1

Define $\{a_n\}$ (a sequence of real numbers) as satisfying the H(m) condition if, for all integers $n$, the following recurrence relation holds:

$$ a_{n-m} + a_{n-m+1} + \dots + a_{n+m} = a_n \cdot (2m + 1) $$

In other words, the average of the $2m + 1$ consecutive terms from $a_{n-m}$ to $a_{n+m}$ is $a_n$.

The task is to prove the following conjecture: If a sequence satisfies both the H(p) and the H(q) condition for integers $p$ and $q$ $(p \neq q, p>1,q > 1)$, then the sequence ${a_n}$ must be an arithmetic progression. In other words, it satisfies H(1) condition.

What I’ve Tried:

I’ve explored using a method similar to the Euclidean algorithm to compute small cases for specific values of $p$ and $q$.

Then I suspect that this problem could be transformed into one involving the roots of the characteristic equation associated with the recurrence relations, where the only common root between the equations is 1. This related to a question left unsolved here.

I've also tried: $$ x^{n-m} + x^{n-m+1} + \dots + x^{n+m} = x^n \cdot (2m + 1) $$ $$ \frac {x^{m+1} - x^{-m}}{x-1} = 2m + 1 $$ let $x = e^{2i\omega}$, ($x \in \mathbb{C}$, so $\omega \in \mathbb{C}$) $$\implies \frac {e^{(2m+2)i\omega} - e^{(-2m)i\omega}}{e^{2i\omega}-1} = 2m + 1 $$ $$\implies \frac {e^{(2m+1)i\omega} - e^{-(2m+1)i\omega}}{e^{i\omega}-e^{-i\omega}} = 2m + 1 $$ $$\implies \sin((2m+1)\omega) = (2m+1)\sin\omega $$

Any guidance would be greatly appreciated!

Reiss
  • 37

1 Answers1

1

DTFT of $a_n = A(z) = \sum_n a_n z^{-n} $

DTFT of $b_n = \left[\frac{1}{(2m+1)},...,\frac{1}{(2m+1)} - 1,...,\frac{1}{(2m+1)} \right] = B_m(z) = \frac{1}{2m+1} \frac{z^{-(m+0.5)}-z^{m+0.5}}{z^{-0.5}-z^{0.5}} - 1$

$$H(m): \text{ convolution}(\{a_n\},\{b_n\}) = 0 = A(z) \times B_m(z) = 0 \implies (B_m(z) \neq 0 \implies A(z) = 0),(A(z) \neq 0 \implies B_m(z) = 0)$$

$$S_m = \{z : B_m(z) = 0 \}$$

Now the condition is true for $m = p,q$. $$z \in S_p \cap S_q \implies \frac{1}{2p+1} \frac{z^{-(p+0.5)}-z^{p+0.5}}{z^{-0.5}-z^{0.5}} - 1 = \frac{1}{2q+1} \frac{z^{-(q+0.5)}-z^{q+0.5}}{z^{-0.5}-z^{0.5}} - 1 = 0$$ We want to prove: $$S_1 \supseteq S_p \cap S_q$$.

Above equations imply: For $z = e^{j\omega}$, $$\omega = 2k\pi \in S_p \cap S_p \text{ but we know that } 2k\pi \in S_1$$

So probably for $z = e^{j \omega}$, $\{a_n\} = a[1,1,....,]$ is the only solution.

Solve for other $z \in S_p \cap S_q$, $z \in \mathbb{C}$ and check those cases.

There could be mistakes in above, please comment below if you find any.

  • Thanks for the detailed response! I have to admit I’m not very familiar with DTFT and how it applies in this context. Could you recommend any resources or provide a simpler explanation of how DTFT is being used to prove the result here? – Reiss Oct 12 '24 at 12:45
  • Condition $H(m)$ is a condition on convolution of $a_n$, $b_n$ and by property of DTFT convolution of $a_n$ and $b_n$ is equal to pointwise product DTFT of $a_n$ and DTFT of $b_n$. That's the only property of DTFT apart from its definition I am using. See wiki page on DTFT. But DTFT exists only for sequences like $l_2$ or sequences which are $l_1$. –  Oct 12 '24 at 13:25
  • I got $B(\omega) = \frac{1}{2m+1} \cdot \frac{\sin\left( \frac{\omega(2m+1)}{2} \right)}{\sin\left( \frac{\omega}{2} \right)}$ - 1, which seems to differ from yours. Did I miss something? – Reiss Oct 13 '24 at 05:09
  • Sorry I missed it ...I will fix it..but u r correct...thanks –  Oct 13 '24 at 05:24
  • @Reiss i fixed the answer based on ur feedback. Please take a look :-) –  Oct 13 '24 at 06:08
  • Now I stuck on another step in your solution ;) $$\implies m\omega = 2k\pi \implies \omega = \frac{2k\pi}{m}$$ Is this deduced from $B(\omega) = \frac{1}{2m+1} \cdot \frac{\sin\left( \frac{\omega(2m+1)}{2} \right)}{\sin\left( \frac{\omega}{2} \right)} - 1 = 0$? Could you offer some explanations? ) – Reiss Oct 13 '24 at 14:25
  • @Reiss $m\omega = 2k\pi$ is deduced from $B(\omega) =0$. Let me see and fix if I made any mistake. –  Oct 13 '24 at 23:17
  • @Reiss i have updated the answer and clearly stated the approach. Try to see if you can get anything from this point onwards. If you cant get anything, ping me here in the comments, i will also try. –  Oct 14 '24 at 01:13
  • @Reiss I have updated with another approach. Please take a look and see if its useful for you. –  Oct 14 '24 at 01:51
  • I’m having some difficulty grasping Approach 1 ... but for the last step, if $\text{gcd}(2p + 1, 2q + 1) = 3$, that implies $p \mod 3 = 1$ and $q \mod 3 = 1$, right?

    As for Approach 2, I’ve been exploring it, but since $omega$ is a complex number, I’m finding it tricky to handle

    – Reiss Oct 14 '24 at 14:51
  • @Reiss yes in last step : $p$ mod $3=1$. Similarly for $q$. In approach 2, in DTFT, $\omega$ is a real number and not a complex number. Give it a try. –  Oct 15 '24 at 07:59
  • Really? I have used another approach as an alternative of your first step which shows w is a complex number. However, if $\omega$ is real, $|\sin(nx)|≤n|\sin(x)| \implies \omega = 2k\pi$ – Reiss Oct 15 '24 at 14:24
  • @Reiss It seems approach 1 is wrong. I have removed it. If you are sure $\omega = 2k\pi$ is the only solution then it means, $a[1,1,1...,1]$ is the only sequence in the solution space for which DTFT exists. There could be other sequences but may be DTFT does not exist for those sequences. What i am saying should be correct, only if $\omega = 2 \pi k$ is the only solution. It means $\sum_n a_n e^{j\omega n}$ does not converge for other solutions ${a_n}$. So try $z$ transform similarly. I will add some details on it now. –  Oct 15 '24 at 16:22
  • @Reiss I updated my answer. See if its helpful. –  Oct 15 '24 at 16:34
  • @Reiss Now that we have generalized, yes we have to go to complex values of $z$. I see you have updated the question with the same approach. –  Oct 15 '24 at 16:53