Heun

experimental.network_dynamics.solvers.Heun()

Heun’s method (improved Euler).

Two-stage method with predictor-corrector structure. For SDEs: Stochastic Heun method.

Note: Uses coupling assumption that coupling is slow relative to dt, so we can reuse the same coupling for both stages.

Methods

Name	Description
step	Heun integration step with predictor-corrector.

step

experimental.network_dynamics.solvers.Heun.step(
    dynamics_fn,
    t,
    state,
    dt,
    params,
    noise_sample=0.0,
)

Heun integration step with predictor-corrector.

Args: dynamics_fn: Dynamics function (t, state, params) -> (derivatives, auxiliaries) Same coupling used for both evaluations t: Current time state: Current state [n_states, n_nodes] dt: Time step params: Parameters noise_sample: Pre-scaled noise increment

Returns: Tuple of (next_state, auxiliaries): - next_state: Next state [n_states, n_nodes] - auxiliaries: Auxiliary variables from FIRST evaluation [n_auxiliaries, n_nodes] (represents valid state at current time t)

# Heun { #tvboptim.experimental.network_dynamics.solvers.Heun }

```python
experimental.network_dynamics.solvers.Heun()
```

Heun's method (improved Euler).

Two-stage method with predictor-corrector structure.
For SDEs: Stochastic Heun method.

Note: Uses coupling assumption that coupling is slow relative to dt,
so we can reuse the same coupling for both stages.

## Methods

| Name | Description |
| --- | --- |
| [step](#tvboptim.experimental.network_dynamics.solvers.Heun.step) | Heun integration step with predictor-corrector. |

### step { #tvboptim.experimental.network_dynamics.solvers.Heun.step }

```python
experimental.network_dynamics.solvers.Heun.step(
    dynamics_fn,
    t,
    state,
    dt,
    params,
    noise_sample=0.0,
)
```

Heun integration step with predictor-corrector.

Args:
    dynamics_fn: Dynamics function (t, state, params) -> (derivatives, auxiliaries)
                Same coupling used for both evaluations
    t: Current time
    state: Current state [n_states, n_nodes]
    dt: Time step
    params: Parameters
    noise_sample: Pre-scaled noise increment

Returns:
    Tuple of (next_state, auxiliaries):
        - next_state: Next state [n_states, n_nodes]
        - auxiliaries: Auxiliary variables from FIRST evaluation [n_auxiliaries, n_nodes]
                      (represents valid state at current time t)