+ The Oktave Forum » Technical » Science and Mathematics » Discrete Mathematics (Moderator: sanket)
|-+ A "covering partition" puzzle and some questions
Username:
Password:

Pages: 1 [2]
Topic Tools  
Read September 11, 2008, 05:07:09 am #15
sanket

Re: A "covering partition" puzzle and some questions

Quote
Am afraid I don't get this either. What are '+' and '-' in terms of network operations?

Let's say we have a network of 13 nodes, with each node having the list P = {1, 3, 9}. Let's say every node connects to a set of other nodes in the following manner: a node with label ni will connect to the node with label ni + Pi; since edges are undirected, ni is also connected to ni + pi*; thus, each node will connect to a maximum of 2*|P| other nodes. The operations are modulo n, of course.

Now, every node knows its direct connections. It also knows that all other nodes have an edge each to their corresponding +/- Pith nodes. In our specific example, node 1 is connected to 2, 13, 4, 11, 10 and 5. If we want to reach node 7 from node 1, node 1 will compute it as folows: 5 + (3 - 1) or, better still, 10 - (3). Because it knows that its neighbours have connections to nodes that are 1, 3 or 9 numbers away from them.

Update:
* I meant, ni is also connected to ni - pi
« Last Edit: September 11, 2008, 05:17:36 am by sanket »
Offline  
Read September 11, 2008, 12:07:18 pm #16
sri

Re: A "covering partition" puzzle and some questions

Quote
t's turning out to be the series 3n

Pmin = {1,3,9,27,81...} and max number that can be generated would be simply sum of all the elements of Pmin.

That's right! For a given n, the k elements of Pmin are, P1 = 30, P2 = 31, ..., Pk = n - summation(P1 through Pk-1) (Or Pk = 3k-1, if n is the biggest number for a given k).

If that is the case, then how is Pmin for 10 = {1,3,6}? 6 is not a power of 3.
Offline  
Read September 11, 2008, 12:14:05 pm #17
sri

Re: A "covering partition" puzzle and some questions

Let's say we have a network of 13 nodes, with each node having the list P = {1, 3, 9}. Let's say every node connects to a set of other nodes in the following manner: a node with label ni will connect to the node with label ni + Pi; since edges are undirected, ni is also connected to ni + pi*; thus, each node will connect to a maximum of 2*|P| other nodes. The operations are modulo n, of course.

Sorry, I'm still not getting the idea..  Huh

What is Pi in the earlier expression? Is it the ith element of P? (If P is considered as an ordered tuple, of course).

If so, how is Pi related to ni? I believe the interval range for i in ni is 0-12 or 1-13. But the range for i in Pi is only 1-3.

I'm quite stuck here and could not proceed further.
Offline  
Read September 11, 2008, 01:30:34 pm #18
sanket

Re: A "covering partition" puzzle and some questions

Quote
t's turning out to be the series 3n

Pmin = {1,3,9,27,81...} and max number that can be generated would be simply sum of all the elements of Pmin.

That's right! For a given n, the k elements of Pmin are, P1 = 30, P2 = 31, ..., Pk = n - summation(P1 through Pk-1) (Or Pk = 3k-1, if n is the biggest number for a given k).

If that is the case, then how is Pmin for 10 = {1,3,6}? 6 is not a power of 3.


As I said, for a given k, there is a maximum n up to which you can construct all numbers using the k elements in Pmin. And the Pmin for this maximum n is, Pmin = {30, 31, 32, ..., 3(k-1)}.

*However*, in the general case, the Pmin starts as 30, 31 etc.. But the last element of Pmin might not be a power of 3. It will be this - Pk = n - summation(all previous elements of Pmin). So, it is basically "whatever is left" after you have divided the number into sums of powers of 3.

So, for 10, Pmin = {1, 3, 10 - (1 + 3)} = {1, 3, 6} or for 51, Pmin = {1, 3, 9, 27, 11}
« Last Edit: September 11, 2008, 04:18:01 pm by sanket »
Offline  
Read September 11, 2008, 02:23:58 pm #19
sanket

Re: A "covering partition" puzzle and some questions

The following is an explanation for the 3^k series. This is not an exact reasoning, but an attempt to show how it works.

Let us say we want to cover numbers from 1 through n. Let us not specify n. Let us simply say we want to generate the largest n possible using k numbers that are elements of Pmin.

Let's begin at the beginning.

When k = 1, We can only have Pmin = {p1 = 1}, and the numbers covered, C = {1}

When k = 2, let's say p2 is the next element. What should it be? We know that using p1 and p2, we can construct 2 more numbers. In addition, p2 itself will be another number it will cover. Now, since we want to maximize the numbers that are covered, we don't want a combination of p1 and p2 to cover a number that is already covered by p1. So, on the number line, p2 should be as far away from p1 as to cover all numbers after p1to its left, and itself, plus another p1 numbers to its right.

More simply, p2 - (the biggest number p1 has covered) = (the biggest number p1 has covered) + 1
Specifically, p2 - 1 = 2. Thus, p2 = 3.

When k = 3, we need to find p3. Now, the set {1, 3} can cover up to 4. So, we can cover from p3 - 4 through p3 + 4. Again, to maximize this, we need to place p3 optimally on the number line. So, p3 should be placed 5 slots away from p2. So, p3 is 9. Another way to look at it is this: p3 covers 4 numbers to its left + it covers itself + it covers 4 numbers to its right. Thus, 9.

When k = 4, p4 = 13 + 1 + 13 = 27  [13 = 1 + 3 + 9]

Similarly, p5 = 40 + 1 + 40 = 81   [40 = 27 + 13]
p6 = 121 + 1 + 121 = 243   [121 = 81 + 40]
p7 = 364 + 1 + 364 = 729   [364 = 243 + 121]
p8 = 2187 etc..
« Last Edit: September 11, 2008, 04:19:58 pm by sanket »
Offline  
Read September 11, 2008, 02:58:43 pm #20
sanket

Re: A "covering partition" puzzle and some questions

Quote
Sorry, I'm still not getting the idea.. 

Okay, guess I am not explaining properly. Plus I mixed up notations. So ignore the previous one. Let me try here.

Let us say there are n nodes, N = {1, 2, ..., n}. An element Ni is any node in this list.

We have a Pmin = {p1, p2, ..., pk}. We have k elements and Pj is any element in this set.

Now, Pmin is a skip-list similar to the Chord skip-list. In case of Chord, every node connects to nodes that are {2, 4, 8, ..., 2log n}, skips away from itself. In our case, every node connects to nodes that are {p1, p2, ..., pk}, skips away from itself. Since, the edges are undirected, a node Ni is effectively connected to nodes Ni +/- Pj (1 <= j <= k).

In the specific example, where n = 13, and Pmin = {1, 3, 9}, node 1, for example, is connected to 1 + 1 = 2, 1 - 1 = 13 (this is modulo 13), 1 + 3 = 4, 1 - 4 = 11, 1 + 9 = 10, 1 - 9 = 5. The other nodes are also similarly connected.

Every node knows its direct connections. It also knows that all other nodes use Pmin to make connections. So, it can work out a routing scheme that is akin to constructing a number using Pmin. For example, I want to reach node 7 from node 1. So, 7 = (1 + 9) - (3) = 10 - 3. Node 1 is connected to 10. And it knows that 10 is connected to 7 as it is 3 numbers away from it. Similarly, 6 = (5) - (1), 8 = 5 + 3 or 11 - 3 and so on.
« Last Edit: September 11, 2008, 04:00:49 pm by sanket »
Offline  
Read September 11, 2008, 03:18:29 pm #21
sri

Re: A "covering partition" puzzle and some questions

Ah! I think I get the idea now.. Somewhat similar to the undirected version of deBruijn graphs. Will analyze in more detail later when time permits.
Offline  
Pages: 1 [2]
Jump to: