[Bug]: Missing variable for submodel 'negative primary sei on cracks': 'Negative primary SEI on cracks thickness [m]'.

[satishrapol]

PyBaMM Version

25.10.1

Python Version

3.11.9

Describe the bug

Hello
I am using the sub model configuration below. Please note I am still interested in using the cracking submodel.

My code

model_options = {
"particle phases": ("2", "1"),
"open-circuit potential": (("single", "current sigmoid"), "single"),
"SEI": "ec reaction limited",
"particle mechanics": ("swelling and cracking","swelling only"),
"loss of active material": "stress-driven",
"SEI on cracks": "true",
"surface form": "algebraic",
}
And I am getting the following error.

Error:

---

KeyError Traceback (most recent call last)
pybamm\models\base_model.py:831, in BaseModel.build_coupled_variables(self)
830 model_var_copy = self.variables.copy()
--> 831 updated_variables = submodel.get_coupled_variables(
832 self.variables
833 )
834 self._variables_by_submodel[submodel_name].update(
835 {
836 key: updated_variables[key]
(...)
839 }
840 )
\pybamm\models\submodels\interface\sei\sei_growth.py:119, in SEIGrowth.get_coupled_variables(self, variables)
114 j = variables[
115 f"X-averaged {domain} electrode total "
116 "interfacial current density [A.m-2]"
117 ]
--> 119 L_sei = variables[f"{Domain} {self.reaction_name}thickness [m]"]
121 R_sei = phase_param.R_sei

KeyError: 'Negative primary SEI on cracks thickness [m]'

The above exception was the direct cause of the following exception:

ModelError Traceback (most recent call last)
Cell In[1], line 53
28 # nominal_value= [
29 # 9.708425e-14, # primary: SEI kinetic rate constant [m.s-1] X[0]
30 # 7.663143e-22, # primary: EC diffusivity [m2.s-1] X[1]
(...)
46 # 10e9 # secondary: negative electrode Young's modulus [Pa] X[17]
47 # ]
49 nominal_value = last_results.Value
---> 53 model = pybamm.lithium_ion.SPMe(
54 options=model_options
55 )
57 # Define the discretization points
58 var_pts = {
59 "x_n": 10, # negative electrode
60 "x_s": 3, # separator
(...)
67 'r_p_sec': 30,
68 }

pybamm\models\full_battery_models\lithium_ion\spme.py:32, in SPMe.init(self, options, name, build)
29 self.x_average = True
31 # Initialize with the SPM
---> 32 super().init(options, name, build)

pybamm\models\full_battery_models\lithium_ion\spm.py:45, in SPM.init(self, options, name, build)
41 options["x-average side reactions"] = "true"
43 super().init(options, name)
---> 45 self.set_submodels(build)
47 if self.class != "MPM":
48 pybamm.citations.register("Marquis2019")

pybamm\models\full_battery_models\lithium_ion\base_lithium_ion_model.py:63, in BaseModel.set_submodels(self, build)
60 self.set_total_interface_submodel()
62 if build:
---> 63 self.build_model()

pybamm\models\full_battery_models\base_battery_model.py:1176, in BaseBatteryModel.build_model(self)
1174 def build_model(self):
1175 # Build model variables and equations
-> 1176 self._build_model()
1178 # Set battery specific variables
1179 pybamm.logger.debug(f"Setting voltage variables ({self.name})")

pybamm\models\base_model.py:904, in BaseModel._build_model(self)
901 if self._built_fundamental is False:
902 self.build_fundamental()
--> 904 self.build_coupled_variables()
906 self.build_model_equations()

pybamm\models\base_model.py:846, in BaseModel.build_coupled_variables(self)
843 except KeyError as key:
844 if len(submodels) == 1 or count == 100:
845 # no more submodels to try
--> 846 raise pybamm.ModelError(
847 f"Missing variable for submodel '{submodel_name}': {key}.\n"
848 + "Check the selected "
849 "submodels provide all of the required variables."
850 ) from key
851 else:
852 # try setting coupled variables on next loop through
853 pybamm.logger.debug(
854 f"Can't find {key}, trying other submodels first"
855 )

ModelError: Missing variable for submodel 'negative primary sei on cracks': 'Negative primary SEI on cracks thickness [m]'.
Check the selected submodels provide all of the required variables.

Steps to Reproduce

Parameter set
param = pybamm.ParameterValues("Chen2020_composite")

model
var_pts = {
"x_n": 10, # negative electrode
"x_s": 3, # separator
"x_p": 10, # positive electrode
"r_n": 30, # negative particle
"r_p": 30, # positive particle
'r_n_prim': 30,
'r_p_prim': 30,
'r_n_sec': 30,
'r_p_sec': 30,
}
experiment_test = pybamm.Experiment(
[
("Rest for 20 min", "Charge at 1.7A until 4.2V", "Discharge at 1.7A until 2.65 V","Rest for 20 min")*2
]
)

solver = pybamm.IDAKLUSolver()

model = pybamm.lithium_ion.SPMe(
options={
"particle phases": ("2", "1"),
"open-circuit potential": (("single", "current sigmoid"), "single"),
"SEI": "ec reaction limited",
"particle mechanics": ("swelling and cracking","swelling only"),
"loss of active material": "stress-driven",
"SEI on cracks": "true",
"surface form": "algebraic",
}
)

sim = pybamm.Simulation(model, parameter_values=param, experiment=experiment_test, var_pts=var_pts, solver=solver)
sol = sim.solve(calc_esoh=True, showprogress=True,initial_soc=0.0)

Relevant log output

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
pybamm\models\base_model.py:831, in BaseModel.build_coupled_variables(self)
    830 model_var_copy = self.variables.copy()
--> 831 updated_variables = submodel.get_coupled_variables(
    832     self.variables
    833 )
    834 self._variables_by_submodel[submodel_name].update(
    835     {
    836         key: updated_variables[key]
   (...)
    839     }
    840 )

pybamm\models\submodels\interface\sei\sei_growth.py:119, in SEIGrowth.get_coupled_variables(self, variables)
    114     j = variables[
    115         f"X-averaged {domain} electrode total "
    116         "interfacial current density [A.m-2]"
    117     ]
--> 119 L_sei = variables[f"{Domain} {self.reaction_name}thickness [m]"]
    121 R_sei = phase_param.R_sei

KeyError: 'Negative primary SEI on cracks thickness [m]'

The above exception was the direct cause of the following exception:

ModelError                                Traceback (most recent call last)
Cell In[1], line 53
     28 # nominal_value= [
     29 #     9.708425e-14,  # primary: SEI kinetic rate constant [m.s-1] X[0]
     30 #     7.663143e-22,  # primary: EC diffusivity [m2.s-1] X[1]
   (...)
     46 #     10e9  # secondary: negative electrode Young's modulus [Pa] X[17]
     47 # ]
     49 nominal_value = last_results.Value
---> 53 model = pybamm.lithium_ion.SPMe(
     54     options=model_options
     55 )
     57 # Define the discretization points
     58 var_pts = {
     59     "x_n": 10,  # negative electrode
     60     "x_s": 3,  # separator
   (...)
     67     'r_p_sec': 30,
     68 }

pybamm\models\full_battery_models\lithium_ion\spme.py:32, in SPMe.__init__(self, options, name, build)
     29 self.x_average = True
     31 # Initialize with the SPM
---> 32 super().__init__(options, name, build)

pybamm\models\full_battery_models\lithium_ion\spm.py:45, in SPM.__init__(self, options, name, build)
     41     options["x-average side reactions"] = "true"
     43 super().__init__(options, name)
---> 45 self.set_submodels(build)
     47 if self.__class__ != "MPM":
     48     pybamm.citations.register("Marquis2019")

pybamm\models\full_battery_models\lithium_ion\base_lithium_ion_model.py:63, in BaseModel.set_submodels(self, build)
     60 self.set_total_interface_submodel()
     62 if build:
---> 63     self.build_model()

pybamm\models\full_battery_models\base_battery_model.py:1176, in BaseBatteryModel.build_model(self)
   1174 def build_model(self):
   1175     # Build model variables and equations
-> 1176     self._build_model()
   1178     # Set battery specific variables
   1179     pybamm.logger.debug(f"Setting voltage variables ({self.name})")

pybamm\models\base_model.py:904, in BaseModel._build_model(self)
    901 if self._built_fundamental is False:
    902     self.build_fundamental()
--> 904 self.build_coupled_variables()
    906 self.build_model_equations()

pybamm\models\base_model.py:846, in BaseModel.build_coupled_variables(self)
    843 except KeyError as key:
    844     if len(submodels) == 1 or count == 100:
    845         # no more submodels to try
--> 846         raise pybamm.ModelError(
    847             f"Missing variable for submodel '{submodel_name}': {key}.\n"
    848             + "Check the selected "
    849             "submodels provide all of the required variables."
    850         ) from key
    851     else:
    852         # try setting coupled variables on next loop through
    853         pybamm.logger.debug(
    854             f"Can't find {key}, trying other submodels first"
    855         )

ModelError: Missing variable for submodel 'negative primary sei on cracks': 'Negative primary SEI on cracks thickness [m]'.
Check the selected submodels provide all of the required variables.

[Stonebanks-js]

Hi @satishrapol 👋,

I'd like to work on this issue.

After investigating the failure, I confirmed that the KeyError originates from the SEI submodel expecting the variable:

Negative primary SEI on cracks thickness [m]

during coupling in sei_growth.py, but this variable is never created when using the combination:

SEI = "ec reaction limited"

"SEI on cracks" = True

multiphase particles: particle phases = ("2", "1")

This appears to be a naming/interface inconsistency in the EC-reaction-limited SEI path, since the solvent-diffusion SEI implementation already supports the cracks variant and generates the corresponding variables.

🔧 Proposed Fix

My planned solution would be implemented in two steps:

• Add a phase- and cracks-aware get_fundamental_variables() implementation in the SEI base class

• Ensures the expected thickness variable is instantiated whenever cracks are enabled.

• Resolves the current KeyError and aligns naming with other SEI submodels.

• Extend the EC-reaction-limited SEI kinetics to correctly update the new cracks-specific SEI thickness variable

• Prevents the variable from remaining static and ensures physically meaningful behaviour.

• A regression test will be added (SPMe + DFN) to validate variable creation and growth.

❓ Clarification Needed

Before proceeding with the PR, could you confirm which approach is preferred:

• A full implementation (support EC-reaction-limited SEI on cracks as a valid configuration),
  or
• A guarded approach (raise a clear OptionError if users combine unsupported options)?

_If this direction sounds good, please assign the issue to me.

Thanks! 🙌_

[satishrapol]

Hello @Stonebanks-js
A full implementation (support EC-reaction-limited SEI on cracks as a valid configuration) seems reasonable for me, as of now I plan to use SEI ec reaction model together with cracking; however, the cracking model working with any SEI model is also helpful.

@DrSOKane can you please confirm if the solution approach is the right direction?

Also it seems i do not have the permission to assign the issue to anyone

[Stonebanks-js]

@satishrapol Are you the maintainer or the participant ? Please ask maintainers to recheck the issue and assign me to it before I start working on it.

Also I need a approval for the approaches I mentioned.

Thank you

[DrSOKane]

Hi @satishrapol 👋,

I'd like to work on this issue.

After investigating the failure, I confirmed that the KeyError originates from the SEI submodel expecting the variable:

Negative primary SEI on cracks thickness [m]

during coupling in sei_growth.py, but this variable is never created when using the combination:

SEI = "ec reaction limited"

"SEI on cracks" = True

multiphase particles: particle phases = ("2", "1")

This appears to be a naming/interface inconsistency in the EC-reaction-limited SEI path, since the solvent-diffusion SEI implementation already supports the cracks variant and generates the corresponding variables.

🔧 Proposed Fix

My planned solution would be implemented in two steps:

• Add a phase- and cracks-aware get_fundamental_variables() implementation in the SEI base class

• Ensures the expected thickness variable is instantiated whenever cracks are enabled.

• Resolves the current KeyError and aligns naming with other SEI submodels.

• Extend the EC-reaction-limited SEI kinetics to correctly update the new cracks-specific SEI thickness variable

• Prevents the variable from remaining static and ensures physically meaningful behaviour.

• A regression test will be added (SPMe + DFN) to validate variable creation and growth.

❓ Clarification Needed

Before proceeding with the PR, could you confirm which approach is preferred:

• A full implementation (support EC-reaction-limited SEI on cracks as a valid configuration),
  or
• A guarded approach (raise a clear OptionError if users combine unsupported options)?

_If this direction sounds good, please assign the issue to me.

Thanks! 🙌_

@Stonebanks-js Thank you so much for working on this! I haven't been able to do it myself because no parameter set that supports this combination of options has been published, and I haven't had time to make one. Your plan looks really good and I look forward to reviewing any resulting pull request.

@satishrapol Is it possible to for you to move to a different SEI model, such as "interstitial-diffusion limited"? This fix may take a while, as the "ec reaction limited" option is sadly prone to bugs, of which this is a very typical example.

[Stonebanks-js]

@DrSOKane and @satishrapol

Hello maintainers,

I spent time investigating the import-related issue assigned to me and wanted to share a clear summary of what I found, along with a small, targeted proposal to prevent this class of problem in the future.

🔍 Root cause analysis

The observed error pattern (e.g. AttributeError: partially initialized module 'pybamm' has no attribute 'Symbol' / 'Array' / 'settings') is caused not by incorrect logic in the expression tree, but by importing PyBaMM directly from the source tree without an editable installation.

During investigation, I reproduced the issue by:

Running python -c "import pybamm" directly against src/pybamm

• Clearing caches to rule out stale bytecode
• Verifying that the errors consistently manifested as circular import symptoms

Once PyBaMM was installed correctly in editable mode:

pip install -e .

the issue disappeared completely, and import pybamm worked as expected, including version reporting.

This indicates that the internal import order in PyBaMM is correct, but that importing directly from source without installation leads to a partially initialised module state and misleading circular-import errors.

✅ Conclusion

**- This is primarily a developer experience / onboarding issue, not a functional bug in PyBaMM itself.

• A new contributor (or user debugging locally) can easily fall into this state without realising that editable installation is required, and the resulting error messages strongly resemble internal circular-import defects.**

🛠 Proposed solution ( For Future Contributors )

I would like to propose:

• A small guard/warning in pybamm/init.py that detects when PyBaMM is imported directly from source without being installed and emits a clear warning suggesting pip install -e ..
• A short documentation note (README / CONTRIBUTING) explicitly calling this out, so contributors don’t end up debugging misleading import errors.

This avoids behavioural changes while preventing significant confusion for future users and contributors.

📁 Small structural observation (non-blocking, for discussion only)

While navigating the expression tree during debugging, I noticed that some functionality related to expression representation/printing lives under expression_tree/printing/.
I wanted to ask (purely as a question, not a claim):
has there ever been consideration of treating these as part of the core expression layer rather than a separate subfolder?
Totally understand if there are historical or design reasons for the current layout — I am only mentioning this as an observation while exploring the import graph.

Please let me know if you’re happy with this direction — if so, I can open a small PR with:

• one guarded warning in init.py
• one documentation addition explaining the correct installation workflow

Best regards,
Aradhya @Stonebanks-js

[Stonebanks-js]

@DrSOKane @satishrapol Kindly approve so that I can proceed with PR