Quasiperiodic Sequences


By Brad Rodgers (P1930) on Thursday, September 14, 2000 - 10:27 pm :

Let Transform S'=P(S) simultaneously replace every 1 in S with a 100, and every 0 in S to a 10. Now, take S0 =1 , and set Sn+1 =P(Sn )

Now let

cn =(the number of ones in Sn )/(the number of digits in Sn )

How can we find what lim(n-> infinity )cn is equal to. It seems to converge to sqrt(2)-1 (try finding cn for Sn up to n=8), but I'm not sure how to prove this.

Thanks,

Brad


By Michael Doré (P904) on Thursday, September 14, 2000 - 11:18 pm :

Hi Brad,

Let an = no. of 1s in Sn
and bn = no. of digits in Sn .

Notice that each digit in Sn will generate exactly one 1 for Sn+1 (because there is exactly one 1 in both 10 and 100). Hence an+1 = bn . [1]

However what will happen to bn+1 , the no. of digits? Well each 1 is replaced by 100, so the length of the chain increases by two, due to each 1 in Sn (as 100 is three-digit). Each 0 produces 10, so each 0 will increase the chain by one digit. So:

bn+1 = bn + 2(an ) + (bn -an )

because there are bn - an 1s. So:

bn+1 = an + 2bn . [2]

Combining [1] and [2] and shuffling about with n gives:

an = 2an-1 + an-2 [3]
bn = 2bn-1 + bn-2 [4]

Now let's first find the limit of bn /bn-1 with n going to infinity (call the limit y). First of all can you see why it has to tend to a limit at all (I'll explain this further if you like - it is probably the hardest part). Then write [4] as:

bn /bn-1 = 2 + bn-2 /bn-1

Now if bn /bn-1 is converging to y, it will be pretty close to bn-1 /bn-2 . With n tending to infinity in fact, these two values will both converge to y. So they we'd have:

y = 2 + 1/y

as bn-2 /bn-1 = 1/[bn-1 /bn-2 ] = 1/y

y2 - 2y - 1 = 0
(y-1)2 = 2
y = sqrt(2) + 1 (because y> 0).

Next rewrite [2] as:

bn+1 /bn = an /bn + 2

an /bn is the value you want. The left hand bit we've already done and it's y = sqrt(2)+1. So for n going off to infinity:

an /bn = sqrt(2) + 1 - 2 = sqrt(2) - 1

like you said.

Michael


By Brad Rodgers (P1930) on Friday, September 15, 2000 - 12:44 am :

Good proof. I had originally wondered whether that result would be true, as it ends up being an irrational density of ones. But, I quess that's why it's a quasiperiodic series rather than a periodic one.

Thanks,

Brad


By Michael Doré (P904) on Friday, September 15, 2000 - 12:54 am :

Now that's a very interesting point you made, about it being an irrational density. But it really is! And the reason that this is OK is that although Sn will have a rational density (for finite n) the limit as n tends to infinity does not have to be rational. This is because in general a rational sequence (i.e. a sequence whose nth term is rational) does not have to tend to an irrational limit.

Example:

3
3.1
3.14
3.141
3.1415
3.14159
...

A rational sequence tending to the irrational pi . In fact of course nearly all rational sequences tend to an irrational limit.

I'm not sure what you mean by quasi-periodic though?

Michael


By Brad Rodgers (P1930) on Friday, September 15, 2000 - 04:04 am :

By quasi-periodic, I mean that there are finite segments within Sinfinity that each occur an infinite number of times at different places in the sequence. Basically, I mean that the series has order to it, but not the sort of order that a periodic sequence, say 10101010..... has to it. If it were periodic like this, then it couldn't have irrational density.

Brad