Engine-owned homogeneous predictor-corrector for diagonal convex QPs with
equality/nonnegative cones, following Goulart and Chen (2024), sections 2-3.
Every solve is cold. It has no clock, random source, runtime module, process
or external numerical service. The supplied workspace holds its iterate and
results copy out before the synchronous borrow ends.
Full termination retains 1e-9 primal/dual feasibility and absolute-or-relative
gap. Certificate, reduced-accuracy, iteration and progress criteria retain
Clarabel 0.11.1 meanings; reduced accuracy is status 4, never status 1.
Finite arithmetic can refuse ill-conditioned inputs. Caller-specific physical
acceptance and subsequent transformations remain independent obligations.
Engine-owned homogeneous predictor-corrector for diagonal convex QPs with equality/nonnegative cones, following Goulart and Chen (2024), sections 2-3. Every solve is cold. It has no clock, random source, runtime module, process or external numerical service. The supplied workspace holds its iterate and results copy out before the synchronous borrow ends.
Full termination retains 1e-9 primal/dual feasibility and absolute-or-relative gap. Certificate, reduced-accuracy, iteration and progress criteria retain Clarabel 0.11.1 meanings; reduced accuracy is status 4, never status 1. Finite arithmetic can refuse ill-conditioned inputs. Caller-specific physical acceptance and subsequent transformations remain independent obligations.