hybrid

Hybrid-system simulation and identification.

ContinuousDynamics(flow, label=None) dataclass

Reusable continuous dynamics definition.

Parameters:

Name	Type	Description	Default
flow	Callable[..., Derivative]	Dynamics function returning the state derivative. Scalar derivative returns are accepted only for single-state systems, both during solver evaluation and when derivatives are captured on the returned trace grid.	required
label	str | None	Optional display label.	None

CrossingDirection

Bases: IntEnum

Direction in which an event surface must cross zero.

Event(time, source_location, target_location, event_surface, reset, state) dataclass

Transition event information for a trace.

EventSurface(fn, direction=CrossingDirection.EITHER, label=None) dataclass

Scalar event surface defining a simulated transition event.

Flowcean transitions fire when fn reaches zero in direction. This is event-surface semantics, not Boolean guard-region semantics.

Parameters:

Name	Type	Description	Default
fn	Callable[..., float]	Root function; transitions when it crosses zero.	required
direction	CrossingDirection	Crossing direction. Defaults to either direction.	EITHER
label	str | None	Optional display label.	None

EventSurfaceFunction

Bases: Protocol

Scalar event-surface callback.

FlowFunction

Bases: Protocol

Continuous dynamics callback.

HybridSystem(locations, transitions, initial_location, initial_state, parameters=dict()) dataclass

Hybrid system with locations and transitions.

Parameters:

Name	Type	Description	Default
locations	Sequence[Location]	Location objects in this system.	required
transitions	Sequence[Transition]	Transition list defining event surfaces and resets.	required
initial_location	Location	Starting location object.	required
initial_state	State	Initial state vector.	required
parameters	Parameters	Global parameter map passed to callbacks.	dict()

transitions_from(location)

Return transitions leaving the given location.

Source code in src/flowcean/hybrid/hybrid_system.py

def transitions_from(self, location: Location) -> list[Transition]:
    """Return transitions leaving the given location."""
    return [
        transition
        for transition in self.transitions
        if transition.source is location
    ]

Location(dynamics, *, label=None, parameters=None) dataclass

Location(
    dynamics: ContinuousDynamics,
    *,
    label: str | None = None,
    parameters: Parameters | None = None,
)

Location(
    dynamics: Callable[..., Derivative],
    *,
    label: str | None = None,
    parameters: Parameters | None = None,
)

Discrete hybrid-automaton location.

Parameters:

Name	Type	Description	Default
dynamics	ContinuousDynamics | Callable[..., Derivative]	Continuous dynamics or bare flow callback active here.	required
label	str | None	Optional display label.	None
parameters	Parameters | None	Location-local parameter map.	None

Source code in src/flowcean/hybrid/hybrid_system.py

def __init__(
    self,
    dynamics: ContinuousDynamics | Callable[..., Derivative],
    *,
    label: str | None = None,
    parameters: Parameters | None = None,
) -> None:
    if isinstance(dynamics, ContinuousDynamics):
        continuous_dynamics = dynamics
    elif callable(dynamics):
        continuous_dynamics = ContinuousDynamics(dynamics)
    else:
        message = (
            "Location requires ContinuousDynamics or a flow callback."
        )
        raise TypeError(message)
    if parameters is not None and not isinstance(parameters, Mapping):
        message = "parameters must be a mapping."
        raise TypeError(message)
    object.__setattr__(self, "dynamics", continuous_dynamics)
    object.__setattr__(self, "label", label)
    object.__setattr__(self, "parameters", dict(parameters or {}))

Reset(fn, label=None) dataclass

State reset applied on a transition.

Parameters:

Name	Type	Description	Default
fn	Callable[..., State]	Reset function applied at the event time.	required
label	str | None	Optional display label.	None

ResetFunction

Bases: Protocol

State reset callback.

Trace(t, x, location, events, u=None, dx=None) dataclass

Simulation trace with time, state, and location labels.

as_dict()

Return a dictionary view of the trace.

Source code in src/flowcean/hybrid/hybrid_system.py

def as_dict(self) -> dict[str, object]:
    """Return a dictionary view of the trace."""
    return {
        "t": self.t,
        "x": self.x,
        "location": self.location,
        "events": self.events,
        "u": self.u,
        "dx": self.dx,
    }

Transition(source, target, event, reset=None) dataclass

Discrete event-triggered transition between locations.

event is a scalar zero-crossing surface.

Parameters:

Name	Type	Description	Default
source	Location	Source location.	required
target	Location	Target location.	required
event	EventSurface | Callable[..., float]	Event surface that triggers the transition.	required
reset	Reset | Callable[..., State] | None	Optional reset applied upon transition.	None

Source code in src/flowcean/hybrid/hybrid_system.py

def __init__(
    self,
    source: Location,
    target: Location,
    event: EventSurface | Callable[..., float],
    reset: Reset | Callable[..., State] | None = None,
) -> None:
    if not isinstance(source, Location):
        message = "source must be a Location."
        raise TypeError(message)
    if not isinstance(target, Location):
        message = "target must be a Location."
        raise TypeError(message)
    if isinstance(event, EventSurface):
        event_surface = event
    elif callable(event):
        event_surface = EventSurface(event)
    else:
        message = "event must be an EventSurface or callable."
        raise TypeError(message)
    if isinstance(reset, Reset) or reset is None:
        transition_reset = reset
    elif callable(reset):
        transition_reset = Reset(reset)
    else:
        message = "reset must be a Reset, callable, or None."
        raise TypeError(message)
    object.__setattr__(self, "source", source)
    object.__setattr__(self, "target", target)
    object.__setattr__(self, "event", event_surface)
    object.__setattr__(self, "reset", transition_reset)

PlotCallback(trace, dims=None, *, trace_index=0, ax=None, show=True, pause=0.001)

Bases: HyDRACallback

Live matplotlib visualization for HyDRA trace analysis progress.

Source code in src/flowcean/hybrid/hydra/callbacks.py

def __init__(
    self,
    trace: Trace,
    dims: Sequence[int] | None = None,
    *,
    trace_index: int = 0,
    ax: Axes | None = None,
    show: bool = True,
    pause: float = 0.001,
) -> None:
    self.trace = trace
    self.dims = list(dims) if dims is not None else [0]
    self.trace_index = trace_index
    self.ax = ax
    self.show = show
    self.pause = pause
    self._finalized_segments: list[tuple[TraceSegment, int]] = []
    self._grouping_segments: list[tuple[TraceSegment, int]] = []
    self._active_segment: TraceSegment | None = None
    self._overlay_artists: list[Artist] = []
    self._base_plotted = False
    self._mode_colors: dict[int, str] = {}

    self._validate_trace()

HyDRALearner(regressor_factory, threshold, start_width=10, step_width=5, selector_learner=None, callback=None, trace_schema=None)

Bases: SupervisedLearner

Identify hybrid-system modes from trace inputs and derivatives.

regressor_factory must create fresh incremental supervised learners. The current learner supports single-output derivative training. When a selector learner is provided, HyDRA labels accurate trace segments and trains a selector to route future rows to learned modes.

Source code in src/flowcean/hybrid/hydra/learner.py

def __init__(
    self,
    regressor_factory: Callable[[], SupervisedIncrementalLearner],
    threshold: float,
    start_width: int = 10,
    step_width: int = 5,
    selector_learner: HybridDecisionTreeLearner | None = None,
    callback: HyDRACallback | None = None,
    trace_schema: HyDRATraceSchema | None = None,
) -> None:
    super().__init__()
    if threshold < 0:
        message = "threshold must be non-negative."
        raise ValueError(message)
    if start_width <= 0:
        message = "start_width must be positive."
        raise ValueError(message)
    if step_width <= 0:
        message = "step_width must be positive."
        raise ValueError(message)
    self.regressor_factory = regressor_factory
    self.threshold = threshold
    self.start_width = start_width
    self.step_width = step_width
    self.selector_learner = selector_learner
    self.callback = callback or NoOpCallback()
    self.trace_schema = trace_schema

HyDRAModel(modes, *, input_features, output_features, selector=None, trace_schema=None)

Bases: Model

Model composed of learned continuous modes and an optional selector.

A single-mode model predicts directly with that mode. A multi-mode model needs a selector for batch prediction. Model persistence uses Flowcean's trusted-only pickle-based model serialization.

Source code in src/flowcean/hybrid/hydra/model.py

def __init__(
    self,
    modes: list[Model],
    *,
    input_features: list[str],
    output_features: list[str],
    selector: HybridDecisionTreeModel | None = None,
    trace_schema: HyDRATraceSchema | None = None,
) -> None:
    super().__init__()
    self.modes = modes
    self.input_features = input_features
    self.output_features = output_features
    self.selector = selector
    self.trace_schema = trace_schema
    self._validate_trace_schema()

HyDRATraceSchema(time, state, derivative, inputs=()) dataclass

Column schema for HyDRA trace-based learning.

Parameters:

Name	Type	Description	Default
time	str	Time column name.	required
state	tuple[str, ...]	State column names used as model inputs.	required
derivative	tuple[str, ...]	Derivative column names used as model outputs.	required
inputs	tuple[str, ...]	Optional external input column names.	()

HyDRA currently expects derivative columns to align with state columns and supports single-output training in HyDRALearner.

SelectorFeatureConfig(state_features=(), input_features=(), derivative_features=(), state_history=0, input_history=0, derivative_history=0, mode_history=0) dataclass

Configuration for selector feature construction.

Parameters:

Name	Type	Description	Default
state_features	tuple[str, ...]	State columns available to the selector.	()
input_features	tuple[str, ...]	Raw input columns available to the selector.	()
derivative_features	tuple[str, ...]	Derivative columns available to the selector.	()
state_history	int	Number of previous state rows used for selector features.	0
input_history	int	Number of previous input rows used for selector features.	0
derivative_history	int	Number of previous derivative rows used for selector features.	0
mode_history	int	Number of previous mode labels used for selector features.	0

save_traces_csv(traces, path, *, trace_metadata=None)

Write traces to a directory with one CSV file per trace.

Source code in src/flowcean/hybrid/io.py

def save_traces_csv(
    traces: Sequence[Trace],
    path: str,
    *,
    trace_metadata: Sequence[Mapping[str, object] | None] | None = None,
) -> None:
    """Write traces to a directory with one CSV file per trace."""
    output_dir = Path(path)
    output_dir.mkdir(parents=True, exist_ok=True)
    metadata_per_trace = _prepare_trace_metadata(trace_metadata, len(traces))
    trace_frames = traces_to_polars(traces)
    for trace_id, (trace_df, metadata) in enumerate(
        zip(trace_frames, metadata_per_trace, strict=True),
    ):
        trace_df.write_csv(output_dir / f"trace_{trace_id}.csv")
        _write_metadata_file(output_dir, trace_id, metadata)

save_traces_parquet(traces, path, *, trace_metadata=None)

Write traces to a directory with one Parquet file per trace.

Source code in src/flowcean/hybrid/io.py

def save_traces_parquet(
    traces: Sequence[Trace],
    path: str,
    *,
    trace_metadata: Sequence[Mapping[str, object] | None] | None = None,
) -> None:
    """Write traces to a directory with one Parquet file per trace."""
    output_dir = Path(path)
    output_dir.mkdir(parents=True, exist_ok=True)
    metadata_per_trace = _prepare_trace_metadata(trace_metadata, len(traces))
    trace_frames = traces_to_polars(traces)
    for trace_id, (trace_df, metadata) in enumerate(
        zip(trace_frames, metadata_per_trace, strict=True),
    ):
        trace_df.write_parquet(output_dir / f"trace_{trace_id}.parquet")
        _write_metadata_file(output_dir, trace_id, metadata)

trace_to_polars(trace, *, state_names=None, derivative_names=None, input_names=None)

Convert a single trace into a Polars DataFrame.

Source code in src/flowcean/hybrid/io.py

def trace_to_polars(
    trace: Trace,
    *,
    state_names: Sequence[str] | None = None,
    derivative_names: Sequence[str] | None = None,
    input_names: Sequence[str] | None = None,
) -> pl.DataFrame:
    """Convert a single trace into a Polars DataFrame."""
    trace_frames = traces_to_polars(
        [trace],
        state_names=state_names,
        derivative_names=derivative_names,
        input_names=input_names,
    )
    return trace_frames[0]

traces_to_polars(traces, *, state_names=None, derivative_names=None, input_names=None)

Convert traces into per-trace Polars DataFrames.

Parameters:

Name	Type	Description	Default
traces	Sequence[Trace]	Sequence of traces to convert.	required
state_names	Sequence[str] | None	Optional names for state dimensions.	None
derivative_names	Sequence[str] | None	Optional names for derivative dimensions.	None
input_names	Sequence[str] | None	Optional names for input dimensions.	None

Returns:

Type	Description
list[DataFrame]	List of per-trace DataFrames in trace order.

Source code in src/flowcean/hybrid/io.py

def traces_to_polars(
    traces: Sequence[Trace],
    *,
    state_names: Sequence[str] | None = None,
    derivative_names: Sequence[str] | None = None,
    input_names: Sequence[str] | None = None,
) -> list[pl.DataFrame]:
    """Convert traces into per-trace Polars DataFrames.

    Args:
        traces: Sequence of traces to convert.
        state_names: Optional names for state dimensions.
        derivative_names: Optional names for derivative dimensions.
        input_names: Optional names for input dimensions.

    Returns:
        List of per-trace DataFrames in trace order.
    """
    trace_frames: list[pl.DataFrame] = []
    for trace in traces:
        inputs = _validated_inputs(trace)
        derivatives = _validated_derivatives(trace)
        state_columns = _column_names(
            "x",
            trace.x.shape[1],
            state_names,
            arg_name="state_names",
        )
        derivative_columns: list[str] = []
        if derivatives is not None:
            derivative_columns = _column_names(
                "dx",
                derivatives.shape[1],
                derivative_names,
                arg_name="derivative_names",
            )
            if derivative_names is None and state_names is not None:
                derivative_columns = [f"dx_{name}" for name in state_columns]
        input_dimension = 0 if inputs is None else inputs.shape[1]
        input_columns = _column_names(
            "u",
            input_dimension,
            input_names,
            arg_name="input_names",
        )
        rows: list[dict[str, object]] = []
        for idx, (time, state, location) in enumerate(
            zip(trace.t, trace.x, trace.location, strict=False),
        ):
            row: dict[str, object] = {
                "step": idx,
                "t": float(time),
                "location": str(location),
            }
            for dim, column in enumerate(state_columns):
                row[column] = float(state[dim])
            for dim, column in enumerate(derivative_columns):
                if derivatives is None:
                    message = "Trace does not contain captured derivatives."
                    raise ValueError(message)
                row[column] = float(derivatives[idx, dim])
            for dim, column in enumerate(input_columns):
                if inputs is None:
                    message = "Trace does not contain captured inputs."
                    raise ValueError(message)
                row[column] = float(inputs[idx, dim])
            rows.append(row)
        trace_frames.append(pl.DataFrame(rows))

    return trace_frames

plot_phase(trace, x_dim=0, y_dim=1, *, location_colors=None, show_location_legend=True, show=False, ax=None)

Plot phase portrait segments colored by location.

Parameters:

Name	Type	Description	Default
trace	Trace	Trace to visualize.	required
x_dim	int	State index on the x-axis.	0
y_dim	int	State index on the y-axis.	1
location_colors	Mapping[str, str] | None	Optional color mapping for locations.	None
show_location_legend	bool	Whether to show location labels in legend.	True
show	bool	Whether to call matplotlib show().	False
ax	Axes | None	Optional axis to draw into.	None

Returns:

Type	Description
Axes	Matplotlib axes containing the plot.

Source code in src/flowcean/hybrid/plotting.py

def plot_phase(
    trace: Trace,
    x_dim: int = 0,
    y_dim: int = 1,
    *,
    location_colors: Mapping[str, str] | None = None,
    show_location_legend: bool = True,
    show: bool = False,
    ax: Axes | None = None,
) -> Axes:
    """Plot phase portrait segments colored by location.

    Args:
        trace: Trace to visualize.
        x_dim: State index on the x-axis.
        y_dim: State index on the y-axis.
        location_colors: Optional color mapping for locations.
        show_location_legend: Whether to show location labels in legend.
        show: Whether to call matplotlib show().
        ax: Optional axis to draw into.

    Returns:
        Matplotlib axes containing the plot.
    """
    if ax is None:
        _, ax = plt.subplots()
    if ax is None:
        message = "Failed to create matplotlib axes."
        raise RuntimeError(message)

    segments, locations = _location_segments(trace)
    colors = _location_color_map(locations, location_colors)

    for start, end, location in segments:
        ax.plot(
            trace.x[start:end, x_dim],
            trace.x[start:end, y_dim],
            color=colors[location],
            label=f"{location}",
        )
    ax.set_xlabel(f"x{x_dim}")
    ax.set_ylabel(f"x{y_dim}")

    if show_location_legend:
        _dedupe_legend(ax)

    if show:
        plt.show()

    return ax

plot_trace(trace, dims=None, *, location_colors=None, show_locations=True, show_location_labels=False, show_events=True, show_event_labels=True, show=False, ax=None)

Plot state trajectories with location shading and event markers.

Parameters:

Name	Type	Description	Default
trace	Trace	Trace to visualize.	required
dims	Sequence[int] | None	State indices to plot.	None
location_colors	Mapping[str, str] | None	Optional color mapping for locations.	None
show_locations	bool	Whether to shade location regions.	True
show_location_labels	bool	Whether to label locations above the trace.	False
show_events	bool	Whether to show transition events.	True
show_event_labels	bool	Whether to label transition events.	True
show	bool	Whether to call matplotlib show().	False
ax	Axes | None	Optional axis to draw into.	None

Returns:

Type	Description
Axes	Matplotlib axes containing the plot.

Source code in src/flowcean/hybrid/plotting.py

def plot_trace(
    trace: Trace,
    dims: Sequence[int] | None = None,
    *,
    location_colors: Mapping[str, str] | None = None,
    show_locations: bool = True,
    show_location_labels: bool = False,
    show_events: bool = True,
    show_event_labels: bool = True,
    show: bool = False,
    ax: Axes | None = None,
) -> Axes:
    """Plot state trajectories with location shading and event markers.

    Args:
        trace: Trace to visualize.
        dims: State indices to plot.
        location_colors: Optional color mapping for locations.
        show_locations: Whether to shade location regions.
        show_location_labels: Whether to label locations above the trace.
        show_events: Whether to show transition events.
        show_event_labels: Whether to label transition events.
        show: Whether to call matplotlib show().
        ax: Optional axis to draw into.

    Returns:
        Matplotlib axes containing the plot.
    """
    if dims is None:
        dims = list(range(trace.x.shape[1]))

    if ax is None:
        _, ax = plt.subplots()
    if ax is None:
        message = "Failed to create matplotlib axes."
        raise RuntimeError(message)

    for dim in dims:
        ax.plot(trace.t, trace.x[:, dim], label=f"x{dim}")

    if show_locations:
        _plot_location_spans(
            trace,
            ax=ax,
            location_colors=location_colors,
            show_labels=show_location_labels,
        )

    if show_events:
        _plot_events(trace, ax=ax, show_labels=show_event_labels)

    ax.set_xlabel("t")
    ax.set_ylabel("state")
    ax.legend(loc="best")

    if show:
        plt.show()

    return ax

generate_traces(system, t_span, initial_states, *, input_stream=None, capture_inputs=None, capture_derivatives=False, max_jumps=256, rtol=1e-07, atol=1e-09, max_step=None, dense_output=False, sample_times=None, sample_dt=None)

Simulate a batch of traces for a set of initial states.

The input stream and capture semantics match :func:simulate, including the requirement that capture_derivatives=True assumes pure flow callbacks under repeated evaluation on the returned trace grid. Scalar derivative returns are accepted only for single-state systems.

Source code in src/flowcean/hybrid/simulator.py

def generate_traces(
    system: HybridSystem,
    t_span: tuple[float, float],
    initial_states: Iterable[Iterable[float]],
    *,
    input_stream: InputStream | None = None,
    capture_inputs: bool | None = None,
    capture_derivatives: bool = False,
    max_jumps: int = 256,
    rtol: float = 1e-7,
    atol: float = 1e-9,
    max_step: float | None = None,
    dense_output: bool = False,
    sample_times: Iterable[float] | None = None,
    sample_dt: float | None = None,
) -> list[Trace]:
    """Simulate a batch of traces for a set of initial states.

    The input stream and capture semantics match :func:`simulate`, including
    the requirement that ``capture_derivatives=True`` assumes pure flow
    callbacks under repeated evaluation on the returned trace grid. Scalar
    derivative returns are accepted only for single-state systems.
    """
    return [
        simulate(
            system,
            t_span,
            x0=state,
            input_stream=input_stream,
            capture_inputs=capture_inputs,
            capture_derivatives=capture_derivatives,
            max_jumps=max_jumps,
            rtol=rtol,
            atol=atol,
            max_step=max_step,
            dense_output=dense_output,
            sample_times=sample_times,
            sample_dt=sample_dt,
        )
        for state in initial_states
    ]

simulate(system, t_span, x0=None, location0=None, *, input_stream=None, capture_inputs=None, capture_derivatives=False, max_jumps=256, rtol=1e-07, atol=1e-09, max_step=None, dense_output=False, sample_times=None, sample_dt=None)

Simulate a hybrid system and return a trace.

Parameters:

Name	Type	Description	Default
system	HybridSystem	Hybrid system to simulate.	required
t_span	tuple[float, float]	Start and end time for integration.	required
x0	Iterable[float] | None	Optional initial state override.	None
location0	Location | None	Optional initial location override.	None
input_stream	InputStream | None	Optional input stream accessor for callbacks.	None
capture_inputs	bool | None	Input capture mode. If None, capture iff an input stream is provided.	None
capture_derivatives	bool	Whether to re-evaluate Location.dynamics.flow on the returned trace grid and store the sampled derivatives in Trace.dx. This assumes pure flow callbacks under repeated evaluation. Scalar derivative returns are accepted only for single-state systems.	False
max_jumps	int	Maximum number of transitions allowed.	256
rtol	float	Relative tolerance for the solver.	1e-07
atol	float	Absolute tolerance for the solver.	1e-09
max_step	float | None	Optional maximum step size.	None
dense_output	bool	Whether to build a continuous solution per segment.	False
sample_times	Iterable[float] | None	Monotone time grid to sample from the dense solution.	None
sample_dt	float | None	Fixed sampling interval to generate a time grid.	None

Returns:

Name	Type	Description
Trace	Trace	The simulation trace with location labels and events.

Source code in src/flowcean/hybrid/simulator.py

def simulate(  # noqa: C901, PLR0912, PLR0915
    system: HybridSystem,
    t_span: tuple[float, float],
    x0: Iterable[float] | None = None,
    location0: Location | None = None,
    *,
    input_stream: InputStream | None = None,
    capture_inputs: bool | None = None,
    capture_derivatives: bool = False,
    max_jumps: int = 256,
    rtol: float = 1e-7,
    atol: float = 1e-9,
    max_step: float | None = None,
    dense_output: bool = False,
    sample_times: Iterable[float] | None = None,
    sample_dt: float | None = None,
) -> Trace:
    """Simulate a hybrid system and return a trace.

    Args:
        system: Hybrid system to simulate.
        t_span: Start and end time for integration.
        x0: Optional initial state override.
        location0: Optional initial location override.
        input_stream: Optional input stream accessor for callbacks.
        capture_inputs: Input capture mode. If ``None``, capture iff an
            input stream is provided.
        capture_derivatives: Whether to re-evaluate ``Location.dynamics.flow``
            on the returned trace grid and store the sampled derivatives in
            ``Trace.dx``. This assumes pure flow callbacks under repeated
            evaluation. Scalar derivative returns are accepted only for
            single-state systems.
        max_jumps: Maximum number of transitions allowed.
        rtol: Relative tolerance for the solver.
        atol: Absolute tolerance for the solver.
        max_step: Optional maximum step size.
        dense_output: Whether to build a continuous solution per segment.
        sample_times: Monotone time grid to sample from the dense solution.
        sample_dt: Fixed sampling interval to generate a time grid.

    Returns:
        Trace: The simulation trace with location labels and events.
    """
    start_location = (
        system.initial_location if location0 is None else location0
    )
    if not isinstance(start_location, Location):
        message = "location0 must be a Location."
        raise TypeError(message)
    location_ids = {id(location) for location in system.locations}
    if id(start_location) not in location_ids:
        message = "location0 must be included in system.locations."
        raise ValueError(message)

    state = ensure_state(x0 if x0 is not None else system.initial_state)
    location = start_location
    effective_input_stream = input_stream or _missing_input_stream
    should_capture = _resolve_capture_inputs(
        capture_inputs=capture_inputs,
        input_stream=input_stream,
    )

    t_segments: list[np.ndarray] = []
    x_segments: list[np.ndarray] = []
    location_segments: list[np.ndarray] = []
    sol_segments: list[Callable[[np.ndarray], np.ndarray] | None] = []
    events: list[Event] = []

    t_current = float(t_span[0])
    t_final = float(t_span[1])
    jumps = 0

    time_epsilon = 1e-12
    sample_grid = _prepare_sample_times(t_span, sample_times, sample_dt)
    needs_dense = dense_output or sample_grid is not None

    while t_current < t_final:
        transitions = system.transitions_from(location)
        event_fns = _build_event_functions(
            transitions,
            system.parameters,
            location.parameters,
            effective_input_stream,
        )

        solve_kwargs = {
            "fun": _wrap_flow(
                location,
                system.parameters,
                effective_input_stream,
            ),
            "t_span": (t_current, t_final),
            "y0": state,
            "events": event_fns or None,
            "rtol": rtol,
            "atol": atol,
            "dense_output": needs_dense,
        }
        if max_step is not None:
            solve_kwargs["max_step"] = max_step

        result = solve_ivp(**solve_kwargs)
        if not result.success:
            message = f"ODE integration failed: {result.message}"
            raise RuntimeError(message)

        t_segments.append(result.t)
        x_segments.append(result.y.T)
        location_segments.append(
            np.full(result.t.shape, location, dtype=object),
        )
        sol_segments.append(result.sol)

        if not result.t_events or all(
            len(event_list) == 0 for event_list in result.t_events
        ):
            break

        triggered_index, event_time, event_state = _first_event(
            result.t_events,
            result.y_events,
        )
        is_zero_time_event = event_time - t_current <= time_epsilon
        transition = transitions[triggered_index]
        jumps += 1
        if jumps > max_jumps:
            message = "Maximum number of transitions exceeded."
            raise RuntimeError(message)

        state, event = _apply_transition(
            transition,
            event_time,
            event_state,
            system.parameters,
            effective_input_stream,
        )
        events.append(event)

        location = transition.target
        while True:
            transitions = system.transitions_from(location)
            immediate_index = _immediate_transition_index(
                transitions,
                location,
                event_time,
                state,
                system.parameters,
                effective_input_stream,
                time_epsilon,
            )
            if immediate_index is None:
                break

            transition = transitions[immediate_index]
            jumps += 1
            if jumps > max_jumps:
                message = "Maximum number of transitions exceeded."
                raise RuntimeError(message)

            state, event = _apply_transition(
                transition,
                event_time,
                state,
                system.parameters,
                effective_input_stream,
            )
            events.append(event)

            location = transition.target

        if is_zero_time_event:
            t_current = min(t_final, t_current + time_epsilon)
        else:
            t_current = min(t_final, event_time + time_epsilon)

    if sample_grid is None:
        t_all = _concat_segments(t_segments)
        x_all = _concat_segments(x_segments)
        location_objects = _concat_segments(location_segments)
        location_all = _location_labels(location_objects)
        u_all = None
        dx_all = None
        if should_capture:
            if input_stream is None:
                message = "Internal error: expected input_stream for capture."
                raise RuntimeError(message)
            u_all = _capture_inputs(t_all, input_stream)
        if capture_derivatives:
            dx_all = _capture_derivatives(
                system=system,
                times=t_all,
                states=x_all,
                locations=location_objects,
                input_stream=effective_input_stream,
            )
        return Trace(
            t=t_all,
            x=x_all,
            location=location_all,
            events=tuple(events),
            u=u_all,
            dx=dx_all,
        )

    rolled = _rollout_segments(
        sample_grid,
        t_segments,
        x_segments,
        location_segments,
        sol_segments,
    )
    u_all = None
    dx_all = None
    if should_capture:
        if input_stream is None:
            message = "Internal error: expected input_stream for capture."
            raise RuntimeError(message)
        u_all = _capture_inputs(rolled.t, input_stream)
    if capture_derivatives:
        dx_all = _capture_derivatives(
            system=system,
            times=rolled.eval_t,
            states=rolled.x,
            locations=rolled.location,
            input_stream=effective_input_stream,
        )
    location_all = _location_labels(rolled.location)
    return Trace(
        t=rolled.t,
        x=rolled.x,
        location=location_all,
        events=tuple(events),
        u=u_all,
        dx=dx_all,
    )