CACTUS crashes

[wmpsmithii-cpu]

Hello CACTUS community,

I'm running CACTUS (REL-0.1-336, dev branch, unmodified except MaxNLIters patched from 10 to 50) on a 5-blade straight-bladed H-Darrieus VAWT and hitting a consistent non-linear iteration convergence failure that terminates the run early. I'd appreciate any guidance on solver settings for high-solidity configurations.

Geometry:

• R = 1.2 m (RefR = 1.2, coordinates normalized by R)
• H = 4.8 m (RefAR = 4.0)
• 5 blades, NACA 6712, diamond taper (chord 0.4m mid, 0.2m tip)
• Pitch axis at 25% chord
• RotN: 0 1 0 (Y-vertical)
• Solidity σ = NB·c_mid/(2πR) ≈ 0.265

Case settings:

• TSR (Ut) = 1.5, RPM = 107.4, rho = 1.225, vis = 1.789e-5 (SI)
• nr = 20, nti = 36, nric = 9, convrg = 0.0001, DSFlag = 1
• NACA 6712 polar: uniform 1° spacing, 7 Reynolds blocks (2.5e4 to 2.0e6)

Symptom:
The simulation completes 4 revolutions cleanly with Cp converging nicely (rev 1: −3.1, rev 2: −1.9, rev 3: −1.74, rev 4: −1.70), then terminates on revolution 5 with:

***** NON-LINEAR ITERATION LOOP DID NOT CONVERGE IN 10 ITERATIONS. PROGRAM TERMINATED. *****

Note: despite the message saying "10 iterations," MaxNLIters is confirmed set to 50 in source — the error message text appears to be a hardcoded string (CACTUS.f90:597).

Floating-point exceptions are also reported at exit: IEEE_INVALID_FLAG, IEEE_DIVIDE_BY_ZERO.

I've confirmed:

• AOA values are physically sane (±10° range at TSR=1.5)
• The same geometry runs cleanly with NACA 0015 polar (RegTest3 reproduces its reference CPave values exactly)
• Changing convrg (0.0001 → 0.001), nric (9 → 2), or nr (20 → 60) does not change the termination point
• The failure always occurs at the same timestep in revolution 5, suggesting a specific numerical singularity rather than a general convergence tightness issue

Questions:

1. Are there recommended solver settings (nric, convrg, iut, ivtxcor, VCRFB/VCRFT/VCRFS) for high-solidity H-Darrieus configurations that are known to be more numerically stable?
2. Is the IEEE_DIVIDE_BY_ZERO likely originating in the dynamic stall model (DSFlag=1, LB model) at high AOA excursions, and if so, is there a known workaround?
3. Does the free-vortex-wake model have known stability issues at low TSR (1.5) with high-solidity rotors where deep stall occurs on part of the revolution?

Any guidance appreciated. Happy to share the full .in and .geom files if helpful.

Bill Smith
Smith's Enterprises LLC