TheWatts–Strogatz model is arandom graph generation model that produces graphs withsmall-world properties, including shortaverage path lengths and highclustering. It was proposed byDuncan J. Watts andSteven Strogatz in their article published in 1998 in theNature scientific journal.^[1] The model also became known as the(Watts)beta model after Watts used${\displaystyle \beta }$ to formulate it in his popular science bookSix Degrees.

The formal study ofrandom graphs dates back to the work ofPaul Erdős andAlfréd Rényi.^[2] The graphs they considered, now known as the classical orErdős–Rényi (ER) graphs, offer a simple and powerful model with many applications.

However theER graphs do not have two important properties observed in many real-world networks:

1. They do not generate local clustering andtriadic closures. Instead, because they have a constant, random, and independent probability of two nodes being connected, ER graphs have a lowclustering coefficient.
2. They do not account for the formation of hubs. Formally, thedegree distribution of ER graphs converges to aPoisson distribution, rather than apower law observed in many real-world,scale-free networks.^[3]

The Watts and Strogatz model was designed as the simplest possible model that addresses thefirst of the two limitations. It accounts for clustering while retaining the short average path lengths of the ER model. It does so by interpolating between a randomized structure close to ER graphs and a regular ringlattice. Consequently, the model is able to at least partially explain the "small-world" phenomena in a variety of networks, such as the power grid, neural network ofC. elegans, networks of movie actors, or fat-metabolism communication inbudding yeast.^[4]

Given the desired number of nodes${\displaystyle N}$, the meandegree${\displaystyle K}$ (assumed to be an even integer), and a parameter${\displaystyle \beta }$, all satisfying${\displaystyle 0\le \beta \le 1}$ and${\displaystyle N\gg K\gg \ln N\gg 1}$, the model constructs anundirected graph with${\displaystyle N}$ nodes and${\displaystyle \frac{NK}{2}}$ edges in the following way:

1. Construct a regular ring lattice, a graph with${\displaystyle N}$ nodes each connected to${\displaystyle K}$ neighbors,${\displaystyle K/2}$ on each side. That is, if the nodes are labeled${\displaystyle 0\dots N-1,}$ there is an edge${\displaystyle (i,j)}$ if and only if
2. For every node${\displaystyle i=0,\dots ,N-1}$ take every edge connecting${\displaystyle i}$ to its${\displaystyle K/2}$ rightmost neighbors, that is every edge${\displaystyle (i,j)}$ such that, and rewire it with probability${\displaystyle \beta }$. Rewiring is done by replacing${\displaystyle (i,j)}$ with${\displaystyle (i,k)}$ where${\displaystyle k}$ is chosen uniformly at random from all possible nodes while avoiding self-loops (${\displaystyle k\ne i}$) and link duplication (there is no edge${\displaystyle (i,k^{\prime })}$ with${\displaystyle k^{\prime }=k}$ at this point in the algorithm).

The underlying lattice structure of the model produces a locally clustered network, while the randomly rewired links dramatically reduce theaverage path lengths. The algorithm introduces about${\displaystyle \beta \frac{NK}{2}}$ of such non-lattice edges. Varying${\displaystyle \beta }$ makes it possible to interpolate between a regular lattice (${\displaystyle \beta =0}$) and a structure close to anErdős–Rényi random graph${\displaystyle G(N,p)}$ with${\displaystyle p=\frac{K}{N-1}}$ at${\displaystyle \beta =1}$. It does not approach the actual ER model since every node will be connected to at least${\displaystyle K/2}$ other nodes.

The three properties of interest are theaverage path length, theclustering coefficient, and thedegree distribution.

For a ring lattice, the average path length^[1] is${\displaystyle \ell (0)\approx N/2K\gg 1}$ and scales linearly with the system size. In thelimiting case of${\displaystyle \beta \to 1}$, the graph approaches a random graph with${\displaystyle \ell (1)\approx \frac{\ln N}{\ln K}}$, while not actually converging to it. In the intermediate region, the average path length falls very rapidly with increasing${\displaystyle \beta }$, quickly approaching its limiting value.

For the ring lattice theclustering coefficient^[5]${\displaystyle C(0)=\frac{3(K-2)}{4(K-1)}}$, and so tends to${\displaystyle 3/4}$ as${\displaystyle K}$ grows, independently of the system size.^[6] In the limiting case of${\displaystyle \beta \to 1}$ the clustering coefficient is of the same order as the clustering coefficient for classical random graphs,${\displaystyle C=K/(N-1)}$ and is thus inversely proportional to the system size. In the intermediate region the clustering coefficient remains quite close to its value for the regular lattice, and only falls at relatively high${\displaystyle \beta }$. This results in a region where the average path length falls rapidly, but the clustering coefficient does not, explaining the "small-world" phenomenon.

If we use the Barrat and Weigt^[6] measure for clustering${\displaystyle C^{\prime }(\beta )}$ defined as the fraction between the average number of edges between the neighbors of a node and the average number of possible edges between these neighbors, or, alternatively,

${\displaystyle C^{\prime }(\beta )\equiv \frac{3\times \text{number of triangles}}{\text{number of connected triples}}}$

then we get${\displaystyle C^{\prime }(\beta )\sim C(0)(1-\beta {)}^3.}$

The degree distribution in the case of the ring lattice is just aDirac delta function centered at${\displaystyle K}$. The degree distribution for a large number of nodes and can be written as,^[6]

${\displaystyle P(k)\approx \sum _{n=0}^{f(k,K)}(\genfrac{}{}{0ex}{}{K/2}{n})(1-\beta {)}^n{\beta }^{K/2-n}\frac{(\beta K/2{)}^{k-K/2-n}}{(k-K/2-n)!}{e}^{-\beta K/2},}$

where${\displaystyle k_i}$ is the number of edges that the${\displaystyle i^{\text{th}}}$ node has or its degree. Here${\displaystyle k\ge K/2}$, and${\displaystyle f(k,K)=min(k-K/2,K/2)}$. The shape of the degree distribution is similar to that of a random graph and has a pronounced peak at${\displaystyle k=K}$ and decays exponentially for large${\displaystyle |k-K|}$. The topology of the network is relatively homogeneous, meaning that all nodes are of similar degree.

The major limitation of the model is that it produces an unrealisticdegree distribution. In contrast, real networks are oftenscale-free networks inhomogeneous in degree, having hubs and a scale-free degree distribution. Such networks are better described in that respect by thepreferential attachment family of models, such as theBarabási–Albert (BA) model. (On the other hand, the Barabási–Albert model fails to produce the high levels of clustering seen in real networks, a shortcoming not shared by the Watts and Strogatz model. Thus, neither the Watts and Strogatz model nor the Barabási–Albert model should be viewed as fully realistic.)

The Watts and Strogatz model also implies a fixed number of nodes and thus cannot be used to model network growth.

• Small-world networks
• Erdős–Rényi (ER) model
• Barabási–Albert model
• Social networks

• Social network analysis
• Random graphs

• CS1 maint: multiple names: authors list
• Articles with short description
• Short description is different from Wikidata