Projector Augmented-Wave Method

• The projector augmented-wave method is a hybrid DFT technique that reconstructs all‐electron wavefunctions by augmenting smooth pseudo solutions with atom‐centered corrections.
• It efficiently computes X‐ray structure factors and other observables by combining FFT on smooth densities with one-dimensional radial grids to capture core features.
• The method delivers benchmark-level accuracy comparable to full all‐electron approaches while significantly reducing computational cost and memory usage.

The projector augmented-wave (PAW) method is a formally exact approach for reconstructing all-electron (AE) electronic structure information from computations that are otherwise based on smooth pseudo-wavefunctions. Developed to overcome the computational inefficiencies and accuracy limitations of norm-conserving pseudopotentials and other pseudization schemes in density functional theory (DFT), PAW operates by augmenting smooth pseudo solutions with atom-centered corrections—restoring both core and valence features of the true AE wavefunctions. This hybrid formalism preserves computational efficiency while providing access to physical observables that are highly sensitive to the details of the electronic density near nuclei, such as X-ray structure factors and core spectroscopies (Shi et al., 2022).

1. Theoretical Foundations and PAW Formalism

The PAW method is fundamentally based on a linear, invertible transformation operator $\mathcal T$ that maps smooth pseudo-wavefunctions, $\{\tilde\psi_n\}$, to corresponding AE wavefunctions, $\{\psi_n\}$. Explicitly,

$$\psi_n(\mathbf r) = \tilde\psi_n(\mathbf r) + \sum_{j,u} \left[\phi_u^j(\mathbf r-\mathbf R_j) - \tilde\phi_u^j(\mathbf r-\mathbf R_j)\right]\, \langle \tilde p_u^j | \tilde\psi_n \rangle\,,$$

where, for each atom $j$, the set $\{\phi_u^j\}$ are AE partial waves, $\{\tilde\phi_u^j\}$ are matching pseudo partial waves, and $\{\tilde p_u^j\}$ are projector functions dual to $\{\tilde\phi_u^j\}$ and biorthogonal within the augmentation spheres: $$\langle \tilde p^j_u | \tilde\phi^k_{u'}\rangle = \delta_{jk}\,\delta_{uu'}$$ (Shi et al., 2022).

The full AE valence density is obtained by combining a smooth pseudo density with atom-centered augmentation contributions and (under the frozen-core approximation) a spherical core density: $$n(\mathbf r) = \tilde n_{\rm val}(\mathbf r) + \sum_{i,j}\left[n^1_{ij}(\mathbf r)-\tilde n^1_{ij}(\mathbf r)\right] D_{ij} + \sum_{j}\rho^j_{\rm core}(|\mathbf r-\mathbf R_j|)\,,$$ with $\tilde n_{\rm val} = \sum_n f_n|\tilde\psi_n|^2$ and $D_{ij}$ density-matrix elements formed from projector overlaps. The differences $(n^1_{ij} - \tilde n^1_{ij})$ systematically restore the atomic-scale AE features lost in the pseudization (Shi et al., 2022).

2. Efficient Computation of Structure Factors

Structure factors, $F(\mathbf G)$—i.e., the Fourier coefficients of the AE density—are fundamental to the interpretation of X-ray and electron diffraction experiments. In PAW, $F(\mathbf G)$ is decomposed as: $$F(\mathbf G) = \tilde F_{\rm val}(\mathbf G) + F_{\rm aug}(\mathbf G) + F_{\rm core}(\mathbf G)\,,$$ where

• $\tilde F_{\rm val}(\mathbf G)$ is obtained via standard FFT of the smooth pseudo-density on the default grid.
• $F_{\rm core}(\mathbf G)$ sums contributions from spherically symmetric core densities, evaluated analytically: $$F_{\rm core}^j(\mathbf G) = 4\pi \int_0^\infty \rho_{\rm core}^j(r)\, r^2\, j_0(2\pi G r) dr$$ with $j_0$ the spherical Bessel function.
• $F_{\rm aug}(\mathbf G)$ comprises the atom-centered augmentation terms, efficiently computed by expanding product harmonics into spherical harmonics and carrying out one-dimensional radial integrals over logarithmic meshes: $$f^j_{u_1u_2}(\mathbf G) = \rho^j_{u_1u_2}\, \sum_L C^{m_1,m_2}_{l_1,l_2,L} Y_{L,M}(\hat{\mathbf G})\, 4\pi\, i^L \int_0^{r_c} j_L(2\pi G r)\, \Delta R^j_{u_1u_2}(r)\, r^2 dr$$ where $\Delta R^j_{u_1u_2}(r)$ is the radial part of the AE-pseudo partial-wave product difference (Shi et al., 2022).

This organization eliminates the need for reconstructing the AE density on ultra-dense three-dimensional real-space grids. Only the smooth component requires FFT on the default grid; all rapidly varying contributions are captured by 1D radial grids concentrated near the nucleus, drastically reducing computational cost and memory.

3. Computational Workflow and Scalability

The PAW structure factor algorithm achieves high efficiency by:

• Avoiding dense 3D grid reconstruction; only $\tilde n_{\rm val}$ is placed on the grid for FFT.
• Evaluating all on-site (core and augmentation) terms on small, atom-centered logarithmic meshes ($N_{\rm rad}\lesssim 200$) and summing over a modest number of projectors.
• Scaling linearly in the number of atoms and number of scattering vectors $N_G$ of experimental interest.
• Cutting peak RAM requirements by orders of magnitude compared to full AE grid-based schemes, as only analytic and 1D radial data must be held in memory.

Timings show PAW structure factor computations are one to two orders of magnitude faster than grid-based AE density evaluation while maintaining all-electron accuracy for diffraction intensities (Shi et al., 2022).

4. Validation, Accuracy, and Limitations

The PAW-based scheme has been validated within CASTEP against AE DFT (APW+lo, WIEN2k) structure factors for elemental and binary crystals (diamond-Si, hcp Mg, rocksalt MgO) with $R$-factors of only $0.02$–$0.08$% when AE augmentation charges are employed (as opposed to pseudized augmentation, which yields errors up to $3$%). Against high-precision X-ray experiment (with Debye–Waller corrections), the $R^{\rm EXP}$ values remain $0.3$–$0.5$% (essentially indistinguishable from AE DFT), while pseudized approximations deteriorate performance ($R^{\rm EXP}>1$%) (Shi et al., 2022).

The method requires full AE one-center expansion data for each PAW dataset (i.e., the AE augmentation functions must be available for all relevant channels). The frozen-core approximation presents a limitation—if core radii are too small to retain deeper (semi-core) states, errors may be introduced (Shi et al., 2022).

5. Comparison to Alternative Approaches

Table: Structure Factor Computation Schemes

Approach	Accuracy	Efficiency	AE charge required?
Grid-based AE DFT (APW+lo)	All-electron	Extremely high cost	Yes
Pseudized USPP/PAW (no AE aug)	Up to 3% error	Lower cost, but inaccurate	No
PAW w/ AE augmentation (this work)	All-electron	10–100× faster vs. grid	Yes

The substantial computational gains are unique to the PAW augmentation scheme exploiting the locality and radial support of the AE corrections. Conventional pseudopotential-based approaches fail to achieve comparable accuracy for structure factors due to missing AE density near nuclei (Shi et al., 2022).

6. Extensions and Impact

The PAW framework is extensible to a wide range of electronic structure observables sensitive to the AE density or wavefunction character—such as core-level spectroscopy, electron momentum densities, and improved electrostatic corrections. PAW's formalism is compatible with plane-wave and real-space DFT codes, including implementations that leverage modern parallel hardware. The analytic separation and radial-grid techniques established in the structure factor context are directly transferable to other observables with similar core-localized features (Shi et al., 2022).

By enabling routine, benchmark-level AE accuracy for experimental quantities at manageable computational cost, the PAW method has established itself as the standard for high-precision DFT calculations in plane-wave codes for materials science and condensed matter physics.