Little Problem About Small World Graphs

Little Problem About Small World Graphs

Consider a graph G:=(V,E)G := (V, E) with |V|=n\lvert V \rvert = n and uvEuv \in E if and only if |uv|=2k\lvert u - v \rvert = 2^k for some k0k \in \mathbb{N}_0. What is E[d(u,v)n]E\left[d\big(u, v) \mid n\right]? I was interested in showing that this is in fact a “small-world” graph, i.e. that Llog(n)L \propto \log\big(n). This is not so difficult, as the maximum distance between two nodes uu and vv cannot exceed log2(n)\log_2\big(n).

My first attempt to find the average involved bounding the average as follows. In this setting, there’s no trip with greater distance than from a node to the node directly behind it, i.e. dn(0,n1)E[dn(0,v)]d_{n}\big(0, n - 1) \geq E[d_{n}\big(0, v)]. But this is just a minor improvement over the maximum.

0nln(u)nln(2)du=log2(n)1ln(2)\begin{equation} \int_{0}^{n}\frac{\ln\left(u\right)}{n \ln\left(2\right)}\ du = \log_2\big(n) - \frac{1}{\ln\big(2)} \end{equation}

Instead, notice that the distance between nodes indexed uu and vv is equal to the number of bits set in uxorvu \ \text{xor} \ v. This gives us an expected distance of 0.5log2(n)0.5 \cdot \log_2\big(n) and a distribution of distances that’s very roughly N(log2(n)/2,log2(n)/4)\operatorname{N}\big(\log_2\big(n)/2, \log_2\big(n)/4).