I'm having troubles figuring out the following problem.
we wanna show that $$\sum_{n=0}^\infty\int_D f_d^n \, \mathrm d x \begin{cases} = \infty, \quad & d=1,2 \\ < \infty, \quad & d \geq 3\end{cases}$$ where $f_d (x) = \frac 1 d \sum_{i=1}^d \cos(2 \, \pi \, x_i)$ and $D = [-\frac 1 2 ,\frac 1 2]^d$.
So the idea is to do this \begin{align} \sum_{n=0}^\infty\int_D f_d^n \, \mathrm d x & \overset{(1)} = \sum_{n=0}^\infty \int_D \lim_{\varepsilon \uparrow 1} \, (\varepsilon \,f_d)^n \, \mathrm d x \overset{(2)} = \sum_{n=0}^\infty \lim_{\varepsilon \uparrow 1} \int_D (\varepsilon \,f_d)^n \, \mathrm d x \\ & \overset{(3)} = \lim_{\varepsilon \uparrow 1} \sum_{n=0}^\infty \int_D (\varepsilon \,f_d)^n \, \mathrm d x \overset{(4)} = \lim_{\varepsilon \uparrow 1} \int_D \sum_{n=0}^\infty (\varepsilon \,f_d)^n \, \mathrm d x \\ & \overset{(5)} = \lim_{\varepsilon \uparrow 1} \int_D \frac 1 {1 - \varepsilon \,f_d} \, \mathrm d x \overset{(6)} = \int_D \frac 1 {1 - f_d} \, \mathrm d x \end{align}
and then the last integral can be bounded from above (for $d \geq 3$) and below (for $d = 1,2$) by the inequality $$\frac {y^2} 6 \leq 1 - \cos(y) \leq \frac {y^2} 2, \quad y \in [-\pi , \pi]$$
That said, we have $$\int_D \frac 1 {1 - f_d} \, \mathrm d x \begin{cases} = \infty, \quad & d=1,2 \\ < \infty, \quad & d \geq 3 .\end{cases}$$
Now my question is, how do I justify the interchanging of limits above?
In particular in the case $d = 1,2$ in (3) and (6).
$\,$
(1) is easy, nothing happens
(2) is understood by the dominant convergence theorem by Lebesgue, since $|f_d| \leq 1$ is bounded
(4) & (5) also easy, since for $0 < \varepsilon < 1$ we have $\int_D (\varepsilon \, f_d)^n \, \mathrm d x < 1$ and therefore the geometric sum converges, and power series can be integrated term by term within their radius of convergence
$\,$
However, I can only justify (6) if $d \geq 3$, otherwise the last integral is unbounded.
For (3) I tried to show that $$h_d(\varepsilon) : = \sum_{n=0}^\infty \int_D (\varepsilon \,f_d)^n \, \mathrm d x = \int_D \frac 1 {1 - \varepsilon \, f_d} \, \mathrm d x$$ is continuous at $\varepsilon = 1$, that is $|h_d(\varepsilon) - h_d(1)| \to 0 \, \,(\varepsilon \to 1)$ for any $d$, but I didn't manage.
$\,$
Is there any easier way to do that?
Any help much appreciated!
(also how about equality (3) ?)
– cesare borgia Oct 19 '17 at 14:07