You are on page 1of 9

Investigations on the Radio Number

Tian-Shun Allan Jiang


North Carolina School of Science and Mathematics
Mentor: Dr. Dan Teague
2nd Trimester Paper
February 12, 2013

1 Overview
For the first half of this trimester, I devoted a lot of energy thinking about and working
on finding the radio number of the grid graph. This was a tough problem as even finding
the maximum hopping distance proved to be a challenge. Eventually, I was able to devise a
method to find the maximum hopping distance on a ladder graph (2 n grid graph), but the
technique seems difficult to extend on a general m n grid graph. The results for the ladder
graph, as well as some attempts at finding the maximum hopping distance for a general grid
graph are given below.
During the second half of the trimester, I spent lots of time working on a Java application
that gives me more flexibility in the graph input type. The program has given me access to
the solutions of any graph of my choosing, as long as it has less than about a dozen vertices.
By working with the program, I have formulated several conjectures about the tightness
graphs of solutions. These conjectures are also given below.
I have also read a paper about the radio number of k-ary trees. The paper gives a nice
overview of the significance of the hopping distance and bumps. Ill go over this first.

2 Hopping Distance and Bumps


Let f be a radio labeling on G. By the condition d(u, v) + |F (u) F (v)| diam(G) + 1, we
know f (u) 6= f (v), as d(u, v) diam(G). Thus, we can order V (G) = {v1 , v2 , . . . vn } such
that
f (v1 ) < f (v2 ) < f (v3 ) < . . . < f (vn )
where n = |V (G)|. By default, we set f (v1 ) = 0. Then, span(f ) = f (vn ) f (v1 ) = f (vn ).
For convenience of notation, let fi := f (vi+1 ) f (vi ) and di := d(vi+1 , vi ). Then, we have

fi D + 1 di .

1
However, we do not know how much greater fi is than the required amount D + 1 di . Thus,
we introduce the bump b(vi+1 , vi ) := fi + di D 1. We will abbreviate this notation with
b(vi+1 , vi ) = bi .
Then, we have fi = D + 1 di + bi . Taking the sum of fi s with 1 i n 1, we get
n1
X
span(F ) = f (vn ) = (n 1)(D + 1) (di bi ). (1)
i=1

Since the radio number is the minimum possible span of F , we want


n1
!
X
rn(G) = min(span(f )) = (n 1)(D + 1) max (di bi ) . (2)
i=1

Note that all of the claims in this section are true for the radio number on a general
graph G. This relation shows that the radio number is determined by maximizing the total
hopping distance minus the total bumps on the graph.
Pn1  To solve the problem, we need to
find the best way to balance the max i=1 (di bi ) quantity.
An interesting question is if there is a property relating the number of bumps to the
distance traveled over some number of vertices. This was the case for the path graph, as if
di + di+1 D+1 2
, then there would be a bump. I suspect that there is no general formula
relating distances to bumps, but there are such relations on specific graphs.
This is an important area of future study, as knowing
Pn1 the relationship between distances
and bumps will help us maximize the quantity i=1 (di bi ), and give solutions to the
problem.
There
Pn1are interesting problems when trying to generalize a method to balance the quan-
Pn1 i=1 (di bi ). At first, I thought using the maximum hopping distance (maximizing
tity
i=1 (di )) would always work, but I realized that the L4 ladder graph is a counter-example.
I then thought that having the maximum possible hopping distance with the condition that
the graph had no bumps would always form a solution. However, when looking at tightness
graphs for the 4-vertex lollipop graph, I found that every solution had a bump. Thus, it is
also not true that there exists a solution for any G without any bumps. I suspect the balance
between hopping and bumps will have to be dealt with on a case-by-case basis. It would be
interesting if I could categorize which graphs balance based on maximum hopping distance,
and which graphs balance based off of minimum bumps,P and which graphs fall in between.
n1 
Without further ado, we determine the value of max i=1 (di bi ) on the ladder graph.

3 Grid Graphs
In the following sections, we give a proof for the ladder graph, and go on a few adventures
in tackling the m n grid.

2
4 The Ladder Graph
4.1 Lower Bound
Pn1 Pn1
From Section 2, we see that rn(G) depends on i=1 di and i=1 bi . We will deal with these
separately.

4.1.1 Hopping Distance


To maximize n1 bi ) on the ladder graph, we want to let n1
P P
i=1 (di P i=1 di take on as large a
n1
value as possible, and i=1 bi to take on as small a value as possible. Pn1 
In this subsection, we find the maximum hopping distance MHD := max i=1 di on
the ladder graph.
Let us label the vertices of the ladder graph with coordinates, as shown in Figure 3.1:

Figure 3.1 - Ladder Graph with Coordinates


We see that the distance between any two points P1 = (x1 , y1 ) and P2 = (x2 , y2 ) is d(P1 , P2 ) =
|x1 x2 | + |y1 y2 |. Let d(vi , vi+1 ) = |xi xi+1 | + |yi yi+1 | = xi + yi Then,
n1
X
di = (x1 + x2 + . . . xn1 ) + (y1 + y2 + . . . yn1 ).
i=1

We may decompose the total hopping distance into its x and y components. Let the maxi-
mum hopping distance in the x direction be MHDx , and the maximum hopping distance in
the y-direction be MHDy . In the following two lemmas, we find MHDx and MHDy .
Lemma 3.1.1  2
4k 1 : n = 2k
MHDx =
4k 2 + 4k : n = 2k + 1
Proof: Let us isolate our attention to the x components of our graph, giving us the x
distance xdi (vi , vi+1 ) = |xi xi+1 |. We are looking for
n1
!
X
MHDx = max xdi = |x1 x2 | + |x2 x3 | + . . . |xn2 xn1 | + |xn1 xn |.
i=1

We see that each of x2 , x3 , . . . , xn1 appears twice, while x1 and xn appear once in the sum
above. Furthermore, in each x distance xdi , either the xi term is positive and the xi+1 term is
negative, or vice versa. Thus, we may assign the set of x-values {x2 , x2 , x3 , x3 , . . . , xn1 , xn1 }
{x1 , xn } an equal number of positive and negative signs such that their total is maximized.

3
Let us use the rearrangement inequality to find the maximum. The rearrangement in-
equality states that for two multisets of ordered real numbers {a1 , a2 , . . . an } and {b1 , b2 , . . . bn },

a1 b1 + a2 b2 + . . . + an bn a(1) b1 + a(2) b2 + . . . + a(n) bn ,

for every permutation a(1) , a(2) , . . . , a(n) .


In our case, we have the multiset of available x values A = {1, 1, 1, 1, 2, 2, 2, 2, . . . n
1, n 1, n 1, n 1, n, n, n, n} {x1 , xn }, and the multiset of positive and negative signs
B = (2n 1){1, 1}. There are two cases: n = 2k and n = 2k + 1.
Case 1: n = 2k. Applying the rearrangement inequality, we get
n1
!
X
MHDx = max xdi
i=1

= 1111. . .kkkk+(k+1)+(k+1)+(k+1)+(k+1)+. . .+n+n+n+nx1 +xn .


Notice that we must remove two terms in this summation, as x1 and xn only appear once.
Without loss of generality, we may remove these terms by x1 + xn above, as x1 and xn
switch places in the labeling F 0 , if F 0 (xi ) = F (xn ) F (xi ) for all i. Note that F 0 has x1 and
xn switched when compared to F , has the same span as F , and is a valid radio labeling.
Thus, we must choose suitable x1 and xn such that 1 1 1 1 . . . k k k k +
(k + 1) + (k + 1) + (k + 1) + (k + 1) + . . . + n + n + n + n x1 + xn is maximized. Clearly, it is
best to remove the largest negative value, and the smallest positive value. Thus, x1 = k + 1
and xn = k.
With some algebra, we find
n1
!
X
MHDx = max xdi
i=1

= 1 1 1 1 . . . k k k + (k + 1) + (k + 1) + (k + 1) + . . . + n + n + n + n.
= 4k 2 1.
Case 2: n = 2k + 1. Applying the rearrangement inequality as in Case 1, we get

MHDx = 1111. . .(k +1)(k +1)+(k +1)+(k +1)+. . .+n+n+n+nx1 +xn .

Again we must remove two terms, the smallest positive term, and the least negative term.
We get

MHDx = 1 1 1 1 . . . (k + 1) + (k + 1) + . . . + n + n + n + n

= 4k 2 + 4k.
Lemma 3.1.2 MHDy = 2n 1
Proof: We use the same techniques as in Lemma 3.1.1. This time, we have the multiset
of available y values A = n{1, 2} {x1 , xn }, and the multiset of positive and negative signs
B = (2n 1){1, 1}.

4
Applying the rearrangement inequality, we get
n1
!
X
MHDy = max xdi
i=1

= 1 1 1 1 . . . + 2 + 2 + 2 + 2 x1 + xn = 2n x1 + x2 .
Clearly, we have x1 = 2 and x2 = 1, so we get

MHDy = 2n 1.

Remark: The sum of the maximum hopping distances in the x direction and the y
direction (as in Lemmas 3.1.1 and 3.1.2) do not represent the maximum hopping distance
of the ladder graph. This is because the ordering of vertices x1 , x2 , . . . xn to achieve MHDx
may be different from the ordering to achieve the MHDy . Thus, we introduce a property of
the ladder graph W(Ln ), which is the total distance that must be wasted to integrate the
x-direction and y-direction orderings. Thus, for a general grid graph we have

MHD = MHDx + MHDy W(Ln ). (3)

Finding the total wasted distance on a m n grid is by no means trivial. In the following
section, we compute the value of W(Ln ).

4.1.2 Wasted Distances


Again we need to break this up into even and odd cases.
Lemma 3.1.1.1 If n = 2k, W(Ln ) = 1.
Proof: By our breakdown of x and y vertex distances, we split the graph into four
quadrants. Quadrant 1 consists of all vertices with y = 2 and x k+1, Quadrant 2 consists
of all vertices with y = 2 and x k, Quadrant 3 consists of all vertices with y = 1 and x k,
and Quadrant 4 consists of all vertices with y = 1 and x k + 1. Since we know from above
that min and max must be in one of the four locations (1, k), (1, k + 1), (2, k), (2, k + 1),
without loss of generality we may place min in Quadrant 1. We dont know if max is in
Quadrant 2 or 3 yet. Notice from our assignment of positive and negative signs above in
4.1.1 that Quadrant 1 contains vertices with positive x and y labels, Quadrant 2 has positive
y labels and negative x labels, Quadrant 3 has negative x and y labels, and Quadrant 4 has
positive x labels and negative y labels. The point of these number assignments was to make
finding distances an easy addition of the distance labels given to the vertices. However, the
sum of these labels must give the correct distance. Thus, we see that we only have correct
distances when we jump from Quadrant 1 to 3, or from Quadrant 2 to 4.
Now, let us hop around. We go from Q1 to Q3 and back to Q1 and to Q3 until we run
out of vertices in Q1. Then, we must go to Q4. However, this is wasting a distance of 1!
After we go to Q4, however, we may continue hopping until we land on max vertex (2, k).
Thus, the maximum hopping distance for even ladder graphs is 1.
Lemma 3.1.1.2 If n = 2k + 1, W(Ln ) = 2.
Proof: Again we split up the graph into quadrants, only this time there are two middle
vertices that have x-value k + 1. We know that these middle vertices are in fact the min

5
and max vertices. So, without loss of generality, let us let (2, k + 1) be the min vertex. Then,
we must hop to Q4. From here, we alternate from Q4 to Q2 until we run out of space in Q4.
Then, we must waste 1 distance by going from Q2 to Q3. From there, we hop from Q3 to
Q1. We finally run out of room in Q1 and are stuck in Q3. The only choice left is to go to
max. However, this wastes another distance of 1, so the total wasted distance in this case is
2.

4.1.3 Total Contribution


Now
Pn1that we have calculated the MHD of the ladder graph, we must take 1into account
i=1 bi . It turns out that for the ladder graph, there always exists a smooth graph con-
structed using the maximum hopping if there are at least 10 vertices. This can be verified
using the bump rule2 for ladder graphs. (I actually ran out of time to write this up! Email me
if you have questions. It is quite similar to the path bump rule which says if di +di+1 D+1,
then there must be a bump.)

4.1.4 Ladder Graph Result


We have finally found MHD for the ladder graph. Combining our results from above, using
MHD = MHDx + MHDy W(Ln ), we get

(4k 2 1) + (2(2k) 1) 1 : n = 2k
MHD(Ln ) = 2
(4k + 4k) + (2(2k + 1) 1) 2 : n = 2k + 1

Doing the simplifications, we get



4k 2 + 4k 3 : n = 2k
MHD(Ln ) =
4k 2 + 8k 1 : n = 2k + 1

The final result for the ladder graph is:


 2
4k 2k + 2 : n = 2k, n > 3
rn(Ln ) =
4k 2 + 2k + 3 : n = 2k + 1, n > 1

4.2 Attempts at the m n


Ive tried several different approaches in search of the maximum hopping distance of the mn
grid. In the end, my attempts can be boiled down to two ideas: a wholistic approach and
a decomposition approach.
The decomposition approach was shown above for the ladder, and I discussed why it was
difficult to extend on an m n grid namely it becomes far more difficult to calculate
W (Lmn ). Thus, I will go over my thoughts in the wholistic approach and show where I
got stuck.
1
a graph with no bumps
2
a rule relating the distance traveled over some set of vertices, and the required number of bumps generated
the graph

6
The wholistic approach involves looking at coordinate pairs to maximize the hopping
distance, instead of breaking up the distances into xs and ys. This has the advantage of
avoiding the whole process of finding W (Lmn ), but has the disadvantage of making it less
clear how we should order the vertices x1 . . . xn .
So far, the best way I have tried to organize a method to order vertices x1 . . . xn is to
assign each vertex two markers: AA, AB, BB, or BA. Two vertices, the min and the max,
will have only 1 marker (either A or B). In total, there must be an even number of A markers
and an even number B markers over all vertices.
So how are we using these A and B markers? For every vertex (xi , yi ), if the vertex has
a marking A it takes on the value of (xi + yi ), called its A-value, and if the vertex has
marking B it takes on the value (xi yi ), called its B-value.
The point of the A and B markers system is to calculate the hopping distance without
tracking down the exact movement of the hopping sequence x1 . . . xn , as each vertex is
assigned a value. Notice that the same technique was used to find the maximum hopping
distance of paths: we found it by assigning each vertex a value, confirmed that the hopping
could actually occur, then summed everything up.
The way A and B markers work is that we may calculate the total distance contribution
from all pairs of A vertices, and add on the total distance contribution from B vertices. We
find the distance between a pair of A vertices by taking the difference of their corresponding
A-values. We see that this only works if the line formed by the two A vertices has a positive
slope. Similarly, we find the distance between a pair of B vertices by taking the difference
of their corresponding B-values. In this case, however, this method only gives the correct
distance if the line through the two B vertices has a negative slope.
So, we now have a grid graph with a whole lot of A and B markers. We want to find a way
to assign A markers and B markers to vertices such that the hopping distance is maximized.
Since A and B markers actually just make a whole lot of A-values and B-values, we turn
back to the rearrangement algorithm.
And yada-yada, we go on our merry way figuring out how to partition the vertices into A-
vertices and B-vertices, and the best way to assign positive and negative labels to A vertices
and B vertices. However, we are horrified to find out that the resulting distance is too great.
At a second check, we actually write the A and B markers onto the grid graph, only to find
out that the hopping could not occur.
Unfortunately, I have not been able to mitigate this problem yet, so this thread of the
problem remains unsolved.

7
5 Tightness Graph Conjectures
Note: k-bumps are bumps where b(vi+1 , vi ) = k. So a 1-bump is where the next vertex is
forced to be loose by 1.

1. Observation: If all bumps are 1-bumps, we can calculate span(f ) if we are given D.

2. Can we show that the following 4 structures are the only structures that appear on the
tightness graph of a solution?

A potentially very important proof!

3. If we can show that there are only 4 structures in the tightness graph for a solution,
can we show that all vertices must be on GT ?

4. Can we even prove that all vertices are on GT if f is a solution?

5. Use the fact that we know distances (small label to big label is small distance) to prove
things. Does there exist a permutation of a subset of vertices of V (G) that decreases
the max needed label? (as a contradiction technique to prove things)

6. Is there an algorithm that improves a

situation by repositioning the labels? (Induction?) From what I have seen, no solution
has ever had a tightness graph where vi vi+3 . The most Ive seen was vi vi+2 and vi vi+1
(as shown in the four structures above).

7. Is there a smoothing algorithm? Given a solution, can we manipulate it such that


all the bumps are smoothed out?

8. Can we start with an arbitrary labeling, and apply an algorithm to decrease its span?
Then there is a necessarily depleted state.

9. F is a solution iff n1
P
i=1 di bi is maximized. If bi = 0, then those corresponding
vertices are tight.

8
10. Tree conjecture: min or max goes on the vertex with greatest degree. This is false:
have one branch extend out really far, and it will want to have a path center. However,
when do the path characteristics take over the large degree vertex? If we make a long
path, and make the degree of the vertex much greater, where will min reside?

11. Edge conjecture: If the diameter is constant, adding an edge will increase rn(G). This
is false. I think it might be true for trees, however. I think it is true that rn(G)
certainly cannot increase, however!

6 Pieces of the Puzzle


Here is some work that I have done while investigating properties of the tightness graph.
Lemma: If F is a solution, then there must exist a walk from x1 to xn on GT .
Proof by contradiction: Assume there is no path from x1 to xn . Let the greatest valued
vertex x1 is connected to on GT be xi . Since xi is not tight with any vertex, other than
vertices with label less than f (xi ), we may increase the value of f (xi ) by 1. Then, we may
increase the values of all labels on the tightness graph that are less than f (xi ) by 1. This
will result in f (x1 ) = 1. However, this is a contradiction, as span(f ) has just decreased, and
we assumed span(f ) was minimal by assuming f was a solution.
In fact, with this proof, we show that there must be a strictly increasing path from x1 to
xn . The other paper that I gave you defined GT as a digraph, which makes it slightly more
clear that the labels must be increasing on the walk from min to max.
I think the next things to prove on the tightness graph of a solution (in order of increasing
difficulty) are the following:

1. Prove that all vertices are on GT

2. Prove that there are no tight edges from xi to xi+3

3. Prove that all bumps on the tightness graph of a solution are 1-bumps

4. Prove that the only possible structures on a tightness graph are the ones shown above.

Unfortunately, these problems have been (unsuccessfully) tackled, and have been quite
difficult to prove. I may have to look into more specific graph types and prove these properties
for certain classes of graphs.

You might also like