Approximating graphs of a class of general Sierpinski triangles and their normalized Laplacian spectra

Abstract

The normalized Laplacian spectrum of a graph is an important tool that one can use to find much information about its topological and structural characteristics and also on some relevant dynamical aspects, specifically in relation to random walks. In this paper we devise an essentially algorithm to obtain the approximating graphs of a class of general Sierpinski triangles and their normalized Laplacian spectra, and illustrate such algorithm by a quasi-program of Matlab. In the meantime, our work also enriches the graphs whose spectrum is known.

Introduction

Spectral graph theory is an interesting topic, in a way, be regarded as the study of relation between graph properties and Laplacian matrix spectrum or adjacency matrix spectrum.^1–4 The spectrum of a graph(the set of eigenvalues of a graph with their multiplicities) is important in the spectral graph theory, because almost all major graph invariants are related to it, such as connectivity, expanding property, isoperimetric number, maximum cut, independence number, genus, diameter, bandwidth-type parameters, etc (see^2,5,6 and the reference therein).

Spectral graph theory has different applications in different scientific fields, such as the eigenvalues were related to the stability of molecules in chemistry, graph spectra occured naturally in minimizing energies of Hamiltonian systems in physics, the recent progress on expander graphs and eigenvalues was initiated by problems of communication networks in computer science, etc. In recent years, there has been a particular interest in the study of the eigenvalues and eigenvectors of normalized Laplacian matrix, since much important information about structural properties and dynamical processes of a graph or network is related to it. In the structural aspect, for example, F. R. Chung ² proved that the number of spanning trees in a network can be determined in terms of the product of all nonzero eigenvalues of a connected network. H. Chen and F. Zhang ⁷ showed the resistance distance can be naturally expressed by the eigenvalues and eigenvectors of normalized Laplacian matrix. With respect to dynamical process, there has been great progress in determining the eigenvalues and eigenvectors of normalized Laplacian matrix, including the first-passage time, ⁸ mixing time ⁹ and Kemeny’s constant ¹⁰ that can be used as a measure of the efficiency of navigation on the network. Particularly, normalized Laplacian eigenvalues and eigenvectors also are relevant in light harvesting, ¹¹ energy or exciton transport, ¹² chemical kinetics, ¹³ and many other problems in chemical physics. ¹⁴

Let

$$△=\{c_1\alpha +c_2\beta :c_1+c_2\le 1\text{ and }0\le c_1,c_2\le 1\}$$

be an equilateral triangle with lower-left coordinate (0, 0) and side 1, where $\alpha =(0,1)$ and $\beta =(1/2,\sqrt{3}/2)$ . Given an integer $n\ge 2$ , write

$$\mathcal{D}(n)=\{l_1\alpha +l_2\beta :l_1+l_2\le n-1\text{ and }{l}_1,l_2\in \mathbb{N}\cup \{0\left\}\right\}$$

For $E,F\subset {ℝ}^2$ and $\lambda \in ℝ$ , we define

$$\begin{array}{l}\lambda E=\{\lambda e:e\in E\},E+F=\{e+f:e\in E,f\in F\}\hfill \end{array}$$

By, ¹⁵ there is a unique non-empty compact subset K(n) of ${ℝ}^2$ satisfying the following set equation:

$$K\left(n\right)=\left[K\right(n)+\mathcal{D}(n\left)\right]/n$$

We shall call K(n) a general Sierpinski triangle throughout this paper. A familiar example of a general Sierpinski triangle is the Sierpinski gasket, in which n = 2, $\mathcal{D}(2)=\{(0,0),(1,0),(1/2,\sqrt{3}/2)\}$ . Recently, evolving networks are usually modeled by the general Sierpinski triangles. For example, J. Guan et al. ¹⁶ and A. Le et al. ¹⁷ used the classical Sierpinski triangle to construct evolving networks.

We define $\Gamma (n,0)=△,\Gamma (n,1)=[△+\mathcal{D}(n)]/n$ , and recurrently

$$\Gamma (n,k+1)=[\Gamma (n,k)+\mathcal{D}(n)]/n$$

for $k\ge 1$ . Then $\Gamma (n,0)$ is an equilateral triangle with side 1, $\Gamma (n,k)(k\ge 1)$ is a union of smaller equilateral triangles with side $1/n^k$ and converges to K(n) as $k\to \infty$ in the Hausdorff metric. ¹⁵ We call $\Gamma (n,k)(k\ge 0)$ the k-th approximating graph of general Sierpinski triangle K(n), in which vertex set and edge set are considered in a usually way (see Figure 1 for n = 2 and k = 0, 1, 2). Another interesting motivation of our work comes form the theory of analysis on fractals. J. Kigami constructed in ¹⁸ a Laplacian on the Sierpinski gasket, and later extended their method to the p.c.f. (post-critically finite) fractals. Kigami’s idea is to approximate the Laplacian on K(n) by the normalized Laplacian on $\Gamma (n,k)$ . This coincide with the known graph Laplacian whose vertex set is V(n, k), where V(n, k) is described as in ‘Some basic structure properties and concepts on approximating graph’ section.

Figure 1.

Approximation graphs of classical Sierpinski triangle $\Gamma (2,k)$ , k = 0, 1, 2.

There has been notable progress on the study of the spectrum of normalized Laplacian matrix associated to fractal, especially in recent years. Yet much is still unknown, and in this direction the existed works concentrate mainly on the theoretical study base on a recursive method (or decimation method ¹⁹ ) as we know of. For example, A.Julaiti et al. in ²⁰ studied the spectra of normalized Laplacian matrices for a family of proposed fractal trees and Cayley trees. M. Dai et al. in ²¹ presented a study on the spectrum of the normalized Laplacian of weighted tree-like fractals. In this paper we will obtain the approximating graphs $\Gamma (n,k)$ and their normalized Laplacian spectra from an algorithmic point of view, which extends the work of B. Rajan et al.. ²² Our method is completely different from the decimation method. Our algorithm need to code the vertices of each approximating graph $\Gamma (n,k)$ base on the theory of iterated function system (IFS), ¹⁵ we call it an IFS coding algorithm throughout this paper. This is of great relevance in studying the structure and some properties of general Sierpinski triangles and related graphs, and dynamical aspects of related networks in a new perspective.

This paper is organized as follows. The next section is preliminaries. We introduce the geometrical construction, normalized Laplacian matrix and some basic structure properties of approximating graph $\Gamma (n,k)$ . In the subsequent section, we present our main result, called IFS coding algorithm, and illustrate our algorithm by a quasi-program of Matlab. Furthermore we also prove the correctness of this program and show the convergence of the algorithm in this section. The paper is concluded with conclusions.

Some basic structure properties and concepts on approximating graph

It is well known that a graph can be represented visually by drawing a dot for every vertex, and drawing a line segment between two vertices if they are connected by an edge. According to, ²³ for any $n\ge 2$ , we can obtain a drawing of $\Gamma (n,k)$ with $k\ge 1$ as follows: We divide the equilateral triangle $△$ into n² congruent triangles by connecting n equipartition points of each pair of edges and replace it with the union of all the upward equilateral triangles. We can now apply this process to each of these upward equilateral triangles, and then repeat the procedure to the k-th step, to obtain a drawing of $\Gamma (n,k)$ with $k\ge 1$ . Obviously, for any $k\ge 0,\Gamma (n,k)$ is an undirected graph without loops and multiple edges, i.e., a simple graph. Let V(n, k) and E(n, k) denote the vertex set and edge set of $\Gamma (n,k)$ , respectively. For any $u,v\in V(n,k)$ , if u and v are connected by an edge in E(n, k), then we denote it by $u\sim v$ , and say that they are adjacent. Write

$$a_{u,v}(n,k)=\{\begin{array}{l}1,\text{ if }u\sim v,\hfill \\ 0,\text{others}.\hfill \end{array}$$

We call $A(n,k)=(a_{u,v}(n,k))$ the adjacency matrix of $\Gamma (n,k)$ . For any $v\in V(n,k)$ , we use $d_v(n,k)$ to denote the number of edges which touch v, and call it the degree of v. Let D(n, k) denote the diagonal matrix in which the (v, v)-th entry is $d_v(n,k)$ . We call $L(n,k):=D(n,k)-A(n,k)$ the Laplacian matrix of $\Gamma (n,k)$ . For the matrix A, let $A^{-1}$ denote the inverse of A and $\sqrt{A}$ raise each element of A to the power of half. For random walk on $\Gamma (n,k)$ , the Markov matrix (or probability transition matrix) is defined as

$$M(n,k):=D{(n,k)}^{-1}A(n,k),$$

which can be normalized to obtain the following real and symmetric matrix

$$P(n,k):={(\sqrt{D(n,k)})}^{-1}A(n,k){(\sqrt{D(n,k)})}^{-1}$$

$$=\sqrt{D(n,k)}M(n,k){(\sqrt{D(n,k)})}^{-1}$$

We see from the definition that the (u, v)-th entry of P(n, k) is

$$p_{u,v}(n,k)=\frac{a_{u,v}(n,k)}{\sqrt{d_u(n,k)d_v(n,k)}}$$

Furthermore, M(n, k) and P(n, k) have the same spectrum since M(n, k) and P(n, k) are similar.

Definition 2.1. Let I be a unit matrix with the same size as P(n, k). We define the normalized Laplacian matrix of $\Gamma (n,k)$ by

$$\mathcal{L}(n,k)=I-P(n,k)=I-\sqrt{D(n,k)}M(n,k){(\sqrt{D(n,k)})}^{-1}$$

Example 2.1. Let n = 2 and k = 0. Then

$$\begin{array}{l}A(2,0)=\left(\begin{array}{ccc}0& 1& 1\\ 1& 0& 1\\ 1& 1& 0\end{array}\right),D(2,0)=\left(\begin{array}{ccc}2& 0& 0\\ 0& 2& 0\\ 0& 0& 2\end{array}\right),\\ M(2,0)={(D(2,0))}^{-1}A(2,0)=\left(\begin{array}{ccc}0& 1/2& 1/2\\ 1/2& 0& 1/2\\ 1/2& 1/2& 0\end{array}\right),\\ L(2,0)=D(2,0)-A(2,0)=\left(\begin{array}{ccc}2& -1& -1\\ -1& 2& -1\\ -1& -1& 2\end{array}\right)\end{array}$$

and

$$\mathcal{L}(2,0)=I-\sqrt{D(2,0)}A(2,0){(\sqrt{D(2,0)})}^{-1}=\left(\begin{array}{ccc}1& -1/2& -1/2\\ -1/2& 1& -1/2\\ -1/2& -1/2& 1\end{array}\right)$$

are the adjacency matrix, diagonal matrix, Markov matrix, Laplacian matrix and normalized Laplacian matrix of $\Gamma (2,0)$ , respectively.

Let $G=(V(G),E(G))$ be a simple graph with vertex set V(G) and edge set E(G), and $H=(V(H),E(H))$ be another simple graph. We say that G and H are isomorphic, and denote it by $G\simeq H$ , if there exists a bijection $f:V(G)\to V(H)$ such that

$$u\sim v\text{if and only if}f(u)\sim f(v),u,v\in V(G)$$

We say that H is a subgraph of G if $V\left(H\right)\subseteq V\left(G\right)$ and $E\left(H\right)\subseteq E\left(G\right)$ . Let $\#A$ denote the cardinality of a set A. For given integer $n\ge 2$ , we see from the geometrical construction of $\Gamma (n,k)$ that for any $k\ge 1,\Gamma (n,k)$ consists of $\#\mathcal{D}(n)$ subgraphs that are isomorphic to $\Gamma (n,k-1)$ . We denote by ${\Gamma }^1(n,k),{\Gamma }^2(n,k)\dots ,{\Gamma }^{\#\mathcal{D}(n)}(n,k)$ these subgraphs of $\Gamma (n,k)$ , respectively, see Figure 2 for n = 2, 3, and let $V^1(n,k),V^2(n,k)\dots ,V^{\#\mathcal{D}(n)}(n,k)$ denote the corresponding vertex sets, respectively. Note that if $V^i(n,k)\cap V^j(n,k)\ne \varnothing$ with $i\ne j$ , then $V^i(n,k)\cap V^j(n,k)$ is a singleton, which is unique common vertex of small triangles ${\Gamma }^i(n,1)$ and ${\Gamma }^j(n,1)$ . It easy to check that $\Gamma (n,k)$ has the following obviously structure properties.

Figure 2.

$\Gamma (n,1)$ and their subgraphs ${\Gamma }^i(n,1)$ . (a) n= 2 and (b) n = 3.

Proposition 2.2. For $n\ge 2$ ,

$$\#\left(\mathcal{D}\right(n\left)\right)=\frac{n(n+1)}{2}.$$

Proposition 2.3. For $n\ge 2$ and $k\ge 0$ , let V(n, k) and $d_v(n,k)$ be defined as above. Then for any $v\in V(n,k),d_v(n,k)\in \{2,4,6\}$ .

Proposition 2.4. For $n\ge 2$ , let d(n, m) denote the total number of vertices with degree m in $\Gamma (n,1)$ , m = 2, 4 and 6. Then

$$d(n,6)=\{\begin{array}{l}0,n=2;\hfill \\ 1,n=3;\hfill \\ 3,n=4;\hfill \\ \#V(n-3,1),n\ge 5,\hfill \end{array}$$

and for $n\ge 2,d(n,2)=3$ and $d(n,4)=\#V(n,1)-d(n,6)-3$ .

Proposition 2.5. Let $n\ge 2$ and $k\ge 1$ .Then

$$\#V(n,k)=(\#\mathcal{D}(n))(\#V(n,k-1))-d(n,4)-2d(n,6),$$

and $\#E(n,k)=3{(\#\mathcal{D}(n))}^k$ .

Example 2.2. Let n = 2. Then $d(n,4)=3,d(n,6)=0,\#V(2,k)=(3^{k+1}+3)/2$ and $\#E(2,k)=3^{k+1}$ for $k\ge 0$ . See Figure 3 for $n=2,k=0,1,2,3$ .

Figure 3.

Coding of $\Gamma (2,k)$ using IFS coding algorithm. (a) k = 0. (b) k = 1. (c) k = 2. (d) k = 3.

Proof. For $k\ge 1$ , we have from the Propositions 2.2, 2.4 and 2.5 that

$$\#V(2,k)=(\#\mathcal{D}(2))(\#V(2,k-1))-d(2,4)-2d(2,6)=3\#V(2,k-1)-3=3(3\#V(2,k-2)-3)-3\cdots =3^k\#V(2,0)-3^k-3^{k-1}-\cdots -3^2-3=3^{k+1}+\frac{3(1-3^k)}{2}=\frac{3^{k+1}+3}{2}$$

Combing the fact that $\#V(2,0)=3$ , we have $\#V(2,k)=(3^{k+1}+3)/2$ for $k\ge 0$ . Using Proposition 2.5 again and the fact $\#\mathcal{D}(2)=3$ , we get that $\#E(2,k)=3^{k+1}$ for $k\ge 0$ . This proof is completed. □

Similarly, we have

Example 2.3. Let n = 3. Then $d(3,4)=6,d(3,6)=1$ , and $\#V(3,k)=(7\times 6^k+8)/5$ and $\#E(3,k)=3\times 6^k$ for $k\ge 0$ . See Figure 4 for $n=3,k=1,2$ .

Figure 4.

Coding of $\Gamma (3,k)$ using IFS coding algorithm. (a) k = 1. (b) k = 2.

IFS coding algorithm and a quasi-program of Matlab

Our aim in this paper is to describe completely the normalized Laplacian spectrum of $\Gamma (n,k)$ for any $k\ge 0$ . In view of Definition 2.1, we first need to find the adjacency matrix A(n, k) and the diagonal matrix D(n, k) of $\Gamma (n,k)$ , which is the most important and most central part. However writing down manually the adjacency matrix and diagonal matrix is unimaginable when n and k are large enough. So we in this section devise an algorithm base on the iterative generation process of $\Gamma (n,k)$ , called IFS coding algorithm, to code the vertices of $\Gamma (n,k)$ , and thus from the adjacency matrix $A(n,0)$ (resp. the diagonal matrix $D(n,0)$ ) find the adjacency matrix A(n, k) (resp. the diagonal matrix D(n, k)) by the recursive relation between $A(n,k-1)$ (resp. $D(n,k-1)$ ) and A(n, k) (resp. D(n, k)).

Let $V^j(n,k),j=1,\dots ,\#\mathcal{D}(n)$ , be described as in ‘Some basic structure properties and concepts on approximating graph’ section. Denote

$$\begin{array}{l}{v}_1=V^1(n,k)\cap V^2(n,k),v_2=V^1(n,k)\cap V^3(n,k),\\ \frac{v_n(n+1)}{2}+i=V^{\frac{n(n-1)}{2}+i+1}(n,k)\cap V^{\frac{n(n-1)}{2}+i+2}(n,k),\\ i=0,\dots ,n-2,\\ \frac{v_l(l+1)}{2}=V^{\frac{l(l-1)}{2}+1}(n,k)\cap V^{\frac{l(l+1)}{2}+1}(n,k),\\ \frac{v_l(l+1)}{2}+i=V^{\frac{l(l-1)}{2}+i}(n,k)\cap V^{\frac{l(l-1)}{2}+i+1}(n,k)\\ \cap V^{\frac{l(l+1)}{2}+i+1}(n,k),i=1,\dots ,l-1,\\ \frac{v_l(l+1)}{2}+l=V^{\frac{l(l-1)}{2}+l}(n,k)\cap V^{\frac{l(l+1)}{2}+1+l}(n,k)\end{array}$$

(1)

where $l=2,\dots ,n-1$ (see Figure 2 for n = 2, 3 and k = 1).

Step 1: Code the vertices of $\Gamma (n,0)=\Delta$ as shown in Figure 3(a).

Step 2: For any $k\ge 1$ , code the vertices of ${\Gamma }^1(n,k)$ as that in $\Gamma (n,k-1)$ . By Proposition 2.5, the code of v₂ in $\Gamma (n,k)$ is $\#V(n,k-1)$ . We define $\#\mathcal{D}(1)=1,d(1,4)=0$ and $d(1,6)=0$ . Note that the uncoded vertices of ${\Gamma }^i(n,k)$ induce a graph isomorphic to ${\Gamma }^1(n,k)-\left\{1\right\}$ for $i\in \{\#\mathcal{D}(m):m=2,\dots ,n\}$ , the uncoded vertices of ${\Gamma }^i(n,k)-\left\{v_{i+m}\right\}$ induce a graph isomorphic to ${\Gamma }^1(n,k)-\{1,v_2\}$ for $m=2,\dots ,n-1,i=\#\mathcal{D}(m-1)+1,\dots ,\#\mathcal{D}(m)-1$ , and the uncoded vertices of ${\Gamma }^i(n,k)-\left\{v_{i+n-1}\right\}$ induce a graph isomorphic to ${\Gamma }^1(n,k)-\{1,v_2\}$ for $i=\#\mathcal{D}(n-1)+1,\dots ,\#\mathcal{D}(n)-1$ . Combing Propositions 2.4 and 2.5, for each vertex j in ${\Gamma }^1(n,k)-\left\{1\right\}$ , we code the corresponding vertex of ${\Gamma }^{\#\mathcal{D}(m)}(n,k)-\left\{v_{\#\mathcal{D}(m)-1}\right\}$ by

$$j+(\#\mathcal{D}(m)-1)\#V(n,k-1)-[d(m,4)+2d(m,6)],$$

where $m=2,\dots ,n.$ For each vertex j in ${\Gamma }^1(n,k)-\{1,v_2\}$ , we code the corresponding vertex of ${\Gamma }^i(n,k)-\left\{v_{i+m}\right\}$ by

$$j+(i-1)\#V(n,k-1)-\left[d\right(m-1,4)+2d(m-1,6\left)\right]-\left[2\right(i-\#\mathcal{D}(m-1))-1],$$

where $m=2,\dots ,n-1,i=\#\mathcal{D}(m-1),\dots ,\#\mathcal{D}(m)-1$ . For each vertex j in ${\Gamma }^1(n,k)-\{1,v_2\}$ , we code the corresponding vertex of ${\Gamma }^i(n,k)-\left\{v_{i+n-1}\right\}$ by j+(i−1) V(n,k−1)−[d(n−1,4)+2d(n−1,6)]−[2(i−D(n−1))−1], where $i=\#\mathcal{D}(n-1)+1,\dots ,\#\mathcal{D}(n)-1$ . We refer to Figure 3 for n = 2 and k = 1, 2, 3, and Figure 4 for n = 3 and k = 1, 2.

Step 3: Input the diagonal matrix and adjacency matrix of $\Gamma (n,0)$ .

Step 4: Output the diagonal matrix and adjacency matrix of $\Gamma (n,k)$ in term of coding rule of Step 2.

Step 5: Compute the normalized Laplacian matrix of $\Gamma (n,k)$ by (2.1) and output eigenvalues.

For simplify, we write

$$a(m,n,k)=(\#\mathcal{D}(m)-1)\#V(n,k-1)-[d(m,4)+2d(m,6)],$$

$$b(i,m,n,k)=(i-1)\#V(n,k-1)-[d(m-1,4)+2d(m-1,6)]-[2(i-\#\mathcal{D}(m-1))-1],$$

where $m=2,\dots ,n,i=\#\mathcal{D}(m-1)+1,\dots ,\#\mathcal{D}(m)-1$ . In the following we illustrate our algorithm by a quasi-program of Matlab, which one need to normalize it according to Matlab syntax rules before using it, see Example 3.2. We denote by the percent-sign % an explanation.

A quasi-program of Matlab base on the IFS coding algorithm

Function Sierpinskispectrum(n,m) % $1/n$ is side length of each small triangle in $\Gamma (n,1)$ and m is iterated depth.

$$v_1=0;$$

$$A=\left(\begin{array}{ccc}0& 1& 1\\ 1& 0& 1\\ 1& 1& 0\end{array}\right);D=\left(\begin{array}{ccc}2& 0& 0\\ 0& 2& 0\\ 0& 0& 2\end{array}\right);$$

% A and D are the adjacency matrix and diagonal matrix of $\Gamma (n,0)=\Delta$ , respectively.

% Output the adjacency matrix and diagonal matrix of $\Gamma (n,m-1)$ .

If m = 1

A; D;

else

for $i=2:m$

$A=$  Adjmatrix(i, A); $D=\text{Diagmatrix}(i,D)$ ;

end

end

% Output the eigenvalues of normalized Laplacian matrix of $\Gamma (n,m-1)$ . spectrum $=\text{eig}\left(\text{eye(size}\right(A\left)\right)-\text{inv}\left(D^{1/2}\right)*A*\text{inv}\left(D^{1/2}\right)$ )

Function $A=$ Adjmatrix(k, A)

% Determine the codes of $v_1,\dots ,v_{(n^2+3n-4)/2}$ which are described as in (1).

$$v_1=2;v_2=\#V(n,k-2);$$

for $j=1:k-2$

$$v_1=v_1+\#\mathcal{D}(n-1)(\#V(n,j-1\left)\right)-\left[\right(d(n-1,4)+2d(n-1,6)]-1;$$

end

for $i=2:n-1$

for $j=(\#\mathcal{D}(i-1)+1):(\#\mathcal{D}(i)-1)$

$$v_{j+i-1}=v_1+b(j,i,n,k-1),$$

end

$$v_{\#\mathcal{D}\left(i\right)+i-1}=v_1+a(i,n,k-1);$$

end

for $j=3:n$

$$v_{\#\mathcal{D}\left(j\right)-1}=v_2+a(j-1,n,k-1)$$

end

for $i=(\#\mathcal{D}(n-1)+1):(\#\mathcal{D}(n)-2)$

$$v_{i+n-1}=v_1+b(i+1,n,n,k-1);$$

end

$$v_{\#\mathcal{D}\left(n\right)+n-2}=v_1+a(n,n,k-1);$$

% Output the adjacency matrix of $\Gamma (n,k-1)$ .

for $i=2:n$

for $j=\#\mathcal{D}(i-1)+1:(\#\mathcal{D}(i)-1)$

$A=$  blkdiag $(A,A(2:v_2-1;2:v_2-1\left)\right)$ ;

end

$A=$  blkdiag $(A,A(2:v_2;2:v_2\left)\right)$ ;

end

$$A(v_1,2+b(2,2,n,k-1\left)\right)=1;A(2+b(2,2,n,k-1),v_1)=1;$$

$$A(v_1,3+b(2,2,n,k-1\left)\right)=1;A(3+b(2,2,n,k-1),v_1)=1;$$

for $i=2:n-1$

$$j=\#\mathcal{D}(i-1)+1;$$

$$A(v_{j+i-1},2+b(\#\mathcal{D}\left(i\right)+1,i+1,n,k-1\left)\right)=1;$$

$$A(2+b(\#\mathcal{D}\left(i\right)+1,i+1,n,k-1),v_{j+i-1})=1;$$

$$A(v_{j+i-1},3+b(\#\mathcal{D}\left(i\right)+1,i+1,n,k-1\left)\right)=1;$$

$$A(3+b(\#\mathcal{D}\left(i\right)+1,i+1,n,k-1),v_{j+i-1})=1;$$

end

for $i=2:n-1$

for $j=1:i-1$ ;

$$A(v_{\frac{i(i+1)}{2}+j},2+b(\frac{i(i+1)}{2}+j+1,i+1,n,k-1\left)\right)=1;A(2+b(\frac{i(i+1)}{2}+j+1,i+1,n,k-1),v_{\frac{i(i+1)}{2}+j})=1;$$

$$A(v_{\frac{i(i+1)}{2}+j},3+b(\frac{i(i+1)}{2}+j+1,i+1,n,k-1\left)\right)=1;A(3+b(\frac{i(i+1)}{2}+j+1,i+1,n,k-1),v_{\frac{i(i+1)}{2}+j})=1;$$

if k = 2

$$A(v_{\frac{i(i+1)}{2}+j},\#\mathcal{D}(i)-(i-j\left)\right)=1;A(\#\mathcal{D}(i)-(i-j),v_{\frac{i(i+1)}{2}+j})=1;$$

$$A(v_{\frac{i(i+1)}{2}+j},\#\mathcal{D}(i+1)-(i+1-j\left)\right)=1;A(\#\mathcal{D}(i+1)-(i+1-j),v_{\frac{i(i+1)}{2}+j})=1;$$

else

$$A(v_{\frac{i(i+1)}{2}+j},\#V(n-1,1)+b(\frac{i(i-1)}{2}+j,i,n,k-1\left)\right)=1;A(\#V(n-1,1)+b(\frac{i(i-1)}{2}+j,i,n,k-1),v_{\frac{i(i+1)}{2}+j})=1;$$

$$A(v_{\frac{i(i+1)}{2}+j},\#V(n,1)-1+b(\frac{i(i-1)}{2}+j,i,n,k-1\left)\right)=1;A(\#V(n,1)-1+b(\frac{i(i-1)}{2}+j,i,n,k-1),v_{\frac{i(i+1)}{2}+j})=1;$$

end

end

end

for $i=2:n$

$$A(v_{\#\mathcal{D}\left(i\right)-1},2+a(i,n,k-1\left)\right)=1;A(2+a(i,n,k-1),v_{\#\mathcal{D}\left(i\right)-1})=1;$$

$$A(v_{\#\mathcal{D}\left(i\right)-1},3+a(i,n,k-1\left)\right)=1;A(3+a(i,n,k-1),v_{\#\mathcal{D}\left(i\right)-1})=1;$$

end

for $i=0:n-2$ % We define $V(1,1)=3$ .if k = 2

$$A(v_{\frac{n(n+1)}{2}+i},\#\mathcal{D}(n)-(n-i-1\left)\right)=1;A(\#\mathcal{D}(n)-(n-i-1),v_{\frac{n(n+1)}{2}+i})=1;$$

$$A(v_{\frac{n(n+1)}{2}+i},\#\mathcal{D}(n+1)-(n-i\left)\right)=1;A(\#\mathcal{D}(n+1)-(n-i),v_{\frac{n(n+1)}{2}+i})=1;$$

else

$$A(v_{\frac{n(n+1)}{2}+i},\#V(n-1,1)+b(\frac{n(n-1)}{2}+i+1,n,n,k-1\left)\right)=1;$$

$$A(\#V(n-1,1)+b(\frac{n(n-1)}{2}+i+1,n,n,k-1),v_{\frac{n(n+1)}{2}+i})=1;$$

$$A(v_{\frac{n(n+1)}{2}+i},\#V(n,1)-1+b(\frac{n(n-1)}{2}+i+1,n,n,k-1\left)\right)=1;$$

$$A(\#V(n,1)-1+b(\frac{n(n-1)}{2}+i+1,n,n,k-1),v_{\frac{n(n+1)}{2}+i})=1;$$

end

end

Function $D=$ Diagmatrix(k, A) % Output the diagonal matrix of $\Gamma (n,k-1)$ .

% Using the same process as that of Function Adjmatrix(k, A), we can determine the codes of $v_1,\dots ,v_{(n^2+3n-4)/2}$ , we omit details.

for $i=2:n$

for $j=\#\mathcal{D}(i-1)+1:(\#\mathcal{D}(i)-1)$

$D=$ blkdiag $(D,D(2:v_2-1;2:v_2-1\left)\right)$ ;

end

$D=$ blkdiag $(D,D(2:v_2;2:v_2\left)\right)$ ;

end

$$D(v_1,v_1)=4$$

for $i=2:n-1$

$$j=\#\mathcal{D}(i-1)+1;$$

$$D(v_{j+i-1},v_{j+i-1})=4;$$

end

for $i=2:n-1$

for $j=1:i-1$ ;

$$D(v_{\frac{i(i+1)}{2}+j},v_{\frac{i(i+1)}{2}+j})=6;$$

end

end

for $i=2:n$

$$D(v_{\#\mathcal{D}\left(i\right)-1},v_{\#\mathcal{D}\left(i\right)-1})=4;$$

end

for $i=0:n-2$

$$D(v_{\frac{n(n+1)}{2}+i},v_{\frac{n(n+1)}{2}+i})=4;$$

end

Using the induction on iterated depth m for given integer $n\ge 2$ , we can give the proof of correctness of above program as follows.

Proof of correctness of program: Without loss of generality, we show that the program is correct for the adjacency matrix. It is easy to see that this is true when m = 1, 2. Assume it is true for the case m – 1 with $m\ge 3$ . Consider the case m. Let A be the adjacency matrix of $\Gamma (n,m-1)$ . Then the adjacency matrix of ${\Gamma }^1(n,m)$ is A since ${\Gamma }^1(n,m)$ is isomorphic to $\Gamma (n,m-1)$ . By Step 2, the uncoded vertices of ${\Gamma }^i(n,m)$ induce a graph isomorphic to ${\Gamma }^1(n,m)-\left\{1\right\}$ for $i\in \{\#\mathcal{D}(l):l=2,\dots ,n\}$ , the uncoded vertices of ${\Gamma }^i(n,m)-\left\{v_{i+l}\right\}$ induce a graph isomorphic to ${\Gamma }^1(n,m)-\{1,v_2\}$ for $l=2,\dots ,n-1,i=\#\mathcal{D}(l-1)+1,\dots ,\#\mathcal{D}(l)-1$ , and the uncoded vertices of ${\Gamma }^i(n,m)-\left\{v_{i+n-1}\right\}$ induce a graph isomorphic to ${\Gamma }^1(n,m)-\{1,v_2\}$ for $i=\#\mathcal{D}(n-1)+1,\dots ,\#\mathcal{D}(n)-1$ . We can get that the adjacency matrix A_i of ${\Gamma }^i(n,m)$ by deleting the first row and column of A for $i\in \{\#\mathcal{D}(l):l=2,\dots ,n\}$ , and the adjacency matrix A_i of ${\Gamma }^i(n,m)$ by deleting the first row and column and the last row and column of A for $l=2,\dots ,n,i=\#\mathcal{D}(l-1)+1,\dots ,\#\mathcal{D}(l)-1$ . Using the function ‘blkdiag’ of Matlab, we obtain a matrix with the form

$$\left(\begin{array}{cccccc}A& *& *& *& *& *\\ *& A_2& *& *& *& *\\ *& *& A_3& *& *& *\\ *& *& *& \ddots & *& *\\ *& *& *& *& \ddots & *\\ *& *& *& *& *& A_{\#\mathcal{D}(n)}\end{array}\right),$$

(2)

where the elements are denoted by $*$ to be accounted for as follows(such as Example 3.1). The elements in $A,A_2,\dots ,A_{\#\mathcal{D}(n)}$ are due to all the edges except $n^2+3n-4+2d(n,6)$ edges (see 16 red edges shown in Figure 5 for n = 3). The elements corresponding to these $n^2+3n-4+2d(n,6)$ edges in $*$ are assigned 1 by the statements below the statement “ $A=$ blkdiag $(A,A(2:v_2;2:v_2))$ ” in function Adjmatrix(k, A) and the others in $*$ are 0. This completes this proof.

Figure 5.

16 missing edges for n = 3 and k = 2.

Example 3.1. Let n = 2 and m = 1. From Example 2.1, we see that

$$A=\left(\begin{array}{ccc}0& 1& 1\\ 1& 0& 1\\ 1& 1& 0\end{array}\right)$$

is the adjacency matrix of $\Gamma (2,0)$ . Then the matrix (2) is

$$\left(\begin{array}{ccc}A& *& *\\ *& A_2& *\\ *& *& A_3\end{array}\right)=\left(\begin{array}{cccccc}0& 1& 1& 0& 0& 0\\ 1& 0& 1& 1& 1& 0\\ 1& 1& 0& 0& 1& 1\\ 0& 1& 0& 0& 1& 0\\ 0& 1& 1& 1& 0& 1\\ 0& 0& 1& 0& 1& 0\end{array}\right),$$

which is the adjacency matrix of $\Gamma (2,1)$ .

Example 3.2. See Figure 6, a Matlab program for n = 2. Input Sirpinskispectrum(3), we can obtain the adjacency matrix, diagonal matrix, normalized Laplacian matrix (see Figure 7), and the spectrum of normalized Laplacian matrix of $\Gamma (2,2)$ :

$$\begin{array}{l}\{0.0141,0.1743,0.2338,0.7500,0.7500,0.8011,1.0679,1.0757\\ 1.2500,1.3831,1.5000,1.5000,1.5000,1.5000,1.5000.\}\end{array}$$

Figure 6.

A Matlab program for n = 2.

Figure 7.

The adjacency matrix, diagonal matrix and normalized Laplacian matrix of $\Gamma (2,2)$ using Matlab. (a) The adjacency matrix A. (b) The diagonal matrix D. (c) The normalized Laplacian matrix L.

Recall that V(n, k) and $\mathcal{L}(n,k)$ denote the vertex set and the normalized Laplacian matrix of the approximating graph $\Gamma (n,k)$ with $k\ge 0$ , respectively. Fix $n\ge 2$ . It is well known that for any $k\ge 1$ , the normalized Laplacian matrix $\mathcal{L}(n,k)$ can be viewed as an operator (called it normalized Laplacian) on the space of functions $f:V(n,k)\mapsto ℝ$ , see. ² Let $-{\Delta }_k^n$ denote the normalized Laplacian corresponding to $\mathcal{L}(n,k)$ .

We remark that our algorithm is convergent. In fact, ${\cup }_{k=0}^{\infty }V(n,k)$ is dense in K(n) in the natural topology ¹⁸ and we have from the [24, Section ‘From graphs to fractals’] that there exists a normalized Laplacian $-{\Delta }^n$ on K(n) such that

$$-{\Delta }^n{f}_0(x)=\underset{k\to \infty }{\lim }-{\Delta }_k^n(f_0|V(n,k)),x\in K(n),$$

where $f_0\in \{f:f\text{maps}K(n)\text{to}ℝ\}$ whose restriction on V(n, k) is $f_0|V(n,k)$ . This implies our algorithm is convergent.

Conclusion

We have developed an approach to compute the normalized Laplacian spectra of approximating graphs of a class of general Sierpinski triangles in this paper. It’s easy to see that we can also get the spectra of adjacency matrix, Laplacian matrix, Markov matrix, normalized Markov matrix and signless Laplacian matrix $D(n,k)+A(n,k)$ ⁶ using same method. Our approach could be also used to obtain the similar approximating graphs associated with some fractals generated by iterated function system and their various spectra, such as, when one replace the Δ in this paper by a regular tetrahedron in ${ℝ}^3$ . Future work should include studying the various topological properties of these approximating graphs like connectedness, toughness, etc; determining the spectra of networks associated with general Sierpinski triangle, and measuring the related structural properties and dynamical processes, specifically in relation to random walk.