9

How can I find the value of this series? $$\dfrac{1}{2} + \dfrac{1}{4} + \dfrac{1}{8} + \dfrac{1}{10} + \dfrac{1}{20} + \dfrac{1}{22} + \dfrac{1}{44} + \dfrac{1}{46} + \dfrac{1}{92}+\cdots$$ The pattern is that the denominator increases by $2$ first and then it gets multiplied by $2$.

I converted this series to this form: $$\dfrac{1}{2} + 3\left(\dfrac{1}{8} + \dfrac{1}{20} + \dfrac{1}{44} + \dfrac{1}{92}+\dots\right)$$ In this version the denominator is multiplied by $2$ and added 4. I have found out that how to find the $n$-th value in the denominator and it is $2^{n+1} \cdot 3 - 4$.

How can I find the exact value of this series?

Sayan Dutta
  • 8,831
  • 2
    There is an explicite expression for the sum, but it's really complicated https://www.wolframalpha.com/input/?i=sum+%282%5E%28n%2B1%29*3-4%29%5E%28-1%29+from+n%3D1+to+infinity It's around 0.219 The link doesn't work, You have to copy it :/ – Jochen Aug 13 '21 at 09:31
  • I can't understand it :/. That expression is too complicated for me – İsmim Yok Aug 13 '21 at 09:35
  • 1
    @İsmimYok There's no simple closed form for such expressions I believe. – Mourad Aug 13 '21 at 09:39
  • I think the best way for you to "close" this problem is to say that the answer is $$ \frac{1}{2}+ 3\cdot (0.219161) \approx 1.157483 $$ – Matti P. Aug 13 '21 at 09:41
  • 3
    @İsmimYok: What is the source of this series? Is it a textbook exercise or contest question for which you have some expectation of a "reasonable" form? Or is it perhaps something that arose in your own work or something, and you're just wondering if (or hoping that) there's a reasonable form? – Blue Aug 13 '21 at 11:24
  • The sequence $4,10,22,46,94,\ldots$ seems to have a fairly rich theory. See https://oeis.org/A033484 – Barry Cipra Aug 13 '21 at 11:25
  • @Jochen Can you tell me what the $\psi_2^{(0)}$ term represents? How is that function defined? – Sayan Dutta Aug 13 '21 at 11:30
  • I can get that the sum is equal to $-\frac{6 \psi_{2}^{(0)} \left( \frac{\ln(3)}{\ln(2)} \right) - \ln(2) - 6\ln(3)}{8\ln(2)}$, where $\psi_{q}$ denotes the q-Polygamma function. Whether this q-Polygamma value has a closed-form, I’m not sure, but I doubt it. – KStarGamer Aug 13 '21 at 11:37
  • No, unfortunately not, I am very sorry. It is something about the digamma function. https://mathworld.wolfram.com/DigammaFunction.html But I don't know anything about it. – Jochen Aug 13 '21 at 11:37
  • 1
    @SayanDutta The q-Polygamma function is defined as follows: $\psi_{q} (z) = \frac{1}{\Gamma_{q} (z)} \frac{\partial \Gamma_{q} (z)}{\partial z}$ where $\Gamma_{q}$ denotes the q-Gamma function defined as $\Gamma_{q} (z) = \frac{(q; q){\infty}}{(q^z; q){\infty}} (1-q)^{1-z}$ where $(z; q)_{\infty}$ is a q-Pochhammer symbol. Feel free to read more here https://mathworld.wolfram.com/q-PolygammaFunction.html https://mathworld.wolfram.com/q-GammaFunction.html There are several related q-Series functions here http://mathworld.wolfram.com/topics/q-Series.html – KStarGamer Aug 13 '21 at 11:57
  • 1
    @Blue I was just hoping that it had a closed form. – İsmim Yok Aug 13 '21 at 14:10

1 Answers1

10

Let the sum to be calculated be

$$s=\sum_{k=1}^{\infty} \frac{1}{a(k)}$$

With the sequence $\{a(k)\} = \{2, 4, 8, 10, 20, 22, 44, 46, 92, ...\}$.

As stated in the OP there is a simple rule to generate this sequence: start with 2, then first multiply by 2, then add 2 to the result, and so on.

This leads to two sequences combined into one.

Now what are the explicit formulas for the terms?

First take the terms with the odd indices 1,3,5,.... These are {2, 8, 20, 44, 92, ...}, and then observe that each term with an even index 2,4,6,... is just 2 more than the previous odd term, i.e. {4,10,22,46, ...}.

Now the first differences of {2, 8, 20, 44, 92, ...}, are {6,12,24,48,...} which is $6\cdot 2^p$. Adjusting $p$ to the counting of terms and adding a constant $(-4)$ to match the initial term we get for the odd terms

$$a(k) = -4 + 3\cdot 2^\frac{k+1}{2}, k=1,3,5,...\tag{1a}$$

and hence, adding $2$, for the even ones

$$a(k) = -2 + 3\cdot 2^\frac{k}{2}, k=2,4,6,...\tag{1b}$$

Letting for odd and even k, respectively, $k=2i-1$ and $k=2 i$ the sum in question can be written as $s = s_o + s_e$ where

$$s_o = \sum_{i=1}^{\infty}\left( \frac{1}{-4+3\cdot 2^{i}} \right)$$

$$s_e = \sum_{i=1}^{\infty}\left( \frac{1}{-2+3\cdot 2^{i}} \right)$$

We shall show below that these can be expressed in terms of the q-polygamma function using

$$ \begin{align}s(a)&:=\sum_{i=1}^{\infty}\left( \frac{1}{-a+3\cdot 2^{i}} \right)\\&= \frac{1}{a \log(2)}\left(- \psi _2^{(0)}\left(\frac{\log \left(\frac{6}{a}\right)}{\log (2)}\right)- \log (a)+\frac{1}{2} \log (18)\right)\end{align}\tag{2}$$

The numerical value is

$$N(s) = 1.157482445913889917496363751936696728866 ...\tag{3}$$

Short excursion to some higher trancendental function appearing here

It turns out that in order to express the simple looking sum $s(a)$ in terms of know function we have to dive into the deeper waters of the family of the so called q-functions.

The q-polygamma function $\psi _q^{(0)}(z)$ is defined in terms of the q-gamma function $\Gamma_q(z)$ which in turn is defined in terms of the q-Pochhammer symbol $(a;q)_n$.

In particular we have

$$\psi _q^{(0)}(z)= \frac{d}{dz} \log(\Gamma_q(z))$$

$$\Gamma_q(z) = \frac{(q;q)_{\infty}}{(q^z;q)_{\infty}}(1-q)^{1-z}$$

$$(a;q)_{\infty}=\prod_{j\ge 0}(1-a q^j)$$

from which follows that

$$\psi _q^{(0)}(z)= -\log(1-q) + \log(q) \sum_{i \ge 0} \frac{q^{i+z}}{1-q^{i+z}}\tag{4}$$

Here appears a sum with the structure we need to identify our sum $s(a)$.

Starting with the index $i=0$ and writing the summand as

$$\frac{1}{-a+6\cdot 2^{i}}=\frac{1}{a}\left(\frac{1}{-1+\frac{6}{a}\cdot 2^{i}}\right)=\frac{1}{a}\left(\frac{\frac{a}{6}(\frac{1}{2})^i}{1-\frac{a}{6}(\frac{1}{2})^i}\right)=\frac{1}{a}\left(\frac{q^{i+z}}{1-q^{i+z}}\right) $$

where we have set $\frac{a}{6}=q^{z}$ or $z=\frac{\log(\frac{a}{6})}{ \log(q)}$ and $q=\frac{1}{2}$.

we find

$$s(a) =\frac{1}{a}- \frac{1}{a \log(2)}\psi _{\frac{1}{2}}^{(0)}\left(\frac{\log(\frac{6}{a})}{ \log(2)}\right)\tag{5}$$

Notice that here, in contrast to the equivalent expression $(2)$, the parameter of the q-polygamma function is $q=\frac{1}{2}$.

Finally, the complete sum is given by $s = s(2)+s(4)$. This can be simplified, using transformation formulas below to

$$s = -\frac{1}{8 \log(2)} \left( 6 \psi _{2}^{(0)}\left(\frac{\log(3)}{\log(2)}\right)-\log(2) - 6 \log(3) \right)\tag{6}$$

This expression was given by @KStarGamer in a comment.

Transformation formulas for q-polygamma

I found for real $z$ and $q \gt 0$ the following transformation formulas

Shifting $z$

$$\psi _{q}^{(0)}(z+1)=\psi _{q}^{(0)}(z) - \log(q) \frac{q^z}{1-q^z}\tag{7a}$$

letting $q \to \frac{1}{q}$

$$\psi _{q}^{(0)}(z)=\psi _{1/q}^{(0)}(z)+\log(q)(z-\frac{3}{2})\tag{7b} $$

It was easy to derive $(7a)$ from $(4)$, but I had to resort to "experimenting" with the equality of $(2)$ and $(5)$ to find $(7b)$ (I would appreciate a derivation from $(4)$ as well).