PTC Benchmark

What PTC computes

All these observables — one engine, one input

PTC is not specialised on atomization energy: it is a PT ab-initio calculator that derives all molecular observables analytically from the single input s = 1/2 + the active-prime backbone {2, 3, 5, 7, 11, 13}. The table below summarises covered observables, their current accuracy and their status. The rest of the page focuses on D_atomization — the most heavily tested observable at scale — but the engine also produces r_e, ω_e, IE, EA, NICS, σ_NMR, etc. without any parameter fitted to data.

Methodological note. All accuracies above are measured under the same constraint: 0 fitting variables, no degrees of freedom, the single input s = 1/2 and the {2, 3, 5, 7, 11, 13} backbone with the PT principles (phase cycles, information balance, cascade, q₊/q₋ bifurcation, anti-double-counting). The engine never receives any data-fitted parameter.

Coverage comparison

PTC vs other quantum-chemistry calculators: who computes what, with how many fitted parameters?

Reproducing observables from first principles is a common sport. Doing it without any data-fitted parameter is rarer — and doing it for D_at, geometry, IE, EA, NICS, σ_NMR all simultaneously, from a single mathematical input, does not appear to exist elsewhere as far as we know. The table below compares PTC against the major families of quantum chemistry methods. A check means the method routinely produces the observable; a grey check means it is in principle possible but rarely used. The last column distinguishes strictly non-empirical methods (built from first principles, never calibrated against the observables they reproduce) from the rest.

How to read. HF and CCSD(T) also technically reach “0 fitted parameter” status but rely on the choice of an atomic basis (an object that itself contains numerically optimised gaussian parameters, whose choice strongly drives accuracy). PTC does not need that choice: connectivity and amplitudes emerge from the variational sieve and the {3, 5, 7} cascade dimensions. Standard DFT methods include the choice of a functional (B3LYP, M06-2X, etc.) whose coefficients are themselves calibrated against experimental thermochemistry — that is the definition of a fitted parameter. ML potentials are trained on millions of DFT data points, so they are derived from a chain of calibrations.

Benchmark 1 — Absolute test

PTC vs experimental measurements

1000 molecules from Active Thermochemical Tables (ATcT), NIST WebBook, JANAF and other standard thermochemistry references. For each molecule, the atomization energy D_at computed by PTC is compared against the experimental value. 0 failures — PTC produces a number for every molecule, including transition metal complexes that B3LYP cannot compute in 6-31G* basis. Each experimental value is automatically cross-validated against the dual CCCBDB convention (ΔfH°(0 K) and ΔfH°(298 K)) — 546 molecules out of 994 pass the audit at less than 2 % drift, the rest is flagged either as moderate drift or non-auditable (no resolved CAS or no CCCBDB entry).

By chemical category

Charts

Table of 994 molecules

Each row is traceable. The "Source" column indicates the database of origin for the experimental value (NIST WebBook, ATcT v1.130, JANAF, Huber-Herzberg…). The "Reliability" column summarises the automatic cross-check against the dual CCCBDB convention (ΔfH°(0 K) and ΔfH°(298 K)). The "Link" column points to the official record.