Answers are grounded in this paper and cite their sources.
This paper derives explicit finite-horizon bounds on the mean-square synchronization error of a drive-response pair of delayed fractional-order stochastic neural networks. The networks have Caputo derivatives of order η∈(12,1), Lipschitz activation functions, a constant delay and a noise intensity that depends on the current and the delayed state, and the response is driven by a linear feedback controller. The linear part of the error system is treated exactly by a Mittag-Leffler variation-of-constants formula, which we prove for arbitrary progressively measurable forcing terms with bounded second moments; these may contain the delayed state. The Itô isometry then gives a delayed integral inequality for the second moment of the error with the singular kernel (t−s)2η−2. A delayed fractional Gronwall estimate based on a known running-supremum argument turns this inequality into an explicit synchronization criterion, a pointwise bound on the mean-square error and a certified horizon defined by one scalar equation. A second criterion follows from a Hölder-type delayed Gronwall inequality, and we prove that the Mittag-Leffler bound is never larger. Both criteria are independent of the delay and of the control gains: they hold for every admissible gain, but they do not improve when the gains increase. Monte Carlo simulations for several fractional orders, delays, gains and network sizes up to 100 neurons confirm the bounds. They also show that the certified horizon is a guaranteed working window, not an estimate of the time at which synchronization is lost.
No takes yet. Share an insight, caveat, or question.
Fakhfakh et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: