11
$\begingroup$

I will show two solutions of this problem.

First solution :

$$\sum_{n=1}^\infty \frac{1}{n(2n-1)}=\sum_{n=1}^\infty \left(\frac{2}{2n-1}-\frac{1}{n}\right)$$

$$=\sum_{n=1}^\infty \left(\int_0^1 (2x^{2n-2}-x^{n-1})dx\right)$$ $$=\int_0^1 \left( \sum_{n=1}^\infty(2x^{2n-2}-x^{n-1})\right)dx$$ $$=\int_0^1 \left( \frac{2}{1-x^2}-\frac{1}{1-x} \right) dx $$ $$=\int_0^1 \frac{1}{1+x}dx $$ $$=\ln 2$$

Second solution :

$$\sum_{n=1}^\infty \frac{1}{n(2n-1)}=\sum_{n=1}^\infty \left(\frac{2}{2n-1}-\frac{1}{n}\right)$$

$$=\sum_{n=1}^\infty \left(\int_0^1 (2x^{2n-2}-2x^{2n-1})dx\right)$$ $$=\int_0^1 \left( \sum_{n=1}^\infty(2x^{2n-2}-2x^{2n-1})\right)dx$$ $$=\int_0^1 \left( \frac{2}{1-x^2}-\frac{2x}{1-x^2} \right) dx $$ $$=\int_0^1 \frac{2}{1+x}dx $$ $$=2\ln 2$$

In fact, $\displaystyle\sum_{n=1}^\infty \frac{1}{n(2n-1)}=2\ln 2$.

Why the first solution is false but the second is true ?

I have known that uniformly convergent series can be integrated term by term.

Is this mean that $\sum_{n=1}^\infty(2x^{2n-2}-2x^{2n-1})$ ; second solution, is uniformly convergent but $\sum_{n=1}^\infty(2x^{2n-2}-x^{n-1})$ ; first solution, is not ?

$\endgroup$
6
  • $\begingroup$ You messed up at one of the steps, so I reccomend comparing them $\endgroup$
    – user253055
    Sep 6, 2015 at 16:39
  • $\begingroup$ @AaronThompson What do you mean? $\endgroup$ Sep 6, 2015 at 16:41
  • 3
    $\begingroup$ @AaronThompson $\int_0^1x^{n-1}=\int_0^12x^{2n-1}=\frac{1}{n}$ $\endgroup$ Sep 6, 2015 at 16:44
  • $\begingroup$ I am saying he made a mistake in one of the calculations and was doing it right in the other calculations. It's not like he's doing it two different ways $\endgroup$
    – user253055
    Sep 6, 2015 at 16:44
  • $\begingroup$ @AaronThompson (S)he is doing it in multiple ways $\endgroup$ Sep 6, 2015 at 16:46

1 Answer 1

9
$\begingroup$

Here is an even more baffling calculation:

\begin{align*} 0 &= \sum_{n=1}^{\infty} \left( \frac{2}{2n} - \frac{1}{n} \right) \\ &= \sum_{n=1}^{\infty} \int_{0}^{1} (2 x^{2n-1} - x^{n-1}) \, dx \\ &= \int_{0}^{1} \sum_{n=1}^{\infty} (2 x^{2n-1} - x^{n-1}) \, dx \\ &= \int_{0}^{1} \left( \frac{2x}{1-x^2} - \frac{1}{1-x} \right) \, dx \\ &= - \int_{0}^{1} \frac{dx}{x+1} \\ &= -\log 2. \end{align*}

So where did we mess up? It is the third step, where we interchanged the order of integration and summation. In this step, we are actually dealing with $\infty - \infty$ type indeterminate, which are cleverly hidden, and we failed to manage it properly. This is more evident if we plot the graph of the partial sum

$$y = \sum_{k=1}^{n} (2x^{2k-1} - x^{k-1})$$

for $n = 1, \cdots, 20$ as follows. (Color changes from red to green to blue as $n$ increases.)

enter image description here

The mass of $\log 2$ is concentrated around the high spike at $x = 1$, which vanishes as $n \to \infty$. This is why in both calculation we end up losing $\log 2$.


Speaking differently, the common mistake we have done is essentially the same as in the following bogus argument:

\begin{align*} 0 &= \left( 1 + \frac{1}{2} + \frac{1}{3} + \cdots \right) - \left( 1 + \frac{1}{2} + \frac{1}{3} + \cdots \right) \\ &= \left( \frac{2}{2} + \frac{2}{4} + \frac{2}{6} + \cdots \right) - \left( 1 + \frac{1}{2} + \frac{1}{3} + \cdots \right) \\ &= -1 + \left(\frac{2}{2} - \frac{1}{2}\right) - \frac{1}{3} + \left( \frac{2}{4} - \frac{1}{4}\right) - \cdots \\ &= -1 + \frac{1}{2} - \frac{1}{3} + \frac{1}{4} - \cdots \\ &= -\log 2. \end{align*}

In our calculation, shifting the order of the summand is achieved by doubling the exponent; that is, by changing $\int_{0}^{1} x^{n-1} \,dx$ by $\int_{0}^{1} 2x^{2n-1} \,dx$.

$\endgroup$
6
  • 1
    $\begingroup$ Thank you. But why the second solution is true? $\endgroup$
    – Bless
    Sep 6, 2015 at 17:17
  • 1
    $\begingroup$ I have known that uniformly convergent series can be interchanged the order of integration and summation. Is this mean that $\sum_{n=1}^\infty(2x^{2n-2}-2x^{2n-1})$ ; second solution, is uniformly convergent but $\sum_{n=1}^\infty(2x^{2n-2}-x^{n-1})$ ; first solution, is not ? $\endgroup$
    – Bless
    Sep 6, 2015 at 17:18
  • $\begingroup$ @Bless, That's right. We have $$ \left| \frac{1}{1+x} - \sum_{k=1}^{n} (2x^{2k-2} - x^{k-1}) \right| = \frac{x^n (1+x-2x^n)}{1-x^2}, $$ and as $x \to 1$ the right hand side converges to $n-\frac{1}{2}$. So the first one does not converge uniformly. $\endgroup$ Sep 6, 2015 at 17:23
  • $\begingroup$ Of course, failure of uniform convergence does not necessarily prevent us from interchanging the order, but in our case this certainly does. $\endgroup$ Sep 6, 2015 at 17:29
  • 1
    $\begingroup$ @Bless, the second one does not converge uniformly, neither. An easy way to see is that, while the partial sum $\sum_{k=1}^{n} (2x^{2k-2} - 2x^{2k-1})$ is zero at $x = 1$ the limiting function $1/(x+1)$ is not zero at $x = 1$. But in this case we have at least two different lifesavers: Abel's theorem and dominated convergence theorem. Either way can be used to justify your second argument. $\endgroup$ Sep 6, 2015 at 17:37

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.