CP2K2026.2 native LS-SCF SIGN/TC2/TRS4 and sparse filtering; installed2025.2/2026.2 schema inventory.
Sample native calculations: P100 CUDA build, GNU13.3/MPICH5.0.1/CUDA12.9/OpenBLAS0.3.33/ScaLAPACK2.2.3. H2O3,8valence electrons, PBE/DZVP-MOLOPT-SR-GTH/GTH-PBE,600/50Ry,12Angstrom nonperiodicMT. Same completed OT reference Hamiltonian. All4 native results included; no performance scaling claim.
A tiny water case teaches input/solver precision only, not linear-scaling efficiency or general metallic applicability. For this exact LS input use the P100 module. Input schema lists syntax, not optional library availability. Native CPUifx startup issue is private maintenance evidence, not an assurance for allCPU builds.
Activate pymatgen/2026.9.24 and bash run.sh to read complete fields and draw plots. bash cli.sh runs parser/tool operations only on CP2K2026.2 CPU; no LOAD/EVAL/--run. bash prepare.sh then sbatch gpu.sh starts genuinely new LS calculations from inputs. prepare.sh copies open CP2K BASIS/POTENTIAL files from the installed module. No vendor data, binary/source, VASP file, user identity or cluster private receipts are bundled.
Schema TSVs are derived native metadata including aliases/removal flags/default pseudo-keywords. Repeated equivalent branches are separate full paths, not separate independent physical capabilities. Actual CP2K XML is generated locally with cp2k.psmp --xml; the native XMLs are not redistributed.
