Update Ewald error formula

[GardevoirX]

Fixes the error formula of Ewald used in tuning. The fixed formula is tested with systems of crystal cells repeated 8 times on each dimension.

No difference in the actual accuracy is observed.

The Ewald calculation time is a bit longer with the fixed formula.

Contributor (creator of pull-request) checklist

• Tests updated (for new features and bugfixes)?
• Documentation updated (for new features)?
• Issue referenced (for PRs that solve an issue)?

Reviewer checklist

• CHANGELOG updated with public API or any other important changes?

---

📚 Documentation preview 📚: https://torch-pme--198.org.readthedocs.build/en/198/

[PicoCentauri]

The docs should also be updated with the change.

[PicoCentauri]

Are you sure this change is correct? I think units still don't match because smearing and the lr_wavelength should cancel. I just re-derived the equation and I got

$$ ΔF_\mathrm{LR} = \frac{Q^2}{\sqrt{N}} \sqrt{\frac{h}{V}} \frac{1}{π\sigma} \mathrm{exp} \left(-\frac{2 π^2 \sigma^2}{h^2}\right) $$

maybe check it again.

I think we should also give the conversion between our convention and the other usual one somewhere in the docs:

$$ α = \frac{1}{\sqrt{2}Σ} K = \frac{2π}{h} $$

where α is the inverse width of Gaussian charge cloud and $K$ is the Fourier space cutoff in inverse units of length.

[PicoCentauri]

Thanks for the fix @GardevoirX!