Harmonic evolutions on graphs

We define the harmonic evolution of states of a graph by iterative application of the harmonic operator (Laplacian over $Z_2$). This provides graphs with a new geometric context and leads to a new tool to analyze them. The digraphs of evolutions are analyzed and classified. This construction can also be viewed as a certain topological generalization of cellular automata.

💡 Research Summary

The paper introduces a novel dynamical system on finite graphs called “harmonic evolution.” The state of a graph is a binary vector assigning 0 (inactive) or 1 (active) to each vertex, i.e., an element of the vector space $\mathbb{Z}_2^{|V|}$. The evolution rule is defined by the harmonic operator $H$, which is the graph Laplacian $L = D - A$ reduced modulo 2. Applying $H$ to a state $s$ yields a new state $Hs$, where each vertex’s new value is the XOR of its current value with the XOR of its neighbours’ values. Because $H$ is linear over $\mathbb{Z}_2$, repeated application produces the sequence $s, Hs, H^2s, \dots$, and the entire dynamics can be captured by a directed graph $D(G)$ whose vertices are all possible states and whose edges point from $s$ to $Hs$.

A central theoretical result is the decomposition of the state space into the kernel $N=\ker H$ and the image $R=\operatorname{im} H$, yielding a direct sum $V = N \oplus R$. On $R$, $H$ acts as a bijection (its eigenvalue is 1), so every state in $R$ belongs to a cycle. On $N$, $H$ annihilates states (eigenvalue 0), leading to fixed points or trees that converge to a zero state. Consequently each connected component of $D(G)$ consists of a single directed cycle (the “loop part”) together with rooted trees feeding into it (the “tree part”). The minimal polynomial of $H$ is $x(x+1)$, guaranteeing that all cycles have length at most 2 and that the nilpotent part has depth at most the dimension of $N$.

The authors illustrate the theory with several families of graphs. For a complete graph $K_n$, $H$ is the all‑ones matrix modulo 2, giving a single large cycle of length 2 covering all $2^n$ states. For a cycle $C_m$, $H$ is invertible, so every state lies on a 2‑cycle. For a path $P_n$, the end vertices belong to $N$, producing tree‑like convergence to the zero state. These examples demonstrate how the structure of $D(G)$ reflects combinatorial properties of the underlying graph.

Beyond pure graph theory, the construction is interpreted as a topological generalization of cellular automata. Traditional cellular automata operate on regular lattices with fixed local rules; here the graph itself supplies the adjacency relation, and the XOR rule is the update function. This framework therefore extends automata concepts to arbitrary networks, opening avenues for modeling diffusion, consensus, error‑correcting codes, and cryptographic permutations on irregular topologies.

The paper concludes with several directions for future work. Extending the field from $\mathbb{Z}_2$ to $\mathbb{Z}_p$ (prime $p>2$) would yield multi‑state automata and richer spectral behavior. The binary spectrum of $H$ (multiplicities of eigenvalues 0 and 1) provides new graph invariants that could aid in graph isomorphism testing. When $H$ is invertible, the evolution digraph becomes a disjoint union of cycles, suggesting applications in designing permutation‑based cryptographic primitives. Finally, quantifying dynamical complexity—such as maximal cycle length, tree depth, and mixing time—in terms of classical graph parameters (e.g., degree distribution, girth, treewidth) is proposed as a promising research agenda.