Is there an arithmetic mean limit on the symmetrical items of the harmonic series?

172 Views Asked by At

Is it possible to determine what is the arithmetic mean of the harmonic series where $n$ and $-n$ are added and divided by two in this manner:

$$f(n) = \frac{ (\frac{3}{4})^n \times 2^{⌈-n \times log_2(\frac{3}{4})⌉} + (\frac{3}{4})^{-n} \times 2^{⌈n \times log_2(\frac{3}{4})⌉} }{2}$$

For each positive plus negative whole number $n$ we have an arithmetic mean. See values in the continuous graph below:

enter image description here

Now I want to sum them and take an average something like:

$$\frac{\sum_{n=1}^{x}{f(n)}}{x}$$

Does this series converge, if x is infinity?

My preliminary spreadsheet calculation gives a value somewhere near to $(6/5)^2 = 1.44$ if $n$ is $20$. Or, maybe a cuberoot of three? See row 20 in the picture. But I'm missing tools to determine if this equation has a limit when $n$ is a whole number approaching infinity?

Possible limit equation

$$ \lim_{h \to \infty}{\frac{1}{h}\sum_{n=1}^{h}{ \frac{ (\frac{3}{4})^n \times 2^{⌈-n \times log_2(\frac{3}{4})⌉} + (\frac{3}{4})^{-n} \times 2^{⌈n \times log_2(\frac{3}{4})⌉} }{2} }} $$

With Wolfram Alpha I got the summation formula simplified to:

$$g(n) = \frac{2^{-\frac{n\log\frac{4}{3}}{\log2}\pmod1} + 2^{\frac{n \log\frac{4}{3}}{\log2}\pmod1}}{2}$$

The distribution of the first 500 values in graph looks like this:

500 values of the average of symmetric harmonized 3/4 series

Spreadsheet

Harmonic series

Col 7 is (col3+col4)/2 corresponding equation in my question. Cols 3, 4, 7 are points (x, y) but x only is applicable now. Col 1 and 2 are positive and negative $n$.

Added references

This seems to relate to Cesaro mean:

https://en.wikipedia.org/wiki/Ces%C3%A0ro_summation

Cesàro means of conditionally convergent series

1

There are 1 best solutions below

7
On

I believe the answer is $\frac12\int_0^1 (2^x+2^{1-x})dx = \frac{1}{\log2}\approx 1.44269504$. This follows from the equidistribution theorem (look for Khinchin's integral and apply to the functions $f_1(x)=2^x$ and $f_2(x)=2^{1-x}$).

With some more steps, we have $(\frac34)^n=2^{n \log_2 \frac34}$ so $$ (\frac34)^n 2^{\lceil -n\log_2 \frac34 \rceil} = 2^{n\log_2 \frac34+\lceil -n\log_2 \frac34 \rceil} $$ and $$(\frac34)^{-n} 2^{\lceil n\log_2 \frac34 \rceil} = 2^{-n\log_2 \frac34+\lceil n\log_2 \frac34 \rceil}. $$ Now for $x\notin\mathbb{Z}$, $x+\lceil -x\rceil = x \bmod 1$ and $-x+\lceil{x}\rceil=(1-x) \bmod 1=1+(-x \bmod 1)$.

Set $\alpha=\log_2{\frac34} = \log_23-2$. Then $\alpha$ is irrational because otherwise, you'd have $3=2^{\frac pq}$ for integers $p$ and $q$, so $3^q=2^p$, clearly impossible.

So we are calculating $$\frac1{2N} \sum_{n=1}^N \left(2^{(n\alpha \bmod 1)}+ 2^{1+(-n\alpha \bmod 1)}\right) $$

Khinchin's identity now says that by the irrationality of $\alpha$, the limit is $$\frac12\int_0^1 \left(2^{x}+2^{1-x}\right) dx = \frac1{2\ln 2}\left[2^x-2^{1-x} \right]_0^1 =\frac1{2\ln2}(2-1-1+2)=\frac{1}{\ln 2}. $$