Chromatic index np-complete

Ian Holyer

Amir Sabag, Daniel Rosenberg, Ori Darshan

may. 2022

1
1 introduction
from Vizing theorem we know: the chromatic index of G(V,E) graph is:

∆(G) ≤ χ (G) ≤ ∆(G) + 1

′

Definition 1 given graph G. its chromatic index number, marked as χ′ (G)

is the minimum edge-colorable number of the graph G.

Definition 2 given graph G. max degree of G is marked by ∆(G)

in this assignment we gonna prove via the article of holyer: to determine
where χ’ (G) stands (the left or right equation) is a NP complete problem.
moreover, we will prove that even for just cubic graphs (3-Reg graphs) its
NP-C problem. to proof if its a NP-complete problem

Definition 3 chromatic index number is NP-C if:

a. chromatic index number ∈ NP (there is validation algorithm)

b. chromatic index number ≤P NP-Hard (there is polynomial reduction)

or in another words, we gonna make polynomial reduction to 3-SAT
problem. BUT in place of proving that for any graph the problem is
NP-hard, we will prove it just for arbitrary cubic graphs (3-reg) which
is enough.

(*) reminder:

Definition 4 3-SAT problem define as follows. A set of clauses C = {C1 , C2 , ..., Cr }

in variables u1 , u2 , ..., us is given.
each clause Ci consisting of three literals li,1 , li,2 , li,3 where a literal li,j is
either a variable uk or its negation uk negative.

Problem 1 the problem is to determine whether C is satisfiable, that is,

whether there is truth assigment to the variables which simultaneously satis-
fies all the clauses in C. A clause is satisfied if one or more of its literals has
the value of ”True”

1
2 Chromatic index problem ∈ NP
given G(V,E) and function f : E → N the validation algorithm is:

1. For every e - edge, check if there is ’adjacent’ edge to e with the same
color number

2. if found at least one pair of adjacent edges: return False, else: return
True (there is no conflict of two color numbers of adjacent edges). □

3 Preparations to build graph from 3-SAT

before we prove there is polynomial reduction from 3-SAT to chromatic index
of cubic graph. we will work about several major things: prove the next
lemma by isacas, define unique components that will compose the graph and
then explain the build of the reduction.

Lemma 1 The parity condition: let G be a cubic, 3-edge-colored graph and

V ′ ⊂ V (G) a set of vertices of G. Let E ′ ⊂ E(G) be the set of edges of G
which connect V’ to the reminder of the graph. If the number of edges of
color i in E’ is ki (i ∈ 1, 2, 3) then

k1 ≡ k2 ≡ k3 (mod2)

Proof: define E12 := {e ∈ E|ke = 1 ∨ ke = 2}

denote: there is at least one cycle at E12 group. (for every vertex in E12
there is 2 edges, means there is |VE12 | = n → |E12 | = 2n ≥ n → there is cycle
and now we prove that E12 consists collection of cycles.
assume for contradiction: w.t.o.l.g there is edge e : ke = 1 (the color of e is
1) that doesnt part of any cycle that in E12 . notice: this edge is connecting
between two vertices call them v, u
from both vertices, there is also edges with color 2 which part of E12 but
since edge e doesnt part of any cycle, both of this edges is no part of any cycle
as well (since e cannot close cycle between them and the graph is 3-edges
colored, and cubic, and E12 holds only edges colored by nums 1, 2)
its means the edges colored num 1 connected to this two edges (color 2)
is not part of cycle as well, and so on.

2
if we got to edge which part of cycle: its contradiction because its means
e is part of this cycle as well. else: there is no cycle at all in E12 but this
contradiction to that - E12 consists at least one cycle. }
we know E12 is collection of cycles. we claim that |E12 ∩ E ′ | ≡ 0 (mod 2)
(is even!) this claim is trivial since E12 consist only cycles, and E’ is all the
edges that connect between 2 diffrent coponents. since E12 consists cycles,
for every path from one component to another, there is way back (otherwise
its a lollipop and then we got an edge part of E12 and not part of a cycle.
because of that, we got: k1 + k2 ≡ 0 (mod 2) which gives k1 ≡ k2 (mod
2). similarly we can define group E23 and return on the proof again, those
means: k2 ≡ k3 (mod 2). □
the components Given an instance C of problem 3-SAT, we aim to show
how to construct a cubic graph G which is 3-edge-colorable iff (if and only if
⇐⇒ ) C is satisfiable.
the graph G will be put together from different components which each
of them carry specific tasks. the information between the components will
be carry via pair of edges. in a 3-coloring-edges of G, such a pair of edges is
said to represent the value of ”T”-True if the pair is within the same color,
”F”-False if their color is distinct.

3.1 The inverting component:

we name to the component in the above image ”the inverting component”

and mark it: Cinvert its purpose is to build from variable uk its negativity

3
via connecting pair of edges from Cinvert to uk then another pair of edges to
another component (which will represent a caluse Ci ). to verify action of
component we will prove the next claim:

Lemma 2 Cinvert is 3-colorable-edges ⇐⇒ from the set of edges {a.b.c.d.e}

only one of the pairs [a,b] ∨ [c,d] is with the same color, and the other three
output edges is with distinct color.

Proof:
⇒ suppose there that Cinvert then from the set of edges {a.b.c.d.e} only one
of the pairs [a,b] ∨ [c,d] is with the same color.
take two Cinvert components call them C1 , C2 . we will connect between their
same output edges: the edge a of C1 , will connect to C2 , and by this way we
connected between their two v1 vertexes. same with the other edges b,c,d,e.
lets mark our new graph:

G = (V, E) :

V (G) = {V (C1 ) ∪ V (C2 )},

E(G) = {all inner edges of C1 , C2 }∪
{(v1C1 , v1C2 ), (v2C1 , v2C2 ), (v3C1 , v3C2 ), (v4C1 , v4C2 ), (v6C1 , v6C2 )}
note: the above line means: the outer edges a,b,c,d,e has been merged with
their adjacent edge at the other component.
so not we can use right now the parity condition on the above new graph.
lets choose V (C1 ) as the group V’ and V (C2 ) as the reminder. since we
have only 5 edges (the merged a,b,c,d,e) from the parity condition we know:
ki (i ∈ 1, 2, 3) then
k1 ≡ k2 ≡ k3 (mod2)
the only combination that works for total edges of 5 is 3-1-1. all is left to
prove that its wont be legit coloring if edge ’e’ - (v6C1 , v6C2 is from the ’3’
group in the combination while both of outer pairs is distinct: we will take
a look on only one of the components, without less of generality take C1 .
case 1: edges a and d colored by 1, b color 2 and c color 3. so [v1,v3] is
colored by 2 and [v2,v4] colored by 3. easily can see that [v2,v7] colored by 1
and [v3,v7] shall be colored by 1 as well. but its a contradiction to 3-coloring
in v7.

4
 case 2: edges a and c colored by 1, b color 2, and d color 3. so [v1,v3] is
colored by 2 or 3 and [v2,v4] is colored by 1. [v4,v5] colored by 2 and [v2,v7]
colored by 3. at this point its can easily seen that the edges [v1,v5] and
[v2,v7] colored same and ∈ {2, 3} colors. but that will make a contradiction
at v5 or v7 depend on which color we will choose for [v1,v3].
little brief to summarize it: we duplicated Cinvert component to not leave
the outer edges ”undefined edges”. then we used the parity condition on the
new graph we built which leaves us just one possible combination (3-1-1) to
color its edges [a,b,c,d,e]. and for the finale we proved that both the pairs
[a,b], [c,d] cannot be distinct together, BUT they also cannot be the same
color since our combination is 3-1-1. which means one of them is same color
and the other three edges are distinct color.

⇐ w.t.o.l.g we say [a,b] same value, and [c,d,e] distinct.

since we can manually color this component, we done with this side. □

5
3.2 The variable setting component:

we call to the component in the above image ”the variable setting compo-
nent”. given uk variable, mark it Cset−uk the component holds as many Cinvert
components as the amount of appearances uk among the clauses of C. that
means, the amount of output edge-pairs is as the the amount of appearances
uk in C. which takes us to the next claim that shall be proved:
lets formulize the construction of Cset−uk :
given n ∈ N, amount of appearances of variable uk at clauses of C the con-
struction will work ’inductively’ via the next steps:

a. create 2n Cinvert components and call each output edge of given Cinvert
e.g ea means the output edge a of the component.

b. split them into pairs of components [Codd , Ceven ] (e.g [1,2], [3,4] .. [2n-1,
2n], ) each pair will be connected their pair of edges: Codd : ec , ed will
connect to ea eb of Ceven .

c. the output edges ec ed of Ceven in every pair of components will connect

to ee pair of the next chronology pair of components (e.g ec ed of C2
from [1,2] will connect to pair ee from [3,4]).

d. the pair of edges ee from [1,2] will connect to pair ec ed of C2n

e. every pair of edges ea eb from Codd remained unused and their roll is
pair output edges

6
Lemma 3 Cset−uk is 3-edge-colorable ⇐⇒ all the output pairs represent
the same value.

reminder: same value means ”F” or ”T” (all output pairs is distinct or same
color).
proof:
⇒: suppose for contradiction there is two output pairs with different value
and the Cset−uk is 3-edge-colored. two different output pairs means two pairs
of components which their output pairs is different.
take two pairs of components not with the same value that are connected
each other.
from the way that we built the whole var-set component (section b.) the
component with the output value of ”T” is connected between his Cinvert via
two pairs of edge with distinct value. so its a component with three pairs of
output, all of them with the same value ”T”. mark this component as A.
on another hand, for components with output value ”F”, there is two
options: 1- two other pairs with value ”F”, 2- one value ”F”, one value ”T”
(granted from pair of [ee , ee ] edges of two Cinvert . mark this component as B.
conclusions:
ˆ each component A or B have three output values, one outer to the
Cset−uk and two inner connection to another components (type A or
B).

ˆ two inner output connections is different pairs of edges, one is from

type [ea , eb ] ∨ [ec , ed ] and another from type [ee , ee ].

ˆ B with output value ”F” from option (1-) cannot connect to A.

ˆ B with output value ”F” from option (2-) can connect to A via connec-
tion between pairs [ee , ee ] to [ec , ed ] (via section d. of the construction)
notice: to be able to connect A to component B we still remains with one
inner output that shall connect to another component (type A or B) from A.
BUT no matter how many components type A we will chain to the reminder
inner output of the first component A, still, we remain with inner connection
type [ee , ee ]. this type of ”T” inner output cannot connect to inner output
of ”T” from B because there is no [ee , ee ] connect to [ee , ee ] at the variable-
setting component.

7
 ⇐ suppose all the output pairs represent the same value. that means we
will prove with induction on the amount of output pairs of the component
that’s 3-edge-colorable: base: n=1, this means we used exact 2 Cinvert com-
ponents. we will color each of the components as we colored in the image
above (leftarrow side of the Cinvert claim). if the output pair is same value.
connect edges type e of both components to a,b of C2 . otherwise, switch the
number of the colors so [ee , ee ] will be distinct same as the distinct color left
at C2 .

induction assumption: the component is 3-edge-colorable for any k≤ n.

now we prove its works for n+1. take the components C2n+1 , C2n+2 out
of the given var component (and connect C2n−1 , C2n as defined in our algo).
now we have only n output pairs. from i.a we hold 3-edge-colorable graph.
but also C2n+1 , C2n+2 is 3-edge-colorable. BUT we have one outer output
pair and two inner output pairs which their types is [ee , ee ] and [ec , ed ] from
C2n . since the value of both pairs is independent, all is left is to set them
exactly as their relatives at [C1 ,C2 ] and [C2n−1 , C2n ]. end of the induction.
end of both sides □

8
3.3 Satisfaction setting component:

we call to the component in the above image ”the satisfaction setting compo-
nent”. the truth of each clause cj will be tested by a satisfaction component
as shown in Fig.3.

Lemma 4 (which is trivial): this component can be 3-edge-colorable ⇐⇒

the three input pairs of edges not all represent ”F” (which means at least one
of the pairs represent ”T”)

note: every input connection to the testing component is via Cinvert compo-
nent which negatives the value from the variable-setting components.
proof:

⇒ suppose the testing component is 3-edge-colorable and all the input

pairs is ”F”. this means after the Cinvert all the pairs represents ”T”. which
means each pairs is the same color. lets look on the input pairs, each pair
share one common edge, and another pair joint edges. lets look on one pair
represent ”T” call it A which connected to edges x and y. x,y must be within
the same color. also, each of the other input pairs shares one edge x or y

9
with A. but that means [z,y] and [w,x] with the same color as well and thats
a contradiction.
⇐ suppose at least one of the input pairs is ”T”. that means after the
Cinvert connecting, its ”F” value. so at least one of the pairs [w,x], [x,y], [y,z]
is distinct. if that ”F” is connected to [x,y] the other two pairs of input is
joint thus the component can be easily 3-edge-colord. if that ”F” without
the loss of generality is connected to [w,x] ([y,z] same case): with more ”T”
inputs, the edge z shall be with same color as x which is distinct to w and
thats. with more ”F” inputs the color of edge y can be either the same as w
or distinct, and since we can control with edge y color, we can also control
edge z color and gain a valid coloring(click here). again, this side can be
easily painted, there is nothing new to be proved. □

3.4 Constructing the graph:

given an instance C of 3-SAT and construct from it a graph G as follows:

for each variable ui take a variable setting component Ui with one output
pair of edges for every appearances of ui or not ui in the clauses of C. take
satisfaction testing component for every clause of C. Suppose literal li,k in
Cj is variable ui then add to the input pair k of Cj an output pair of Ui . if
on the other hand li,k represent the variable not ui then insert an inverting
component between the k input pair of Cj and the associated output pair
of Ui . lets call the resulting graph - H. we still have the edges marked as
ee of the inverting component, and also the output edges from the testing
component which not associated anywhere yet. then, copy H, and add every
orphan edge to its twin at her copy graph. call the result graph G.

4 Chromatic index problem ∈ NP-hard

proof of the phase: chromatic index number ≤P NP-Hard (there is polyno-
mial reduction)
in another words: graph G is 3-edge-colorable ⇐⇒ C ∈ 3 − SAT is satis-
faction.

⇒ due to claims above we know that every ’satisfaction testing compo-

nent’ is 3-edge-colorable iff at least one of pairs of input is within the same

10
color. so all is left to do is to check every pair of inputs from testing compo-
nent and mark the variable its represent via the next options:
a. pair is T and there is no inverting component connected to this input:
set the corresponds variable ui to T.
b. pair is T and there is inverting component connected as input: set the
corresponds variable ui to F.
c. pair is F and there is no inverting component connected as input: set
the corresponds variable ui to F.
d. pair is F and there is inverting component connected as input: set the
corresponds variable ui to T.
since we proved the claim ”Cset−uk is 3-edge-colorable ⇐⇒ all the output
pairs represent the same value, there is no confilct between the setting of
each variable. every testing component represent clause Cj so we done with
this side. □
⇐ this side is trival. since every testing component represnt a clause Cj
and is 3 colorable. also, there is at least one input of pairs which have the
same colors (”T”). and also we proved every setting component is 3 colored
and its output edges is all synced colors. the only way to conflict its at the
inverting component which is fine because the only cases is:
1- we connected F from setting component and its output to testing com-
ponent is T (wont do any problem)
2- we connected T from setting component and its output to testing com-
ponent is F,
BUT there is no conflict in coloring of the testing component as we mentioned
above, end of this side as well. □

5 Summarize:
along this article we managed to prove the chromatic index number to be
NP-Complete problem.
we used the construction of the components of Holyer to make a reduction
from 3-SAT problem.
and we proved that the chromatic index number ∈ NP-C.

11
6 special thanks:
our teacher doctor Elad aigner horev.

12