Spherically Averaged Ewald Interaction

• Spherically Averaged Ewald Interaction is a radial effective potential obtained by averaging the anisotropic periodic Coulomb interaction over all orientations, yielding an isotropic pair potential.
• It accurately reproduces the Ewald energy for homogeneous systems, such as the uniform electron gas, and significantly reduces computational cost in Monte Carlo simulations.
• Its integration in ab initio path integral Monte Carlo accelerates free-energy calculations while maintaining high precision with effective finite-size corrections.

The spherically averaged Ewald interaction is a radial effective interaction derived from the periodic Ewald Coulomb potential by averaging over the orientation of the interparticle separation at fixed scalar distance. In recent warm-dense-matter and uniform-electron-gas work, the term denotes in particular the Yakub–Ronchi (YR) potential, a spherically symmetric, distance-dependent pair potential that reproduces the Ewald energy of a periodic Coulomb system to high accuracy for homogeneous systems while being much cheaper to evaluate numerically than a full Ewald sum (Svensson et al., 17 Jul 2025, Dornheim et al., 1 Apr 2025). Its central role is methodological: it replaces the anisotropic, lattice-dependent finite-size Ewald kernel by a scalar pair interaction that is especially effective in bosonic or classical sectors of Monte Carlo calculations, but it does not eliminate the need for the full Ewald interaction when long-wavelength, small-$q$, or explicitly anisotropic physics must be resolved.

1. Definition and mathematical construction

For a periodic Coulomb system in a cubic box of side length $L$ and volume $V=L^3$, the standard Ewald interaction can be written in compact form as

$$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$

with

$$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$

At finite system size this kernel is not spherically symmetric, because the periodic image sum retains the anisotropy of the simulation cell (Svensson et al., 17 Jul 2025).

The spherical average is obtained by averaging $v_{\text{Ew}}(\mathbf r)$ over all orientations $\hat{\mathbf r}$ at fixed $r=|\mathbf r|$,

$$v_{\text{YR}}(r)=\frac{1}{4\pi}\int d\Omega_{\hat{\mathbf r}}\, v_{\text{Ew}}(\mathbf r).$$

Yakub and Ronchi then fit this angle-averaged potential by a simple analytic function of $r$, yielding an effective pair potential

$L$0

with a suitable system-size-dependent constant chosen so that $L$1 for homogeneous systems (Svensson et al., 17 Jul 2025).

In the explicit warm-dense-electron-gas implementation, the YR pair potential is written as

$L$2

with

$L$3

This construction preserves the short-range Coulomb behavior, enforces a finite range comparable to the box size, and replaces the cubic anisotropy of the periodic Ewald potential by a central interaction (Dornheim et al., 1 Apr 2025).

The underlying assumptions are explicit: spatial homogeneity and isotropy, periodic boundary conditions, overall neutrality, and a thermodynamic-limit regime in which the YR potential tends to the pure Coulomb form $L$4 and the difference from the full Ewald interaction vanishes (Svensson et al., 17 Jul 2025).

2. Relation to standard Ewald and to other radial Ewald reformulations

The spherically averaged Ewald interaction is not a new physical interaction; it is an alternative representation of periodic Coulomb electrostatics designed to suppress finite-cell anisotropy that is unphysical for homogeneous systems. In this sense it sits between the full periodic Ewald kernel and purely real-space truncation schemes. The key distinction from standard Ewald is therefore not the underlying Coulomb law but the treatment of the long-range periodic image contribution: standard Ewald preserves the full lattice geometry, whereas spherical averaging replaces that geometry by an isotropic effective environment (Svensson et al., 17 Jul 2025).

Related real-space constructions make the same idea explicit. A purely real-space electrostatic summation in a uniform neutralizing background can be written as a damped, spherically truncated pair sum plus local analytic corrections. There the short-range interaction is the real-space Ewald kernel $L$5, while the omitted long-range part is represented by a spherically averaged correction associated with a neutralizing sphere or shell (Pickard, 2018). This provides a concrete geometric interpretation: the long-range Ewald tail is replaced by an isotropic, sphere-based representation rather than an explicit reciprocal-space lattice sum.

Generalized Ewald kernels for $L$6 interactions admit the same operation. In three dimensions, the reciprocal-space part of the generalized kernel can be angle-averaged to a radial factor $L$7, yielding an explicitly radial effective interaction built from incomplete gamma functions in real space and exponential-integral coefficients in reciprocal space (Osychenko et al., 2011). Extended Ewald constructions based on linear combinations of Gaussian screening functions likewise preserve a radial real-space kernel and improve $L$8-space convergence without changing the conceptual status of the spherical average (Kylänpää et al., 2014).

A common misconception is that spherical averaging is equivalent to uncontrolled truncation. In the uniform-electron-gas free-energy framework this is incorrect: the YR interaction is introduced only as an intermediate artificial reference, and the remaining YR–Ewald difference is corrected exactly in a separate extended-ensemble step (Svensson et al., 17 Jul 2025).

3. Role in ab initio path integral Monte Carlo

The most developed recent use of the spherically averaged Ewald interaction is in ab initio PIMC for the uniform electron gas. There the physical Hamiltonian retains the full Ewald interaction,

$L$9

while the artificial interaction is chosen as

$V=L^3$0

The purpose is to accelerate free-energy estimation in the bosonic sector, where the cost of evaluating the Ewald interaction dominates nearly every Monte Carlo move (Svensson et al., 17 Jul 2025).

The free-energy calculation is formulated with an extended ensemble

$V=L^3$1

An $V=L^3$2-ensemble connects the ideal and interacting systems through

$V=L^3$3

and in the accelerated scheme the entire bosonic interaction contribution is evaluated with $V=L^3$4. The remaining difference between YR and Ewald is then recovered through an $V=L^3$5-ensemble,

$V=L^3$6

with

$V=L^3$7

found to be sufficient because the two interactions are energetically very close (Svensson et al., 17 Jul 2025).

The full exchange–correlation free energy is decomposed as

$V=L^3$8

so the spherical average enters only in the bosonic contribution. The fermionic sign contribution is still computed with the physical Ewald interaction, and a separate $V=L^3$9-extrapolation technique is used to alleviate the fermion sign problem for large $$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$0 (Svensson et al., 17 Jul 2025).

Implementation is carried out in the ISHTAR PIMC code with the canonical version of the worm algorithm and primitive factorization,

$$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$1

In this setting, the YR interaction is a classical pair potential at each imaginary-time slice, and because $$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$2 is a simple algebraic function of one scalar distance, its evaluation is far cheaper than the combined real- and reciprocal-space Ewald calculation (Svensson et al., 17 Jul 2025).

4. Accuracy, finite-size behavior, and thermodynamic limit

The central validation result is that the accelerated YR-based free-energy route and the Ewald-only route give indistinguishable total free energies within statistical error for small systems. In particular, the decomposition through $$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$3 introduces no systematic bias relative to direct Ewald evaluation (Svensson et al., 17 Jul 2025).

The correction that measures the difference between the spherical average and the full Ewald interaction,

$$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$4

is about three orders of magnitude smaller than the total interaction contribution and vanishes with increasing $$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$5. For $$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$6, its finite-size decay is reported as $$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$7 with $$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$8, while the exponent increases at stronger coupling, such as $$V_{\text{Ew}} = \frac12 \sum_{i\neq j} q_i q_j\, v_{\text{Ew}}(\mathbf r_{ij}) + V_{\text{self}} + V_{\text{background}},$$9 (Svensson et al., 17 Jul 2025).

Finite-size behavior is especially important because the YR potential is constructed from the periodic Ewald interaction itself. The practical consequence is that standard finite-size correction strategies remain effective. Using the GDSMFB free-energy correction scheme, the finite-size correction removes about $$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$0 of the finite-size error already at the smallest $$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$1 used for $$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$2, and the residual effect decays approximately as $$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$3 (Svensson et al., 17 Jul 2025).

Combined with $$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$4-extrapolation, this framework enabled free-energy evaluation for $$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$5 electrons, with finite-size and statistical errors below chemical accuracy. After finite-$$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$6 and finite-size corrections, the results agree with the GDSMFB parametrization to within about $$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$7, and the residual finite-size error at $$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$8 is below $$v_{\text{Ew}}(\mathbf r) = \sum_{\mathbf n\in\mathbb Z^3} \frac{\mathrm{erfc}\!\left(\alpha |\mathbf r+\mathbf nL|\right)}{|\mathbf r+\mathbf nL|} + \frac{4\pi}{V}\sum_{\mathbf k\neq 0}\frac{1}{k^2} \exp\!\left(-\frac{k^2}{4\alpha^2}\right)\cos(\mathbf k\cdot \mathbf r) -\frac{\alpha}{\sqrt{\pi}}.$$9 per electron (Svensson et al., 17 Jul 2025). This strongly supports the interpretation of the spherically averaged Ewald interaction as an efficient intermediate representation rather than a different thermodynamic limit.

5. Observable-dependent strengths and limitations

The most important limitation of the spherically averaged Ewald interaction emerges for $v_{\text{Ew}}(\mathbf r)$0-resolved observables. In warm-dense-electron-gas PIMC, integrated quantities such as the kinetic energy and potential energy are reproduced very accurately with respect to Ewald reference data, with deviations of order $v_{\text{Ew}}(\mathbf r)$1–$v_{\text{Ew}}(\mathbf r)$2 for the systems studied. By contrast, wave-number-resolved observables fluctuate for small $v_{\text{Ew}}(\mathbf r)$3 (Dornheim et al., 1 Apr 2025).

For the static structure factor $v_{\text{Ew}}(\mathbf r)$4 and the static linear density response $v_{\text{Ew}}(\mathbf r)$5, the YR interaction produces pronounced oscillations around a characteristic wave number

$v_{\text{Ew}}(\mathbf r)$6

associated with the finite interaction range and with unavoidable double counting of interactions when particles and their images both lie within the volume-equivalent sphere. Ewald results remain smooth and essentially size-independent, whereas YR results show suppression and overshoot near $v_{\text{Ew}}(\mathbf r)$7 (Dornheim et al., 1 Apr 2025).

The same distinction appears dynamically. Analytic continuation of the imaginary-time density–density correlation function shows that the dynamic structure factor $v_{\text{Ew}}(\mathbf r)$8 can differ appreciably between Ewald and YR at small and intermediate $v_{\text{Ew}}(\mathbf r)$9. In the collective regime, the plasmon peak can be more strongly damped or shifted relative to the Ewald result, while at larger $\hat{\mathbf r}$0 the differences essentially vanish within reconstruction uncertainties (Dornheim et al., 1 Apr 2025).

Nonlinear response is even more sensitive. The static quadratic density response $\hat{\mathbf r}$1 shows marked size dependence with YR and significant deviations from Ewald near the maximum of $\hat{\mathbf r}$2, indicating that the spherical average is much less reliable for subtle three-body and long-wavelength correlations (Dornheim et al., 1 Apr 2025).

This yields a clear division of labor. The spherical average is adequate for equation-of-state quantities, integrated thermodynamic properties, and $\hat{\mathbf r}$3-resolved observables in the non-collective regime. A full Ewald treatment is mandatory for compressibility sum rules, small-angle x-ray scattering, optical and transport properties, and the detailed collective structure of $\hat{\mathbf r}$4, $\hat{\mathbf r}$5, and $\hat{\mathbf r}$6 at small $\hat{\mathbf r}$7 (Dornheim et al., 1 Apr 2025).

6. Computational significance and broader methodological context

The computational appeal of the spherically averaged Ewald interaction is direct. In the accelerated free-energy scheme, the interaction contribution was evaluated up to 18 times faster than with the Ewald-only method, with the speedup essentially independent of the number of imaginary-time slices $\hat{\mathbf r}$8 and saturating at large $\hat{\mathbf r}$9 (Svensson et al., 17 Jul 2025). In the PIMC implementation of the YR Hamiltonian itself, the reported acceleration is $r=|\mathbf r|$0, reflecting the replacement of real- plus reciprocal-space Ewald sums by a finite-range analytic pair potential (Dornheim et al., 1 Apr 2025).

This efficiency advantage arises because the dominant cost in PIMC scales with repeated interaction evaluations over beads and particles. In the ISHTAR cost model,

$r=|\mathbf r|$1

the interaction prefactor $r=|\mathbf r|$2 is much smaller for YR than for Ewald (Svensson et al., 17 Jul 2025). The gain is therefore especially large in bosonic sectors, extensive parameter scans, and large-$r=|\mathbf r|$3 calculations.

The broader methodological significance is that the YR construction exemplifies a general two-step reference-system strategy: choose an artificial interaction that is much cheaper to evaluate than the physical interaction but remains close enough energetically that the correction can be performed in one or a few extended-ensemble steps. The free-energy study states explicitly that this strategy can be extended to other long-range interactions, to inhomogeneous systems such as electrons in ionic backgrounds, and to systems relevant for planetary interiors and inertial confinement fusion with low to moderate quantum degeneracy (Svensson et al., 17 Jul 2025).

More generally, several Ewald variants exploit radial kernels or spherical averaging in related ways. Interpolated real-space Ewald strategies such as ANKH are built on the radial screened Coulomb kernel $r=|\mathbf r|$4 and use interpolation-based acceleration rather than a fitted YR-type pair potential (Chollet et al., 2022). Model-density approaches recast periodic electrostatics in terms of multipole operators acting on an isotropic Ewald kernel, clarifying how spherical contributions and anisotropic corrections are separated (Ribaldone et al., 29 Jan 2026). These developments indicate that the spherically averaged Ewald interaction is best understood not as a single formula, but as a class of radial representations of periodic Coulomb electrostatics whose utility depends on whether the target observable is controlled primarily by integrated energetics or by long-wavelength collective structure.