Low-scaling GW: The space-time formalism

Available as of VASP.6 are low-scaling algorithms for ACFDT/RPA. This page describes the formalism of the corresponding low-scaling GW approach. A theoretical description of the ACFDT/RPA total energies is found here. A brief summary regarding GW theory is given below, while a practical guide can be found here.

Theory

The GW implementations in VASP described in the papers of Shishkin et al. avoid storage of the Green's function [math]\displaystyle{ G }[/math] as well as Fourier transformations between time and frequency domain entirely. That is, all calculations are performed solely on the real frequency axis using Kramers-Kronig transformations for convolutions in the equation of [math]\displaystyle{ \chi }[/math] and [math]\displaystyle{ \Sigma }[/math] in reciprocal space and results in a relatively high computational cost that scales with [math]\displaystyle{ N^4 }[/math] (number of electrons).

The scaling with system size can, however, be reduced to [math]\displaystyle{ N^3 }[/math] by performing a so-called Wick-rotation to imaginary time [math]\displaystyle{ t\to i\tau }[/math].

Following the low scaling ACFDT/RPA algorithms the space-time implementation determines first, the non-interacting Green's function on the imaginary time axis in real space

[math]\displaystyle{ G({\bf r},{\bf r}',i\tau)=-\sum_{n{\bf k}}\phi_{n{\bf k}}^{(0)}({\bf r}) \phi_{n{\bf k}}^{*(0)}({\bf r}') e^{-(\epsilon_{n{\bf k}}-\mu)\tau}\left[\Theta(\tau)(1-f_{n{\bf k}})-\Theta(-\tau)f_{n{\bf k}}\right] }[/math]

Here [math]\displaystyle{ \Theta }[/math] is the step function and [math]\displaystyle{ f_{n{\bf k}} }[/math] the occupation number of the state [math]\displaystyle{ \phi_{n{\bf k}}^{(0)} }[/math]. Because the Green's function is non-oscillatory on the imaginary time axis it can be represented on a coarse grid [math]\displaystyle{ \tau_{m} }[/math], where the number of time points can be selected in VASP via the NOMEGA tag. Usually 12 to 16 points are sufficient for insulators and small band gap systems.

Subsequently, the irreducible polarizability is calculated from a contraction of two imaginary time Green's functions

[math]\displaystyle{ \chi({\bf r},{\bf r}',i\tau_m) = -G({\bf r},{\bf r}',i\tau_m)G({\bf r}',{\bf r},-i\tau_m) }[/math]

Afterwards, the same compressed Fourier transformation as for the low scaling ACFDT/RPA algorithms is employed to obtain the irreducible polarizability in reciprocal space on the imaginary frequency axis [math]\displaystyle{ \chi({\bf r},{\bf r}',i\tau_m) \to \chi_{{\bf G}{\bf G}'}({\bf q},i \omega_n) }[/math].

The next step is the computation of the screened potential

[math]\displaystyle{ W_{{\bf G}{\bf G}'}({\bf q},i\omega_m)=\left[\delta_{{\bf G}{\bf G}'}-\chi_{{\bf G}{\bf G}'}({\bf q},i\omega_m)V_{{\bf G}{\bf G}'}({\bf q})\right]^{-1}V_{{\bf G}{\bf G}'}({\bf q}) }[/math]

followed by the inverse Fourier transform [math]\displaystyle{ W_{{\bf G}{\bf G}'}({\bf q},i \omega_n) \to \chi({\bf r},{\bf r}',i\tau_m) }[/math] and the calculation of the self-energy

[math]\displaystyle{ \Sigma({\bf r},{\bf r}',i\tau_m) = -G({\bf r},{\bf r}',i\tau_m)W({\bf r}',{\bf r},i\tau_m) }[/math]

From here, several routes are possible including all approximations mentioned above, that is the single-shot, EVG₀ and QPEVG₀ approximation. All approximations have one point in common.

In contrast to the real-frequency implementation, the low-scaling GW algorithms require an analytical continuation of the self-energy from the imaginary frequency axis to the real axis. In general, this is an ill-defined problem and usually prone to errors, since the self-energy is known on a finite set of points. VASP determines internally a Padé approximation of the self-energy [math]\displaystyle{ \Sigma(z) }[/math] from the calculated set of NOMEGA points [math]\displaystyle{ \Sigma(i\omega_n) }[/math] and solves the non-linear eigenvalue problem

[math]\displaystyle{ \left[ T+V_{ext}+V_h+\Sigma(z) \right]\left|\phi_{n\bf k}\right\rangle = z\left| \phi_{n\bf k} \right\rangle }[/math]

on the real frequency axis [math]\displaystyle{ z=\omega }[/math].

Because preceding Fourier transformations have been carried out with exponentially suppressed errors, the analytical continuation [math]\displaystyle{ \Sigma(z) }[/math] of the self-energy can be determined with high accuracy. The analytical continuation typically yields energies that differ less than 20 meV from quasi-particle energies obtained from the real-frequency calculation.

In addition, the space-time formulation allows to solve the full Dyson equation for [math]\displaystyle{ G({\bf r,r'},i\tau) }[/math] with decent computational cost. This approach is known as the self-consistent GW approach (scGW) and is available as of VASP6.

Finite temperature formalism

The zero-temperature formalism of many-body perturbation theory breaks down for metals (systems with zero energy band-gap) as pointed out by Kohn and Luttinger. This conundrum is lifted by considering diagrammatic perturbation theory at finite temperature [math]\displaystyle{ T\gt 0 }[/math], which may be understood by an analytical continuation of the real-time [math]\displaystyle{ t }[/math] to the imaginary time axis [math]\displaystyle{ -i\tau }[/math]. Matsubara has shown that this Wick rotation in time [math]\displaystyle{ t\to-i\tau }[/math] reveals an intriguing connection to the inverse temperature [math]\displaystyle{ \beta=1/T }[/math] of the system. More precisely, Matsubara has shown that all terms in perturbation theory at finite temperature can be expressed as integrals of imaginary time quantities (such as the polarizability [math]\displaystyle{ \chi(-i\tau) }[/math]) over the fundamental interval [math]\displaystyle{ -\beta\le\tau\le\beta }[/math].

As a consequence, one decomposes imaginary time quantities into a Fourier series with period [math]\displaystyle{ \beta }[/math] that determines the spacing of the Fourier modes. For instance the imaginary polarizability can be written as

[math]\displaystyle{ \chi(-i\tau)=\frac1\beta\sum_{m=-\infty}^\infty \tilde \chi(i\nu_m)e^{-i\nu_m\tau},\quad \nu_m=\frac{2m}\beta\pi }[/math]

and the corresponding random-phase approximation of the correlation energy at finite temperature becomes a series over (in this case, bosonic) Matsubara frequencies

[math]\displaystyle{ \Omega_c^{\rm RPA}=\frac12\frac1\beta \sum_{m=-\infty}^\infty {\rm Tr}\left\lbrace \ln\left[ 1 -\tilde \chi(i\nu_m) V \right] -\tilde \chi(i\nu_m) V \right\rbrace,\quad \nu_m=\frac{2m}\beta\pi }[/math]

The Matsubara formalism has the advantage that all contributions to the Green's function and the polarizability are mathematically well-defined, including contributions from states close to the chemical potential [math]\displaystyle{ \epsilon_{n{\bf k}}\approx \mu }[/math], such that Matsubara series also converge for metallic systems.

Although formally convenient, the Matsubara series converges poorly with the number of considered terms in practice. VASP, therefore, uses a compressed representation of the Fourier modes by employing the Minimax-Isometry method. This approach converges exponentially with the number of considered frequency points.

References

</references>

---