Make LAM/crack degradation submodels well-posed under 1+1D current collectors (#5296)

[gaoflow]

Description

Fixes #5296.

Discretising any lithium_ion model with "dimensionality": 1 plus the degradation stack ("loss of active material": "stress-driven", "particle mechanics": "swelling and cracking", …) fails with

ModelError: rhs and initial conditions must have the same shape after
discretisation but rhs.shape = (10, 1) and initial_conditions.shape = (1, 1)
for variable 'Loss of lithium due to loss of active material in negative electrode [mol]'.

Two independent shape gaps are responsible, both activated only when the current-collector mesh has more than one node.

Root cause

(a) submodels/active_material/loss_active_material.py:192 — the LAM loss-of-lithium accumulator is declared as a scalar (pybamm.Variable(...) with no domain, IC pybamm.Scalar(0)). Its rhs was

-V * pybamm.x_average(c_s_rav * deps_solid_dt)

x_average averages through the cell thickness but keeps the current-collector axis. For "dimensionality": 1 the rhs is (z_n, 1) while the IC is (1, 1).

(b) submodels/particle_mechanics/crack_propagation.py:175 — for DFN (x_average is False) the crack-length state is declared

pybamm.Variable(
    "<Domain> particle crack length [m]",
    domain="<dom> electrode",
    auxiliary_domains={"secondary": "current collector"},
)

but its IC was pybamm.PrimaryBroadcast(l_cr_0, "<dom> electrode"), which only broadcasts across the electrode mesh. Under "dimensionality": 1 the rhs is (x_n, z_n) while the IC is (x_n, 1). PyBaMM's discretisation check raises:

ModelError: ... rhs.shape = (40, 1) and initial_conditions.shape = (5, 1)
for variable 'Negative particle crack length [m]'.

The SPMe reproducer in the issue hits (a) first; DFN with the same options additionally hits (b).

Fix

Two surgical changes.

• In loss_active_material.py:192 wrap the existing x_average in pybamm.yz_average so the rhs collapses to a scalar in 1+1D and 2+1D current collectors. yz_average is the identity on 0-D current collectors, so 0-D models are unaffected. Total lithium loss is preserved because V = L * A_cc already carries L_y * L_z (see interface/lithium_plating/base_plating.py:87 and interface/sei/base_sei.py:132 for the same pattern on plating and SEI accumulators).
• In crack_propagation.py:175 replace PrimaryBroadcast(l_cr_0, "<dom> electrode") with FullBroadcast(l_cr_0, "<dom> electrode", "current collector") so the IC spans both the electrode and the current-collector mesh. FullBroadcast is the identity on 0-D current collectors.

Reproducer

Exact options from the issue (only the Simulation.solve t_eval was shortened to [0, 600] to keep CI fast):

import pybamm
param = pybamm.ParameterValues("OKane2022")
L_y = param["Electrode width [m]"]
L_z = param["Electrode height [m]"]
options = {
    "current collector": "potential pair",
    "dimensionality": 1,
    "thermal": "x-lumped",
    "SEI": "solvent-diffusion limited",
    "SEI porosity change": "true",
    "lithium plating": "partially reversible",
    "lithium plating porosity change": "true",
    "particle mechanics": ("swelling and cracking", "swelling only"),
    "SEI on cracks": "true",
    "loss of active material": "stress-driven",
}
model = pybamm.lithium_ion.SPMe(options=options)
# … thermal/tab parameter overrides per the issue body …
sim = pybamm.Simulation(model, parameter_values=param,
                        var_pts={"x_n": 5, "x_s": 5, "x_p": 5, "r_n": 30, "r_p": 30, "z": 10},
                        solver=pybamm.IDAKLUSolver())
sim.solve([0, 600])  # ModelError on main, OK on this branch

Verification

• The exact issue reproducer (SPMe + dimensionality=1 + full degradation stack) now solves end-to-end. Loss of lithium due to loss of active material in negative electrode [mol] evolves to ~3.0e-7 mol after 600 s.
• DFN and SPM with the same options also now solve (their LLI_LAM_n[-1] matches SPMe within solver tolerance: ~3.0e-7 mol).
• 0-D regression (SPM/SPMe/DFN with the same degradation options but no high-dim current collector) produces totals indistinguishable from main (~3.0e-7 mol after 600 s).
• pytest tests/unit/test_models/test_full_battery_models/test_lithium_ion/ — 637 passed, 12 skipped (+4 new regression tests).
• ruff check + ruff format --check clean.

Regression test

New tests/unit/test_models/test_full_battery_models/test_lithium_ion/test_lam_lli_dimensionality_5296.py pins four guards:

1. SPMe + dimensionality=1 + full degradation stack discretises and solves (locks fix (a)).
2. Direct shape guard: after Discretisation every LAM-LLI variable's rhs must match its IC shape (catches future regressions in either direction).
3. DFN + dimensionality=1 + full degradation stack discretises and solves (locks fix (b)).
4. 0-D non-regression: stress-driven LAM with the standard current collector still produces a strictly positive but bounded total lithium loss.

Checklist

• No style issues (ruff check, ruff format --check)
• All tests passed
• CHANGELOG.md updated (Bug fixes / Unreleased)
• Updated docs if needed — n/a, no public-API change
• Issue reference in the PR description and the commit message