ACFDT/RPA calculations

The adiabatic-connection-fluctuation-dissipation theorem (ACDFT) can be used to derive the random-phase approximation (RPA). The ACFDT-RPA ground-state energy $E_{\mathrm{RPA}}$ is the sum of the ACFDT-RPA correlation energy $E_{c}$ and the Hartree-Fock (HF) energy $E_{\mathrm{EXX}}$ evaluated non self-consistently using Kohn-Sham orbitals computed within density-functional theory (DFT):

$E_{\mathrm{RPA}}=E_{\mathrm{c}}+E_{\mathrm{EXX}}$.

Note that, here $E_{\mathrm{EXX}}$ includes also the Hartree energy, the kinetic energy, as well as the Ewald energy of the ions, whereas often in literature $E_{\mathrm{EXX}}$ refers only to the exact-exchange (EXX) energy evaluated using DFT orbitals.

If ALGO=RPA is set in the INCAR file, VASP calculates the correlation energy within RPA. To this end, VASP first calculates the independent-particle response function, using the unoccupied states given in the WAVECAR file, and then determines the correlation energy using the plasmon-fluctuation equation:

$E_{c} = \frac{1}{2 \pi} \int_{0}^{\infty} \mathrm{Tr} \, \ln(1-\chi( i \omega) V) + \chi( i \omega) V \, d \omega$.

More information about the theory behind the RPA is found here.

General recipe to calculate ACFDT-RPA total energies

The ACFDT-RPA total energies can be computed in a four-step procedure or in an all-in-one approach. The all-in-one approach is available as of VASP.6 and can be done in one single step. Both procedures comprise a self-consistent DFT calculation, diagonalizing the KS Hamiltonian, and the RPA calculation itself. While the all-in-one approach is very convenient, there are several caveats discussed below.

Four-step procedure to calculate ACFDT-RPA total energy

This four-step procedure is the standard procedure in VASP.5, and is also available in VASP.6 by setting NBANDS in the RPA step.

Step 1: a standard DFT run.

All occupied orbitals (and as usual in VASP, a few unoccupied orbitals) of the DFT Hamiltonian are calculated. For instance,

EDIFF = 1E-8
ISMEAR = 0 ; SIGMA = 0.05

Tip: Use the Perdew-Burke-Ernzerhof (PBE) pseudopotentials with tight convergence criteria, and avoid ISMEAR>0.

The DFT run can be done with your favorite setup, but we recommend attaining very high precision (small EDIFF tag) and to use a small smearing width (SIGMA tag), and to avoid higher-order Methfessel-Paxton smearing (see also ISMEAR). That is because Methfessel-Paxton smearing can lead to negative one-electron occupancies, which can result in unphysical correlation energies. We suggest using PBE orbitals as input for the ACFDT-RPA run, but other choices, e.g., local-density-approximation (LDA) or hybrids, are possible as well. For hybrid functionals, we suggest carefully considering the caveats mentioned in Ref. . Specifically, the RPA dielectric matrix yields too weak screening for hybrid functionals, which potentially deteriorates RPA results.

Step 2: Compute the HF energy $E_{\mathrm{EXX}}$ using the DFT orbitals. That is, e.g.,

ALGO  = EIGENVAL ; NELM = 1
LWAVE = .FALSE.                ! WAVECAR not written     
LHFCALC = .TRUE. ; AEXX = 1.0  ! ALDAC = 0.0 
ISMEAR = 0 ; SIGMA = 0.05

Here, LWAVE = .FALSE. avoids an accidental update of the WAVECAR file. Also, VASP automatically sets ALDAC = 1-AEXX, i.e., for AEXX = 1.0, ALDAC = 0.0.

Tip: For insulators and semiconductors with a sizable gap, faster k-point convergence of the HF energy can be obtained by setting HFRCUT=-1, although this slows convergence for metals.

Step 3: Compute a sufficient number of unoccupied states.

In the OUTCAR file of Step 1, search for maximum number of plane-waves: and run VASP again with the following INCAR file to determine the unoccupied states by exact diagonalization of the KS Hamiltonian. The KS Hamiltonian must be the same as in step 1, which might be DFT or hybrid.

NBANDS = maximum number of plane-waves (times 2 for vasp_gam)
ALGO = Exact    ! exact diagonalization
NELM = 1        ! one step suffices since WAVECAR is pre-converged
LOPTICS = .TRUE.
ISMEAR = 0 ; SIGMA = 0.05

For calculations using vasp_gam, that is the gamma-point only version of VASP, NBANDS must be set to twice the maximum number of plane-waves: found in the OUTCAR file of Step 1. For metals, we recommend to avoid setting LOPTICS=.TRUE., since this slows down k-point convergence .

Step 4: Calculate the ACFDT-RPA correlation energy.

NBANDS =  maximum number of plane-waves
ALGO = ACFDT
NOMEGA = 12     ! default
ISMEAR = 0 ; SIGMA = 0.05 

The number of imaginary-frequency and imaginary-time-grid points NOMEGA is usually set to 8-24. For large gap insulators NOMEGA = 8 will suffice, while for semiconductors 10-12 suffices.

Important: For VASP 6.X, NBANDS must be set in this step.

All-in-one approach to calculate ACFDT-RPA total energy

VASP will read the WAVECAR file, perform a self-consistent DFT calculation by iterating until convergence is reached, diagonalize the DFT Hamiltonian in the basis set spanned by all KS orbitals, and finally proceed with the RPA calculation. The all-in-one approach is only available as of VASP.6, and it is particularly convenient for the RPAR and GWR algorithms. An example INCAR file is

ALGO = RPAR

EDIFF = 1E-7
ISMEAR = 0 ; SIGMA = 0.05
 
NOMEGA = 8-24    ! for large gap insulators NOMEGA = 8 will suffice, for semiconductors 10-12 suffices
   
LOPTICS = .TRUE. ! optional 
LPEAD = .TRUE.   ! required for 6.1 and 6.2 if LOPTICS = .TRUE.

Warning: The tag NBANDS must not be set in the INCAR file.

The head of the dielectric function can be calculated only for insulators by setting the LOPTICS tag and only the Perturbation-Expression-After-Discretization (PEAD) method is supported, see LPEAD. Furthermore, EDIFF should be small to calculate the DFT orbitals with high precision. Although it is not strictly required, we recommend starting the calculation from a well-converged DFT WAVECAR file.

There are several caveats to the all-in-one approach:

• There is no general support for ACFDT/RPA on top of hybrid functionals. In VASP $\leq$ 6.2, only LDA and gradient corrected functionals are supported. As of VASP 6.2, the all-in-one approach can be used for Hartree-Fock-type calculations, as well as hybrid functionals without range separation. However, range-separated hybrid functionals (including HSE, LMODELHF) yield erroneous results or potentially even crash. Please compare the total energies in a standard DFT/hybrid functional calculation against the energies obtained for the mean-field step in the all-in-one approach: The energies must be exactly the same. If they differ, fall back to the four-step procedure that is explained above.
• To select the new all-in-one approach, it is important "not to set NBANDS" in the RPA step. If NBANDS is set in the INCAR file, VASP.6 proceeds with reading the WAVECAR file found in the directory (if not present, random orbitals are used!), and then calculates the correlation energy using these orbitals and one-electron energies.
• The basis set used in the diagonalization is strictly given by all plane waves. That is, there is no option to reduce the basis set size as one can by setting NBANDS in Step 3 and Step 4 of the four-step procedure.
• The exchange energy is only calculated if the low-scaling RPA is selected.
• As of VASP 6.3, the head of the dielectric function (G=0 component) can be calculated by setting LOPTICS = .TRUE.
• In the all-in-one procedure, VASP automatically sets LMAXFOCKAE and NMAXFOCKAE. This changes the energies for hybrid functionals and Hartree-Fock slightly. Hence, you need to set LMAXFOCKAE = 4 also in Step 1 of the four-step procedure to obtain the same results as in the all-in-one procedure.

Output of ACFDT-RPA total energy

The ACFDT-RPA total energy is calculated for 8 different cutoff energies and a linear regression is used to extrapolate the results to the infinite cutoff limit, see section below. A successful RPA calculation writes the following lines into the OUTCAR file:

      cutoff energy     smooth cutoff   RPA   correlation   Hartree contr. to MP2
---------------------------------------------------------------------------------
            316.767           316.767      -17.5265976349      -26.2640927215
            301.683           301.683      -17.3846505665      -26.0990489039
            287.317           287.317      -17.2429031341      -25.9344769084
            273.635           273.635      -17.0686574017      -25.7325162480
            260.605           260.605      -16.8914915810      -25.5277026697
            248.195           248.195      -16.7202601717      -25.3302982602
            236.376           236.376      -16.5559849344      -25.1415392478
            225.120           225.120      -16.3635400223      -24.9210737434
  linear regression    
  converged value                          -19.2585393615      -28.2627347266

Here, the third and fourth columns correspond to the correlation energy (for that specific cutoff energy) in the RPA and the direct second-order Møller–Plesset (MP2) approximation, i.e., the second-order term in RPA. The corresponding results of the linear regression are found in the line starting with converged value.

Low-scaling ACFDT/RPA algorithm

Virtually the same tags and procedures apply to the low-scaling RPA algorithm implemented in VASP 6 . However, ALGO=ACFDT or ALGO=RPA needs to be replaced by either ALGO=ACFDTR or ALGO=RPAR.

With this setting VASP calculates the independent-particle polarizability $\chi(i\omega)$ using Green's functions $G(i\tau)$ on the imaginary time axis $i\tau$ by the contraction formula

$\chi(i\tau_m) = -G(i\tau_m)G(-i\tau_m)$

Subsequently, a compressed Fourier transformation on the imaginary axes $\tau_m,\omega_n$ yields

$\chi(i\omega_n) = \sum_{m=1}^{N} \gamma_{nm} \chi(i\omega_m)$

The remaining step is the evaluation of the correlation energy and is the same as described above.

Crucial to this approach is the accuracy of the Fourier transformation from $\chi(i\tau)\to \chi(i\omega)$, which in general depends on two factors: First, the grid order $N$ that can be set by NOMEGA in the INCAR file. Here, similar choices as for the ALGO=ACFDT are recommended. Second, the grid points $\tau_m,\omega_n$ and Fourier matrix $\gamma_{nm}$ have to be optimized for the same interval as spanned by all possible transition energies in the polarizability. The minimum (maximum) transition energy can be set with the OMEGAMIN (OMEGATL) tag and should be smaller (larger) than the bandgap, i.e., the maximum transition energy, of the previous DFT calculation. VASP determines these values automatically and writes it in the OUTCAR after the lines

Response functions by GG contraction: 

These values should be checked for consistency. Furthermore, we recommend inspecting the grid and transformation errors by looking for the following lines in the OUTCAR file

nu_ 1=  0.1561030E+00 ERR=   0.6327933E-05 finished after   1 steps    
nu_ 2= ...
Maximum error of frequency grid:  0.3369591E-06

Every frequency point will have a similar line as shown above for the first point. The value after ERR= corresponds to the maximum Fourier-transformation error and should be of similar order as the maximum integration error of the frequency grid.

Output of low-scaling ACFDT/RPA

Selecting the low-scaling RPA algorithm, VASP computes the total energy in the RPA and writes the following output

 -----------------------------------------------------
 HF-free energy      FHF    =       -25.11173505 eV
 HF+RPA corr. energy TOTEN  =       -36.96463791 eV
 HF+E_corr(extrapolated)    =       -37.70506951 eV

The line HF+RPA corr. energy TOTEN contains the total energy calculated with the largest cutoff ENCUTGW. The line HF+E_corr(extrapolated) contains the total energy with the extrapolated value for the RPA correlation energy.

Optional: RPA Forces

Optionally, RPA forces can be calculated by adding following line to the INCAR file:

 LRPAFORCE = .TRUE. 

For RPA forces the change in the one-electron density is required. This is automatically performed with the linear-response routine within VASP. After a successful run, the following block of data is found in the OUTCAR file.

POSITION                                       TOTAL RPA FORCE (eV/Angst)
-----------------------------------------------------------------------------------
     0.17542     -0.22348      0.17542        -0.292069      7.581315     -0.292069
     1.12850      1.31044      1.12850         0.304683     -7.605527      0.304683
-----------------------------------------------------------------------------------
   total drift:                                0.012614     -0.024212      0.012614

SUGGESTED UPDATED POSCAR (direct coordinates)  step
-----------------------------------------------------------------------------------
 -0.00958461  -0.00958461   0.13485779         0.04179056   0.04179056   0.00283088
  0.25787833   0.25787833   0.22191754        -0.04337198  -0.04337198   0.00431513

Warning: The RPA forces computed for metallic systems are currently not supported.

Memory bottleneck and Parallelization

Memory requirements of low-scaling GW and RPA algorithms

Some Issues Particular to ACFDT-RPA Calculations on Metals

For metals, the RPA groundstate energy converges the fastest with respect to k-points, if the exchange (Eq. (12) in reference ) and correlation energy are calculated on the same k-point grid, HFRCUT is not set, and the long-wavelength contributions from the polarizability are not considered (see reference ).

To evaluate Eq. (12), a correction energy for $E_{\mathrm{EXX}}$ related to partial occupancies has to be added to the RPA groundstate energy:

$E_{\mathrm{RPA}}=E_{\mathrm{c}}+E_{\mathrm{EXX}}+E_{\mathrm{HFc}}$.

In vasp.5.4.1, this value is calculated for any HF type calculation (step 2) and can be found in the OUTCAR file after the total energy (in the line starting with exchange ACFDT corr. =).

To neglect the long-wavelength contributions, simply set LOPTICS=.FALSE. in the ALGO=Exact step (third step), and remove the WAVEDER files in the directory.

Possible tests and known issues

Basis set convergence

The expression for the ACFDT-RPA correlation energy written in terms of reciprocal lattice vectors reads:

$E_{\rm c}^{\rm RPA}=\int_{0}^{\infty} \frac{\mathrm{d}\omega}{2\pi} \sum_{{\mathbf{q}}\in \mathbf{BZ} }\sum_{{\mathbf{G}}} \left\{(\mathrm{ln}[1-\tilde\chi^0({\mathbf{q}},\mathrm{i}\omega)V({\mathbf{q}})])_{{\mathbf{G,G}}} +V_{{\mathbf{G,G}}}({\mathbf{q}})\tilde\chi^0({\mathbf{q}},{\mathrm{i}}\omega) \right\}$.

The sum over reciprocal lattice vectors has to be truncated at some $\mathbf{G}_{\mathrm{max}}$, determined by $\frac{\hbar^2|{\mathbf{G}}+{\mathbf{q}}|^2}{2\mathrm{m}_e}$ < ENCUTGW, which can be set in the INCAR file. The default value is $\frac{2}{3}\times$ ENCUT, which experience has taught us not to change. For systematic convergence tests, instead increase ENCUT and repeat steps 1 to 4, but be aware that the "maximum number of plane-waves" changes when ENCUT is increased. Note that it is virtually impossible, to converge absolute correlation energies. Rather concentrate on relative energies (e.g. energy differences between two solids, or between a solid and the constituent atoms).

Since correlation energies converge very slowly with respect to $\mathbf{G}_{\rm max }$, VASP automatically extrapolates to the infinite basis set limit using a linear regression to the equation:

$E_{\mathrm{c}}({\mathbf{G}})=E_{\mathrm{c}}(\infty)+\frac{A}{{\mathbf{G}}^3}$.

Furthermore, the Coulomb kernel is smoothly truncated between ENCUTGWSOFT and ENCUTGW using a simple cosine like window function (Hann window function). Alternatively, the basis set extrapolation can be performed by setting LSCK=.TRUE., using the squeezed Coulomb kernel method.

The default for ENCUTGWSOFT is 0.8$\times$ENCUTGW (again we do not recommend to change this default).

The integral over $\omega$ is evaluated by means of a highly accurate minimax integration. The number of $\omega$ points is determined by the flag NOMEGA, whereas the energy range of transitions is determined by the band gap and the energy difference between the lowest occupied and highest unoccupied one-electron orbital. VASP determines these values automatically (from vasp.5.4.1 on), and the user should only carefully converge with respect to the number of frequency points NOMEGA. A good choice is usually NOMEGA=12, however, for large gap systems one might obtain $\mu$eV convergence per atom already using 8 points, whereas for metals up to NOMEGA=24 frequency points are sometimes necessary, in particular, for large unit cells.

Strictly adhere to the steps outlines above. Specifically, be aware that steps two and three require the WAVECAR file generated in step one, whereas step four requires the WAVECAR and WAVEDER file generated in step three (generated by setting LOPTICS=.TRUE.).

Convergence with respect to the number of plane waves can be rather slow, and we recommend to test the calculations carefully. Specifically, the calculations should be performed at the default energy cutoff ENCUT, and at an increased cutoff (ideally the default energy cutoff $\times 1.3$). Another issue is that energy volume-curves are sometimes not particularly smooth. In that case, the best strategy is to set

ENCUT = 1.3 times default cutoff energy
ENCUTGWSOFT = 0.5 times default cutoff energy

where the default cutoff energy is the usual cutoff energy (maximum ENMAX in POTCAR files). The frequency integration also needs to be checked carefully, in particular for small gap systems (some symmetry broken atoms) convergence can be rather slow, since the one-electron band gap can be very small, requiring a very small minimum $\omega$ in the frequency integration.

Related Tags and Sections

• ALGO for response functions and ACFDT calculations
• NOMEGA, NOMEGAR number of frequency points
• LHFCALC, switches on HF calculations
• LOPTICS, required in the DFT step to store head and wings
• ENCUTGW, to set cutoff for response functions
• ENCUTGWSOFT
• PRECFOCK controls the FFT grids in HF, GW, RPA calculations
• NTAUPAR controls the number of imaginary time groups in space-time GW and RPA calculations
• NOMEGAPAR controls the number of imaginary frequency groups in space-time GW and RPA calculations
• MAXMEM sets the available memory per MPI rank on each node
• LFINITE_TEMPERATURE switches on Matsubara (finite temperature) formalism

---