Improve handling of transport_model="None"

[ischoegl]

Problem description

In the current implementation, specifying a transport model of "None" (C++) or None (Python) prevents the instantiation of a transport manager object, with various internal pointers being set to NULL. At least in Python, there is a band-aid in place to make handling safe (using a newDefaultTransportMgr), but this sometimes fails - see example below. Also, it is not ideal to display transport_model = ‘Transport’, where the meaning is not clear to a user.

A safer implementation would use an empty Transport object (similar to a Kinetics object with zero reactions added). A Solution object with no Transport created in C++ is unsafe as it contains NULL as the Transport reference.

Steps to reproduce

The following should create equivalent objects (using mock_ion.xml, which does not include reactions):

In [5]: _ = ct.ThermoPhase("mock_ion.xml")

In [6]: _ = ct.Solution("mock_ion.xml", transport_model=None)
---------------------------------------------------------------------------
CanteraError                              Traceback (most recent call last)
<ipython-input-6-5d233d73507d> in <module>()
----> 1 _ = ct.Solution("mock_ion.xml", transport_model="None")

/src/interfaces/cython/cantera/thermo.pyx in cantera._cantera.ThermoPhase.__init__()
    301     # The signature of this function causes warnings for Sphinx documentation
    302     def __init__(self, *args, **kwargs):
--> 303         super().__init__(*args, **kwargs)
    304         if 'source' not in kwargs:
    305             self.thermo_basis = mass_basis

/src/interfaces/cython/cantera/transport.pyx in cantera._cantera.Transport.__init__()
    175                 if not model:
    176                     model = 'None'
--> 177                 self.transport = newTransportMgr(stringify(model), self.thermo)
    178             self._transport.reset(self.transport)
    179 

CanteraError: 
*******************************************************************************
CanteraError thrown by IonsFromNeutralVPSSTP::calcNeutralMoleculeMoleFractions:
neutral molecule calc error - anion
*******************************************************************************

Behavior

While the former will prevent Kinetics and Transport methods from being called, the second command should result in an equivalent object with appropriate error handling if a user attempts to access transport properties.

System information

• Cantera version: 2.6
• OS: any
• Python/MATLAB/other software versions: Python 3.6.9

Other thoughts

• Prior to the implementation of Solution.h, there was no central object managing Transport (making this less of an issue)
• KineticsFactory does include an empty object, but there is no obvious way to reach it
• The issue surfaced in Read YAML using C++ newSolution in Python / access root level data #1129: the same complication becomes apparent in the C++ interface

[ischoegl]

A secondary observation is the following:

In [1]: import cantera as ct

In [2]: gas = ct.Solution("h2o2.yaml", transport_model=None)

In [3]: gas.input_data
Out[3]: 
{'elements': ['O', 'H', 'Ar', 'N'],
 'kinetics': 'gas',
 'name': 'ohmech',
 'species': ['H2', 'H', 'O', 'O2', 'OH', 'H2O', 'HO2', 'H2O2', 'AR', 'N2'],
 'state': {'T': 300.0, 'Y': {'H2': 1.0}, 'density': 0.08189392763801234},
 'thermo': 'ideal-gas',
 'transport': 'mixture-averaged'}

Here, the expectation would be to see transport being either None or omitted. Further, the following is not what is expected:

In [6]: gas.transport_model
Out[6]: 'Multi'

In [7]: gas.input_data
Out[7]: 
{'elements': ['O', 'H', 'Ar', 'N'],
 'kinetics': 'gas',
 'name': 'ohmech',
 'species': ['H2', 'H', 'O', 'O2', 'OH', 'H2O', 'HO2', 'H2O2', 'AR', 'N2'],
 'state': {'T': 300.0, 'Y': {'H2': 1.0}, 'density': 0.08189392763801234},
 'thermo': 'ideal-gas',
 'transport': 'mixture-averaged'}

[bryanwweber]

Further, the following is not what is expected:

In [6]: gas.transport_model
Out[6]: 'Multi'

In [7]: gas.input_data
Out[7]: 
{...
 'transport': 'mixture-averaged'}

I'm not sure I agree that this is unexpected. Depending on your definition of "input", you might expect it to remain 'mixture-averged' which was set by the input file (presuming that that case was working properly 😛 ).

[ischoegl]

Based on the docstring, it is unexpected, as it refers to the current state 😉

cantera/interfaces/cython/cantera/base.pyx

Lines 230 to 235 in 0d7373f

	property input_data:
	"""
	Get input data corresponding to the current state of this Solution,
	along with any user-specified data provided with its input (YAML)
	definition.
	"""

The way it is currently implemented, this isn't meant to guarantee round-trip conversion. The method name is probably a bit ambiguous, as discussed in #984 (and documented in #1053).

[bryanwweber]

Based on the docstring, it is unexpected, as it refers to the current state 😉

Ha! Fair enough 😄 I thought we had discussed this before, I just couldn't remember where. Thanks for the links!