History dependence in synchronization from rate asymmetry and phase weight feedback in adaptive networks

In the natural world, from the flashing of fireflies to the beating of a heart, separate units often fall into step with one another. This phenomenon, known as synchronization, occurs when independent rhythms align to create a unified, collective motion. Scientists have long studied how this happens, usually imagining a fixed set of connections between the units, like a rigid scaffold holding a group of dancers together. However, many real-world systems are not so static. In adaptive networks, the connections themselves change based on how the units are behaving. If two units move in harmony, the link between them might grow stronger; if they fall out of step, the link might weaken. This creates a feedback loop where the behavior of the group shapes the connections, and the connections, in turn, shape the behavior. The question researchers have been asking is whether this changing web of connections allows a system to "remember" its past. If a group of oscillators has been synchronized for a long time and then the conditions change, does it stay synchronized? Or does it snap back to chaos? The answer depends on the specific rules governing how fast these connections grow versus how fast they fade.

A new study by Yousef Jamali at Tarbiat Modares University investigates this memory effect in a network of adaptive oscillators. The researchers built a computer model of a large group of units, each with its own natural rhythm, connected by a sparse web of links. The strength of these links was not fixed; instead, they evolved over time. When two connected units were close to moving in sync, the link between them strengthened. When they were far apart in their rhythm, the link weakened. The crucial variable in this experiment was the speed of these changes. The team wanted to know what happens if the links strengthen much faster than they weaken, versus the opposite scenario where they weaken much faster than they strengthen. To ensure a fair comparison, they kept the total amount of "activity" in the system constant, simply shifting the balance between the speed of growth and the speed of decay.

The results revealed a striking dependence on the direction of this imbalance. When the system was set up so that links weakened much faster than they strengthened, the network exhibited a massive memory effect. The researchers found that the system could exist in two completely different states depending on its history. If the group started in a disordered, chaotic state and the connections were slowly strengthened, the system remained disordered even when the conditions were favorable for synchronization. However, if the group started in a synchronized state and the connections were slowly weakened, the group stayed synchronized even when the conditions had become unfavorable. This created a wide gap, or hysteresis loop, where the final state of the system was determined entirely by the path it took to get there. In this weakening-dominated regime, the system could remain in a synchronized state long after the external conditions that created it had vanished.

In contrast, when the researchers flipped the script and made the links strengthen much faster than they weakened, this memory effect almost disappeared. The system became forgetful. Whether it started in a synchronized or disordered state, it quickly settled into the same outcome based on the current conditions, ignoring its past. The "history-dependent" interval, where the past mattered, shrank to nearly nothing. This suggests that the balance between reinforcement and decay is not just a minor detail but a fundamental control knob for memory in these systems. If the system is too eager to build new connections, it forgets its history. If it is too quick to let go of connections, it holds onto its past states with surprising tenacity.

To understand where this memory was actually stored, the researchers performed a clever experiment. They took the synchronized state of the network, which consists of two parts: the rhythm of each individual unit and the strength of the connections between them. They then mixed and matched these parts. They took the rhythms from a synchronized group and paired them with the connection strengths from a disordered group, and vice versa. They found that neither the rhythms alone nor the connection strengths alone were enough to recreate the synchronized state. The system only remained synchronized when both the rhythms and the connection strengths were taken from a synchronized history and kept together. This revealed that the memory is not stored in the individual parts but in the specific relationship between them. It is the precise alignment of strong connections with the right rhythmic partners that preserves the synchronized state.

The study also looked at how long this memory lasts. By holding the system in a specific state for a very long time—much longer than the time it takes for the connections to change—the researchers observed that the two different histories remained distinct. The system that started synchronized stayed synchronized, while the one that started disordered stayed disordered, even after thousands of time units. While the disordered state showed a very slow drift toward synchronization, the separation between the two states remained clear and robust. This indicates that the system does not just temporarily remember its past; it settles into persistent regimes that are stable over long periods.

The findings offer a clear picture of how local rules can create global memory. The system works through a positive feedback loop: when units synchronize, their connections strengthen, which makes it easier for them to stay synchronized. The researchers showed that the speed of this loop depends on the ratio of strengthening to weakening. When weakening is fast, the system can easily fall apart if it starts disordered, but once it is together, it is hard to break apart because the connections decay quickly if they are not constantly reinforced by synchronization. This creates a wide range of conditions where the system's fate is decided by its history. When strengthening is fast, the system builds connections so quickly that even a disordered group can become synchronized, erasing the distinction between different starting points.

This research highlights that memory in complex networks is a relational property. It is not enough to have strong connections or synchronized rhythms in isolation; the two must be organized in a specific way. The study suggests that in any adaptive system, from biological neural networks to engineered power grids, the balance between how fast things grow and how fast they fade determines whether the system can retain a record of its past. If the decay of connections is rapid compared to their growth, the system becomes a guardian of its history, capable of maintaining distinct states that would otherwise be impossible. If growth dominates, the system becomes fluid and forgetful, responding only to the immediate present. The work provides a fundamental insight into how the simple rules of adaptation can shape the complex, memory-laden behavior of the natural world.