A Conjecture about spectral distances between cycles, paths and certain trees
Alireza Abdollahi
Department of Pure Mathematics, Faculty of Mathematics and Statistics, University of Isfahan, Isfahan 81746-73441, Iran; and School of Mathematics, Institute for Research in Fundamental Sciences (IPM), P.O. Box 19395-5746, Tehran, Iran
[email protected]
and
Niloufar Zakeri
Department of Pure Mathematics, Faculty of Mathematics and Statistics, University of Isfahan, Isfahan 81746-73441, Iran
[email protected]
Abstract.
We confirm the following conjecture which has been proposed in [Linear Algebra and its Applications , 436 (2012), No. 5,
1425-1435.]:
0.945 ≈ lim n ⟶ ∞ σ ( P n , Z n ) = lim n ⟶ ∞ σ ( W n , Z n ) = 1 2 lim n ⟶ ∞ σ ( P n , W n ) ; lim n ⟶ ∞ σ ( C 2 n , Z 2 n ) = 2 , 0.945\approx\displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},Z_{n})=\displaystyle\lim_{n\longrightarrow\infty}\sigma(W_{n},Z_{n})=\frac{1}{2}\displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},W_{n});\ \displaystyle\lim_{n\longrightarrow\infty}\sigma(C_{2n},Z_{2n})=2,
where σ ( G 1 , G 2 ) = ∑ i = 1 n | λ i ( G 1 ) − λ i ( G 2 ) | \sigma(G_{1},G_{2})=\sum_{i=1}^{n}|\lambda_{i}(G_{1})-\lambda_{i}(G_{2})| is the spectral distance between n n vertex non-isomorphic graphs G 1 G_{1} and G 2 G_{2} with adjacency spectra λ 1 ( G i ) ≥ λ 2 ( G i ) ≥ ⋯ ≥ λ n ( G i ) \lambda_{1}(G_{i})\geq\lambda_{2}(G_{i})\geq\cdots\geq\lambda_{n}(G_{i}) for i = 1 , 2 i=1,2 , and P n P_{n} and C n C_{n} denote the path and cycle on n n vertices, respectively; Z n Z_{n} denotes the coalescence of P n − 2 P_{n-2} and P 3 P_{3} on one of the vertices of degree 1 of P n − 2 P_{n-2} and the vertex of degree 2 2 of P 3 P_{3} ; and W n W_{n} denotes the coalescence of Z n − 2 Z_{n-2} and P 3 P_{3} on the vertex of degree 1 of Z n − 2 Z_{n-2} which is adjacent to a vertex of degree 2 2 and the vertex of degree 2 2 of P 3 P_{3} .
Key words and phrases: Adjacency spectra of graphs; Spectral distance of graphs; Path; Cycle
2010 Mathematics Subject Classification: 05C50; 05C31
1. Introduction and Results
All graphs considered here are simple, that is finite and undirected without loops and multiple edges. By the eigenvalues of G G , we mean those of its adjacency matrix.
In [4 ] , the following conjecture has been proposed:
0.945 ≈ lim n ⟶ ∞ σ ( P n , Z n ) = lim n ⟶ ∞ σ ( W n , Z n ) = 1 2 lim n ⟶ ∞ σ ( P n , W n ) ; lim n ⟶ ∞ σ ( C 2 n , Z 2 n ) = 2 , 0.945\approx\displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},Z_{n})=\displaystyle\lim_{n\longrightarrow\infty}\sigma(W_{n},Z_{n})=\frac{1}{2}\displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},W_{n});\ \displaystyle\lim_{n\longrightarrow\infty}\sigma(C_{2n},Z_{2n})=2,
where σ ( G 1 , G 2 ) = ∑ i = 1 n | λ i ( G 1 ) − λ i ( G 2 ) | \sigma(G_{1},G_{2})=\sum_{i=1}^{n}|\lambda_{i}(G_{1})-\lambda_{i}(G_{2})| is the spectral distance between n n vertex non-isomorphic graphs G 1 G_{1} and G 2 G_{2} with adjacency spectra λ 1 ( G i ) ≥ λ 2 ( G i ) ≥ ⋯ ≥ λ n ( G i ) \lambda_{1}(G_{i})\geq\lambda_{2}(G_{i})\geq\cdots\geq\lambda_{n}(G_{i}) for i = 1 , 2 i=1,2 .
In this paper we completely prove the conjecture.
The spectral distance between graphs is studied based on different matrix representations of graphs and norms in many papers for example see [1 , 3 , 5 , 6 ] .
Let us first introduce some notations.
The coalescence of two graphs G G and H H is obtained by identifying a vertex u u of G G with a vertex v v of H H ; Z n Z_{n} denotes the coalescence of P n − 2 P_{n-2} and P 3 P_{3} on one of the vertices of degree 1 of P n − 2 P_{n-2} and the vertex of degree 2 2 of P 3 P_{3} ; and W n W_{n} denotes the coalescence of Z n − 2 Z_{n-2} and P 3 P_{3} on the vertex of degree 1 of Z n − 2 Z_{n-2} which is adjacent to a vertex of degree 2 2 and the vertex of degree 2 2 of P 3 P_{3} . They are depicted in Figure 1 (Their spectra can be found in [2 ] , p. 77).
Our main result is as follows.
Theorem 1.1 .
(1)
lim n ⟶ ∞ σ ( C 2 n , Z 2 n ) = 2 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(C_{2n},Z_{2n})=2.
(2)
lim n ⟶ ∞ σ ( P n , Z n ) = 8 − 8 2 + 2 π π ≈ 0.945 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},Z_{n})=\frac{8-8\sqrt{2}+2\pi}{\pi}\approx 0.945.
(3)
lim n ⟶ ∞ σ ( W n , Z n ) = 8 − 8 2 + 2 π π ≈ 0.945 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(W_{n},Z_{n})=\frac{8-8\sqrt{2}+2\pi}{\pi}\approx 0.945.
(4)
lim n ⟶ ∞ σ ( P n , Z n ) = lim n ⟶ ∞ σ ( W n , Z n ) = 1 2 lim n ⟶ ∞ σ ( P n , W n ) = 8 − 8 2 + 2 π π ≈ 0.945 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},Z_{n})=\displaystyle\lim_{n\longrightarrow\infty}\sigma(W_{n},Z_{n})=\frac{1}{2}\displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},W_{n})=\frac{8-8\sqrt{2}+2\pi}{\pi}\approx 0.945.
Figure 1. Graphs Z n Z_{n} and W n W_{n} .
2. Proof of the conjecture
To prove the following results, we need the Table 1.
Table 1. Some specific graphs and their spectra
Proof of Theorem 1.1 .
(1)
Let n ≥ 2 n\geq 2 . By Table 1,
λ 1 ( C 2 n ) > λ k ( C 2 n ) = λ k + 1 ( C 2 n ) > λ k + 2 ( C 2 n ) = λ k + 3 ( C 2 n ) > λ 2 n ( C 2 n ) , k = 2 , 4 , 6 , … , 2 n − 4 . \lambda_{1}(C_{2n})>\lambda_{k}(C_{2n})=\lambda_{k+1}(C_{2n})>\lambda_{k+2}(C_{2n})=\lambda_{k+3}(C_{2n})>\lambda_{2n}(C_{2n}),\ \ k=2,4,6,\ldots,2n-4.
Assume that n n is odd. We have
λ k ( C 2 n ) > λ k ( Z 2 n ) > λ k + 1 ( Z 2 n ) > λ k + 1 ( C 2 n ) , k = 1 , 3 , 5 , … , 2 n − 1 . \lambda_{k}(C_{2n})>\lambda_{k}(Z_{2n})>\lambda_{k+1}(Z_{2n})>\lambda_{k+1}(C_{2n}),\ \ k=1,3,5,\ldots,2n-1.
If n n is even, then λ k ( C 2 n ) = λ k ( Z 2 n ) \lambda_{k}(C_{2n})=\lambda_{k}(Z_{2n}) for k = n , n + 1 k=n,n+1 and the other eigenvalues are the same as the case that n n is odd.
Since C 2 n C_{2n} and Z 2 n Z_{2n} are bipartite, by using the above (in)equalities, we have
σ ( C 2 n , Z 2 n ) \displaystyle\sigma(C_{2n},Z_{2n})
= \displaystyle=
2 ( λ 1 ( C 2 n ) − λ 1 ( Z 2 n ) ) + ∑ k = 2 2 n − 1 | λ k ( C 2 n ) − λ k ( Z 2 n ) | \displaystyle 2(\lambda_{1}(C_{2n})-\lambda_{1}(Z_{2n}))+\displaystyle\sum_{k=2}^{2n-1}|\lambda_{k}(C_{2n})-\lambda_{k}(Z_{2n})|
= \displaystyle=
2 ( λ 1 ( C 2 n ) − λ 1 ( Z 2 n ) ) + ∑ k = 2 , k is even 2 n − 2 λ k ( Z 2 n ) − λ k ( C 2 n ) + λ k + 1 ( C 2 n ) − λ k + 1 ( Z 2 n ) \displaystyle 2(\lambda_{1}(C_{2n})-\lambda_{1}(Z_{2n}))+\displaystyle\sum_{{k=2}\;{\text{, $k$ is even}\;}}^{2n-2}\lambda_{k}(Z_{2n})-\lambda_{k}(C_{2n})+\lambda_{k+1}(C_{2n})-\lambda_{k+1}(Z_{2n})
= \displaystyle=
2 ( λ 1 ( C 2 n ) − λ 1 ( Z 2 n ) ) + 2 ∑ k = 2 n − 1 ( − 1 ) k λ k ( Z 2 n ) \displaystyle 2(\lambda_{1}(C_{2n})-\lambda_{1}(Z_{2n}))+2\displaystyle\sum_{k=2}^{n-1}(-1)^{k}\lambda_{k}(Z_{2n})
= \displaystyle=
2 λ 1 ( C 2 n ) + 2 ∑ k = 1 n − 1 ( − 1 ) k λ k ( Z 2 n ) \displaystyle 2\lambda_{1}(C_{2n})+2\displaystyle\sum_{k=1}^{n-1}(-1)^{k}\lambda_{k}(Z_{2n})
= \displaystyle=
4 + 4 ∑ k = 1 n − 1 ( − 1 ) k cos ( ( 2 k − 1 ) π 4 n − 2 ) . \displaystyle 4+4\displaystyle\sum_{k=1}^{n-1}(-1)^{k}\cos(\frac{(2k-1)\pi}{4n-2}).
Since
lim n → ∞ ∑ k = 1 n − 1 ( − 1 ) k cos ( ( 2 k − 1 ) π 4 n − 2 ) = − 1 2 , \displaystyle\lim_{n\rightarrow\infty}\displaystyle\sum_{k=1}^{n-1}(-1)^{k}\cos(\frac{(2k-1)\pi}{4n-2})=\frac{-1}{2}, we have
lim n → ∞ σ ( C 2 n , Z 2 n ) = 2 . \lim_{n\rightarrow\infty}\sigma(C_{2n},Z_{2n})=2.
(2)
Let n ≥ 4 n\geq 4 and a n := σ ( P n , Z n ) a_{n}:=\sigma(P_{n},Z_{n}) . We prove
lim n ⟶ ∞ a 4 n + 1 = lim n ⟶ ∞ a 4 n + 2 = lim n ⟶ ∞ a 4 n + 3 = lim n ⟶ ∞ a 4 n = 8 − 8 2 + 2 π π , \displaystyle\lim_{n\longrightarrow\infty}a_{4n+1}=\displaystyle\lim_{n\longrightarrow\infty}a_{4n+2}=\displaystyle\lim_{n\longrightarrow\infty}a_{4n+3}=\displaystyle\lim_{n\longrightarrow\infty}a_{4n}=\frac{8-8\sqrt{2}+2\pi}{\pi},
which follows that a n a_{n} is convergent to 8 − 8 2 + 2 π π \frac{8-8\sqrt{2}+2\pi}{\pi} .
Suppose that λ 1 ( P n ) ≥ ⋯ ≥ λ n ( P n ) \lambda_{1}(P_{n})\geq\cdots\geq\lambda_{n}(P_{n}) and λ 1 ( Z n ) ≥ ⋯ ≥ λ n ( Z n ) \lambda_{1}(Z_{n})\geq\cdots\geq\lambda_{n}(Z_{n}) are the eigenvalues of P n P_{n} and Z n Z_{n} , respectively. If n ≡ 1 ( mod 4 ) n\equiv 1\pmod{4} , then
λ k ( Z n ) > λ k ( P n ) , k = 1 , … , n − 1 4 , \lambda_{k}(Z_{n})>\lambda_{k}(P_{n}),\ \ \ k=1,\ldots,\frac{n-1}{4},
λ k ( Z n ) < λ k ( P n ) , k = n + 3 4 , … , n − 1 2 , \lambda_{k}(Z_{n})<\lambda_{k}(P_{n}),\ \ \ k=\frac{n+3}{4},\ldots,\frac{n-1}{2},
and
λ n + 1 2 ( Z n ) = λ n + 1 2 ( P n ) . \lambda_{\frac{n+1}{2}}(Z_{n})=\lambda_{\frac{n+1}{2}}(P_{n}).
Thus, by Table 1 and the bipartiteness of P n P_{n} and Z n Z_{n} , we have
σ ( P n , Z n ) \displaystyle\sigma(P_{n},Z_{n})
= \displaystyle=
2 ( ∑ k = 1 n − 1 4 ( λ k ( Z n ) − λ k ( P n ) ) + ∑ k = n + 3 4 n − 1 2 ( λ k ( P n ) − λ k ( Z n ) ) ) \displaystyle 2\big{(}\displaystyle\sum_{k=1}^{\frac{n-1}{4}}(\lambda_{k}(Z_{n})-\lambda_{k}(P_{n}))+\displaystyle\sum_{k=\frac{n+3}{4}}^{\frac{n-1}{2}}(\lambda_{k}(P_{n})-\lambda_{k}(Z_{n}))\big{)}
= \displaystyle=
4 ( ∑ k = 1 n − 1 4 ( cos ( ( 2 k − 1 ) π 2 n − 2 ) − cos ( k π n + 1 ) ) + ∑ k = n + 3 4 n − 1 2 ( cos ( k π n + 1 ) − cos ( ( 2 k − 1 ) π 2 n − 2 ) ) ) . \displaystyle 4\big{(}\displaystyle\sum_{k=1}^{\frac{n-1}{4}}(\cos(\frac{(2k-1)\pi}{2n-2})-\cos(\frac{k\pi}{n+1}))+\displaystyle\sum_{k=\frac{n+3}{4}}^{\frac{n-1}{2}}(\cos(\frac{k\pi}{n+1})-\cos(\frac{(2k-1)\pi}{2n-2}))\big{)}.
By using a simple code (e.g., see 2 ) in symbolic computational software Maple [7 ] , we have
Figure 2. The Maple code
lim n ⟶ ∞ σ ( P n , Z n ) = 8 − 8 2 + 2 π π ≈ 0.945 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},Z_{n})=\frac{8-8\sqrt{2}+2\pi}{\pi}\approx 0.945.
If n ≡ 3 ( mod 4 ) n\equiv 3\pmod{4} , then
λ k ( Z n ) > λ k ( P n ) , k = 1 , … , n − 3 4 , \lambda_{k}(Z_{n})>\lambda_{k}(P_{n}),\ \ \ k=1,\ldots,\frac{n-3}{4},
λ k ( Z n ) < λ k ( P n ) , k = n + 5 4 , … , n − 1 2 , \lambda_{k}(Z_{n})<\lambda_{k}(P_{n}),\ \ \ k=\frac{n+5}{4},\ldots,\frac{n-1}{2},
and
λ k ( Z n ) = λ k ( P n ) , k = n + 1 4 , n + 1 2 . \lambda_{k}(Z_{n})=\lambda_{k}(P_{n}),\ \ \ k=\frac{n+1}{4},\frac{n+1}{2}.
It follows from Table 1 and the bipartiteness of P n P_{n} and Z n Z_{n} that
σ ( P n , Z n ) \displaystyle\sigma(P_{n},Z_{n})
= \displaystyle=
2 ( ∑ k = 1 n − 3 4 ( λ k ( Z n ) − λ k ( P n ) ) + ∑ k = n + 5 4 n − 1 2 ( λ k ( P n ) − λ k ( Z n ) ) ) \displaystyle 2\big{(}\displaystyle\sum_{k=1}^{\frac{n-3}{4}}(\lambda_{k}(Z_{n})-\lambda_{k}(P_{n}))+\displaystyle\sum_{k=\frac{n+5}{4}}^{\frac{n-1}{2}}(\lambda_{k}(P_{n})-\lambda_{k}(Z_{n}))\big{)}
= \displaystyle=
4 ( ∑ k = 1 n − 3 4 ( cos ( ( 2 k − 1 ) π 2 n − 2 ) − cos ( k π n + 1 ) ) + ∑ k = n + 5 4 n − 1 2 ( cos ( k π n + 1 ) − cos ( ( 2 k − 1 ) π 2 n − 2 ) ) ) . \displaystyle 4\big{(}\displaystyle\sum_{k=1}^{\frac{n-3}{4}}(\cos(\frac{(2k-1)\pi}{2n-2})-\cos(\frac{k\pi}{n+1}))+\displaystyle\sum_{k=\frac{n+5}{4}}^{\frac{n-1}{2}}(\cos(\frac{k\pi}{n+1})-\cos(\frac{(2k-1)\pi}{2n-2}))\big{)}.
By using Maple [7 ] ,
lim n ⟶ ∞ σ ( P n , Z n ) = 8 − 8 2 + 2 π π ≈ 0.945 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},Z_{n})=\frac{8-8\sqrt{2}+2\pi}{\pi}\approx 0.945.
If
n ∗ = { n 4 n ≡ 0 ( mod 4 ) , n − 2 4 n ≡ 2 ( mod 4 ) , n^{*}=\left\{\begin{array}[]{ll}\frac{n}{4}&n\equiv 0\pmod{4},\\
\\
\frac{n-2}{4}&n\equiv 2\pmod{4},\end{array}\right.
then
λ k ( Z n ) > λ k ( P n ) , k = 1 , … , n ∗ , \lambda_{k}(Z_{n})>\lambda_{k}(P_{n}),\ \ \ k=1,\ldots,n^{*},
and
λ k ( Z n ) < λ k ( P n ) , k = n ∗ + 1 , … , n 2 . \lambda_{k}(Z_{n})<\lambda_{k}(P_{n}),\ \ \ k=n^{*}+1,\ldots,\frac{n}{2}.
Therefore
σ ( P n , Z n ) \displaystyle\sigma(P_{n},Z_{n})
= \displaystyle=
2 ( ∑ k = 1 n ∗ ( λ k ( Z n ) − λ k ( P n ) ) + ∑ k = n ∗ + 1 n 2 ( λ k ( P n ) − λ k ( Z n ) ) ) \displaystyle 2\big{(}\displaystyle\sum_{k=1}^{n^{*}}(\lambda_{k}(Z_{n})-\lambda_{k}(P_{n}))+\displaystyle\sum_{k=n^{*}+1}^{\frac{n}{2}}(\lambda_{k}(P_{n})-\lambda_{k}(Z_{n}))\big{)}
= \displaystyle=
4 ( ∑ k = 1 n ∗ ( cos ( ( 2 k − 1 ) π 2 n − 2 ) − cos ( k π n + 1 ) ) + ∑ k = n ∗ + 1 n 2 ( cos ( k π n + 1 ) − cos ( ( 2 k − 1 ) π 2 n − 2 ) ) ) . \displaystyle 4\big{(}\displaystyle\sum_{k=1}^{n^{*}}(\cos(\frac{(2k-1)\pi}{2n-2})-\cos(\frac{k\pi}{n+1}))+\displaystyle\sum_{k=n^{*}+1}^{\frac{n}{2}}(\cos(\frac{k\pi}{n+1})-\cos(\frac{(2k-1)\pi}{2n-2}))\big{)}.
In both cases, by using Maple [7 ] , we have
lim n ⟶ ∞ σ ( P n , Z n ) = 8 − 8 2 + 2 π π ≈ 0.945 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},Z_{n})=\frac{8-8\sqrt{2}+2\pi}{\pi}\approx 0.945.
Therefore a n = σ ( P n , Z n ) a_{n}=\sigma(P_{n},Z_{n}) is convergent to 8 − 8 2 + 2 π π \frac{8-8\sqrt{2}+2\pi}{\pi} .
(3)
Let n ≥ 6 n\geq 6 and b n := σ ( P n , Z n ) b_{n}:=\sigma(P_{n},Z_{n}) . We prove
lim n ⟶ ∞ b 4 n + 1 = lim n ⟶ ∞ b 4 n + 2 = lim n ⟶ ∞ b 4 n + 3 = lim n ⟶ ∞ b 4 n = 8 − 8 2 + 2 π π , \displaystyle\lim_{n\longrightarrow\infty}b_{4n+1}=\displaystyle\lim_{n\longrightarrow\infty}b_{4n+2}=\displaystyle\lim_{n\longrightarrow\infty}b_{4n+3}=\displaystyle\lim_{n\longrightarrow\infty}b_{4n}=\frac{8-8\sqrt{2}+2\pi}{\pi},
which follows that b n b_{n} is convergent to 8 − 8 2 + 2 π π \frac{8-8\sqrt{2}+2\pi}{\pi} .
Suppose that λ 1 ( W n ) ≥ ⋯ ≥ λ n ( W n ) \lambda_{1}(W_{n})\geq\cdots\geq\lambda_{n}(W_{n}) and λ 1 ( Z n ) ≥ ⋯ ≥ λ n ( Z n ) \lambda_{1}(Z_{n})\geq\cdots\geq\lambda_{n}(Z_{n}) are the eigenvalues of
W n W_{n} and Z n Z_{n} , respectively. By similar arguments given in part (2), if n ≡ 1 ( mod 4 ) n\equiv 1\pmod{4} , then
σ ( W n , Z n ) \displaystyle\sigma(W_{n},Z_{n})
= \displaystyle=
2 ( ∑ k = 1 n − 1 4 ( λ k ( W n ) − λ k ( Z n ) ) + ∑ k = n + 3 4 n − 1 2 ( λ k ( Z n ) − λ k ( W n ) ) ) \displaystyle 2\big{(}\displaystyle\sum_{k=1}^{\frac{n-1}{4}}(\lambda_{k}(W_{n})-\lambda_{k}(Z_{n}))+\displaystyle\sum_{k=\frac{n+3}{4}}^{\frac{n-1}{2}}(\lambda_{k}(Z_{n})-\lambda_{k}(W_{n}))\big{)}
= \displaystyle=
4 ( ∑ k = 1 n − 1 4 ( cos ( ( k − 1 ) π n − 3 ) − cos ( ( 2 k − 1 ) π 2 n − 2 ) ) + ∑ k = n + 3 4 n − 1 2 ( cos ( ( 2 k − 1 ) π 2 n − 2 ) − cos ( ( k − 1 ) π n − 3 ) ) ) . \displaystyle 4\big{(}\displaystyle\sum_{k=1}^{\frac{n-1}{4}}(\cos(\frac{(k-1)\pi}{n-3})-\cos(\frac{(2k-1)\pi}{2n-2}))+\displaystyle\sum_{k=\frac{n+3}{4}}^{\frac{n-1}{2}}(\cos(\frac{(2k-1)\pi}{2n-2})-\cos(\frac{(k-1)\pi}{n-3}))\big{)}.
We have
σ ( W n , Z n ) \displaystyle\sigma(W_{n},Z_{n})
= \displaystyle=
2 ( ∑ k = 1 n − 3 4 ( λ k ( W n ) − λ k ( Z n ) ) + ∑ k = n + 5 4 n − 1 2 ( λ k ( Z n ) − λ k ( W n ) ) ) \displaystyle 2\big{(}\displaystyle\sum_{k=1}^{\frac{n-3}{4}}(\lambda_{k}(W_{n})-\lambda_{k}(Z_{n}))+\displaystyle\sum_{k=\frac{n+5}{4}}^{\frac{n-1}{2}}(\lambda_{k}(Z_{n})-\lambda_{k}(W_{n}))\big{)}
= \displaystyle=
4 ( ∑ k = 1 n − 1 4 ( cos ( ( k − 1 ) π n − 3 ) − cos ( ( 2 k − 1 ) π 2 n − 2 ) ) + ∑ k = n + 5 4 n − 1 2 ( cos ( ( 2 k − 1 ) π 2 n − 2 ) − cos ( ( k − 1 ) π n − 3 ) ) ) . \displaystyle 4\big{(}\displaystyle\sum_{k=1}^{\frac{n-1}{4}}(\cos(\frac{(k-1)\pi}{n-3})-\cos(\frac{(2k-1)\pi}{2n-2}))+\displaystyle\sum_{k=\frac{n+5}{4}}^{\frac{n-1}{2}}(\cos(\frac{(2k-1)\pi}{2n-2})-\cos(\frac{(k-1)\pi}{n-3}))\big{)}.
where n ≡ 3 ( mod 4 ) n\equiv 3\pmod{4} . In both cases, by using Maple [7 ] ,
lim n ⟶ ∞ σ ( W n , Z n ) = 8 − 8 2 + 2 π π ≈ 0.945 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(W_{n},Z_{n})=\frac{8-8\sqrt{2}+2\pi}{\pi}\approx 0.945.
If
n ∗ = { n 4 n ≡ 0 ( mod 4 ) , n − 2 4 n ≡ 2 ( mod 4 ) , n^{*}=\left\{\begin{array}[]{ll}\frac{n}{4}&n\equiv 0\pmod{4},\\
\\
\frac{n-2}{4}&n\equiv 2\pmod{4},\end{array}\right.
then
λ k ( W n ) > λ k ( Z n ) , k = 1 , … , n ∗ , \lambda_{k}(W_{n})>\lambda_{k}(Z_{n}),\ \ \ k=1,\ldots,n^{*},
λ k ( W n ) < λ k ( Z n ) , k = n ∗ + 1 , … , n 2 − 1 , \lambda_{k}(W_{n})<\lambda_{k}(Z_{n}),\ \ \ k=n^{*}+1,\ldots,\frac{n}{2}-1,
and
λ n 2 ( W n ) = λ n 2 ( Z n ) . \lambda_{\frac{n}{2}}(W_{n})=\lambda_{\frac{n}{2}}(Z_{n}).
Therefore
σ ( W n , Z n ) \displaystyle\sigma(W_{n},Z_{n})
= \displaystyle=
2 ( ∑ k = 1 n ∗ ( λ k ( W n ) − λ k ( Z n ) ) + ∑ k = n ∗ + 1 n 2 − 1 ( λ k ( Z n ) − λ k ( W n ) ) ) \displaystyle 2\big{(}\displaystyle\sum_{k=1}^{n^{*}}(\lambda_{k}(W_{n})-\lambda_{k}(Z_{n}))+\displaystyle\sum_{k=n^{*}+1}^{\frac{n}{2}-1}(\lambda_{k}(Z_{n})-\lambda_{k}(W_{n}))\big{)}
= \displaystyle=
4 ( ∑ k = 1 n ∗ ( cos ( ( k − 1 ) π n − 3 ) − cos ( ( 2 k − 1 ) π 2 n − 2 ) ) + ∑ k = n ∗ + 1 n 2 − 1 ( cos ( ( 2 k − 1 ) π 2 n − 2 ) − cos ( ( k − 1 ) π n − 3 ) ) ) . \displaystyle 4\big{(}\displaystyle\sum_{k=1}^{n^{*}}(\cos(\frac{(k-1)\pi}{n-3})-\cos(\frac{(2k-1)\pi}{2n-2}))+\displaystyle\sum_{k=n^{*}+1}^{\frac{n}{2}-1}(\cos(\frac{(2k-1)\pi}{2n-2})-\cos(\frac{(k-1)\pi}{n-3}))\big{)}.
In both cases, by using Maple [7 ] ,
lim n ⟶ ∞ σ ( W n , Z n ) = 8 − 8 2 + 2 π π ≈ 0.945 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(W_{n},Z_{n})=\frac{8-8\sqrt{2}+2\pi}{\pi}\approx 0.945.
Thus b n = σ ( W n , Z n ) b_{n}=\sigma(W_{n},Z_{n}) is convergent to 8 − 8 2 + 2 π π \frac{8-8\sqrt{2}+2\pi}{\pi} .
(4)
The eigenvalues of P n P_{n} and W n W_{n} satisfy the same (in)equalities as the eigenvalues of P n P_{n} and Z n Z_{n} . So, by the parts (2) and (3), we obtain
σ ( P n , W n ) = σ ( P n , Z n ) + σ ( W n , Z n ) . \sigma(P_{n},W_{n})=\sigma(P_{n},Z_{n})+\sigma(W_{n},Z_{n}).
Since
lim n ⟶ ∞ σ ( P n , Z n ) = lim n ⟶ ∞ σ ( W n , Z n ) , \lim_{n\longrightarrow\infty}\sigma(P_{n},Z_{n})=\lim_{n\longrightarrow\infty}\sigma(W_{n},Z_{n}),
we have
lim n ⟶ ∞ σ ( P n , Z n ) = lim n ⟶ ∞ σ ( W n , Z n ) = 1 2 lim n ⟶ ∞ σ ( P n , W n ) = 8 − 8 2 + 2 π π ≈ 0.945 . \displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},Z_{n})=\displaystyle\lim_{n\longrightarrow\infty}\sigma(W_{n},Z_{n})=\frac{1}{2}\displaystyle\lim_{n\longrightarrow\infty}\sigma(P_{n},W_{n})=\frac{8-8\sqrt{2}+2\pi}{\pi}\approx 0.945.
This completes the proof. □ \hfill\Box
Acknowledgments
The research of the first author was in part supported by a grant from School of Mathematics, Institute for Research in Fundamental Sciences (IPM).
References
[1]
A. Abdollahi, Sh. Janbaz and M. R. Oboudi, Distance between spectra of graphs, Linear Algebra Appl., 466 (2015), 401–408.
[2]
D. M. Cvetković, M. Doob and H. Sachs, Spectra of Graphs, Theory and Application, Academic Press, Inc., New York, 1980.
[3]
K. Ch. Das and S. Sun, Distance between the normalized Laplacian spectra of two graphs, Linear Algebra Appl., 530 (2017), 305–321.
[4]
I. Jovanović and Z. Stanić, Spectral distances of graphs, Linear Algebra Appl., 436 (2012), No. 5, 1425–1435.
[5]
I. Jovanović and Z. Stanić, Spectral distances of graphs based on their different matrix representations, Filomat, 28 (2014), No. 4, 723–734.
[6]
H. Lin, D. Li and K. Ch. Das, Distance between distance spectra of graphs, Linear Multilinear Algebra, 65 (2017), No. 12, 2538–2550.
[7]
Maple, Computer Algebra System, Numeric Computation Waterloo Maple, www.maplesoft.com/products/maple/