As of the 2020 release, the DFTB engine supports two different classes of model Hamiltonians, Grimme’s extended tight-binding, and the classic Slater-Koster based DFTB. All of these model Hamiltonians are obtained by applying tight-binding approximations to the DFT total energy expression.
Slater-Koster based DFTB¶
The efficiency of Slater-Koster based DFTB stems from its use of an optimized minimum valence orbital basis that reduces the linear algebra operations, and a two center-approximation for the Kohn-Sham potential that allows precalculation and storage of integrals using the Slater-Koster technique. This makes DFTB orders of magnitude faster than DFT, but requires parameter files (containing the integrals) for all pair-wise combinations of atoms in a molecule. Many elements can be handled with the parameter sets included in the distribution. Alternatively, sets of parameters in the SKF format can be downloaded and used from third party sources.
There are three flavors of Slater-Koster based DFTB available in our implementation:
- The “plain” DFTB Hamiltonian as introduced by Porezag and Seifert without a self-consistency cycle.
- The second order self-consistent charge extension SCC-DFTB (recently also called DFTB2), which accounts for density fluctuations and improves results on polar bonds. Note that the self-consistent calculations is about an order of magnitude slower than calculations with the “plain” DFTB Hamiltonian.
- The third order extension known as DFTB3, which improve the description of hydrogen-bonded complexes and proton affinities. Note that DFTB3 calculations are only marginally slower than SCC-DFTB based calculations.
Note that since these methods have been respectively parametrized, it is important to specify a matching parameter set when applying one of these models.
Extended tight-binding (xTB)¶
The extended tight-binding (xTB) model Hamiltonian as recently been introduced by Grimme and coworkers. It makes similar approximations as Slater-Koster based DFTB, but instead of using precalculated integrals, xTB employs a (small) basis of Slater-type orbitals and uses an extended Hückel-like approximation for the Hamiltonian.
The DFTB Engine supports the GFN1-xTB parameterization of xTB, which is optimized for geometries, frequencies and non-covalent interactions and covers all elements of the periodic table up to radon.
The following keys allow you to select a model Hamiltonian and control different aspects of how the stationary Schroedinger equation is solved.
Model [DFTB | SCC-DFTB | DFTB3 | GFN1-xTB | NonSCC-GFN1-xTB]
Type: Multiple Choice Default value: GFN1-xTB Options: [DFTB, SCC-DFTB, DFTB3, GFN1-xTB, NonSCC-GFN1-xTB] Description: Selects the Hamiltonian used in the DFTB calculation: - DFTB/DFTB0/DFTB1 for classic DFTB without a self-consistent charge cycle - SCC-DFTB/DFTB2 with a self-consistency loop for the Mulliken charges - DFTB3 for additional third-order contributions. - GFN1-xTB for Grimme’s extended tight-binding model in the GFN1 version. - NonSCC-GFN1-xTB for a less accurate but faster version of GFN1-xTB without a self-consistency cycle The choice has to be supported by the selected parameter set.
Different parameters may be suitable for different model Hamiltonians. It is important to choose the appropriate parameter set for the type of calculation and molecular system under study, see parameter sets.
Type: String Description: The directory containing the parameter files. The path can be absolute or relative. Relative paths starting with ./ are considered relative to the directory in which the calculation is started, otherwise they are considered relative to $AMSRESOURCES/DFTB. This key is required for the Slater-Koster based DFTB models, but optional for xTB.
- Uses the resource directory $AMSRESOURCES/DFTB/Dresden.
- Uses the specified path /home/myusername/myparamsdir as the resource directory.
NOTE: Each resource directory must contain a file called metainfo.yaml, which specifies the capabilities of the parameter set. For details see metainfo.yaml.
The selected model Hamiltonian can be extended with dispersion correction:
DispersionCorrection [None | Auto | UFF | ULG | D2 | D3-BJ | D4]
Type: Multiple Choice Default value: None Options: [None, Auto, UFF, ULG, D2, D3-BJ, D4] GUI name: Dispersion Description: This key is used to specify an empirical dispersion model. Please refer to the DFTB documentation for details on the different methods. By default no dispersion correction will be applied. Setting this to auto applies the dispersion correction recommended in the DFTB parameter set’s metainfo file. Note that the D3-BJ dispersion correction is enabled by default when using the GFN1-xTB model Hamiltonian, but can be disabled manually by setting this keyword to None.
The newest and most accurate dispersion correction is D4. We recommend both the D3-BJ and D4 dispersion corrections as good defaults, depending on their availability for the specific combination of the model Hamiltonian and parameterization. Note that the D4 dispersion corrections is computationally more expensive than D3-BJ for bulk periodic systems (it scales as O(N3) with the number of atoms and is not parallelized), thus the user may first want to evaluate if the increased accuracy justifies the increased computational cost.
Solvation effects can be included via the implicit GBSA solvation model. We gratefully acknowledge the Grimme’s group in Bonn for their contribution of the GBSA solvation method code.
To enable the GBSA method, specify the desired solvent:
Solvation Solvent [...] End
Type: Block Description: Generalized Born solvation model with Solvent Accessible Surface Area (GBSA).
Type: Multiple Choice Default value: None Options: [None, Acetone, Acetonitrile, CHCl3, CS2, DMSO, Ether, H2O, Methanol, THF, Toluene] Description: Solvent used in the GBSA implicit solvation model.
More options can be specified in the
Solvation UseGSASA Yes/No GSolvState [Gas1BarSolvent | Gas1MSolvent1M | Gas1BarSolvent1M] Temperature float SurfaceGrid [230 | 974 | 2030 | 5810] End
Type: Bool Default value: Yes GUI name: Solvation Free Energy Description: Include shift term and G(SASA) terms in the energy and gradient.
Type: Multiple Choice Default value: Gas1MSolvent1M Options: [Gas1BarSolvent, Gas1MSolvent1M, Gas1BarSolvent1M] Description: Reference state for solvation free energy shift.
Type: Float Default value: 298.15 Unit: Kelvin Description: The temperature used when calculating the solvation free energy shift. Only used for ‘Gas1BarSolvent’ and ‘Gas1BarSolvent1M’ GSolvState options.
Type: Multiple Choice Default value: 230 Options: [230, 974, 2030, 5810] Description: Number of angular grid points for the construction of the solvent accessible surface area. Usually the default number of grid point suffices, but in case of suspicious behaviors you can increase the number of points.
SCC details and spin-polarization¶
SCC AdaptiveMixing Yes/No Converge Charge float End DIIS Enabled Yes/No MaxSamples integer MaximumCoefficient float MinSamples integer MixingFactor float End HXDamping Yes/No Iterations integer OrbitalDependent Yes/No Unrestricted Yes/No End
Type: Block Description: This optional section configures various details of the self-consistent charge cycle. If the model Hamiltonian does not need a self-consistent solution (e.g. plain DFTB0), none of this information is used and the entire section will be ignored.
Type: Bool Default value: Yes Description: Change the mixing parameter based on the monitored energy. A significant increase of energy will strongly reduce the mixing. Then it will slowly grow back to the SCC%Mixing value.
Type: Block Description: Controls the convergence criteria of the SCC cycle.
Type: Float Default value: 1e-08 GUI name: Charge convergence Description: The maximum change in atomic charges between subsequent SCC iterations. If the charges change less, the SCC cycle is considered converged.
Type: Block Description: Parameters influencing the DIIS self-consistency method
Type: Bool Default value: Yes Description: If not enabled simple mixing without DIIS acceleration will be used.
Type: Integer Default value: 20 Description: Specifies the maximum number of samples considered during the direct inversion of iteration of subspace (DIIS) extrapolation of the atomic charges during the SCC iterations. A smaller number of samples potentially leads to a more aggressive convergence acceleration, while a larger number often guarantees a more stable iteration. Due to often occurring linear dependencies within the set of sample vectors, the maximum number of samples is reached only in very rare cases.
Type: Float Default value: 10.0 Description: When the diis expansion coefficients exceed this threshold, the solution is rejected. The vector space is too crowded. The oldest vector is discarded, and the expansion is re-evaluated.
Type: Integer Default value: -1 Description: When bigger than one, this affects the shrinking of the DIIS space on linear depence. It will not reduce to a smaller space than MinSamples unless there is extreme dependency.
Type: Float Default value: 0.15 Description: The parameter used to mix the DIIS linear combination of previously sampled atomic charge vectors with an analogous linear combination of charge vectors resulting from population analysis combination. It can assume real values between 0 and 1.
Type: Bool Description: This option activates the DFTB3 style damping for H-X bonds. Note that this is always enabled if the DFTB%Model key is set to DFTB3. Not used with xTB.
Type: Integer Default value: 500 Description: Allows to specify the maximum number of SCC iterations. The default should suffice for most standard calculations. Convergence issues may arise due to the use of the Aufbau occupations for systems with small HOMO-LUMO gaps. In this case the use of a Fermi broadening strategy may improve convergence. Choosing a smaller mixing parameter (see DFTB%SCC%Mixing) may also help with convergence issues: it often provides a more stable but slower way to converge the SCC cycle.
Type: Bool Description: Activates or disables orbital resolved calculations. If this key is absent the recommended settings from the parameter file’s metainfo.
Type: Bool Default value: No Description: Enables spin unrestricted calculations. Only collinear spin polarization is supported, see Theor Chem Acc (2016) 135: 232, for details. Must be supported by the chosen parameter set. Not yet compatible with DFTB3, k-space sampling periodic calculations or the xTB models.
Occupation KT float NumBoltz integer Strategy [Auto | Aufbau | Fermi] Temperature float End
Type: Block Description: Configures the details of how the molecular orbitals are occupied with electrons.
Type: Float Unit: Hartree Description: (KT) Boltmann constant times temperature, used for electronic temperature with strategy is auto. The default value is the default value for Temperature*3.166815423e-6. This key and Temperature are mutually exlusive.
Type: Integer Default value: 10 Description: The electronic temperature is done with a Riemann Stieltjes numerical integration, between zero and one occupation. This defines the number of points to be used.
Type: Multiple Choice Default value: Auto Options: [Auto, Aufbau, Fermi] GUI name: Occupation Description: This optional key allows to specify the fill strategy to use for the molecular orbitals. Can either be ‘Aufbau’ for simply filling the energetically lowest orbitals, or ‘Fermi’ for a smeared out Fermi-Dirac occupation. By default the occupation strategy is determined automatically, based on the other settings (such as the number of unpaired electrons).
Type: Float Default value: 300.0 Unit: Kelvin GUI name: Fermi temperature Description: The Fermi temperature used for the Fermi-Dirac distribution. Ignored in case of aufbau occupations.
Type: Integer Default value: 0 GUI name: Spin polarization Description: This specifies the number of unpaired electrons (not the multiplicity!). This number will then be used in the orbital-filling strategy. Has to be compatible with the total number of electrons, meaning it must be an even number if the total number of electrons is even and odd if the total number is odd. Must be an integer value. Note that this does not activate spin polarization, it only affects the filling of the orbitals.
As of the 2019 release, the k-space integration is unified between BAND and DFTB and uses the same keys as input, and the same defaults. See the page on k-space integration in the BAND manual for details and recommendations.
KSpace Quality [GammaOnly | Basic | Normal | Good | VeryGood | Excellent] Regular NumberOfPoints integer_list End Symmetric KInteg integer End Type [Regular | Symmetric] End
Type: Block Description: Options for the k-space integration (i.e. the grid used to sample the Brillouin zone).
Type: Multiple Choice Default value: Normal Options: [GammaOnly, Basic, Normal, Good, VeryGood, Excellent] GUI name: K-space Description: Select the quality of the K-space grid used to sample the Brillouin Zone. If ‘GammaOnly’, only one point (the gamma point) will be used. For the other options, the actual number of K points generated depends on the size of the unit cell. The larger the real space cell, the fewer K points will be generated. The CPU-time and accuracy strongly depend on this option.
Type: Block Description: Options for the regular k-space integration grid.
Type: Integer List Description: Use a regular grid with the specified number of k-points along each reciprocal lattice vector. For 1D periodic systems you should specify only one number, for 2D systems two numbers, and for 3D systems three numbers.
Type: Block Description: Options for the symmetric k-space integration grid.
Type: Integer GUI name: Accuracy Description: Specify the accuracy for the Symmetric method. 1: absolutely minimal (only the G-point is used) 2: linear tetrahedron method, coarsest spacing 3: quadratic tetrahedron method, coarsest spacing 4,6,… (even): linear tetrahedron method 5,7…. (odd): quadratic method The tetrahedron method is usually by far inferior.
Type: Multiple Choice Default value: Regular Options: [Regular, Symmetric] GUI name: K-space grid type Description: The type of k-space integration grid used to sample the Brillouin zone (BZ) used. ‘Regular’: simple regular grid. ‘Symmetric’: symmetric grid for the irreducible wedge of the first BZ (useful when high-symmetry points in the BZ are needed to capture the correct physics of the system, graphene being a notable example).
xTB specific keywords¶
A few keywords only apply to the xTB model Hamiltonian.
XTBConfig SlaterRadialThreshold float useXBTerm Yes/No End
Type: Block Description: This block allows for minor tweaking.
Type: Float Default value: 1e-05 Description: Threshold determining the range of the basis functions. Using a larger threshold will speed up the calculation, but will also make the results less accurate.
Type: Bool Default value: No Description: Whether to use the Halogen bonding (XB) term. This is not advised as it has a non-continuous PES.
The GFN1-xTB implementation in AMS currently does not implement the electronic entropy term from the article by Grimme et al. It therefore gives slightly different energies (but not gradients!) for systems with partially occupied molecular orbitals.
Technical AnalyticalStressTensor Yes/No EwaldSummation CellRangeFactor float Enabled Yes/No Tolerance float End MatricesViaFullMaxSize integer Parallel nCoresPerGroup integer nGroups integer nNodesPerGroup integer End ReuseKSpaceConfig Yes/No Screening dMadel float rMadel float End UseGeneralizedDiagonalization Yes/No End
Type: Block Description: This optional section is about technical aspects of the program that should not concern the normal user.
Type: Bool Default value: Yes Description: Whether to compute the stress tensor analytically. Note: This can only be used together with Ewald summation as it will give (slightly) wrong results with Madelung screening.
Type: Block Description: Configures the details of the Ewald summation of the Coulomb interaction.
Type: Float Default value: 2.0 Description: Smaller values will make the Ewald summation less accurate but faster.
Type: Bool Default value: Yes Description: Whether to use Ewald summation for the long-range part of the Coulomb interaction. Otherwise screening is used.
Type: Float Default value: 1e-10 Description: Larger values will make the Ewald summation less accurate but faster.
Type: Integer Default value: 2047 Description: Matrices smaller than this size are constructed via a full matrix. This is faster, but uses more memory in the construction.
Type: Block Description: Calculation of the orbitals in several k-points is trivially parallel.
Type: Integer Description: Number of cores in each working group.
Type: Integer Description: Total number of processor groups. This is the number of tasks that will be executed in parallel.
Type: Integer GUI name: Cores per task Description: Number of nodes in each group. This option should only be used on homogeneous compute clusters, where all used compute nodes have the same number of processor cores.
Type: Bool Default value: Yes Description: Keep the number of k-points constant during a lattice optimization. Otherwise the PES might display jumps, because the number of points depends on the lattice vector sizes. If this option is on it will always use the number of k-points that was used from a previous result.
Type: Block Description: For SCC-DFTB in periodic systems the Coulomb interaction can (instead of using Ewald summation) be screened with a Fermi-Dirac like function defined as S(r)=1/(exp((r-r_madel)/d_madel)+1). This section allows to change some details of the screening procedure. Note that Coulomb screening is only used if the Ewald summation is disabled.
Type: Float Unit: Bohr Description: Sets the smoothness of the screening function. The default is 1/10 of [rMadel].
Type: Float Unit: Bohr Description: Sets the range of the screening function. The default is 2x the norm of the longest lattice vector.
Type: Bool Default value: Yes Description: Whether or not to use generalized diagonalization. Does not affect the results, but might be faster or slower.
Type: Bool Default value: No Description: Determines whether the Hamiltonian and overlap matrices are stored in the binary result file.