.
...
A few inequalities related to spectral excess theorem
Chih-wen Weng (翁志文)
(joint work with Guang-Siang Lee(李光祥))
Department of Applied Mathematics National Chiao Tung University
August 19, 2012
f (eigenvalues) ≤ g
(graph parameters eigenvectors)
≤ h(eigenvalues).
quasi-spectral characterization
?
quasi-spectral characterization
?
spectral characterization
I
(still working)
Preliminaries
.
1
..
Throughout let G = (V G, EG) be a simple connected graph of order n and diameter D..
2
..
Assume that adjacency matrix A has d + 1 distinct eigenvalues λ0 >
λ1 > . . . >
λd
with corresponding multiplicities1 = m
0 , m 1 , ··· ,m d
. d is called thespectral diameter
of G..
3
..
It is well-known that D≤ d.
.
4
..
Z(x) :=
∏ d i=0
(x
−
λi
) is theminimal polynomial
of A.Inner product
Consider the vector spaceR
d
[x] ∼=R[x]/⟨Z(x)⟩ with the inner product⟨p(x),q(x)⟩ :=
1n
tr(p(A)q(A)) =1n ∑
i, j
(p(A)
◦ q(A)) i j ,
for p(x), q(x)
∈ R d
[x], where◦ is the entrywise product of matrices.
Predistance polynomials
.
Definition 1.1
.....
(i) The orthogonal polynomials 1 = p
0
(x), p1
(x), . . . , pd
(x) in Rd
[x] satisfyingdeg p
i
(x) = i and⟨p i
(x), pj
(x)⟩ =
δi j p i
(λ0
) are called thepredistance polynomials
of G.(ii) The polynomial
H(x) := n
∏ d i=1
x −
λi
λ
0 −
λi
is called the
Hoffman polynomial
of G. Moreover, G is regular iff H(A) = J, the all 1’s matrix.The sum of all predistance polynomials gives the Hoffman polynomial
H(x) = p 0
(x) + p1
(x) +··· + p d
(x) andH(A) = p 0
(A) + p1
(A) +··· + p d
(A).Three -term recurrence
The predistance polynomials satisfy a three-term recurrence:
xp i
(x) = c′ i+1 p i+1
(x) + a′ i p i
(x) + b′ i −1 p i−1
(x) 0≤ i ≤ d,
where c′ i+1 , a ′ i , b ′ i−1 ∈ R with b ′ −1
= c′ d+1
:= 0.Three -term recurrence for bipartite graph
.
1
..
If G is bipartite, then the predistance polynomials satisfy a three-term recurrence of the formx 2 p i
(x) = Xi+2 p i+2
(x) +Yi p i
(x) + Zi −2 p i −2
(x) 0≤ i ≤ d,
(1) whereX i+2
= c′ i+1 c ′ i+2 , Y i
= b′ i c ′ i+1
+ b′ i−1 c ′ i , Z i −2
= b′ i −2 b ′ i −1 . .
2
..
Moreover, if G is bipartite, then for 0≤ j ≤ d, a ′ j
= 0, andp j
(x) is even or odd depending on whether j is even or odd.Distance polynomials
.
1
..
Let α be the eigenvector of A corresponding to λ0
such that αt
α = n and all entries are positive. Note thatα = (1,1,...,1)
t
iff G is regular..
2
..
The matrix Ai
, indexed by V G, satisfying (Ai
)uv
={ α
u
αv ,
if∂(u,v) = i;0, else. is called the
i-th (weighted) distance matrix
of G..
3
.. A 0
= p0
(A)(= I) iff G is regular..
4
.. A 0
+ A1
+··· + A D
= H(A) = p0
(A) + p1
(A) +··· + p d
(A).The spectral excess and the excess
.
1
.. p d
(λ0
) is called thespectral excess
of G..
2
..
δd
:=1 n
∑i, j
(Ad ◦ A d
)i j
is called theexcess
of G, where Ad
:= 0 if D < d.If G is regular, thenδ
d
is the average number of vertices to have distance d to a vertex.The spectral excess theorem (SET)
.
Theorem 1.2
.....
δ
d ≤ p d
(λ0
) with equality iff G is a distance-regular graph.[FG1997] M.A. Fiol and E. Garriga, From local adjacency polynomials to local pseudo-distance-regular graphs, J. Combin.
Theory Ser. B 71 (1997), 162–183.
Related definitions
Define
δ
i
:=1n ∑
u,v
(A
i ◦ A i
)uv ,
δ≥i
:=δi
+δi+1
+··· ,
p ≥i
(λ0
) := pi
(λ0
) + pi+1
(λ0
) +··· , p even
(λ0
) := p0
(λ0
) + p2
(λ0
) +··· ,
p odd ≥i
(λ0
) := pi
(λ0
) + pi+2
(λ0
) +···
for odd i,A odd
:= A1
+ A3
+···
...
Modify the proof of SET
.
Proposition 1.3
.....
δ
≥i ≤ p ≥i
(λ0
) with equality iff A≥i
= p≥i
(A)..
Proposition 1.4
.....
δ
≤i ≥ p ≤i
(λ0
) with equality iff A≤i
= p≤i
(A).G is bipartite
.
Proposition 1.5
.....
If G is bipartite and
∗ ∈ {even,odd} then
δ≥i ∗ ≤ p ∗ ≥i
(λ0
) with equality iff A∗ ≥i
= p∗ ≥i
(A)..
Proposition 1.6
..If G is bipartite and
∗ ∈ {even,odd} then
δ≤i ∗ ≥ p ∗ ≤i
(λ0
) with equality iff A∗
= p∗
(A).Questions?
δ
i ≤ p i
(λ0
) or δi ≥ p i
(λ0
)?Which graphs have the property thatδ
i
= pi
(λ0
)?The case i = 0
.
Proposition 2.1
.....
δ
0 ≥ 1(= p 0
(λ0
)) ( i.e. α1 4
+α2 4
+··· +
αn 4 ≥ n),
and the following are equivalent..
1
..
δ0
= 1. (i.e. α1 4
+α2 4
+··· +
αn 4
= n.).
2
.. A 0
= I..
3
.. G is regular.
.
4
..
The entries ofα are all 1.The above inequality is fromδ
≤0 ≥ p ≤0
(λ0
) which we mentioned earlier, but also follows from the Cauchy-Schwarz inequalityThe case i = 1
A bipartite graph with bipartition V (G) = X
∪Y is biregular
if there exist distinct integers k̸= k ′
such that every x∈ X has degree k,
and every y∈ Y has degree k ′ .
.
Proposition 3.1
.....
δ
1 ≥ p 1
(λ0
), and the following statements are equivalent.(i) δ
1
= p1
(λ0
) (or equivalently δ1 k =
λ0 2
), (ii) A1
= p1
(A),(iii) G is regular or G is bipartite biregular.
.
Corollary 3.2
.....
There is no bipartite biregular graph G with exactly four distinct eigenvalues.
The idea of the proof is to modify a proof in
[DDFGG2011] C. Dalfó, E.R. van Dam, M.A. Fiol, E. Garriga and B.L. Gorissen, On almost distance-regular graphs, J. Combin.
Theory Ser. A 118 (2011), 1094–1113.
The case i = 2
.
1
..
If d = 2 thenδ2 ≤ p 2
(λ0
)..
2
..
If G is regular bipartite, thenδ////0
+δ2 ≥ p
///////// p0
(λ0
)+2
(λ0
)..
3
..
There is no hope to determine the order ofδ2
and p2
(λ0
) uniformly..
Definition 4.1
.....
Let G be a graph and i is a nonnegative integer. We say the numbers c
i , a i , b i
respectively arewell-defined
in G if for anyx, y ∈ V(G) with
∂(x,y) = i, the numbersc i
:=|G 1
(x)∩ G i −1
(y)|, a i
:=|G 1
(x)∩ G i
(y)|, b i
:=|G 1
(x)∩ G i+1
(y)|,
respectively are independent of the choice of x, y.d y
d x
c i b i
.
Definition 4.2
.....
A graph G is
t-partially distance-regular
if for 2≤ i ≤ t and the
numbers ci , a i −1 , b i −2
are well-defined..
Lemma 4.3
.....δ
0
= p0
(λ0
) andδ2
= p2
(λ0
) iff G is 2-partially distance-regular..
Definition 4.4
.....
For a connected bipartite graph G with bipartition X
∪Y, the halved graphs G X
and GY
are the two connected components of the distance-2 graph of G.G :
'
&
$
% G X
'
&
$
% G Y
.
Theorem 4.5 (BCN, Prop 4.2.2, p.141)
.....
The halved graphs of a bipartite distance-regular graph are again distance-regular and, in the case where G is vertex transitive, the two halved graphs are isomorphic.
.
Lemma 4.6
.....
If G = (X,Y ) is a connected regular bipartite graph with
δ
2
= p2
(λ0
) (so 2–partially distance-regular by previous lemma), then the halved graphs GX
and GY
have the same spectrum.Bipartite graphs with δ d −1 = p d −1 ( λ 0 )
.
Lemma 5.1
.....
Let G be bipartite andδ
d −1
= pd −1
(λ0
). ThenA d −1
= pd −1
(A), Ad −3
= pd −3
(A), Ad −5
= pd −5
(A), . . . . In particular G is regular (if A0
= p0
(A)) or bipartite biregular (ifA 1
= p1
(A))..
Theorem 5.2
..Let G be a connected bipartite graph with bipartition X
∪Y, odd d.
Then the following are equivalent.
(i)
δd −1
= pd −1
(λ0
);(ii) G is 2-partially distance-regular and both halved graphs G X
The following example shows that a bipartite graph satisfying Theorem 5.2(i)-(ii) and D = d needs not to be distance-regular graph.
.
Example 5.3
.....
Consider the regular bipartite graphs G on 20 vertices obtained from the Desargues graph (the bipartite double of the Petersen graph) by the GM-switching (a way to produce cospectral
nonisomorphic graphs). One can check (by Maple) that D = d = 5, sp G =
{3 1 , 2 4 , 1 5 , (−1) 5 , (−2) 4 , (−3) 1 }, p 0
(x) = 1, p1
(x) = x,p 2
(x) = x2 − 3, p 3
(x) = (x3 − 5x)/2, p 4
(x) = (x4 − 9x 2
+ 12)/4,p 5
(x) = (x5 − 11x 3
+ 22x)/12, Ai
= pi
(A) for i∈ {0,1,2,4}. Hence
δ0
= p0
(λ0
) = 1,δ1
= p1
(λ0
= 3,δ2
= p2
(λ0
) = 6,δ
4
= p4
(λ0
) = 3),δ3
= 32/5,δ5
= 3/5 , p3
(λ0
) = 6,p 5
(λ0
) = 1̸= 3/5 =
δ5
. Then G is not distance-regular.The following example provides a graph G satisfying Theorem 5.2(i)-(ii) with D = d
− 1.
.
Example 5.4
..Consider the Möbius-Kantor graph G, i.e., the generalized Petersen graph GP(8, 3) with vertex set
{u 0 , u 1 , . . . , u 7 , v 0 , v 1 , . . . , v 7 } and
edge set{u i v i , u i u i+1 , v i v i+3 |0 ≤ i ≤ 7} with arithmetic modulo 8.
One can check (by Maple) that D = 4 < 5 = d, sp G =
{3 1 , √
3
4 , 1 3 , ( −1) 3 , ( − √
3)
4 , ( −3) 1 }, p 0
(x) = 1, p1
(x) = x,p 2
(x) = x2 − 3, p 3
(x) = 2(x3 − 5x)/5, p 4
(x) = (x4 − 10x 2
+ 15)/6,p 5
(x) = (x5 − 56x 3 /5 + 21x)/18, A i
= pi
(A) for i∈ {0,1,2,4}
(δ
0
= p0
(λ0
) = 1,δ1
= p1
(λ0
) = 3, δ2
= p2
(λ0
) = 6,δ
4
= p4
(λ0
) = 1),δ3
= 5, p3
(λ0
) = 24/5, p5
(λ0
) = 1/5. Note thatG 2
= 2X , where X is the 16-cell graph(http://mathworld.wolfram.com/16-Cell.html), which is distance-regular with sp X =
{6 1 , 0 4 , ( −2) 3 }.
The parity of d makes things different
.
Theorem 5.5
.....
Let G be a connected bipartite graph with bipartition X
∪Y and even d. Then the following are equivalent.
(i) G is distance-regular;
(ii) G is 2-partially distance-regular and both of the halved graphs G X
and GY
are distance-regular of diameter d/2.In the next two pages, we provide two examples of
non-distance-regular graphs that satisfy p
d−1
(λ0
) =δd−1
when d is even. The first one is bipartitle biregular and the second one is regular..
Example 5.6
..Consider the bipartite graphs G on 25 vertices obtained from the Petersen graph by subdividing each edge once. One can check (by Maple) that D = d = 6,
sp G =
{ √
6
1 , 2 5 , 1 4 , 0 5 , ( −1) 4 , ( −2) 5 , ( − √
6)
1 }, the
Perron-Frobenius vectorα = (√5/4,
··· ,
√| {z 5/4}
10
,
√5/6,
··· ,
√| {z 5/6}
15
)
t
,p 0
(x) = 1, p1
(x) = 5√
6x/12, p
2
(x) = 15(x2 − 12/5)/16, p 3
(x) = 5√
6(x
3 − 4x)/12, p 4
(x) = 25(x4 − 21x 2 /4 + 3)/28, p 5
(x) = 5√
6(x
5 − 7x 3
+ 10x)/24,p 6
(x) = 5(x6 −65x 4 /7 + 22x 2 −48/7)/24, e A i
= pi
(A) for i∈ {1,3,5}
(δ
1
= p1
(λ0
) = 5/2,δ3
= p3
(λ0
) = 5,δ5
= p5
(λ0
) = 5), δ0
= 25/24, δ2
= 85/24,δ4
= 85/12,δ6
= 5/6, p0
(λ0
) = 1, p2
(λ0
) = 27/8,p 4
(λ0
) = 375/56, p6
(λ0
) = 10/7. Note that G2
is the disjoint union of the Petersen graph X and the line graph Y of X. We have.
Example 5.7
.....
Let G be the Hoffman graph (A graph nonisomorphic but cospectral to 4-cube). Then sp G =
{4 1 , 2 4 , 0 6 , ( −2) 4 , ( −4) 1 }, D = d = 4, p 0
(x) = 1, p1
(x) = x, p2
(x) = (x2 − 4)/2,
p 3
(x) = (x3 − 10x)/6, p 4
(x) = (x4 − 16x 2
+ 24)/24, and A3
= p3
(A).Note that G