For each $n\in\mathbb{N}$ consider $$W_n=\int_0^{\pi/2}\cos^n{t} dt$$ We know the that $$\lim_{n\to\infty}\dfrac{W_n}{\sqrt{\dfrac{\pi}{2n}}}=1$$ Find the radius of convergence and the sum of the series: $$\sum_{n=0}^{\infty}W_nx^n$$ What I have obtained is that $W_n=\dfrac{n-1}{n}\dfrac{n-3}{n-2}...\dfrac{2}{3}$
Radius of convergence of sum $\sum_{n=0}^{\infty}W_nx^n$
1
$\begingroup$
integration
sequences-and-series
limits
convergence
-
1Since $W_n \sim \sqrt{\frac{\pi}{2n}}$ the convergence radius of the series is the same as that of $\sum \frac{x^n}{\sqrt{n}}$. – 2017-02-01
-
0For evaluating the sum one approach that often works is by starting with the recursion formula that lead you to the last result. Then multiply it by $x^n$ and get it on the form $f(n)W_nx^n = g(n) W_{n-2}x^n$ for some (linear I think) polynomials $f,g$. Then sum over all $n$ and relate the different terms to $F(x) = \sum W_n x^n$ and it's derivatives. This will give you a ODE for $F(x)$ which hopefully can be solved. – 2017-02-01
-
1...and a simpler way is to use the integral in the series: $$F(x) = \sum_{n=0}^\infty \int_0^{\pi/2}(\cos(t) x)^n{\rm d}t = \int_0^{\pi/2} \frac{1}{1-x\cos(t)}{\rm d}t$$ where we interchanged the summation and integration and summed a geometrical series. – 2017-02-01
-
0Very ingenious solutions. But why are you able to commute summation with the integral? – 2017-02-01
-
0See [this question](http://math.stackexchange.com/questions/83721/when-can-a-sum-and-integral-be-interchanged) for some useful criteria. In this case we justify it from the uniform convergence of $\sum f_n(t)$ with $f_n(t) = (\cos(t)x)^n$ for $|x|<1$. – 2017-02-01
-
0I see. Thank you very much! – 2017-02-01
1 Answers
3
As suggested by Winther, $$\begin{eqnarray*} \sum_{n\geq 0} x^n\int_{0}^{\pi/2}\cos^n(t)\,dt &=& \int_{0}^{\pi/2}\frac{dt}{1-x\cos(t)}\\&=&2\int_{0}^{\pi/4}\frac{dt}{(1+x)-2x\cos^2(t)}\\&=&2\int_{0}^{1}\frac{dt}{(1+x)(1+t^2)-2x}\\&=&\frac{2}{\sqrt{1-x^2}}\,\arctan\sqrt{\frac{1+x}{1-x}}\\&=&\frac{\pi+2\arcsin(x)}{2\sqrt{1-x^2}}=\frac{\pi-\arccos(x)}{\sqrt{1-x^2}}\end{eqnarray*}$$ whose radius of convergence is clearly $1$, due to the singularity at $x= 1$. On the other hand we do not need to explicitly locate such singularities to be able to state the same, since: $$ \int_{0}^{\pi/2}\cos^n(t)\,dt \sim\sqrt{\frac{\pi}{2n}} $$ ensures $\rho=1$ through the definition of the radius of convergence as a $\limsup$.