CDW-EIS calculations deliver a database of atomic ionization cross sections for about 130 projectile-target systems, with a neural-network surrogate extending coverage to systems not calculated directly

The work provides an open CDW-EIS database of about 130 projectile-target systems and 6239 subshell-resolved total ionization cross sections, covering atomic targets from H to Kr, bare ions and antiprotons, at roughly twelve projectile velocities per pair. K-, L-, and M-shell ionization cross sections needed for PIXE analysis have mainly come from measurements or semiempirical models such as ECPSSR; this work supplies systematic ab initio values and gives particular attention to period 3 and 4 elements relevant to pigment analysis and materials characterization. Each database entry records projectile and target atomic numbers, subshell quantum numbers, occupation number, binding energy, projectile velocity, and cross section; total cross sections were obtained by integrating differential cross sections over ejected-electron energy and emission angle. Differential cross sections are not currently in the online database and are available from the authors on request.

In reduced variables, proton- and antiproton-induced K-shell ionization curves for different targets collapse onto a nearly universal behavior, indicating that the K-shell binding energy sets the leading target-dependent scale, while a residual target dependence remains at low and high reduced velocities, reflecting outer-shell electronic structure. The work attributes a substantial part of the target dependence to the binding energy of the ionized subshell and defines reduced projectile velocity and reduced cross section so that results for different targets and projectile species can be compared on a common scale. The conclusion rests on the reduced representation of proton and antiproton K-shell cross sections in the database; the text reports a pronounced reduction of dispersion, especially around the cross-section maximum, with residual target dependence at low and high reduced velocities.

Inner-shell ionization shows a charge-sign dependence: at intermediate and high velocities proton and antiproton cross sections approach one another, while at lower velocities the curves cross and antiproton-induced ionization becomes larger, which the authors regard as a Barkas-type effect associated with higher-order collision effects. The proton-antiproton comparison is extended from light systems such as H and He to inner-shell sums for metallic targets Al, Si, Ni, and Cu, and the cross sections are positioned as input for shell-resolved descriptions of proton and antiproton stopping in metals. The comparison uses calculated sums of inner-shell cross sections for Al, Si, Ni, and Cu; the text notes that antiproton calculations are currently available for targets with specific atomic numbers and that experimental antiproton stopping-power data exist for these metallic targets.

The five-member neural-network surrogate CDWEIS-NN achieves a median relative error of 3.1% in group-held-out validation, with 99.5% of held-out cross sections within 30% of the CDW-EIS values, and its predictions for K-shell ionization of Cl, K, Ti, Fe, and Cu by C6+ and O8+ reproduce the overall magnitude and energy dependence of measurements for pairs absent from the training database. The IKEBANA neural-network strategy previously applied to electron-impact K-shell ionization is extended to ion-atom collisions, using an antiproton flag so that a single network represents both proton and antiproton collisions without recomputing CDW-EIS. The training set comprises 6089 entries over 119 projectile-target pairs; the held-out set comprises 47 of the 119 pairs and 2371 cross sections, 38.9% of the filtered dataset; the five members have MAEs between 0.01822 and 0.02171. The authors state that validation pairs also determine the stopping epoch, so the procedure is repeated group-held-out validation rather than a completely independent final test.

The resource is aimed at ion-beam analysis and materials researchers who need K-, L-, and M-shell ionization cross sections: the database covers atomic targets from H to Kr under bare-ion and antiproton impact, gives cross sections for all occupied subshells at roughly twelve projectile velocities per pair, and is accessible through the online repository and the CDWEISNN.ipynb notebook. CDWEIS-NN is restricted to antiprotons and bare ions within a specified projectile-charge range and to targets from H to Kr, so it interpolates within that physical domain and predicts projectile-target combinations not calculated directly; the text states that its purpose is not to replace the collision model but to extend the practical usefulness of the available CDW-EIS results. The shell-resolved cross sections are positioned as input for shell-resolved descriptions of proton and antiproton stopping in metallic targets, particularly in the low-velocity region where charge-sign effects are most pronounced.

The CDWEIS-NN ensemble spread reflects only consistency among the five members; the authors state explicitly that it does not include the discrepancy of the underlying CDW-EIS model with experiment and should not be interpreted as a complete uncertainty estimate. The comparison for K-shell ionization of Cl, K, Ti, Fe, and Cu by C6+ and O8+ cannot separate the neural-network approximation from the accuracy of the underlying CDW-EIS description, because direct CDW-EIS calculations are not available for these pairs. The validation pairs also determine the stopping epoch, making the procedure repeated group-held-out validation rather than a completely independent final test. In addition, the online database does not currently include differential cross sections, which are available from the authors, and the L-shell comparison uses a compilation whose original experimental references the text says should be identified in the final version.