r2SCAN-D4: Dispersion-Corrected Meta-GGA DFT

• r2SCAN-D4 is a dispersion-corrected meta-GGA density functional that integrates the regularized r2SCAN framework with D4 dispersion corrections to accurately model both molecular and solid-state systems.
• It restores all SCAN constraints through minimal regularization, ensuring numerical stability on standard grids while maintaining efficient computational cost.
• Benchmark studies show r2SCAN-D4 delivers competitive accuracy across thermochemical, noncovalent, and materials properties, offering improved transferability over conventional GGAs.

r2SCAN-D4 is a dispersion-corrected meta-generalized gradient approximation (meta-GGA) density functional designed for general chemical, solid-state, and molecular applications. It combines the regularized r2SCAN functional—which is based on the SCAN meta-GGA but with improved numerical stability and exact constraint satisfaction—with the atom-pairwise and three-body D4 London dispersion correction. This construction aims to deliver the reliability, non-empirical constraint satisfaction, and transferability of the SCAN family, augmented by accurate long-range van der Waals interactions, at a computational cost comparable to standard GGAs (Ehlert et al., 2020, Ning et al., 2023, Attarian et al., 2024).

1. Theoretical and Mathematical Foundations

r2SCAN-D4’s total energy is the sum of the r2SCAN exchange–correlation functional, the D4 dispersion energy, and standard Kohn–Sham terms:

$$E_{\text{r}^2\text{SCAN-D4}} = E_{\text{r}^2\text{SCAN}}^{\text{XC}}[\rho] + E_{\text{disp}}^{\text{D4}} + E_{\text{other KS terms}}$$

where $E_{\text{other KS terms}}$ accounts for $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$.

r2SCAN Exchange–Correlation

The r2SCAN functional writes its semilocal exchange–correlation energy in the meta-GGA form:

$$E_{xc}[\rho] = \int \rho(\mathbf{r})\, \epsilon_{xc} (\rho, \nabla\rho, \tau)\, d^3r$$

where $$\tau(\mathbf{r}) = \frac{1}{2} \sum_n |\nabla \phi_n|^2$$ (positive-definite kinetic energy density). The “iso-orbital indicator” $\bar{\alpha}$, regularized in r2SCAN, distinguishes between single-orbital and uniform-electron-gas limits:

$$\bar{\alpha} = \frac{\tau - \tau_W}{\tau_U + \eta\,\tau_W}$$

with

$$\tau_W = \frac{|\nabla\rho|^2}{8\rho}, \quad \tau_U = \frac{3}{10}(3\pi^2)^{2/3}\rho^{5/3}, \quad \eta=10^{-3}$$

All pieces of SCAN’s exchange and correlation enhancement factors are retained, but all dependencies on $\alpha$ use the regularized $\bar{\alpha}$, and polynomial interpolation is restored for correct gradient expansion recovery.

D4 Dispersion Correction

The D4 model (Caldeweyher et al., 2019) adds atom-pairwise and three-body London (van der Waals) terms. Its total dispersion energy is:

$E_{\text{other KS terms}}$0

where $E_{\text{other KS terms}}$1 are geometry-adapted dispersion coefficients computed from atomic polarizabilities, $E_{\text{other KS terms}}$2 is the interatomic distance, $E_{\text{other KS terms}}$3 is the Becke–Johnson damping, and the three-body term involves the Axilrod-Teller-Muto form. The D4 parameters for r2SCAN are typically $E_{\text{other KS terms}}$4, $E_{\text{other KS terms}}$5, $E_{\text{other KS terms}}$6, $E_{\text{other KS terms}}$7, $E_{\text{other KS terms}}$8 (Ning et al., 2023, Attarian et al., 2024).

2. Regularization and Numerical Stability

The original SCAN meta-GGA is known for exact satisfaction of 17 known density functional constraints but exhibits numerical instabilities—particularly due to piecewise interpolation over the iso-orbital indicator $E_{\text{other KS terms}}$9. The rSCAN version addressed this by ad hoc smoothing but at the cost of violating some constraints. r2SCAN restores all SCAN constraints by applying a minimal regularization ($$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$0) in the denominator of $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$1 and correcting the enhancement factors to recover the second-order slowly-varying gradient expansion limit. r2SCAN’s smoother interpolation leads to robust numerical behavior on medium grids and standard SCF convergence criteria, eliminating the need for exceptionally dense numerical quadrature setups typically required by SCAN (Ehlert et al., 2020).

3. Computational Implementation

Integration Grids and SCF Convergence

r2SCAN-D4 achieves nearly GGA-level stability and efficiency:

• In Turbomole, “grid m4” (6 radial × up to ~50 angular points) with the default radial size suffices for routine tasks; grid size only requires escalation in rare cases with extreme density features.
• Self-consistent-field convergence of $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$2 Hartree (or $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$3 eV in VASP) is recommended throughout.

Basis Sets and Plane-Wave Cutoffs

• The def2-QZVP basis yields near basis-set convergence for molecular tasks.
• For plane wave DFT (e.g., VASP), cutoffs of 600 eV ensure convergence with r2SCAN-D4 for transition metals and ionic systems (Ning et al., 2023, Attarian et al., 2024).

Performance and Cost

• On large molecular complexes, r2SCAN-D4 delivers results $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$4 faster than SCAN for identical grid parameters while retaining accuracy.
• The D4 correction introduces negligible overhead, while r2SCAN meta-GGA is only $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$5 more expensive per SCF step than PBE (Attarian et al., 2024).

4. Benchmark Accuracy Across Chemical Space

r2SCAN-D4 demonstrates benchmark-level accuracy for diverse chemical and condensed-matter systems, summarized in the following table (all errors are mean absolute deviations, unless noted):

Property/Class	Error/Metric	Comparator Performance
Main group bonds (LMGB35/HMGB11)	0.7 pm	Competitive w/ PBE0-D4, TPSS-D4
Transition metal complexes (TMC32)	1.9 pm	Outperforms hybrid functionals
General thermochemistry (GMTKN55)	WTMAD2 = 7.5 kcal/mol	Best non-hybrid meta-GGA
Organometallic (MOR41)	3.3 kcal/mol	Nearly matches PBE0-D4
Large noncovalents (S30L, L7, X40×10)	0.36–1.8 kcal/mol	Superior to SCAN/rSCAN-D4
Molecular crystals (X23+ICE10)	0.7 kcal/mol	Within chemical accuracy
Molten salt (e.g., NaCl density, 1200K)	–2.7% (RMSE ≈ 0.03 g/cm³)	Superior to PBE-D3/D4, rVV10

For highly correlated solids (e.g., YBa₂Cu₃O₆), r2SCAN-D4, possibly augmented with a modest Hubbard $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$6, achieves phonon frequencies and lattice constants within 1 meV and 0.01 Å, respectively, of experimental reference values (Ning et al., 2023). For thermophysical properties of molten salts (NaCl, MgCl₂), r2SCAN-D4 yields densities and viscosities typically within 3% and 10% of experiment, outperforming alternatives such as PBE-D3/D4 and r2SCAN-rVV10 (Attarian et al., 2024).

5. Application Scope and Comparative Analysis

Recommended Use-cases:

• Main group and transition metal molecules, including organometallic and catalytic reactions;
• Large supramolecular complexes, non-covalent interactions in solution and crystal environments;
• Molecular crystals, polymorph energy ranking, and lattice properties;
• Solid-state phases requiring meta-GGA-level correlation with minimal computational burden.

A key advantage is r2SCAN-D4’s constraint-based, fully non-empirical construction, leading to robust transferability across a wide range of chemical bonding motifs. The D4 dispersion model’s geometry-adaptive coefficients allow reliable treatment from ions and metals to neutral organic and inorganic systems.

Limitations and Caveats:

• r2SCAN-D4 systematically underestimates band gaps, being a semilocal functional without exact exchange.
• Persistent overbinding (by ~1 kcal/mol) remains for strongly hydrogen-bonded ices and possibly for certain other self-interaction–driven phases.
• In rare circumstances—such as systems with exceptionally sharp density variations—grid refinement may be required.

For molten salts, r2SCAN-D4 provides the best available accuracy among tested GGAs/meta-GGAs, though implementation cost is increased compared to PBE. A plausible implication is that its broader adoption in materials modeling is facilitated by the balance of speed and predictive accuracy (Attarian et al., 2024).

6. Extensions and Practical Guidance

Best practices include:

• In machine-learning interatomic potential fitting protocols, data generated at r2SCAN-D4 level for end members and two intermediate compositions suffice for chemically transferable potential construction.
• For “ab initio” lattice-dynamical studies, augmenting r2SCAN-D4 with moderate on-site $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$7 on correlated $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$8-shells yields optimal agreement with experiment, with lower $$T_s[\{\phi_i\}] + E_{\text{Coulomb}} + E_{\text{ext}}$$9 required than in PBE-based protocols (Ning et al., 2023).
• The D4 parameters do not require further tuning; standard values are transferable for both molecular and periodic systems.

Practical DFT+ML-AIMD workflows using r2SCAN-D4–level data are recommended for studies targeting experimental accuracy in thermodynamic and transport properties of ionic melts and molecular crystals (Attarian et al., 2024).

7. Comparative Summary Table

Functional	D4 Dispersion	Numerical Cost	Best/Typical Use-cases	Limitations
r2SCAN-D4	Yes	3–5× PBE	Thermochemistry, NCIs, solids	Bandgap underestimate, rare grids
PBE-D4	Yes	1× PBE	Standard DFT tasks; reference GGA	NaCl densities underbound
r2SCAN-rVV10	rVV10 kernel	5× PBE	Some ionic/organic crystals	Density overestimates in MgCl₂
SCAN-D4	Yes	10–20× PBE	Accurate but slow, problematic	Numerically unstable on grids

This comparative view reinforces r2SCAN-D4 as a reliable, broadly applicable meta-GGA for molecular, materials, and condensed-phase computational studies, with particular strengths in systems where both semilocal correlation and long-range dispersion must be treated accurately and efficiently (Ehlert et al., 2020, Ning et al., 2023, Attarian et al., 2024).