Scribd
Upload
309 views4 pages

A Theorem On Coloring The Lines of A Network

1) The document presents a theorem stating that the lines of any network can be colored using at most the ceiling of the maximum number of lines touching one junction. 2) It provides examples to show this number of colors is necessary, such as connecting pairs of junctions with lines in a simple network. 3) The proof of the theorem involves inductively coloring one network by first coloring a similar network with one less junction, and then showing the remaining junction's lines can be properly colored through operations like color and row interchanges in a table representing the network.

Uploaded by

castrojp
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
309 views4 pages

A Theorem On Coloring The Lines of A Network

1) The document presents a theorem stating that the lines of any network can be colored using at most the ceiling of the maximum number of lines touching one junction. 2) It provides examples to show this number of colors is necessary, such as connecting pairs of junctions with lines in a simple network. 3) The proof of the theorem involves inductively coloring one network by first coloring a similar network with one less junction, and then showing the remaining junction's lines can be properly colored through operations like color and row interchanges in a table representing the network.

Uploaded by

castrojp
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

A THEOREM ON COLORING THE LINES OF A NETWORK

By

CLAUDI:

E.

SHANNON

A topological coloring problem has been suggested in connection with the


color coding of wires in electrical units such as relay panels. In these units there
are a number of relays, switches, and other devices A, B, .. , E to be interconnected. The connecting wires are first fonned in a cable with the wires to
be connected to A coming out at one point, those to B at another, etc., and it is
necessary, in order to distinguish the different wires, that all those coming out of
the cable at the same point be differently colored. There may be any number
of wires joining the same two points but no wire joins more than two points.
Assuming that not more than m wires end at anyone point, the question arises
as to the least number of different color codings that is sufficient for any network.
Theorem: The lines of any network can be colored so that no two lines with a
common junction have the same color using at most [1m] colors, where m is the
maximum number of lines touching one junction. This number of colors is
necessary for some networks.
.
Simple networks requiring [1m] colors can be constructed as follows. For
m = 2n, let each pair of the three junctions A, B, C be connected with n lines.
Since all the lines must obviously be different, 3n = [1m] colors are necessary.
For m = 2n + 1, connect AB with n lines, Be with n lines and AC with n + 1
lines. Here again all lines must be different and there are 3n + 1 = [1(2n +
1)] = [1m] lines. Another example for m = 3 is furnished by the cross connection of two pentagons abcde and ABCDE by lines aA, bl), cB, dE, eC.
For the sufficiency proof let us first suppose m even. N ow if N is our given
network it is well known that we may add lines and junction points to get a
regular network N' of degree m, i.e., one in which exactly m lines end at each
junction point. If we can color N' we can surely color N. A theorem due to
Peterson states that any regular network of even degree m = 2n can be factored
into n regular second degree graphs. In our case let the factors of N' be N, ,
N2 , , N A Each of these is merely a collection of polygons which do not
touch each other, and each N" therefore, can be colored with at most three
colors. This gives a total of 3n = 1m colors.*
Peterson has conjectured that any regular bridgeless network of odd degree
2n + 1 can be factored into one first and n second degree graphs, and if this is
true the theorem follows easily in the odd case. However this conjecture has
never been proved for m > 3 and we will use a different attack.
The theorem will be proved by induction, making the coloring of N depend on
coloring a network with one less junction point. Let us eliminate from None
junction point P and the m = 2n + 1 lines coming to it and assume the remaining network to be satisfactorily colored with 3n + 1 colors. Let the junctions
that were connected to P in the original network be numbered 1, 2, .. , 8, and
This proof for m even was suggested by R. M. Foster.
Reprinted from JOURNAL or MATRE.IATICS AND
Vol. XXVIII, No.2, July, 1949

PSTSICS

584

A Theorem on Coloring the Lines of a Network

585

suppose there were PI parallel lines in the first group G1 connecting P to junction
1, etc. N ow after coloring the reduced network we have left available at junction 1 at least [(3n + 1) - (2n + I - PI) = n + PI colors, at junction 2, at
least (n + 112) colors, etc. By choosing properly from these available colors and
by suitable interchanges of certain colors in the part of the network already
colored we will show that the lines from P can be satisfactorily colored.
Let us arrange the data in a table as follows.
Colors
1
2

123
1 1 0 1
1 0 1 0

(3n

1)

lines

(2n

1)
FIG.

In this array the 2n + 1 lines from P are listed vertically, the an + 1 colors
horizontally. If a certain color is available for a certain line a 1 is placed at the
intersection of the corresponding row and column, otherwise a o. In a row
corresponding to a line in Gi there will be (n + Pi) 1'8. By the use of three
operations on this array we will arrange to obtain a series of 1'8 along the main
diagonal and this will represent a coloring scheme for the network. These
operations are:
1. Interchange of columns. This corresponds to merely renumbering the
colors.
2. Interchange of rows. This corresponds to renumbering the lines from P.
3. Interchange of colors in a chain of two colors. Two junctions will be said
to be chained together for two colors if we can go from one junction to the other
along lines which alternate these colors. If we have a satisfactorily colored network and interchange the two colors in a chain of that network over its entire
length (note that in a correctly colored network a chain cannot branch out) then
it is clear that the network will still be satisfactorily colored. We will also use
the fact that if only one of the two colors in question appears at each of three
distinct junctions then only one pair of the junctions (at most) can be chained
together for these two colors, since a chain can only have two ends. Now consider the array of Fig. 1. Suppose that in the first row there is a 0 in one column
and a 1 in another. Interchanging these two colors in the chain starting from
the junction that line one is connected to will be seen to be equivalent to interchanging these two columns for this row and all other rows that are chained to it.
Let us suppose that we have arranged to get 1'8 on the main diagonal D down
to a certain point. We will show that we can get another 1 on the diagonal.

586

C.E.Shannon

(3n

1) Colors

1----0---

(2n
1)
lines

1-1--0---

1
1

X---

T
Flo. 2

Referring to Fig. 2, if there are any 1's in the next line a, on or to the right of T,
one of these may be brought by an interchange of columns, to position X.
Assuming this is not true there will be n + Pi 1's to the left of column T in a
(assuming a is in Gi ) . Hence there are n + p, rows above a having a 1 in D in
the same column as a 1 in a. At least n + 1 of these rows are not in Gi , since
G, has Pi members and we have accounted for one already, namely . Let {j be
one of these, belonging, say, to Gt If ~ has a 1 on or to the right of T, by an
interchange first of columns, then of a and {J this may be moved to X without
affecting the 1 's along D. Assuming this is not true, there are n + Pi 1'8 on {j
to the left of T and hence n + Pi rows above a have a 1 in D in the same column
as a 1 in {j, and of these at least n do not belong to G; (as it only has Pi members).
N ow there are not more than 2n rows above a and therefore the n rows we have
associated with ~ and the n
1 we have associated with a must have at least
one in common, i.e., there exists a row not belonging to G. or Gi and having a 1
in D in the same column that Q and {3 also have a 1. Call this row v, and suppose
it belongs to G". If 'Y has a 1 on or to the right of T it may be moved to X by
first interchanging columns and then rows a and 'Y 8S before. Assuming this is
not true, there are O's and l's at the intersections of a, {j, 1', and T, S, as shown
in Fig. 2 and X is o. Hence at least one of a, P, 'Y is not chained to either of the
others by the t\VO colors of T and S. If it is a, interchange the chain starting at
i and the 1 at the as intersection moves to X without affecting D. If it is p,

A Theorem on Coloring the Lines of a Network

587

interchange the chain starting at j and then rows a and s. This moves the 1 at
the {jS intersection to X and its place is taken by the 1 at
If it is 'Y, inter..
change the chain starting at k and then the rows a and "I so that the 1 at "IS moves
to X and the 1 at aR takes its place.

as.

BELL TELEPHONE LABORATORIES,

INC.,

NEW YORK CITY.

(Received September 14. 1948)

You might also like