CasADi performance bottlenecks overview

[valentinsulzer]

Issue to add any minimum working examples related to improving the casadi solver

[valentinsulzer]

For very simple models, it takes much longer to evaluate the outputs than it does to solve the model

import pybamm
import numpy as np


class Decay(pybamm.BaseModel):
    def __init__(self):
        super().__init__()
        x = pybamm.Variable("x")
        self.rhs = {x: -x}
        self.initial_conditions = {x: 1}
        self.variables = {"x": x}


model = Decay()
solver = pybamm.CasadiSolver()
t_data = np.linspace(0, 10, 1000)
# profile this line:
solution = solver.solve(model, t_data)["x"]
# currently about 10% solver / 90% evaluate

[valentinsulzer]

SPMe with differential surface form and power mode fails to converge

import pybamm

pybamm.set_logging_level("INFO")
model = pybamm.lithium_ion.SPMe(
    {"surface form": "differential", "operating mode": "power"}
)
parameter_values = pybamm.ParameterValues("Chen2020")
parameter_values.update({"Power function [W]": 4}, check_already_exists=False)
sim = pybamm.Simulation(model, parameter_values=parameter_values)
sim.solve([0, 3600])

[valentinsulzer]

#2173 (it now fails instead of hanging indefinitely)

[valentinsulzer]

An example that takes a long time because it struggles near an event (sharp voltage drop-off)

import pybamm

pybamm.set_logging_level("INFO")

model = pybamm.lithium_ion.DFN()
parameter_values = pybamm.ParameterValues("Chen2020")
parameter_values["Positive electrode active material volume fraction"] = 0.1
sim = pybamm.Simulation(model, parameter_values=parameter_values)
sim.solve([0, 3600])
sim.plot(["Voltage [V]"])

[valentinsulzer]

Linking #1996 , we probably don't want to use vertcat at all, instead store the ys separately

[valentinsulzer]

Linking #1482 which only seems to happen with the casadi solver

[valentinsulzer]

#3943

[valentinsulzer]

#3960