace{(f(\alpha) - f(-\alpha))}_{=0} \\ - \int_0^1f'(x)(f(x+\alpha) - f(x-\alpha)) \, dx = - \int_0^1f'(x)(f(x+\alpha) - f(x-\alpha)) \, dx.
Dado que $f$ es decreciente y diferenciable en $[0,1]$, entonces $f'(x) \le 0$ en la integral anterior. Queda por probar que \begin{equation*> g(x,\alpha) = f(x+\alpha) - f(x-\alpha) \le 0 \end{equation*> en el cuadrado $0 \le x \le 1, \, 0 \le \alpha \le 1$. Pero \begin{equation*> g(\alpha,x) = f(\alpha + x) - f(\alpha - x) = [(3)] = f(x+\alpha) - f(x-\alpha) = g(x,\alpha). \end{equation*> Consequently it is sufficient to examine $g$ on the triangle $0 \le \alpha \le x \le 1.$ To do that we study $g$ on that part of the straight line $\alpha = x-k$, that runs inside the triangle. Here $k$ is constant in $[0,1]$. To be more precise look at \begin{equation*> g(x,x-k) = f(2x-k)-f(k), \quad 0 \le k \le 1, \: k \le x \le 1.
The second part of the $x$-interval is $\dfrac{k+1}{2} \le x \le 1$. Then $1 \le 2x-k \le 2-k \le 2$. Furthermore $f(k) = f(2-k)$ and $1 \le 2-k \le 2.$ But on the interval $[1,2]$ the function $f$ is increasing. Since $2x-k \le 2-k$ then $f(2x-k ) \le f(2-k)$ and $g(x,2x-k) \le 0.$
We have proved that $F(1)$ is the smallest value under the additional assumption that $f$ is differentiable with a continuous derivative.
Now we will prove that the same is true if $f$ only is continuous. Define a smooth substitute $\tilde{f}$ for $f$ via \begin{equation*> \tilde{f}(x) = \dfrac{1}{2h}\int_{x-h}^{x+h}f(t) \, dt = \dfrac{1}{2h}\int_{-h}^{h}f(x+t) \, dt. \tag{4}
Then $\tilde{f}$ will be differentiable with the continuous derivative $\dfrac{f(x+h)-f(x-h)}{2h}$. Furthermore it inherits a lot of properties from $f$. It is obvious that $\tilde{f}$ will be periodic with $T=2$. It will also fulfil the condition $\tilde{f}(x) = \tilde{f}(2-x) = \tilde{f}(-x)$. It will also be decreasing on $[0,1]$. To realize that we need a little argument. Assume that $0 < x_1 < x_2 < 1$. If $h$ in (4) is small enough then $|t|$ will be so small that \begin{equation*> 0
From (4) we obtain that $\tilde{f}(x_1) \ge \tilde{f}(x_2)$. We have proved that $\tilde{f}$ is decreasing on the open interval $0 < x < 1$. But since $\tilde{f}$ is differentiable it will also be decreasing on the closed interval $0 \le x \le 1$ (use the intermediate value theorem). Analogously we prove that $\tilde{f}$ is increasing on $1\le x \le 2.$
Since $f$ is continuous and since we work on a compact interval $f$ will be uniformly continuous. Thus to a given $\varepsilon$ there exists a $\delta$ such that \begin{equation*> |f(x_1)-f(x_2)| < \varepsilon \text{ if } |x_1-x_2|< \delta.
With these $\varepsilon$ and $\delta$ and with $0 |f(x)-\tilde{f}(x)| \le \dfrac{1}{2h}\int_{-h}^{h}|f(x)-f(x+t)| \, dt \le \dfrac{1}{2h}\int_{-h}^{h}\varepsilon \, dt = \varepsilon. \tag{5}
We are now prepared to study the function
\begin{equation*> \tilde{F}(\alpha) = \int_0^2\tilde{f}(x)\tilde{f}(x+\alpha) \, dx.
According to what we have done above we know that \begin{equation*> \tilde{F}(\alpha) \ge \tilde{F}(1)
and that $\tilde{F}(1)$ is the smallest value.
From (5) we get \begin{gather*> |F(\alpha) - \tilde{F}(\alpha)| \le \int_0^2|f(x)f(x+\alpha)-\tilde{f}(x)\tilde{f}(x+\alpha)| \, dx \\ = \int_0^1|(f(x)-\tilde{f}(x)) f(x+\alpha)+\tilde{f}(x)(f(x+\alpha)-\tilde{f}(x+\alpha))| \, dx \\ \le \int_0^2(\epsilon C + C\epsilon) \, dx = 2C\epsilon
where $C = \max_{0 \le x \le 2}|f(x)|$. Consequently, \begin{equation*> F(\alpha) = F(\alpha) - \tilde{F}(\alpha) + \tilde{F}(\alpha) \ge -2C\epsilon + \tilde{F}(1).
But $\epsilon$ can be arbitrariamente pequeño. En el límite tenemos que \begin{equation*> F(\alpha) \ge F(1)
i.e. $F(1)$ es el valor más pequeño de $F(\alpha)$.
Doy la bienvenida a soluciones más cortas.
0 votos
¿Quieres decir $\int_0^2{f(x)f(\alpha+x)dx}$, ¿verdad?
0 votos
@mwomath: Oh, por supuesto. Gracias
0 votos
¿Varía $\alpha$ o $f(x)$?
0 votos
Puede comprobar bastante fácilmente que $F$ también es 2-periódico y que $F$ es par. Por lo tanto, basta con considerar $\alpha \in [0,1]$. Creo que $F$ tendrá un máximo en $\alpha=0$ y un mínimo en $\alpha=1.