Basic theoretical concepts behind DFT-LDA-based ab-initio ...

3 downloads 2591 Views 1MB Size Report
Apr 30, 2008 - Spin degeneracy: electrons always occur pairwise, occupying both spin ... Adiabatic approximation: separate treatment of nuclei and electrons.
Welcome to the coffee seminar:

Basic theoretical concepts behind DFT-LDA-based ab-initio calculations Jan-Martin Wagner 30 April 2008

(DFT: density-functional theory, LDA: local-density approximation)

1

Overview •

General aim; starting point



Basic approximations (solid-state theory): – Spin degeneracy, periodic boundary conditions, adiabatic approximation



Density-functional theory (DFT) and local density approximation (LDA) – Hohenberg–Kohn theorem – The universal F functional – Variational principle – Kohn–Sham equation – Local-density approximation – XC functionals beyond LDA



Further approximations (for numerical treatment): – Frozen-core approximation, pseudopotentials, k-point summation



Excursion: How to overcome the gap problem



Summary

2

General aim; basic approximations •

General aim: Calculate all materials! properties on fully quantummechanical grounds (i.e., without use of experimental parameters)



Starting point: Schrödinger equation for a solid, containing approx. 1023 particles (number of variables of the wavefunction ! ) ! unsolvable

Let!s try to tackle the most simple case, described in the following. •

Basic approximations (solid-state theory): – Spin degeneracy: electrons always occur pairwise, occupying both spin states ! no magnetism, no currents – Periodic boundary conditions: calculate representative part of an infinite bulk system ! no surfaces – Perfect crystal: no defects – Adiabatic approximation: separate treatment of nuclei and electrons ! “frozen-lattice” calculation

3

Adiabatic approximation •

Hamilton operator: ________________________

! Frozen-lattice eigenvalue equation: •

Expansion of the total wave function using the frozen-lattice functions (X: all nuclear coordinates, x: all electronic coordinates):



Adiabatic approximation: keep only one term ! Schrödinger equation of the crystal becomes equation for the nuclei:

! Efrozen determines their equilibrium positions R:

4

Density-functional theory (DFT) •

Search for the frozen-lattice ground-state! ! Zero temperature without zero-point motion



Separate the electronic contributions:

! Ground-state energy: •

, where

Electronic problem to be solved (ground-state only): , where V ext depends on the nuclear coordinates:

5

DFT: Hohenberg–Kohn theorem •

Hohenberg–Kohn theorem: Local external potentials which differ by more than a constant lead to different ground-state densities. Vice versa, the external potential is determined by the ground-state density (up to a c.).



Proof via reductio ad absurdum [Phys. Rev. 136, B864 (1964)]: Let V with ground-state "0 lead to the ground-state density n. Assume V ! ! V + const with ground-state "0! leads to the same density n. Being the primed gound-state means:

E0" = #0" H "#"0 < #0 H "#0 = # 0 [H + V " $ V ]# 0 E0" < E0 + # 0 [V " $ V ]#0 = E0 + % [V "(r) $ V (r)] n(r)dr In the same way for the unprimed ground-state:

!

!

6

E0 < E0" + # "0 [V $ V "]#0" = E0" + % [V(r) $ V "(r)] n"(r)dr Adding the inequalities and using n’ = n leads to a contradiction (q. e. d.):

E0" + E0 < E0 + E0"

DFT: Hohenberg–Kohn theorem •

7

What does the Hohenberg–Kohn theorem mean? Hamilton operator

External potential

H=T+W+V

H

V ext

V’

H"=#" H–K theorem

"0 Ground-state

n = |"|2

n0

"0 ’

Ground-state density

Just this: The wavefunction is determined by the density! ! Use the density as basic quantity; it depends only on one variable r! But: How shall this work? Information contained only in " will be lost! ! Well, we!ll see …

DFT: Variational principle •

8

Towards the variational principle: The universal F functional Remember the eigenvalue equation to be solved:

For a certain type of electron–electron interaction, this depends only on the external potential ! functional notation (for given V ext):



How to transfer this into a functional of n? Let "[n] be the gound-state wavefunction determined by n (via HK), then … – doesn!t work: The potential is determined only up to a constant! Instead, define a universal functional (for given n):

DFT: Variational principle •

The variational principle: How to find the ground-state density? Since F does not depend on V ext, one can construct an energy functional where n and V ext can be chosen independently:

ext

Since E[n,V ] = "[n ] H "[n ] , one finds for a given V ext the corresponding ground-state density from the following variation:

!

! Use this instead of the eigenvalue problem of the frozen-lattice Hamiltonian for the ground-state determination!



But: No explicit expression is known for F!



And: How to vary n in a systematic way? ! Kohn–Sham approach!

9

DFT: Kohn–Sham approach

10



Basic idea: Apply the variational principle to a formally non-interacting, auxiliary many-electron system to obtain an effective external potential leading to the same ground-state density and energy as the true system.



Main advantage: The many-particle wavefunction can be composed of single-particle ones; the problem is separable into single-particle equations [Phys. Rev. 140, A1133 (1965)] ! Index “s”



What happens to F? Remember:

! For vanishing W e-e it reduces to Ts: •

How to link this to the true, interacting system? First consider the Hartree (classical Coulomb) interaction as functional of n:

Use this to define the exchange and correlation (XC) functional:

DFT: Kohn–Sham approach •

What does “exchange and correlation functional” mean? – Exchange energy: non-classical energy contribution due to the electrons being Fermions ! antisymmetric wavefunctions; but here: density ! explicit correction needed!

– Correlation energy: Electrons aren!t independent classical particles but can show interference effects (! phase of the common wavefunction) ! “correlation”: all nonclassical behaviour beyond exchange effects ! The XC functional takes care of all nonclassical effects which are lost due to the use of a density instead of a wavefunction! •

11

Define the corresponding XC potential:

In this way, XC effects enter the formally non-interacting auxiliary system, described by the …

DFT: Kohn–Sham equation •

Kohn–Sham equation: single-particle equation of the non-interacting auxiliary system (i.e., it has no physical meaning of its own!), …

… which contains an effective potential (the KS potential) taking care of the many-particle interactions (Hartree and XC interaction), …

… and which needs to be solved iteratively, since everything depends on the density (N is assumed even; factor 2 due to spin degeneracy):



Good news: This self-consistent scheme is mathematically exact, it provides the ground-state energy and density of the interacting system.



Bad news: No explicit expression is known for the exact XC functional.

12

Local-density approximation (LDA)

13



Define the Kohn–Sham XC energy density as an integral kernel:



Homogeneous electron gas: The XC energy density is spatially constant, depends just on the electron density n. It is known from quantum Monte Carlo calculations [Ceperley & Alder, Phys. Rev. Lett. 45, 566 (1980)].



Local-density approximation (LDA): Replace the true XC energy density by that of the homogeneous electron gas, taken at the local density:



Advantage: LDA works surprisingly well, also for systems with significantly varying density; reason: (V˜ XC + V H ) fulfills the same sum rule as the exact expression ! a partial cancellation of errors can be expected.



Drawback: LDA overestimates the bonding ! too small lattice constants, too high bulk moduli ! and cohesive energies.

XC functionals beyond LDA

14



Corrections to LDA (minor improvements; increased effort worth only in certain cases, mainly when exchange splittings are underestimated): – Self-interaction correction (SIC): In LDA, the Coulomb self-interaction is not properly cancelled. – Exact exchange: Calculate the exchange contribution from the KS wavefunctions, not from the density.



Concepts beyond LDA (mainly tailored for specific problems): – Generalized gradient approximation (GGA): includes dependence on the gradient of the density ! semilocal functional; improves the cohesive energy, but worsenes the lattice constants; various construction schemes known (PW86, BP, LYP, PW91, PBE, …). – Extensions of GGA: include the second derivative of the density, include the kinetic energy density; motivation: modifications needed to correctly describe the |r| ! " limit of XC energy and XC potential simultaneously. – Hybrid functionals: mixing exact exchange and GGA parts; e.g. B3LYP:

Further approximations

15



Frozen-core approximation: Independent treatment of core and valence electrons; possible if their densities don!t overlap (and no overlap between cores of different atoms) ! KS equation for the valence electron density only ! significantly reduced numerical effort



Pseudopotential approximation: Further reduce numerical effort for valence electrons in the core region by avoiding the oscillations of their wavefunctions; acheived by replacing the singular atomic Coulomb potential with a finite potential, roughly like this: Wavefunction Potential true

pseudo

– Outside the “cutoff radius”, pseudopotential and pseudo wavefunction are identical to the true potential/wavefunction. – The generation of ab-initio pseudopotentials (from atomic all-electron calculations) is a “science of its own”; different schemes (NC, US, PAW).

Numerical treatment

16



Choice of basis functions: Represent all quantities by vectors/matrices, obtained as coefficients relative to a certain basis; most popular: plane waves, Gauß-like functions. – Advantage of localized functions: “natural choice” for molecules. – Advantage of plane waves: Fourier expansion, systematic convergence; choice of cutoff energy determines number of Fourier coefficients (k: inside first Brillouin zone, G: reciprocal lattice):



Brillouin zone summation: KS energy functional, to be minimized with respect to the density n in order to obtain E0el:

– For a solid: sum over KS eigenvalues ! band-structure energy

Numerical treatment – Band-structure energy: sum over all occupied states in the first Brillouin zone:

– Periodic boundary conditions ! dense mesh of “allowed” k points; too many for numerical calculations ! use special k points (SP), adapted to the crystal symmetry. Then: !

, with

• “Specifications” of DFT-LDA ab-initio calculations:

– name of the code (SIESTA, VASP, PWSCF, ABINIT, CASTEP, …) – type of pseudopotentials – numerical parameters (k-points: type and set size, cutoff energy, XC functional)

17

Excursion: The gap problem •

What is “the gap problem”? DFT calculations based on the KS equation do not reproduce the band-gap energy of semiconductors – because DFT is a ground-state theory, and the eigenvalues of the KS equation have no physical meaning! Therefore, it is a wrong expectation if one wants to get the correct gap here. (Note: Even with the correct XC functional, the KS equation cannot produce the correct gap ! it!s not due to the LDA!)



18

How to calculate a correct gap? Use Green!s functions (Dyson equation, GW approximation) or timedependent DFT (TD-DFT)! Then it looks like this:

Excursion: The gap problem

From: Aulbur et al., “Quasiparticle calculations …”, Solid State Physics, Vol. 54 (2000), p. 1

19

Summary (what to take along)

21



DFT (HK theorem, variational principle) is an exact theory for the ground-state of an arbitrary many-electron system.



The KS equation is an exact effective single-particle equation for an auxiliary system, its iterative solution provides the ground-state energy and density of the real system. The KS eigenvalues and wavefunctions have no physical meaning!



No excited states can be obtained from DFT (Hohenberg–Kohn–Sham theory), therefore also the band-gap energy cannot be obtained. It is a misuse of HKS theory to take the eigenvalue difference for the gap.



To obtain the gap, use a GW or a TD-DFT calculation.



Only using ab-initio pseudopotentials makes a DFT calculation a true ab-initio calculation.



Literature: – R. M. Dreizler, E. K. U. Groß: “Density Functional Theory”, Springer 1990 – H. Eschrig, “The Fundamentals of Density Functional Theory”, Teubner 1996 – J.-M. Wagner, Ph. D. thesis, 2004 (www.db-thueringen.de/servlets/DocumentServlet?id=9461)