Systematically Improvable Numerical Atomic Orbital Basis Using Contracted Truncated Spherical Waves

This paper presents a systematically improvable numerical atomic orbital basis set constructed by contracting truncated spherical waves to minimize kinetic operator spillage, which enhances transferability and eliminates spurious periodic interactions while achieving high precision across diverse molecular and bulk properties.

The Gist

The Big Picture: Building Better Lego Sets for Atoms

Imagine you are trying to build a perfect model of a house using Lego bricks. In the world of computer science and chemistry, we do something similar when we try to simulate how atoms behave. We use "basis sets"—which are essentially pre-made sets of mathematical shapes (like Lego bricks)—to build the electron clouds around atoms.

For decades, scientists have used Numerical Atomic Orbitals (NAOs). Think of these as custom-made, flexible Lego bricks that fit perfectly around an atom. They are great because they are small, fast, and efficient. However, they have a flaw: it's hard to know if you have enough bricks to build a perfect model. You might build a house that looks okay, but if you zoom in, the roof is slightly crooked. You don't know if adding more bricks would fix it, or if you just need different types of bricks.

On the other hand, there is a method called Plane Waves (PW). Imagine this as using an infinite supply of tiny, identical, square tiles to cover the whole house. You can make the tiles as small as you want to get a perfect picture. The problem? It takes a massive amount of computer power to use enough tiles to get a good result, especially for big systems.

The Goal of This Paper:
The authors (Yike Huang, Mohan Chen, and colleagues) wanted to create a "Super-Lego Set." They wanted a system that is:

1. Fast and small (like the custom NAOs).
2. Systematically perfectible (like the infinite tiles), meaning you can always add more pieces to get a better result without guessing.
3. Accurate for both simple molecules and complex solid materials.

---

The Solution: "Truncated Spherical Waves" (The Magic Balls)

To solve the problem, the authors used a clever trick. Instead of using standard Lego bricks, they started with a different kind of building block called Truncated Spherical Waves (TSWs).

The Analogy: The Soundproof Room
Imagine you are in a spherical room (a ball-shaped room). If you clap your hands, the sound waves bounce off the walls.

• TSWs are like specific patterns of sound waves that fit perfectly inside this ball-shaped room.
• The authors take these sound waves and "cut them off" (truncate) at the edge of the room so they don't spill out. This makes them behave like the custom NAO bricks (localized to one atom).

The Secret Sauce: "Contraction"
You can't just use every single sound wave; that would be too many. So, they use a process called contraction.

• Imagine you have a huge bag of different colored marbles (the TSWs).
• You want to mix them together to create a few "super-marbles" that capture the most important features of the bag.
• The authors developed a mathematical recipe to mix these waves together. They minimize the "spillage"—a fancy way of saying they ensure no important information is left behind in the mixing process. They do this by focusing on the kinetic energy (how fast the electrons are moving), ensuring the new "super-marbles" (the new NAOs) are the most efficient possible.

---

Why Is This Better? (The "No Ghosts" Rule)

In the past, when scientists tried to improve these basis sets, they sometimes ran into a weird problem called spurious interactions.

The Analogy: The Hall of Mirrors
Imagine you are in a room with mirrors on all sides. If you stand in the middle, you see your reflection. But if the room is too small, you might see a reflection of your reflection, and it looks like there is a second person standing right next to you. In computer simulations, this is called a "periodic image."

If you use the old "infinite tiles" (Plane Waves) to build your basis set, the computer sometimes gets confused and thinks the atom is interacting with its own ghost reflection across the empty space. This creates fake chemical bonds that don't exist in reality.

The Fix:
Because the authors used Truncated Spherical Waves (the sound waves cut off at the wall), the "ghosts" are cut off too. The waves simply stop at the edge of the atom's neighborhood. This eliminates the fake interactions, making the simulation much more reliable, especially when looking at how atoms bond in different environments.

---

The Results: A Swiss Army Knife for Chemistry

The authors tested their new "Super-Lego Set" on two types of things:

1. Molecules: Like pairs of atoms stuck together (e.g., Oxygen gas).
2. Bulk Systems: Like solid crystals (e.g., Silicon chips or Salt).

They checked everything:

• Total Energy: How stable is the structure?
• Bond Lengths: How far apart are the atoms?
• Band Gaps: How does electricity flow through the material? (Crucial for making solar cells and computer chips).

The Verdict:

• Precision: Their new basis sets were incredibly accurate. They could predict bond lengths and energy levels with errors so small they were almost invisible (measured in "meV," which is a tiny fraction of an electron's energy).
• Transferability: This is the most important word. It means the Lego set works well whether you are building a tiny molecule or a giant crystal. Previous sets often worked great for one but failed for the other. This new set works for both.
• The "Unoccupied" States: Usually, these sets are great at describing electrons that are there, but bad at describing empty spots where electrons could go (which is needed for understanding light absorption and conductivity). By including "virtual states" (imaginary extra electrons) in their training, they made the set excellent at predicting these high-energy behaviors too.

Summary

Think of this paper as the invention of a universal, self-improving toolkit for simulating matter.

1. Old Way: You had to choose between a fast, rough tool (NAOs) or a slow, perfect tool (Plane Waves).
2. New Way: They built a tool that starts with a perfect, infinite set of waves, cuts them down to be small and fast, and then mathematically optimizes them so they never lose accuracy.
3. The Result: Scientists can now simulate complex materials (like new battery chemicals or solar cells) with high speed and high precision, without worrying about "ghost" atoms messing up the math.

It's like upgrading from a sketch artist who draws a house based on memory, to a 3D printer that can print a perfect house at the push of a button, using a blueprint that gets better every time you ask for more detail.

Technical Summary

1. Problem Statement

Numerical Atomic Orbitals (NAOs) are widely used in Density Functional Theory (DFT) calculations due to their computational efficiency and ability to enable linear-scaling methods. However, they suffer from a lack of systematic convergence compared to plane-wave (PW) or real-space grid bases.

• Limitations of Existing Schemes: Previous NAO generation methods (e.g., CGH, LRH) often rely on splitting valence shells and adding polarization orbitals with restricted angular momentum (e.g., $l+1$). This restriction hinders completeness, leading to energy gaps of tens of meV compared to PW bases.
• Transferability Issues: Including more reference states improves transferability, but using PW as an expansion basis for these references introduces spurious interactions between periodic images in vacuum regions, creating artificial bonding and distorting reference states.
• Unoccupied States: Standard NAOs often fail to accurately describe conduction bands and high-lying unoccupied states, which are critical for optical properties and transport calculations.

2. Methodology

The authors propose a new scheme to construct NAOs by contracting Truncated Spherical Waves (TSWs). The workflow involves four main steps:

A. Parametrization: Truncated Spherical Waves (TSW)

Instead of using Slater-type or pseudo-atomic orbitals as primitives, the method uses TSWs as building blocks.

• Form: The radial part is a linear combination of spherical Bessel functions $j_l(kr)$ truncated at a cutoff radius $r_c$.
• Smoothness: To avoid derivative discontinuities at $r_c$, the authors construct a subspace of TSWs where the first $M$ derivatives vanish at the cutoff (specifically $M=2$). These are termed Near-Complete Spherical Waves (NSW).
• Systematic Improvability: The NSW basis approaches the Complete Basis Set (CBS) limit as $r_c$, maximum angular momentum ($l_{max}$), and kinetic energy cutoff increase.

B. Optimization: Generalized Spillage Minimization

The contraction coefficients for transforming NSWs into NAOs are determined by minimizing a generalized spillage function:
$$\tilde{S} = \sum_{nk} \langle \psi^{NSW}_{nk} | [1 - \hat{P}_k] \hat{O} [1 - \hat{P}_k] | \psi^{NSW}_{nk} \rangle$$

• Operator ($\hat{O}$): The kinetic energy operator ($\hat{p}^2$) is chosen to minimize the trace of the kinetic operator in the residual space. This generalizes previous "spillage" minimization schemes.
• Reference States ($\psi^{NSW}_{nk}$): The method uses high-quality NSW wavefunctions calculated for a set of reference systems (homonuclear dimers with various bond lengths). Crucially, the reference states are expanded using NSW, not PW, to eliminate spurious periodic image interactions.
• Virtual States: To improve transferability for conduction bands, the scheme allows the inclusion of unoccupied (virtual) states in the spillage minimization.

C. Basis Set Hierarchy

The authors construct a hierarchical set of NAOs following Dunning's hierarchy:

• Minimal: Pseudopotential valence configuration.
• pVDZ: Polarized valence double-zeta.
• pVTZ: Polarized valence triple-zeta (includes higher angular momentum, e.g., $f$-orbitals).
• pVQZ=: Restricted angular momentum quadruple-zeta (for comparison).
• Variants: Special variants (pVTZ-2V, pVTZ-3V) are created by including 2x or 3x the number of virtual states in the spillage optimization.

3. Key Contributions

1. Novel Basis Construction: Introduction of Contracted TSWs as the primitive basis, bridging the gap between reference states and NAOs more effectively than PW-based schemes.
2. Elimination of Spurious Interactions: By using NSW (truncated in real space) rather than PW for generating reference states, the method avoids artificial bonding across vacuum boundaries, significantly improving transferability.
3. Generalized Spillage Minimization: A new objective function minimizing the kinetic energy trace in the residual space, which is more robust than previous overlap-based spillage definitions.
4. Systematic Improvability Strategy: A clear protocol for selecting $r_c$ and $l_{max}$ to achieve chemical accuracy (0.1 kcal/mol) for specific elements.
5. Enhanced Conduction Band Description: Demonstration that including virtual states in the spillage optimization significantly improves the description of unoccupied states without necessarily increasing the basis set size (though higher angular momentum is still preferred for certain systems).

4. Results and Benchmarks

The method was benchmarked against Plane Waves (PW) using the ABACUS code on 11 diatomic molecules and 26 bulk systems (covalent, ionic, semiconductors).

• Total Energy:
  • The pVTZ basis achieves chemical accuracy (< 0.042 eV/atom) for almost all systems.
  • The error decreases systematically from pVDZ $\to$ pVTZ $\to$ NSW.
  • NSW (primitive) shows negligible error (< 3 meV/atom) compared to PW, validating the truncation parameters.
• Structural Properties (Bond Lengths & Lattice Constants):
  • Mean Absolute Errors (MAE) for bond lengths are < 0.01 Å.
  • Lattice constant errors are < 0.05% for pVTZ, comparable to experimental standard deviations.
• Thermodynamic Properties:
  • Atomization/Cohesive Energies: pVTZ achieves chemical accuracy. The inclusion of higher angular momentum functions (e.g., $f$-orbitals in pVTZ) is shown to be more critical than simply adding more radial functions of lower angular momentum (pVQZ=).
  • Bulk Moduli: MAE is ~1 GPa, near experimental precision.
• Electronic Properties (Band Gaps & Band Structures):
  • Band Gaps: pVTZ reduces MAE to < 0.04 eV.
  • Conduction Bands ($\eta_{10}$ metric):
    • Standard pVTZ (occupied states only) struggles with high-lying conduction bands.
    • pVTZ-2V/3V (including virtual states) significantly narrows the error distribution for high-energy states.
    • However, pVQZ= (higher angular momentum) often remains the most accurate for conduction bands, suggesting that both angular momentum expansion and virtual state inclusion are valuable strategies.
  • Case Study (NaCl & GaN): Band structure comparisons show that pVQZ= and virtual-state-included bases provide qualitatively correct predictions up to 10–20 eV above the Fermi level, whereas minimal bases fail.

5. Significance

This work presents a robust, systematically improvable framework for generating NAO basis sets that rivals the accuracy of plane-wave methods while retaining the computational efficiency of localized bases.

• Transferability: The elimination of spurious periodic interactions makes these basis sets highly transferable across diverse chemical environments.
• Versatility: The method is effective for both ground-state properties (energies, structures) and excited-state properties (band structures, optical spectra) by leveraging virtual state inclusion.
• Practicality: The availability of the implementation in the open-source ABACUS code allows for large-scale, high-precision simulations in materials science and chemistry, facilitating the discovery of new materials with reliable electronic structure predictions.

In summary, the paper successfully addresses the historical trade-off between the efficiency of atomic orbitals and the systematic convergence of plane waves, providing a "best of both worlds" solution for modern computational materials science.