5

I have to give low and high bounds for the following:

$$ \sum_{n=1}^\infty \frac{1}{2^n - 3^n } $$

How do I determine an upper bound? How can I show this sum exists?

edit: removed erroneous conclusion.

klaufir
  • 209
  • 1
    What if you remove $2^n$ from the denominator in a similar spirit? (And as the other comments say, you might want to check that you're actually bounding the sum from above/below.) – Stahl Jun 06 '13 at 19:26
  • 2
    Every term being negative, the whole sum (which indeed exists) is negative hence 1 cannot be a lower bound. One may suggest more study of this series before continuing "in a similar spirit". – Did Jun 06 '13 at 19:27
  • 1
    $2^n < 3^n$ when $n \geq 1$. Are you sure $1$ is a lower bound? – Patrick Jun 06 '13 at 19:28

4 Answers4

7

I strongly suggest not working with the sum as it stands. The terms are obviously negative. Change signs. Your intuition will get much better. For sure mine would.

Find upper and lower bounds for the sign-altered sum, and then use the result to draw conclusions about upper and lower bounds for the original sum.

To show that the sign-altered sum exists, we can note that $2^n\le \frac{2}{3}3^n$, and therefore $\frac{1}{3^n-2^n} \le \frac{3}{3^n}$. From this you can also get an upper bound for $\sum_1^\infty \frac{1}{3^n-2^n}$.

As for a lower bound for $\sum_1^\infty \frac{1}{3^n-2^n}$, a very easy one is $0$! Almost as easy is to use the first term, which is $1$.

André Nicolas
  • 507,029
5

Let $x_n=3^n-2^n$, then $2^n\leqslant2\cdot3^{n-1}=3^n-3^{n-1}$ hence $x_n\geqslant3^{n-1}$ for every $n\geqslant1$. Thus, every $x_n$ is positive and $$ \sum\limits_{n\geqslant1}\frac1{x_n}\leqslant\sum\limits_{n\geqslant0}\frac1{3^n}=\frac32. $$ In particular, the series $\displaystyle\sum\limits_{n}\frac1{x_n}$ converges (absolutely) and its sum is at most $\dfrac32$.

To get a lower bound, use $x_1=1$ and $x_2=5$ to get $$ \sum\limits_{n\geqslant1}\frac1{x_n}\geqslant1+\frac15=\frac65. $$

Did
  • 279,727
  • Isn't it more practical to just use the first two terms to obtain the lower bound? Gives you $1+\frac15 = \frac65 > \frac76$. – Myself Jun 07 '13 at 05:47
  • @Myself It is. Good point. I will modify my post, thanks. – Did Jun 07 '13 at 05:50
  • (Actually for any $\ell$ the one may compute a lower bound $\sum_{i=1}^\ell \tfrac1{3^i-2^i} + \sum_{i=\ell+1}^\infty\tfrac1{3^i} = \sum_{i=1}^\ell \tfrac1{3^i-2^i} + 3^\ell/2$. Taking $\ell=1$ gives your first bound $7/6$, whereas taking $\ell=2$ but deleting the second term gives you $6/5$.) – Myself Jun 07 '13 at 05:52
  • @Myself There are misprints in your last comment but I see what you mean and what you mean is correct. – Did Jun 07 '13 at 05:54
5

The sum $$S = \sum_{n\ge 1} \frac{1}{3^n-2^n}$$ is a special evaluation of a more general harmonic sum and may be treated by Mellin transforms. The base function $f(x)$ is $$f(x) = \frac{1}{3^x-2^x}$$ which has the following Mellin transform: $$\mathfrak{M}(f(x); s) = \int_0^\infty \frac{1}{3^x-2^x} x^{s-1} dx = \int_0^\infty \frac{1}{3^x} \frac{1}{1-(2/3)^x} x^{s-1} dx \\= \int_0^\infty \frac{1}{3^x} \sum_{q\ge 0} \left(\frac{2}{3}\right)^{qx} x^{s-1} dx = \sum_{q\ge 0} \int_0^\infty \left(\frac{2^q}{3^{q+1}}\right)^x x^{s-1} dx \\ = \sum_{q\ge 0} \int_0^\infty e^{(q\log(2/3)-\log 3)x} x^{s-1} dx = \Gamma(s) \sum_{q\ge 0} \frac{1}{(\log 3 - q\log(2/3))^s} \\= \Gamma(s) \sum_{q\ge 0} \frac{1}{(\log 3 + q\log(3/2))^s} = \frac{\Gamma(s)}{(\log(3/2))^s} \sum_{q\ge 0} \frac{1}{(\log (3)/\log (3/2) + q)^s} \\ = \frac{\Gamma(s)}{(\log(3/2))^s} \zeta\left(s,\log (3)/\log (3/2) \right). $$ It follows that the Mellin transform $g^*(s)$ of the harmonic sum $$ g(x) = \sum_{n\ge 1} \frac{1}{3^{nx}-2^{nx}}$$ is given by $$ g^*(s) = \mathfrak{M}(g(x); s) = \frac{\Gamma(s)}{(\log(3/2))^s} \zeta\left(s,\log (3)/\log (3/2) \right) \zeta(s).$$ Looking at the first few poles we perform Mellin inversion to get $$ \operatorname{Res}(g^*(s)/x^s; s = 1) = -\frac{\log x + \psi(\log (3)/\log (3/2)) + \log\log(3/2)}{x\log(3/2)} \\ \operatorname{Res}(g^*(s)/x^s; s = 0) = \frac{1}{4} \frac{2\log(3)-\log(3/2)}{\log(3/2)} \\ \operatorname{Res}(g^*(s)/x^s; s = -1) = -{\frac {1}{144}}\,{\frac {x \left( \left( \log \left( 3/2 \right) \right) ^{2} +6\, \left( \log \left( 3 \right) \right) ^{2}-6\,\ln \left( 3 \right) \log \left( 3/2 \right) \right) }{\log \left( 3/2 \right) }} \\ \operatorname{Res}(g^*(s)/x^s; s = -3) \\= {\frac {1}{86400}}\,{\frac {{x}^{3} \left( - \left( \log \left( 3/2 \right) \right) ^{4}+30\, \left( \log \left( 3 \right) \right) ^{2} \left( \log \left( 3/2 \right) \right) ^{2}+30\, \left( \log \left( 3 \right) \right) ^{4}-60\, \left( \log \left( 3 \right) \right) ^{3}\log \left( 3/2 \right) \right) }{ \log \left( 3/2 \right) }}.$$ Finally set $x=1$ and switch to numerics, including a few more terms, to obtain the approximation $$ S \approx 0.2508048329 + 1.104755646 - 0.08106948472 + 0.0004958137737 \\ - 0.5355881093 \times 10^{-5} + 0.6368945126 \times 10^{-7} = 1.274981516.$$ This agrees with the exact value to the extend of the precision used.

Marko Riedel
  • 61,317
2

$\begin{align} \sum_{n=1}^\infty \frac{1}{3^n - 2^n } &=\sum_{n=1}^\infty \frac{1}{3^n(1 - (2/3)^n) }\\ &=\sum_{n=1}^\infty \frac{1}{3^n} \sum_{k=0}^\infty(2/3)^{nk}) \\ &=\sum_{n=1}^\infty \frac{1}{3^n}+\sum_{n=1}^\infty \frac{1}{3^n} \sum_{k=1}^\infty(2/3)^{nk}) \\ &= \frac{1/3}{1-1/3} + \sum_{m=1}^{\infty} (2/3)^m \sum_{d|m} (1/3)^d\\ &= \frac{1}{2} + \sum_{m=1}^{\infty} (2/3)^m \sum_{d|m} (1/3)^d\\ \end{align} $

Looking at the inner sum, let $S(m, r) = \sum_{d|m}r^d$, where $0 < r < 1$.

$S(m, r) > r$, since the term $d=1$ always occurs.

$S(m, r) < \sum_{d=1}^{m}r^d = \frac{r}{1-r} $, since more terms are here than in the original sum.

So $S(m, 1/3) > 1/3$ and $S(m, 1/3) < 1/2$.

Therefore

$\begin{align} \sum_{n=1}^\infty \frac{1}{3^n - 2^n } &= \frac{1}{2} + \sum_{m=1}^{\infty} (2/3)^m S(m, 1/3)\\ &> \frac{1}{2} + \sum_{m=1}^{\infty} (2/3)^m (1/3)\\ &> \frac{1}{2} + \frac1{3}\sum_{m=1}^{\infty} (2/3)^m\\ &= \frac{1}{2} + \frac1{3}\frac{2/3}{1-2/3}\\ &= \frac{1}{2} + \frac1{3}2\\ &= \frac{1}{2} + \frac{2}{3}\\ &= \frac{7}{6}\\ \end{align} $

and

$\begin{align} \sum_{n=1}^\infty \frac{1}{3^n - 2^n } &< \frac{1}{2} + \frac1{2}2\\ &= \frac{3}{2} \end{align} $

More accurate bounds for $S(m, r)$ would result in more accurate bounds for the sum.

marty cohen
  • 107,799