AbstractSolver

experimental.network_dynamics.solvers.AbstractSolver()

Base class for all solver types.

Attributes

Name	Type	Description
stage_time_centroid	float	The method’s stage-time centroid `sum_i b_i * c_i`: where in the step the method’s drift contribution is effectively evaluated, as a fraction of `dt`. 0 for a single-stage method evaluated at the step start (Euler), 1/2 for any method satisfying the order-2 condition (Heun, RK4). Delayed couplings read it to undo the bias that freezing the coupling across stages introduces: holding the coupling at its step-start value is the same as evaluating it on time with every delay lengthened by `stage_time_centroid * dt`, so the read subtracts that back. A custom solver that leaves this at the default 0.0 gets the old (unshifted) read, which is correct for a step-start evaluation and merely first-order accurate otherwise.

# AbstractSolver { #tvboptim.experimental.network_dynamics.solvers.AbstractSolver } ```python experimental.network_dynamics.solvers.AbstractSolver() ``` Base class for all solver types. ## Attributes {.doc-section .doc-section-attributes} | Name | Type | Description | |---------------------|------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | stage_time_centroid | [float](`float`) | The method's stage-time centroid ``sum_i b_i * c_i``: where in the step the method's drift contribution is effectively evaluated, as a fraction of ``dt``. 0 for a single-stage method evaluated at the step start (Euler), 1/2 for any method satisfying the order-2 condition (Heun, RK4). Delayed couplings read it to undo the bias that freezing the coupling across stages introduces: holding the coupling at its step-start value is the same as evaluating it on time with every delay lengthened by ``stage_time_centroid * dt``, so the read subtracts that back. A custom solver that leaves this at the default 0.0 gets the old (unshifted) read, which is correct for a step-start evaluation and merely first-order accurate otherwise. |