adiabatic_scan

experimental.network_dynamics.analysis.adiabatic_scan

Adiabatic parameter scan: a network bifurcation diagram.

Slowly ramp one parameter from low to high (and optionally back down to catch hysteresis), carrying the settled state forward between steps, and record summary statistics of an observed network signal at each value. This traces a bifurcation-diagram-like picture of how the network’s activity changes with the swept parameter.

The swept parameter is addressed with a lens accessor applied through eqx.tree_at (e.g. lambda c: c.coupling.instant.G), so any nested config field can be scanned without the function knowing its name.

Classes

Name	Description
AdiabaticScanResult	Result of an :func:adiabatic_scan.

AdiabaticScanResult

experimental.network_dynamics.analysis.adiabatic_scan.AdiabaticScanResult(
    p,
    n_up,
    stats,
)

Result of an :func:adiabatic_scan.

Attributes

Name	Type	Description
p	jax.Array	The swept parameter values, in scan order. Length 2*n when bothways (up then down), else n.
n_up	int	Number of values in the ascending branch. p[:n_up] is the up-branch and p[n_up:] the down-branch.
stats	Bunch of str -> jax.Array	One array per recorded statistic, stacked along the scan axis and reachable by attribute (stats.mean) or key (stats["mean"]). Shape is [len(p)] for scalar reducers and [len(p), ...] for vector-valued reducers (e.g. [len(p), n_nodes] for a per-node statistic).

Functions

Name	Description
adiabatic_scan	Ramp one parameter and record network statistics (bifurcation diagram).

adiabatic_scan

experimental.network_dynamics.analysis.adiabatic_scan.adiabatic_scan(
    network,
    solver=None,
    *,
    accessor,
    low,
    high,
    n,
    t=2000.0,
    skip=1000.0,
    dt=1.0,
    t0=0.0,
    bothways=True,
    observe=None,
    statistics=None,
)

Ramp one parameter and record network statistics (bifurcation diagram).

Parameters

Name	Type	Description	Default
network	Network		required
solver	solver instance, optional (default: Heun())		None
accessor	callable	Lens onto the swept leaf, used as eqx.tree_at(accessor, config, value). Example: lambda c: c.coupling.instant.G.	required
low	float	Bounds of the swept parameter.	required
high	float	Bounds of the swept parameter.	required
n	int	Number of values per branch.	required
t	float	Simulation duration per step in ms.	2000.0
skip	float	Transient duration in ms discarded before computing statistics.	1000.0
dt	float	Integration timestep in ms.	1.0
t0	float	Simulation start time.	0.0
bothways	bool	If True, scan up then back down to expose hysteresis.	True
observe	callable	result -> [n_time, n_nodes] signal to summarise. Defaults to the first variable across all nodes.	None
statistics	dict of str -> callable	Maps a name to a reducer [n_time, n_nodes] -> scalar or array. Vector-valued reducers (e.g. a per-node [n_nodes] statistic) are supported as long as the output shape is the same at every scan point. Defaults to mean/min/max of the per-node temporal mean across the network.	None

Returns

Name	Type	Description
	AdiabaticScanResult

Notes

The settled state is carried forward between steps (the slow, adiabatic ramp). For networks with delayed coupling the delay history buffer is not carried, so this is only exact for instantaneous (non-delayed) coupling.