Flows in Complex Networks: Theory, Algorithms, and Application to Lennard-Jones Cluster Rearrangement

Maria Cameron, Eric Vanden-EijndenarXiv 2014 · DOI · OpenReview

Abstract

A set of analytical and computational tools based on transition path theory (TPT) is proposed to analyze flows in complex networks. Specifically, TPT is used to study the statistical properties of the reactive trajectories by which transitions occur between specific groups of nodes on the network. Sampling tools are built upon the outputs of TPT that allow to generate these reactive trajectories directly, or even transition paths that travel from one group of nodes to the other without making any detour and carry the same probability current as the reactive trajectories. These objects permit to characterize the mechanism of the transitions, for example by quantifying the width of the tubes by which these transitions occur, the location and distribution of their dynamical bottlenecks, etc. These tools are applied to a network modeling the dynamics of the Lennard-Jones cluster with 38 atoms (LJ38) and used to understand the mechanism by which this cluster rearranges itself between its two most likely states at various temperatures.

Introduction

In fields ranging from molecular biology to social sciences, complex systems are frequently modeled as networks where nodes represent states and edges represent transitions. While the static structure of these networks is often well-characterized, understanding their dynamics—specifically how a system transitions from one metastable state to another—remains a significant challenge. Traditional approaches, such as Large Deviation Theory (LDT), are effective at near-zero temperatures where transitions follow a single "minmax" path of least resistance. However, real-world systems often operate at finite temperatures where thermal fluctuations allow for a multitude of pathways.

This paper addresses the need for tools that can quantify these "reactive trajectories"—the segments of a long equilibrium simulation that successfully navigate from a source state (A) to a target state (B). By extending Transition Path Theory (TPT) to discrete networks modeled as Markov Jump Processes (MJPs), the authors provide a framework to visualize and analyze the "transition tubes" through which these flows occur.

Methodology: Transition Path Theory on Networks

The core of the proposed framework is the adaptation of TPT for discrete networks. The central object is the committor function ($q$), which assigns a value to every node representing the probability that a process starting at that node will reach the target state B before returning to state A.

Building on the committor function, the authors introduce two novel sampling algorithms for generating reactive trajectories:

  1. Reactive Process with Detours: This method captures the actual paths taken in a long equilibrium run by lumping all non-reactive time (time spent wandering within A or B) into a virtual state.
  2. No-Detour Process: This algorithm generates "idealized" trajectories where the committor value increases monotonically. By stripping away back-and-forth wandering, it reveals the productive probability current of the reaction.

Additionally, the authors use isocommittor cuts to measure the "width" of the transition channel. By counting the number of edges carrying the current at specific stages of the reaction (defined by $q$ values), they can identify dynamical bottlenecks where the flow is constrained.

Case Study: The Lennard-Jones 38-atom Cluster

To demonstrate these tools, the authors analyze the rearrangement of a Lennard-Jones cluster with 38 atoms ($LJ_{38}$). This system is a classic benchmark in chemical physics due to its "double-funnel" energy landscape.

Structural states of the LJ38 cluster: (a) FCC and (b) ICO.

The $LJ_{38}$ network is highly complex, consisting of 71,887 nodes (local energy minima) and 119,853 edges (transition states). The system must navigate between two primary structural motifs: a deep, narrow octahedral structure (Face-Centered Cubic, or FCC) and a slightly higher-energy but much broader icosahedral structure (ICO).

Key Insights and Results

The application of TPT to the $LJ_{38}$ cluster reveals how the transition mechanism evolves with temperature:

  • Entropic Broadening: At low temperatures ($T=0.05$), the transition current is concentrated on a single dominant path. As temperature increases to $T=0.18$, the "minmax" path becomes largely irrelevant. Instead, the system accesses thousands of energetically less favorable but entropically numerous pathways, forming a broad "transition tube."
  • Rate and Mechanism: While the overall transition rates calculated via TPT match previous specialized studies and follow an Arrhenius-like law, the TPT analysis shows that the underlying physical mechanism changes drastically. The "tube" of pathways widens, and the flow becomes distributed across a vast web of transitions.
  • Identification of Traps: Visualizations of reactive trajectories reveal that the system often wanders into "quasi-deadends"—intermediate structures that do not lead directly to the goal but are frequently visited during the transition process.

Discussion and Future Directions

While the proposed tools provide deep mechanistic insights, some challenges remain. At extremely low temperatures, the vast scale separation between transition rates can lead to numerical precision issues, requiring network truncation or specialized solvers. Furthermore, the current framework assumes "detailed balance" (time-reversibility). Extending these sampling tools to non-equilibrium networks, such as directed biological signaling pathways or social networks, represents a promising avenue for future research.

Summary updated

Abstract

A set of analytical and computational tools based on transition path theory (TPT) is proposed to analyze flows in complex networks. Specifically, TPT is used to study the statistical properties of the reactive trajectories by which transitions occur between specific groups of nodes on the network. Sampling tools are built upon the outputs of TPT that allow to generate these reactive trajectories directly, or even transition paths that travel from one group of nodes to the other without making any detour and carry the same probability current as the reactive trajectories. These objects permit to characterize the mechanism of the transitions, for example by quantifying the width of the tubes by which these transitions occur, the location and distribution of their dynamical bottlenecks, etc. These tools are applied to a network modeling the dynamics of the Lennard-Jones cluster with 38 atoms (LJ38) and used to understand the mechanism by which this cluster rearranges itself between its two most likely states at various temperatures.
BibTeX
@misc{cameron2014flows,
  author = {Maria Cameron and Eric Vanden-Eijnden},
  title = {{Flows in Complex Networks: Theory, Algorithms, and Application to Lennard-Jones Cluster Rearrangement}},
  year = {2014},
  eprint = {1402.1736},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1402.1736},
}
Figure