Verification

Every FLUME capability is backed by a verification case in src/tests/flume/verification/: a check.sh that runs the case (CPU by default; FLUME_EXE=exe/adam_flume_fnl for the GPU backend) and an oracle that asserts the physics (an exact solution, a reference, a symmetry or a conservation property), not a stored golden. The stored goldens are the job of the regression suite; the building blocks below the solver are pinned by the unit tests.

Each case has its own page with the same sections: the case and its scope, the models and numerics with references to the code, the initial and boundary conditions, the input files, the results on CPU and FNL, and the findings. All runs use 2 MPI ranks and WENO5 unless a page says otherwise. The figures are drawn by src/tests/flume/verification/make_doc_figures.py, from the work directories of the check.sh scripts or from the re-runs of make_doc_runs.sh.

Run any case from its directory:

cd src/tests/flume/verification/sod && ./check.sh                  # CPU backend
FLUME_EXE=$PWD/../../../../../exe/adam_flume_fnl ./check.sh        # FNL backend (GPU)

The verification cases are complete simulations, and they are run with a focused scope: the legs a change touches, not the whole catalog. The regression suite is the layer that runs on every change. Every script therefore selects its legs the same way, through the shared legs.sh:

./check.sh --legs vv1,vv3          # the listed legs (commas or blanks)
./check.sh --legs all              # every leg, including those the default leaves out
./check.sh                         # the default selection (the script's header lists it)

Each script lists its legs and its default in its Usage: header. An unknown leg, or a selection that names no leg, stops the script with exit status 2 before anything runs, so a misspelt leg can never pass by running nothing. Each log opens with the resolved selection (>> legs: ...). The older spellings --leg and --cases still work as aliases of --legs.

In the catalog, Figure "no" marks a contract or bitwise leg (an input refusal, CPU = FNL, a restart, a round trip): a setup table documents it better than a plot of a difference that is zero.

Euler

Case	ID	Milestone	Backends	Figure
Sod shock tube	V1, RV-3	M1 (#35), M3 (#47)	CPU, FNL	yes
Lax problem	RV-3 (lax leg)	M3 (#47)	CPU (all schemes), FNL (weno-riemann HLLC)	yes
Shu–Osher shock / entropy-wave interaction	RV-3 (shu-osher leg)	M3 (#47)	CPU (all schemes), FNL (weno-riemann HLLC)	yes
Isentropic vortex: order of accuracy	V2, RV-2	M1 (#35), M3 (#47)	CPU (all schemes), FNL (weno, weno-riemann LLF and HLLC)	yes
Shock over a cylinder	V6	M1 (#35), M3 (#47)	CPU, FNL	yes
Conservation across AMR seams	V3, RV-4	M1 (#35), M3 (#47)	CPU, FNL	yes
Multi-realm runs	multirealm legs 1–6	M2 entry (#37); #40, #51, #52, #54; M6 (#68)	CPU, FNL	yes
Forward-facing step: a three-realm forest on quadtrees	EV-step	— (#46, quadtree AMR; #66)	CPU, FNL (trees leg); CPU only (full leg)	yes

Ideal MHD

Case	ID	Milestone	Backends	Figure
MHD on weno-riemann	RV-5, RV-6, RV-7	M3 (#47, M3-P3c)	CPU, FNL	yes
Ryu–Jones 2a: all seven waves	MV-4	M2 (#41)	CPU, FNL	yes
Brio–Wu and Ryu–Jones 4d	— (mhd/riemann, M2-P3 exit)	M2 (#41)	CPU, FNL	yes
MHD plumbing	— (mhd/plumbing, M2-P1)	M2 (#41, P1)	CPU, FNL	no
MHD with a zero field	MV-2	M2 (#41, P3)	CPU, FNL	no
GLM divergence cleaning	MV-3, MV-10	M2 (#41); the MV-10 monitor M6 (#68, G1)	CPU, FNL	yes
Order of accuracy: linear waves, CPAW, magnetised vortex	MV-5, MV-6, MV-7	M2 (#41)	CPU, FNL	yes
MHD conservation across AMR seams	MV-11	M2 (#41, P7a)	CPU, FNL	yes
Orszag–Tang vortex	MV-12	M2 (#41)	CPU, FNL	yes
MHD rotor	MV-13	M2 (#41)	CPU, FNL	yes
Field loop advection	MV-9	M2 (#41); --agree #79	CPU, FNL	yes
Rotated shock tube	MV-8	M2 (#41)	CPU, FNL	yes
EGLM cleaning	EV-2, EV-3, EV-4	M3 (#47, M3-P4)	CPU, FNL	yes
Positivity limiter	PV-1 to PV-5	M3 (#47, M3-P5); PV-5 #50	CPU, FNL (PV-1 as a sphere: FNL only)	yes
Quadtree AMR	MV-15	#46 (no milestone, after M3)	CPU, FNL	yes

Infrastructure

Case	ID	Milestone	Backends	Figure
Ghost probe: every face and edge ghost on a linear field	GP	M4 (#65 P0, P1), M6 (#68 C2, D1)	CPU, FNL	yes
Scaling covariance and the reference layer	NV-0, NV-4, NV-5, NV-6, NV-8, NV-9	— (#49, non-dimensional form, N0–N2)	CPU, FNL	no
Ranks without blocks	DV-1	— (#82)	CPU, FNL	yes
Non-finite state guard	— (nonfinite)	— (#45)	CPU, FNL	no

Dissipative terms (M4)

Case	ID	Milestone	Backends	Figure
Input contract of the dissipative terms	DC	M4 (#65)	CPU, FNL	no
Viscous waves	VV-1, VV-2	M4 (#65)	CPU, FNL	yes
Compressible Couette flow	VV-3	M4 (#65)	CPU, FNL	yes
Becker's viscous shock	VV-4	M4 (#65)	CPU, FNL	yes
Ohmic decay and the visco-resistive Alfvén wave	VV-5	M4 (#65)	CPU, FNL	yes
Ohmic heating and the divergence constraint	VV-6	M4 (#65)	CPU, FNL	no
Conservation across 2:1 seams and forest seams	VV-7	M4 (#65)	CPU, FNL	yes
Accuracy across 2:1 seams	VV-8	M4 (#65)	CPU, FNL	yes
CPU and FNL agree run by run	VV agree	M4 (#65)	CPU, FNL	no

Runtime AMR (M5)

Case	ID	Milestone	Backends	Figure
Regrid round trip and input contract	RG (rg0, rg1, rg2)	M5 (#74)	CPU (rg0, a host program); rg1 and rg2 CPU, FNL	no
Runtime regrids in FLUME	RG (rg3 to rg6)	M5 (#74)	CPU, FNL	yes
Accuracy of runtime AMR	AV (av2 to av8)	M5 (#74)	CPU, FNL	yes
MHD blast symmetry on the flux-splitting path	#77 study, bounded by av3c	M5 (#74), #77	CPU, FNL	yes

Seam accuracy and divergence (M6)

Case	ID	Milestone	Backends	Figure
Order across static 2:1 seams	AO (A1, C1)	M6 (#68)	CPU, FNL	yes
Divergence across seams	AD (A2)	M6 (#68), #78	FNL (tables); CPU for the loop leg and the D_flux cross-check	yes
Shock crossing a static 2:1 face	AS (A3, C3, C3b, C3c)	M6 (#68)	CPU, FNL (Sod); FNL only (RJ2a, C3, C3c)	yes
Integral drift and the WENO sensor near seams	AL (A4)	M6 (#68)	CPU, FNL (the two backends log the same values)	yes

Seam studies (M6)

Case	ID	Milestone	Backends	Figure
Quintic coarse-to-fine fill	D1	M6 (#68)	CPU, FNL (table: FNL)	yes
Limited coarse-to-fine fills	D2	M6 (#68)	FNL (crossings); CPU, FNL (AO legs)	yes
Shock-crossing defect is a capturing error	E0	M6 (#68)	FNL; CPU re-runs agree to the printed digits	yes
Reflux policy: troubled against always	E2 (E1)	M6 (#68)	FNL (flags checked equal to the CPU's); AO euler-tq: CPU, FNL	yes
#78 is the reflux, not the restricted ghosts	G3a	M6 (#68), #78	FNL only	yes
GLM damping length on a seam	G2	M6 (#68)	FNL only	yes
EGLM reflux leaves B alone	G3b	M6 (#68), #78	CPU, FNL	yes
2:1 caps in a 2-D prototype	B1	M6 (#68)	NumPy prototype (no ADAM run)	yes
Sensor-gated reflux prototype	B2	M6 (#68)	NumPy prototype (no ADAM run)	yes
Footprint of a high-order restriction	B3	M6 (#68)	NumPy prototype (no ADAM run)	yes

External gravity (M7)

Case	ID	Milestone	Backends	Figure
Stratified atmospheres and the hydrostatic boundary	GV (G0)	M7 (#73)	CPU, FNL	yes
The gravity force	GV-1, GV-2, GV-3, GV-6, GV-7 (G1)	M7 (#73)	CPU, FNL	yes
The order of the energy source	GV-9 (G2a)	M7 (#73)	CPU, FNL	yes
The well-balanced mode	GV-2b, GV-5, GV-8, GV-10, GV-7 (G2b)	M7 (#73)	CPU, FNL	yes
Rayleigh–Taylor	GV-4 (G3)	M7 (#73); symmetry study #83	CPU, FNL	yes

Self-gravity (M8)

Case	ID	Milestone	Backends	Figure
Gravity from a cell potential	SV-1	M8 (#81), P1	CPU, FNL	yes
The FFT Poisson solver	SV-2	M8 (#81), P2	CPU, FNL	yes
Jeans modes under self-gravity	SV-3, SV-4, SV-5	M8 (#81), P3	CPU, FNL	yes
The isolated Poisson solver	SV-6	M8 (#81), P4	CPU, FNL	yes
The n = 1 polytrope: equilibrium and collapse	SV-7, SV-8, SV-14	M8 (#81), P4; #87, P3	CPU, FNL	yes
The multigrid Poisson solver on AMR grids	SV-9, SV-12, SV-13	M8 (#81), P5a; #87, P1, P2	CPU only	yes
Jeans modes on a static AMR grid	SV-10	M8 (#81), P5b, P5c	CPU, FNL	yes
Self-gravity under runtime AMR	SV-11	M8 (#81), P5c	CPU, FNL	yes