Imagine a crowded room where everyone is trying to reach a consensus on a topic. In a classical computer simulation, we might model this using a global clock: at tick $t$, everyone listens to their neighbors, and at tick $t+1$, everyone simultaneously updates their opinion.
This is the synchronous update assumption. It is a cornerstone of classical network modeling, systems biology, and distributed computing because it makes the mathematics beautifully tractable. When every node in a network updates at the exact same time, we can represent the system's state transitions using clean matrix algebra or deterministic mappings.
But real life does not have a global synchronizing clock. Neurons fire when they reach an individual electrical threshold; social media users post when they wake up or feel inspired; molecules collide stochastically at varying speeds. By forcing an artificial global clock onto these systems, we do not just lose precision—we introduce severe mathematical artifacts that fundamentally distort how information, opinions, and biological signals actually propagate.
The Unintended Artifacts of Synchronization
When we force a naturally asynchronous system into a synchronous execution model, the system can exhibit behaviors that disappear entirely under more realistic conditions.
First, synchronous models create artificial clusters of simultaneous events. In the study of complex contagions, such as rumor or disease propagation, Yijun Ran et al. demonstrate in A generalized linear threshold model for an improved description of the spreading dynamics that synchronous batch updating forces nodes to wait for arbitrary discrete time steps before changing state. This artificial synchronization creates unrealistic, massive clusters of simultaneous infections, obscuring the true stochastic sequence of individual transitions.
Second, synchronization can cause networks to instantly lock into an incorrect consensus. In social learning environments, Maryam Bahrani et al. show in Asynchronous Majority Dynamics in Preferential Attachment Trees that perfectly synchronized updates can cause a hierarchical network to converge instantly on an incorrect majority opinion if the initial signals are noisy. Conversely, asynchronous updates act as a temporal buffer, allowing correct information to stabilize in small pockets before spreading across the network.
Third, synchronous updates can actively destroy structural memory. When analyzing majority-like dynamics on clustered graphs, Emilio Cruciani et al. discover in On the Metastability of Quadratic Majority Dynamics on Clustered Graphs and its Biological Implications that asynchronous environments naturally preserve "metastable" states. In these states, distinct communities maintain internal agreement but remain different from one another, effectively "remembering" the network's cluster topology. A synchronous update rule can easily disrupt this local reinforcement, wiping out the system's structural memory.
Modeling Real-World Bursts and Delays
To capture the true behavior of decentralized systems, researchers are replacing global clocks with event-driven, continuous-time frameworks that accommodate irregular timing and communication delays.
In human social networks, interactions are notoriously "bursty" rather than regular. As Buddhika Nettasinghe argues in In-Group Love, Out-Group Hate: A Framework to Measure Affective Polarization via Contentious Online Discussions, models relying on regular, synchronous intervals fail to capture the erratic nature of online polarization. To address this, frameworks like SLANT, developed by Abir De et al. in Learning and Forecasting Opinion Dynamics in Social Networks, treat opinion evolution as a continuous-time process where updates occur at irregular, user-driven intervals.
Synchronous: |---Update---|---Update---|---Update---| (Rigid, simultaneous)
Asynchronous: |--Node A--|----Node B----|--Node A--| (Event-driven, bursty)
Accounting for the physical time it takes for information to travel is equally critical. In A predictive model for the temporal dynamics of information diffusion in online social networks, the authors incorporate explicit time-delay parameters onto every edge of a social graph. This allows them to simulate the variable latency between a user receiving information and subsequently interacting with it, replicating real-world viral propagation patterns that synchronous models miss.
This shift to asynchrony also reveals unexpected structural resilience in economic and optimization frameworks. For example, Yoav Kolumbus et al. investigate market dynamics in Asynchronous Proportional Response Dynamics: Convergence in Markets with Adversarial Scheduling. They prove that decentralized bidding heuristics guarantee convergence to market equilibrium even under "adversarial scheduling"—where an opponent decides exactly who updates when—provided every player eventually gets a turn.
Non-Determinism and Control in Biological Circuits
In systems biology, modeling gene regulatory networks (GRNs) using discrete Boolean states is a powerful paradigm. However, assuming that every gene in an organism flips its state simultaneously is biologically unrealistic.
When we transition to asynchronous Boolean networks, we introduce a massive challenge: non-determinism. Because different cellular processes occur at varying timescales, a single state can branch into multiple potential trajectories depending on which gene updates first. As Cui Su and Jun Pang 0001 explain in Cabean 2.0: Efficient and Efficacious Control of Asynchronous Boolean Networks, this non-determinism makes predicting and controlling a cell's trajectory toward a healthy stable state (attractor) exceptionally difficult.
To bridge the gap between mathematical simplicity and biological reality, researchers look for properties that remain invariant across both update schemes. In Logic Programming for Boolean Networks, Katsumi Inoue demonstrates that by using cardinality constraints within logic programs, one can ensure that point attractors (steady states) remain consistent regardless of whether the network updates synchronously or asynchronously. This mathematical bridge ensures that stable states identified in simplified synchronous models remain valid under realistic, asynchronous biological conditions.
Similarly, in artificial neural architectures, moving away from a global clock requires proving that the network remains stable despite communication delays. In Asynchronous Dynamics of Continuous Time Neural Networks, the authors prove that contractive and monotone networks will reliably converge to the correct stable state even when operating with independent local clocks and variable signal propagation delays.
The Computational Cost of Reality
If asynchronous modeling is demonstrably more accurate, why has the synchronous update assumption persisted for so long? The answer comes down to a steep explosion in computational complexity.
In a synchronous system, every state has exactly one deterministic successor. The state space is a straightforward directed graph where paths never branch. In an asynchronous system, because any node or subset of nodes might update next, the state space explodes into a dense, highly branched nondeterministic network.
Synchronous State Space: (State 1) -------> (State 2) -------> (State 3)
+---> (State 2A via Node A)
|
Asynchronous State Space: (State 1) ---+---> (State 2B via Node B)
|
+---> (State 2C via Both)
The theoretical boundaries of this complexity are severe:
- Finding Stable States: Christopher L. Barrett et al. prove in Gardens of Eden and Fixed Points in Sequential Dynamical Systems that when node updates are sequential rather than parallel, determining whether a system even possesses a stable fixed point is generally NP-complete.
- Predicting Long-Term Cycles: When moving from single-layer networks to multi-agent interactions across multiple layers, the complexity compounds. Daniel Rosenkrantz et al. demonstrate in On Some Fundamental Problems for Multi-Agent Systems Over Multilayer Networks that asynchronous update models over multilayer networks trigger a "complexity explosion," resulting in exponentially long behavioral cycles and making it NP-complete to verify if two seemingly identical systems will behave the same way.
Ultimately, the research direction across these diverse fields highlights a shared realization: while the synchronous update assumption served as a vital training wheel for early network science, understanding the true behavior of markets, societies, and living organisms requires embracing the messy, non-deterministic reality of asynchronous time.
Key Foundational Work
Optimal Signalling in Attractor Neural Networks (1993) — contributed 1 concept elements to this direction.
Asynchronous Dynamics of Continuous Time Neural Networks (1993) — contributed 1 concept elements to this direction.
Gardens of Eden and Fixed Points in Sequential Dynamical Systems (2001) — contributed 1 concept elements to this direction.