Range Separation in HSE06

• Range separation in HSE06 is a method that divides the Coulomb operator into short- and long-range components to balance accuracy and computational efficiency.
• The tuning of parameters, like the screening parameter ω and exchange fraction a, is key to optimizing predictions such as band gaps and lattice constants.
• Implementations leverage error function-based screening, confining exact (HF) exchange to short ranges, thereby significantly reducing computational demands.

Range separation in HSE06 refers to the systematic partitioning of the Coulomb operator into short-range and long-range contributions, combined with selective mixing of Hartree–Fock (HF) exchange and semilocal exchange, in order to optimize both the accuracy and computational tractability of hybrid density functionals for solids and molecules. HSE06 leverages the error function to confine the exact exchange to short interelectronic distances, while relegating long-range exchange to efficient semilocal approximations, significantly accelerating simulations and improving band-gap predictions relative to global hybrids. The parameterization and performance trade-offs of this approach—most notably the choice of the screening parameter ω and the exchange mixing fraction a—have been widely characterized and refined in the literature, providing a rigorous framework for hybrid functional development.

1. Mathematical Framework of Range Separation

In HSE06, the electron–electron Coulomb interaction is split as:

$$\frac{1}{r} = \frac{\operatorname{erfc}(\omega r)}{r} + \frac{\operatorname{erf}(\omega r)}{r}$$

• $\operatorname{erfc}(\omega r)/r$ defines the short-range (SR) part.
• $\operatorname{erf}(\omega r)/r$ defines the long-range (LR) part.

The HSE06 exchange–correlation energy becomes:

$$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$

• $a$: Fraction of short-range Fock exchange.
• $E_{x}^{\text{HF,SR}}(\omega)$: HF exchange energy calculated with the SR kernel.
• $E_{x}^{\text{PBE,SR}}(\omega)$ and $E_{x}^{\text{PBE,LR}}(\omega)$: PBE exchange energy evaluated on the SR and LR kernels, respectively.
• $E_c^\text{PBE}$: Full-range PBE correlation.

This structure excludes any Fock exchange at long range, fully delegating it to the semilocal approximation (Yang et al., 2023, Moussa et al., 2012). For HSE06, the canonical parameters are $a = 0.25$ (matching PBE0 at $\operatorname{erfc}(\omega r)/r$0) and $\operatorname{erfc}(\omega r)/r$1 (screening length $\operatorname{erfc}(\omega r)/r$2), or equivalently $\operatorname{erfc}(\omega r)/r$3 (Moussa et al., 2012, Yang et al., 2023).

2. Parameterization and Physical Rationale

Parameter selection in HSE06 is guided by accuracy in benchmark properties and computational efficiency.

• Exchange mixing ($\operatorname{erfc}(\omega r)/r$4): Retained at $\operatorname{erfc}(\omega r)/r$5, corresponding to 25% SR Fock exchange, aligning with PBE0 for correct on-top exchange.
• Screening parameter ($\operatorname{erfc}(\omega r)/r$6): Set at $\operatorname{erfc}(\omega r)/r$7 in HSE06 following optimization for semiconductor lattice constants and band gaps. The screening length ($\operatorname{erfc}(\omega r)/r$8) delineates the spatial range where exact exchange is retained (Moussa et al., 2012).

Optimizing these parameters involves sampling the two-dimensional $\operatorname{erfc}(\omega r)/r$9 space and quantifying accuracy with metrics such as mean absolute error (MAE), mean error (ME), and root mean square error (RMSE) across benchmarks for band gaps, lattice constants, formation energies, reaction barriers, and bond lengths.

The relationship between parameter choices and computed properties is nontrivial: decreasing $\operatorname{erf}(\omega r)/r$0 extends the range of HF mixing, which typically narrows band-gap error up to a limit; excessive HF exchange at long range leads to gap overestimation (Moussa et al., 2012).

3. Influence on Computational Cost and Accuracy

The core efficiency gain from range separation is that only the SR component of Fock exchange, which decays rapidly with distance, is computed explicitly and scales favorably with system size. The computational cost for evaluating $\operatorname{erf}(\omega r)/r$1 scales with the effective cutoff radius $\operatorname{erf}(\omega r)/r$2 (empirically $\operatorname{erf}(\omega r)/r$3–$\operatorname{erf}(\omega r)/r$4). Halving the screening length (doubling $\operatorname{erf}(\omega r)/r$5) reduces $\operatorname{erf}(\omega r)/r$6 accordingly and typically yields a $\operatorname{erf}(\omega r)/r$7–$\operatorname{erf}(\omega r)/r$8 reduction in calculation time for the Fock exchange, as observed in both plane-wave (VASP, $\operatorname{erf}(\omega r)/r$9) and Gaussian-based codes ($$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$0) (Moussa et al., 2012, Kokott et al., 5 May 2025, Sun, 2023).

The accuracy–cost trade-off is explicit:

• Lower $$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$1 (less screening): improved description for properties sensitive to long-range exchange (e.g., molecular formation energies), but increased cost and risk of spurious long-range effects.
• Higher $$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$2 (more screening): reduced cost and improved convergence in periodic systems, acceptable overall accuracy for band gap and structural predictions (Moussa et al., 2012, Kokott et al., 5 May 2025).

4. Optimal Tuning and Alternative Range-Separation Prescriptions

The canonical HSE06 employs fixed parameters, but recent advances define optimal, system-dependent $$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$3 values. In periodic RSH functionals, $$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$4 can be tuned by seeking stationary points where the DFT gap and $$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$5-corrected gap respond identically to $$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$6 adjustments:

• Stationary-point tuning: Solve $$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$7, where $$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$8. This yields material-specific $$E_{xc}^{\text{HSE}}(\omega, a) = a E_{x}^{\text{HF,SR}}(\omega) + (1-a) E_{x}^{\text{PBE,SR}}(\omega) + E_{x}^{\text{PBE,LR}}(\omega) + E_c^\text{PBE}$$9 aligning RSH and quasiparticle gaps, often delivering agreement with self-consistent GW at lower computational cost (Li et al., 2021).

Numerical examples illustrate that the optimal $a$0 in semiconductors (e.g., diamond, Si, SiC) is substantially larger than the HSE06 default, indicating a need for more aggressive SR exchange to open the gap correctly. For wide-gap insulators, $a$1 is closer to the standard value (Li et al., 2021).

5. Implementation in Electronic Structure Codes

Implementing range separation in HSE06 within atom-centered and plane-wave frameworks requires specialized handling of SR and LR Coulomb kernels in electron-repulsion integrals (ERIs):

• SR ERIs: Calculated analytically with modified Rys quadrature or as full-range minus LR integrals (Sun, 2023).
• LR ERIs: Efficiently approximated via density fitting (DF) methods, using a small auxiliary basis due to the smoothness of the LR kernel (Sun, 2023).
• Exchange–correlation assembly: Only SR HF exchange is mixed ($a$2), with semilocal exchange for all LR contributions; correlation is treated without range separation (Yang et al., 2023, Sun, 2023).
• Control of truncation and auxiliary basis: Auxiliary set sizes and eigenvalue truncation thresholds can be tuned to balance cost and accuracy; error bounds of $a$3 per atom are achievable (Sun, 2023).
• Pseudopotentials: All-electron atomic equations are solved with the range-separation kernel embedded, requiring multipole expansions for SR Fock terms (Yang et al., 2023).

6. Variants and Extensions of HSE06 Range Separation

Systematic exploration of the $a$4 parameter space revealed optimal and range-minimized variants:

• HSE12: $a$5 minimizes aggregate MAE across molecular and solid-state benchmarks.
• HSE12s: $a$6, halves the SR Fock screening length relative to HSE06, increasing speed by $a$7 at minimal accuracy loss. For HSE12s vs. HSE06: gap MAE changes from $a$8 to $a$9, but formation-energy MAE slightly increases ($E_{x}^{\text{HF,SR}}(\omega)$0) (Moussa et al., 2012).

There is no globally optimal $E_{x}^{\text{HF,SR}}(\omega)$1; selection must consider application-specific priorities.

7. Approximations and Accelerated Hybrid Schemes

Further computational acceleration can be achieved by analytical approximations to the LR kernel. The extended screening function of Kokott et al. replaces $E_{x}^{\text{HF,SR}}(\omega)$2 in Fock integrals with

$E_{x}^{\text{HF,SR}}(\omega)$3

recovering the bulk of missing LR exchange with nearly the speed of HSE06. For semiconductors and organic crystals, this approximation (denoted PBE0′) yields band gap errors $E_{x}^{\text{HF,SR}}(\omega)$4 and total-energy errors $E_{x}^{\text{HF,SR}}(\omega)$5 with $E_{x}^{\text{HF,SR}}(\omega)$6–$E_{x}^{\text{HF,SR}}(\omega)$7 speedup relative to full-range hybrid calculations (Kokott et al., 5 May 2025).

Table: Summary of Key HSE06-Like Functionals and Parameter Regimes

Functional	ω (Å⁻¹)	a (Fock frac.)	Screening Length (Å)	Main Features
HSE06	0.208	0.25	4.8	Standard; band gap/lattice
HSE12	0.185	0.313	5.4	Lower MAE in molecules/solids
HSE12s	0.408	0.425	2.45	Half-range, accelerated

All error metrics, implementation recipes, and parameter rationales are detailed in (Moussa et al., 2012).

---

The conceptual framework and parameterization of range separation in HSE06 have established the method as a benchmark for balanced hybrid exchange in electronic structure, enabling improved predictive performance for solids and molecules under computational constraints, and continue to motivate systematic extensions and optimal tuning protocols for advanced materials simulation (Moussa et al., 2012, Yang et al., 2023, Li et al., 2021, Kokott et al., 5 May 2025, Sun, 2023).