Scroll radial leakage: _radial_leakage_angles raises ValueError on ~2-ULP rounding (phi_min > phi_max) in multilump example

[ibell]

Summary

_radial_leakage_angles in PDSim/scroll/symm_scroll_geo.pyx raises a ValueError because of a floating-point rounding artifact: it does a strict if phi_min > phi_max: raise check, and for a degenerate (zero-length) radial-leakage overlap the two angles come out mathematically equal but differ by ~2 ULP, so phi_min ends up infinitesimally greater than phi_max.

Observed for the radial-leakage key pair (2001, 1) at theta = 0.0:

phi_max(15.148966328953213) < phi_min(15.148966328953215)   # differ by ~2e-15

This is pure scroll geometry — no CoolProp involvement, version-independent.

Reproduction

examples/scroll_compressor_multilump.py hits it once it gets into the solver. (Note: that example also has a separate default_timer typo that aborts it earlier; fixed in #108. With that fix in place, this geometry error is what stops the run.) The other scroll examples (scroll_compressor.py, _disc, _valves, _w_VI) do not trigger it — it's specific to the multilump example's geometry/parameters.

File "PDSim/scroll/_scroll.pyx", line 41, in radial_leakage_area
File "PDSim/scroll/symm_scroll_geo.pyx", line 405, in radial_leakage_area
File "PDSim/scroll/symm_scroll_geo.pyx", line 515, in _radial_leakage_angles
ValueError: For the keys (2001,1) @theta = 0.0 max < min (error because phi_max(15.148966328953213) < phi_min(15.148966328953215)

Root cause

In _radial_leakage_angles (around line 514):

if phi_min > phi_max:
    raise ValueError('For the keys ('+str(key1)+','+str(key2)+') @theta = '+...)

phi_min / phi_max are built from involute-angle arithmetic (geo.phi_fie - theta - 2*pi*alpha, the overlap() helper, etc.). When the overlap genuinely collapses to zero length, the two expressions are equal in exact arithmetic but round to values a couple of ULP apart, and the strict > comparison treats that as an error rather than a zero-area leakage path.

Suggested fix

Tolerate a negligible difference instead of raising — e.g. only raise when the inversion exceeds a small tolerance, and otherwise clamp to a zero-length (zero-area) overlap:

if phi_min > phi_max:
    if phi_min - phi_max < 1e-10:   # rounding noise on a degenerate overlap
        phi_min = phi_max
    else:
        raise ValueError(...)

(Tolerance value up to the maintainers; the point is to not fail on ~ULP-level noise.) Worth confirming the (2001,1) @ theta=0 overlap is in fact meant to be zero-length here.

Environment

• PDSim @ master (also present on the coolprop-abi-guard / v8 work, which touches this file)
• Python 3.9, macOS arm64; reproduced against CoolProp 7.2.1.dev20260610124924 but the code path is CoolProp-independent