Input contract of the dissipative terms

ID	Group	Milestone	Backends	Figure	Regression
DC	Dissipative terms (M4)	M4 (#65)	CPU, FNL	no	—

Case and scope

M4 reads each dissipative coefficient either as a coefficient or as its dimensionless number (#49 N4), refuses the positivity limiter with any of them (D-M4-5) and adds the no-slip walls (D-M4-7). The case is a contract test, not a flow: a list of 25 inputs, each a base input with some keys set, each with an expected outcome.

What the case catches:

• an invalid combination that runs: it must stop the run with a message that names it;
• a wrong implied coefficient: a valid combination must imply exactly the coefficient it states ($\mu =1/Re$, $k=\mu c_p/Pr$, $\eta =1/Rm$);
• a key dropped silently: a coefficient that is read but not used, or a dimensional key not converted back;
• a side effect of the dissipative path on an ideal run: zero coefficients must leave the run bit for bit equal to the base run.

The keys are listed in input.

Models and numerics

Item	Choice	Code
Coefficients and numbers	viscosity or reynolds, conductivity or prandtl, resistivity or magnetic_reynolds or lundquist; viscosity_law = power-law with viscosity_exponent, reference_temperature	load_from_file, laws, adam_flume_dissipation_object.F90
Order of the dissipative fluxes	[numerics] dissipative_order, 2 or 4; 4 needs ngc >= 3	load_from_file, adam_flume_numerics_object.F90; check_dissipation, adam_flume_common_object.F90
Positivity limiter	refused with any dissipative coefficient (D-M4-5)	check_positivity_limiter, adam_flume_common_object.F90
No-slip walls	wall-noslip (tangential wall_u/v/w, a normal component is fatal), wall-isothermal (wall_temperature > 0)	load_from_file, adam_flume_bc_object.F90; boundary conditions
Diffusive time-step limit	logged at every run (dissipative dt limit (issue #65))	log_dt_limit, adam_flume_common_object.F90; time-step bound
Dimensional input	[reference] section; lundquist needs the Alfvénic velocity preset	classify, adam_flume_reference_object.F90; units and scaling

Initial and boundary conditions

The cases modify two regression inputs; the flow itself is not judged.

Item	Value
Euler base	regression/sod-x: the Sod tube along $x$, 64 blocks of $50\times 4\times 4$ cells, extrapolation on every face
MHD base	regression/uniform-amr-mhd: a uniform MHD state on a periodic octree, $8^3$ cells per block, level 1 with the box $[0,0.5{]}^3$ refined to level 2, GLM, 10 steps
Ranks	2
Steps	the log and convert cases stop after 10 steps (it_max = 10); the ideal case runs the base input to its end
Dimensionalisation (convert)	scaling.py dimensionalize --j 1 --k -1 --m 1: $L_0=2$, $u_0=1/2$, ${\rho }_0=4$

The 25 cases (make_contract.py --list):

Kind	Cases	Oracle
refuse (18)	both keys of a term (3: viscosity + reynolds, conductivity + prandtl, magnetic_reynolds + lundquist), prandtl without a viscosity, resistivity on Euler, a negative viscosity, a zero Reynolds number, the power law without a viscosity, its keys without the law, a missing reference temperature, an unknown law, dissipative_order = 3, the limiter with a viscosity, a normal wall velocity, an isothermal wall without (or with a negative) temperature, dissipative_order = 4 with ngc = 2, a negative power-law exponent	the run stops with the expected message
log (3)	reynolds + prandtl (64, 0.75); viscosity + conductivity (0.015625, 0.03125); magnetic_reynolds + reynolds (128, 32, MHD)	the logged $\mu$, $k$, $\eta$ equal $1/Re$, $\mu c_p/Pr$, $1/Rm$ exactly; every run takes ten steps and logs its diffusive time-step limit
ideal (1)	zero coefficients, dissipative_order = 2	the run equals the base run bit for bit
convert (2)	the Euler case with viscosity, conductivity, the power law and an isothermal moving wall; the MHD case with resistivity; dimensionalised with $L_0=2$, $u_0=1/2$, ${\rho }_0=4$	every dimensional key logged converted back exactly (scaling.py check-log), the conductivity and the temperatures with the gas constant
convert-refuse (1)	lundquist under a [reference] without the Alfvénic preset	refused

Input files

File	Role
src/tests/flume/verification/dissipation/make_contract.py	the 25 cases (--list), writes each input, judges the logged coefficients (--check-log)
src/tests/flume/verification/dissipation/check.sh	runs every case and applies its oracle (--np N, --legs LIST: dc-refuse, dc-log, dc-ideal, dc-convert, dc-convert-refuse, each the cases of that kind, default all; FLUME_EXE selects the backend)
src/tests/flume/verification/scaling/scaling.py	dimensionalize and check-log of the convert cases
src/tests/flume/regression/sod-x/input.ini, src/tests/flume/regression/uniform-amr-mhd/input.ini	the base inputs

cd src/tests/flume/verification/dissipation && ./check.sh             # CPU
FLUME_EXE=$PWD/../../../../../exe/adam_flume_fnl ./check.sh           # FNL

No figure: a contract leg has nothing a plot would show (#85).

Results

Check	Bound	CPU	FNL
18 refusals	expected message	pass	pass
3 logged coefficients	exact	pass	pass
ideal run against the base run	bitwise	pass	pass
2 conversions	exact	pass (34 keys on the Euler case, 30 on the MHD one)	pass
1 conversion refusal	expected message	pass	pass

All 25 cases pass, on the CPU (2 ranks) and on FNL.

Related cases: the coefficients are exercised by viscous waves and Ohmic decay; the dimensional input by scaling covariance.

Findings

• Every invalid combination of D-M4-5, D-M4-6 and D-M4-7 stops the run with a message that names the keys; none runs with a coefficient dropped.
• The coefficient a number implies is logged bit-exactly, so the conversion $1/Re$, $\mu c_p/Pr$, $1/Rm$ has no hidden rounding step.
• The dissipative path is inert at zero coefficients: the run is bitwise the base run.