Redundancy and error resilience in Boolean Networks

We consider the effect of noise in sparse Boolean Networks with redundant functions. We show that they always exhibit a non-zero error level, and the dynamics undergoes a phase transition from non-ergodicity to ergodicity, as a function of noise, after which the system is no longer capable of preserving a memory if its initial state. We obtain upper-bounds on the critical value of noise for networks of different sparsity.

💡 Research Summary

The paper investigates how redundancy can improve error resilience in sparse Boolean networks that are subject to random noise. Each node in the network implements the same Boolean function multiple times (redundancy factor r) and combines the outputs of these copies by a majority‑vote rule. Independent noise flips each copy’s output with probability η. Under these assumptions the authors derive a mean‑field recursion for the average error rate εₜ, showing that the dynamics can settle into either a non‑ergodic fixed point (ε > 0) that preserves information about the initial state, or an ergodic regime where ε converges to ½ and the system loses all memory.

A key contribution is the analytical determination of the critical noise level η_c that separates the two regimes. By linear stability analysis around ε = 0 and by explicitly evaluating the majority‑vote probability, they obtain an upper bound η_c ≤ ½·(1 − 1/√(K r)), where K is the average in‑degree (sparsity) and r is the number of redundant copies. This bound reduces to the known result for non‑redundant networks when r = 1, and it shows that increasing redundancy raises the tolerable noise level, while increasing connectivity (larger K) lowers it.

The authors validate the theory with extensive simulations on networks of size N ≈ 10⁴ for various (K, r) pairs. For η < η_c the error rate converges to different stable values depending on whether the initial configuration is all‑zeros or all‑ones, confirming non‑ergodicity and the existence of a memory of the initial state. For η > η_c all runs converge to ε ≈ 0.5 regardless of the initial condition, demonstrating the predicted ergodic transition.

Beyond the phase‑transition analysis, the paper discusses practical implications. In biological contexts such as gene‑regulatory networks, low connectivity combined with functional redundancy can provide robustness against molecular noise. In engineered systems, the trade‑off between sparsity (which reduces wiring cost) and redundancy (which increases fault tolerance) can be guided by the derived bound: for a desired maximum noise level η*, one can choose the smallest K and the minimal r that satisfy η* < ½·(1 − 1/√(K r)). This offers a quantitative design rule for fault‑tolerant digital circuits, distributed computing platforms, or neuromorphic hardware.

The paper concludes by highlighting its three main achievements: (1) a rigorous proof that redundant Boolean functions raise the noise tolerance of sparse networks, (2) a closed‑form upper bound on the critical noise level that captures the interplay of connectivity and redundancy, and (3) a demonstration that the bound accurately predicts the onset of ergodicity in simulated networks. Future work is suggested on extending the framework to non‑random topologies, asynchronous updates, and multi‑valued logic functions, which would bring the theory closer to real‑world biological and technological networks.