Pseudopotentials and Core Approximations

Introduction

Electrons near a nucleus experience a sharply varying potential. Accurately resolving their rapid wavefunction oscillations with plane waves would require an enormous basis, even when the chemistry of interest mostly concerns outer electrons. Pseudopotentials and related core approximations make calculations practical by replacing or transforming the difficult inner region. Their speed comes with a responsibility: the selected core treatment must remain valid for the oxidation states, bond lengths and properties being studied. “Using a pseudopotential” does not identify a single universal approximation.

Core explanation

In an all-electron treatment, both core and valence electrons are represented explicitly. A frozen-core approximation holds selected core electrons fixed while treating the valence response. A pseudopotential replaces the action of the nucleus plus chosen core electrons on valence electrons with a smoother effective potential. The pseudo-valence wavefunctions need not reproduce all rapid oscillations close to the nucleus, so fewer plane waves can describe them. Outside a chosen core region, the effective model is designed to reproduce relevant all-electron scattering or valence behavior. Different pseudopotential families make different tradeoffs among smoothness, norm conservation, cutoff and transferability.

The projector-augmented-wave (PAW) approach uses smooth auxiliary wavefunctions for the plane-wave calculation plus localized information that reconstructs all-electron-like valence behavior near nuclei. It commonly retains a frozen-core approximation. PAW is often grouped with pseudopotential approaches in software workflows, but its reconstruction machinery is distinctive; one should not assume that every pseudo-density is identical to an all-electron density at the core. The VASP developers' PAW formalism description explains the smooth partial waves, all-electron partial waves and projectors. The selected potential dataset, version, valence electron count and recommended cutoff are essential reproducibility details.

Choosing which shells count as core is a chemical decision. Semicore states can overlap or polarize enough to influence short bonds, pressure response, magnetic energetics or spectroscopy. A potential that treats a transition metal's 3p shell as core may be cheaper than one including it explicitly as valence, yet the latter may be needed for a demanding oxide or high-pressure environment. Soft potentials with a larger core radius need a lower plane-wave cutoff but may be less transferable to short bonds; harder potentials generally cost more. VASP's dataset guidance provides concrete examples of semicore and soft/hard tradeoffs and recommends prototype comparisons when accuracy matters. These are implementation-specific recommendations, so a calculation should identify its actual library rather than copy a rule blindly.

Core approximations may also include relativistic effects. A heavy-element potential may encode scalar-relativistic behavior, and some versions include spin–orbit terms. One must not assume that a scalar potential will reproduce SOC-sensitive spectra, or that a potential prepared under one exchange–correlation convention is automatically ideal for another. For core-level X-ray spectroscopy, removing the very core electrons of interest from explicit treatment can require a special reconstruction or all-electron method. For ordinary valence bonding, a good core model may give excellent efficiency. The appropriate choice follows the property, not an abstract ranking of “harder” versus “softer.”

Validation is comparative. First converge the plane-wave cutoff for the chosen potential, because the required cutoff depends on it. Then compare a chemically relevant test—bond length, reaction energy, magnetic gap, bulk modulus or response—with an all-electron or harder-core benchmark when available. If two competing phases differ by only a few meV per atom, a small potential-dependent shift can reverse their order. Use the same potential family and valence partition consistently across the reaction or phase comparison. Mixing potentials across calculations can insert a methodological difference into the supposed chemical energy change.

Step-by-step reasoning

1. Define the target chemistry and observable, including charge states, bond lengths and any heavy-element or core-level effects. 2. Inspect candidate potential libraries for valence electron count, semicore treatment, relativistic terms and exchange–correlation compatibility. 3. Select a potential and converge its plane-wave cutoff for the target energy, force or stress. 4. Check k-point and cell-size convergence independently from the potential choice. 5. Compare a prototype with an alternative potential or all-electron reference if the result is sensitive or high stakes. 6. Report the exact potential identifiers and use one consistent set across compared systems.

Visual explanation

Draw a steep nuclear potential at the center with rapidly oscillating all-electron valence wavefunction nearby. Overlay a smoother pseudo-wavefunction that matches appropriate behavior outside a marked core radius. In a second panel, show PAW's smooth auxiliary function and its reconstructed all-electron-like shape within an augmentation sphere. A third panel lists two potential choices: “core includes semicore shell” and “semicore treated as valence,” with arrows to cost and possible property sensitivity.

Real-world analogy

To model traffic across a country, a map may replace the complex street pattern inside each city with a simplified interchange that reproduces how cars enter and leave. The map becomes faster to use, but an interchange calibrated for ordinary traffic may fail during a special event that depends on local streets. A pseudopotential similarly preserves selected valence behavior while simplifying the core. The analogy is about model transferability, not actual electron paths.

Real-world example

A team compares two magnetic phases of a transition-metal oxide. A standard potential treats certain lower shells as core; a second includes them as valence. The predicted magnetic energy difference changes from +3 to −1 meV per metal under otherwise matched settings. The team cannot claim a robust ordering from the cheaper model. It tests cutoff and k points for both potential choices and checks a small all-electron reference if possible. In contrast, for a large qualitative geometry screen where both potentials agree within the needed tolerance, the cheaper model may be an appropriate choice.

Why?

Why does a softer potential usually reduce cost? Smoothing the sharp core-region wavefunction reduces the highest spatial frequencies needed in its plane-wave expansion, so a lower kinetic-energy cutoff can suffice. But smoothing removes detail and can make the model less reliable when valence density penetrates the core region more than it did during construction. A lower required cutoff is therefore a computational advantage, not direct evidence of better chemistry.

Common misconception

“A pseudopotential means the core has no effect.” Its effect is encoded in the effective operator and frozen-core assumptions. “A PAW calculation explicitly optimizes every electron.” It may reconstruct all-electron-like valence behavior while retaining a frozen core. “The library's recommended cutoff proves convergence.” It is a starting value for a property-specific test. “A potential that works for one oxidation state works for all compounds of the element” overlooks transferability limits.

Worked example

Suppose two potential datasets for a hypothetical element produce an oxide formation energy of −250 and −244 kJ mol⁻¹ per formula unit, respectively, after each has its own converged plane-wave cutoff. Their 6-kJ difference is a potential-choice uncertainty indicator for this test, not a measured error bound. If an intended conclusion depends on a 2-kJ difference between candidate reaction pathways, this spread is too large to ignore. A comparison with an all-electron benchmark or a more transferable potential is needed. Increasing the cutoff of the first potential further cannot in general transform it into the second, because their core and valence definitions differ.

Quick check

1. Why are pseudopotentials especially useful with plane waves? Answer: They smooth difficult core-region behavior, reducing the plane-wave basis required for valence calculations. 2. Does converging cutoff for one potential establish that a different potential is equally accurate? Answer: No. Numerical convergence and the core model's physical transferability are separate issues.

Exam focus

Explain the purpose of a core approximation and distinguish numerical basis error from pseudopotential transferability. State what a semicore-in-valence choice changes. Describe PAW as smooth auxiliary valence calculation with local reconstruction, usually with a frozen core. In a comparative calculation, insist on consistent datasets and converged cutoffs. For a specific property such as core spectroscopy or spin–orbit splitting, ask whether the chosen potential actually represents the relevant physics.

Advanced insight

Potential datasets are generated from atomic reference configurations, but crystals can have different oxidation and pressure conditions. Transferability tests therefore probe whether scattering and response behavior remains adequate outside the reference. Core polarization, nonlinear core corrections and relativistic construction can matter in particular cases. A pseudopotential and an all-electron calculation may use different energy zeros for absolute totals; compare balanced energy differences and documented conventions rather than raw total energies across frameworks. The most credible benchmark isolates potential error from functional and structural errors.

Summary

Core approximations make periodic electronic-structure calculations tractable by replacing or transforming rapidly varying inner-electron behavior. The choice of valence and semicore shells, potential hardness, relativistic content and PAW or pseudopotential dataset can affect chemical results. Plane-wave cutoff convergence tests numerical representation; prototype comparison tests transferability. Document exact datasets and validate the approximation for the actual property and chemical environment.

Practice questions

1. Why might a potential that freezes a semicore shell fail for a short, highly polar bond? Answer: The semicore electrons may contribute to polarization or overlap that the frozen model cannot represent adequately. 2. What is the consequence of increasing the plane-wave cutoff while leaving the pseudopotential unchanged? Answer: The pseudo-valence basis becomes more complete, but the core approximation itself is unchanged. 3. Can a scalar-relativistic potential alone resolve a spin–orbit-split spectrum? Answer: No. Explicit spin–orbit treatment is required for that splitting. 4. What should be reported so another researcher can reproduce a PAW calculation? Answer: The specific potential dataset/version, valence configuration, method, cutoff and relevant convergence settings.