In this work we introduce a Linearized version of the Koringa Khon and Rostoker method (LKKR) and show it to be equivalent to the Linearized Muffin Tin Orbitals method (LMTO). We then present higher derivative versions of both methods, e.g. HDKKR and HDMTO and show them to be partially distinct (not equivalent). In particular HDKKR basis set does not have an equivalent ground state for the Khon Sham (KS) Hamiltonian as the HDKKR basis set and has greater variational power than the HDMTO one. Because the KS method, for Density Functional Theory (DFT), is variational HDKKR will give better ground state energies than HDMTO. However HDKKR is much harder to work with then HDMTO requiring much greater computer resources so HDMTO can often be preferred.
Introduction
The choice of basis set is fundamental for an efficient solution of a Density Functional Theory problem (DFT) - the diagonalization of the Khon Sham (KS) Hamiltonian. The basis set must be efficient, in that a small number of basis elements well approximate the exact Khon Sham (KS) wavefunctions at least in the middle of the valence band. The basis must also be simple to manipulate - with various practical calculations associated with finding matrix element of the KS Hamiltonian with respect to the basis easily implementable on computer. The basis set must be transferable, that is one does not want to use a significantly different basis for every single many body problem or equivalently for diagonalizing every KS Hamiltonian. The Koringa Khon Rostoker (KKR) and the Augmented plane Wave (APW) basis sets are such basis sets, however they are deficient in that the basis elements, of both basis sets, depend explicitly on the energy of the eigenstate, to be computed, which leads to a self consistent calculation of the basis set thereby increasing computational costs by easily an order of magnitude [1, 2]. The key idea to overcome this difficulty is due to O. K. Andersen who linearized the basis set (with the linearization energy being chosen in the middle of the valance band) thereby obtaining a fixed basis for each iteration of the self consistency loop for the solution of the self consistent KS problem [1, 2, 3, 4], each step requires simply the diagonalization of a matrix. This lead to the introduction of the Linearized Augmented Plane Wave (LAPW) [1, 2, 3] and Linearized Muffin Tin Orbitals (LMTO) methods [5, 6]. In both these methods the solid is divided into a Muffin Tin (MT) sphere part (with spheres centered around atomic nuclei) and an interstitial part [1, 2, 3]. In the interstitial part the KS potential is assumed to be sufficiently smooth that plane waves or spherical Bessel functions form a good basis for the region while the basis set is adapted to the MT part to be the solution of the spherically averaged KS Hamiltonian. The solution to the spherically averaged KS Hamiltonian is chosen at some linearization energy (typically in the middle of the valence band) and the solution ψlμE(r) as well as
ψ˙lμE(r)=∂E∂ψlμE(r) - its derivative with respect to energy - are used as a basis set inside the MT spheres. Further accuracy may be obtained for the HDLO method where ψ¨lμE(r)=∂E2∂2ψlμE(r) is used to augment the basis set [2, 7]. In this work after a brief review of the usual KKR method we linearize the KKR method to LKKR and show it to be equivalent to LMTO. We then extend these ideas to higher derivative versions HDKKR and HDMTO and show that HDKKR basis set has greater variational power then the HDMTO basis set. However HDKKR is much harder to implement then HDLMTO and requires much greater computer resources so often HDMTO is preferred.
For the some structure constants Bνμl′m′;lm{rμ}(k)[5]. Now we wish to eliminate all tails in all basis functions that is Bessel like wavefunction components inside all MT spheres. As such we want for ∣r−rμ∣≤Sμ:
That is all the terms proportional to Bessel and Neumann Functions vanish inside the spheres ∣r−rμ∣≤Sμ (see Eq. (14)). Let us pick a basis of solutions of Eq. (17)
for some Aνl′m′lmΘ(k) and Bνl′m′lmΘ(k). As such we would see that the LMTO method is spanned by the LKKR method, however by counting the total number of basis set elements we would then see the two are equivalent. However Eq. (24) is a direct check - see Appendix A.
HDKKR
We now write the HDKKR wavefunctions. We write the following basis set wavefunctions:
so that the wavefunction along with its first and second derivatives are continuous. The derivation is identical to the one in Appendix A. Now we write:
χμ,lmMTO(k)=R∑exp(ik⋅R)Θlmμ(E,r−R,k)(34)
And study this basis set.
Differences and similarities between HDMTO and HDKKR
We note that HDKKR basis set includes HDMTO as HDKKR only demands continuity of the wavefunction and its derivative while HDMTO also demands continuity of the second derivative. Furthermore because of the similarities of the form of the HDMTO and LMTO basis wavefunctions the HDMTO basis set eliminates all Bessel like wavefunction components inside all MT spheres so HDMTO basis wavefunctions are a type of HDKKR basis wavefunctions. As such HDKKR has greater variational power then HDMTO. We note that this observation is true, despite the fact that the exact solution of the KS Hamiltonian have all derivatives continuous, as wavefunctions with discontinuous second derivatives may help approximate ones with continuous second derivatives better. As the KS Equations are variational HDKKR gives a better estimate of the ground state energy then HDMTO. However the HDMTO method is much more efficient as the basis is smaller and it does not require solving Eq. (29) on computer as the basis set automatically satisfies it as there are no Bessel like functions inside any MT spheres so in many case HDMTO is preferable to HDKKR.
Conclusion
In this work we have studied a linearized version of KKR (LKKR) and shown it to be exactly equivalent to LMTO. We have extended these ideas to higher derivative HDKKR and HDMTO basis sets. These two basis sets are not equivalent. The HDKKR basis set has greater variational power then HDMTO basis set as it includes it as a subset. However HDMTO is much easier to work with, requires significantly less computational power and no auxiliary equations to solve. As such in many cases, since HDMTO is easier to implement, it is likely the preferred basis set method. In the future it would be of interest to set up practical HDMTO calculations for real solid state crystal systems.
Technical Calculations
As such for the interstitial region to match between LKKR and LMTO we must have that:
Aνl′m′lmΘ(k)+Bνl′m′lmΘ(k)=δμνδl,l′δm,m′(35)
Now we match the MT region. From which we read of:
As such Eq. (17) is verified for the LMTO basis set as it is a continuous continuously differentiable wavefunction in the span of the KKR basis set with no components of the form of Bessel functions inside the MT spheres.
References
[1]Singh, D. J., & Nordstrom, L. (2005). Planewaves, Pseudopotentials and the LAPW Method.
[2]Martin, R. M. (2020). Electronic Structures Basic Theory and Practical Methods.
[3]O. K. Andersen, Phys. Rev. B 12, 3060 (1975).
[4]Marx, D., & Hutter, J. (2009). Ab Initio Molecular Dynamics Basic Theory and Advanced Methods.
[5]Wills, J. M., Alouani, M., Anderson, P., Dellin, A., Eriksson, O., & Grechnyev, O. (2010). Full-Potential Electronic Structure Method Energy and Force Calculations with Density Functional Theory and Dynamical Mean Field Theory.
[6]H. L. Skriver, The LMTO method Muffin-Tin Orbitals and Electronic Structure (Springer, New York, 1984).DOI
[7]G. Michalicek, M. Betzinger, C. Freidrich and S. Blugel, Comp. Phys. Comm. 184, 2670 (2013).