Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
RDMFTNumerics
Reduced Density Matrix Functional Theory forMany Electron Systems
S. Sharma1,2, J. K. Dewhurst3, N. N. Lathiotakis2,4
and E. K. U. Gross2
1 Fritz Haber Institute of the Max Planck Society, Berlin, Germany2 Institut fur Theoretische Physik, Freie Universitat Berlin, Germany
3 School of Chemistry, The University of Edinburgh, Scotland4 Theoretical and Physical Chemistry Institute, The National Hellenic Research
Foundation, Greece
24 July 2008
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Many-body quantum theory
Begin with the Schrodinger equation for N particles:
HΨi(x1,x2 . . . ,xN ) = EiΨi(x1,x2, . . . ,xN )
x ≡ {r, σ}
H = −12
N∑i
∇2i +
N∑i
N∑j 6=i
1|ri − rj |
+N∑i
vext(ri)
vext(ri) = −M∑ν
Zν
|Rν − ri|
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Many-body quantum theory
Many-body theory involves solving for functions in 3N coordinates:Ψ(r1, r2, . . . , rN )
Element of a anti-symmetrised N -body Hilbert spaceH ⊗H ⊗ · · · ⊗H
Neon atom (10 electrons) wavefunction stored with 100 points ineach dimension would require ∼ 106 tons of DVDs for storage
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Reduced density matrix functional theory
Density
ρ(r) = N
∫Ψ(r, r2, . . . , rN )Ψ∗(r, r2, . . . , rN ) d3r2 . . . d
3rN .
E[ρ] = T [ρ] + Eext[ρ] + Eee[ρ]
One-reduced density matrix (1-RDM)
γ(r, r′) = N
∫Ψ(r, r2, . . . , rN )Ψ∗(r′, r2, . . . , rN ) d3r2 . . . d
3rN .
E[γ] =∫d3r′d3rδ(r− r′)
[−∇
2
2
]γ(r, r′) + Eext[γ] + Eee[γ]
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Reduced density matrix functional theory
Density
ρ(r) = N
∫Ψ(r, r2, . . . , rN )Ψ∗(r, r2, . . . , rN ) d3r2 . . . d
3rN .
E[ρ] = T [ρ] + Eext[ρ] + Eee[ρ]
One-reduced density matrix (1-RDM)
γ(r, r′) = N
∫Ψ(r, r2, . . . , rN )Ψ∗(r′, r2, . . . , rN ) d3r2 . . . d
3rN .
E[γ] =∫d3r′d3rδ(r− r′)
[−∇
2
2
]γ(r, r′) + Eext[γ] + Eee[γ]
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Reduced density matrix functional theory
One-Reduced density matrix functional theory (1-RDMFT)requires solving for functions in 6 coordinates: γ(r, r′)
Diagonalising the density matrix gives the natural orbitals andoccupation numbers
γ(r, r′) =∑
i
niφi(r)φ∗i (r′)
∫γ(r, r′)φi(r′)d3r′ = niφi(r)
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Reduced density matrix functional theory
Gilbert’s Theorem [PRB 12, 2111 (1975)] (HK for 1-RDM)
Total energy is a unique functional E[γ] of the 1-RDM
Ground-state energy can be calculated by minimizing
F [γ] ≡ E[γ]− µ
[∫γ(r, r) d3r −N
]Must ensure that γ is N -representable!Hence an additional constraint that 0 ≤ ni ≤ 1Proof by A. J. Coleman [RMP 35, 668 (1963)]
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Reduced density matrix functional theory
Three major differences from DFT
Exact kinetic-energy functional is known explicitly
T [γ] = −12
∫δ(r− r′)∇2γ(r, r′) d3r d3r′
so no kinetic energy in Exc (≡ Eee − EH)
There exists no Kohn-Sham system reproducing the exact γ(because γKS is idempotent)
There exists no variational equation
δF [γ]γ(r, r′)
= 0
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
No simple variational equation
N -representability condition 0 ≤ ni ≤ 1 leads to border minimum
One can still minimise but δF [γ]δγ(r,r′) 6= 0 at minimum
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Functionals
Given the exact 1-RDM we can compute the exact kinetic energy,Hartree energy and external potential energy explicitly
BUT the exchange-correlation interaction energy is an implicitfunctional of γ
Bother.
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Functionals
Given the exact 1-RDM we can compute the exact kinetic energy,Hartree energy and external potential energy explicitly
BUT the exchange-correlation interaction energy is an implicitfunctional of γ
Bother.
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Functionals of γ
Conditions to be satisfied by approximate functionals
xc hole integrates to -1
Lieb’s Conjecture: for Coulomb systems E(N) is upwardlyconvex
Lieb Oxford bound Exc ≥ −1.68∫ρ4/3(r)d3r
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Hartree-Fock
Exc[γ] = −12
∫|γ(r, r′)|2
|r− r′|d3r d3r′
xc hole integrates to -1 but E(N) is not upwardly convex.
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Muller functional [PRL 105A, 446 (1984)]
Exc[γ] = −12
∫(γp(r, r′))∗γ1−p(r, r′)
|r− r′|d3r d3r′ (p = 1/2)
xc hole integrates to -1 and energy is upward convex. Lieb-Oxfordbound violated for the homogeneous electron gas.
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Our functional (Sharma et al. cond-mat/0801.3787)
Exc[γ] = −12
∫|γα(r, r′)|2
|r− r′|d3r d3r′ (0.5 ≤ α ≤ 1)
Energy is upward convex. Lieb-Oxford bound satisfied for thehomogeneous electron gas. But xc hole does not integrate to -1.
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Results
Homogeneous electron gas
Finite systems
Solids
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Results for the homogeneous electron gas
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Results for the homogeneous electron gas
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Correlation energy for atoms and molecules
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Mott insulators
Solid TN Band gap
MnO 118 3.7FeO 198 2.5CoO 292 2.4NiO 523 4.1
Table: Neel temperature (in K) and band gap (in eV) above Neeltemperature for transition metal oxides.
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
functionalsresults
Band gaps for solids
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Numerical issues
Full-potential linearised augmented planewaves (FP-LAPW)
potential is fully described without any shape approximation
core is treated as Dirac spinors and valence as Pauli spinors
space divided into interstitial and muffin-tin regions
this is one of the most precise methods available
MT
I
II
I
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Numerical issues
Full-potential linearised augmented planewaves (FP-LAPW)
potential is fully described without any shape approximation
core is treated as Dirac spinors and valence as Pauli spinors
space divided into interstitial and muffin-tin regions
this is one of the most precise methods available
MT
I
II
I
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Numerical issues
Full-potential linearised augmented planewaves (FP-LAPW)
potential is fully described without any shape approximation
core is treated as Dirac spinors and valence as Pauli spinors
space divided into interstitial and muffin-tin regions
this is one of the most precise methods available
MT
I
II
I
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Numerical issues
Full-potential linearised augmented planewaves (FP-LAPW)
potential is fully described without any shape approximation
core is treated as Dirac spinors and valence as Pauli spinors
space divided into interstitial and muffin-tin regions
this is one of the most precise methods available
MT
I
II
I
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Numerical issues
Full-potential linearised augmented planewaves (FP-LAPW)
potential is fully described without any shape approximation
core is treated as Dirac spinors and valence as Pauli spinors
space divided into interstitial and muffin-tin regions
this is one of the most precise methods available
MT
I
II
I
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Natural-orbital minimisation
E[γ] = E[ni,Φi]
Self-consistent Kohn-Sham calculation is performed (anyfunctional can be used) : ψKS
j (r)1 Natural orbitals are expanded in Kohn-Sham wave functions
ΦRDMi (r) =
∑j
cij ψKSj (r)
2 Compute gradients of the total energy w.r.t. cij3 Use steepest-descent along the gradient cij → cij + λdE/dcij4 Use Gramm-Schmidt to orthogonalise the natural-orbitals
5 Goto step 1, or exit once convergence is achieved
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Occupation number minimisation
Constraints:∑
i ni = N and 0 ≤ ni ≤ 1
Define gi(κ) ≡ dE/dni − κ and
gi(κ) ≡{gi(1− ni) gi > 0gini gi ≤ 0
1 Compute dE/dni
2 Find κ such that∑
i gi(κ) = 03 Make change in occupation number: ni → ni + λgi(κ), for
largest λ which keeps occupancies in [0, 1]4 Goto step 1, or exit once convergence is achieved
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
(q −→ 0 term)
Exchange correlation energyinvolves integrals:∫ Φ∗
ik(r)Φjk′(r)Φ∗jk′(r
′)Φik(r′)
|r− r′|d3rd3r′
Require integral over k-pointdifferences (q = k− k′):
I =∫
BZ
f(q)q2
d3q
Integrable but very slowconvergence w.r.t. k-points.
Precalculate weights usingmany q-points (millions)
Wq =∫
Bq
1q2
d3q,
then
I '∑q
Wqf(q).
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Code used: EXCITING
J. K. Dewhurst, S. Sharma and E. K. U. GrossThe code is released under GPL and is freely available at:http://exciting.sourceforge.net/
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Summary
RDMFT for periodic solids is implemented within a FP-LAPWcode.
Produces very good results for wide range of systems.
New algorithm for minimisation of energy with respect tooccupation numbers.
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Atomisation energy for molecules
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Band gap using chemical potential
Perdew et al. PRL 49 1691 (82), Helbig et al. EPL 77 67003 (07)
µ(M) =δE(M)δM
= −I(N) N − 1 < M < N
µ(M) =δE(M)δM
= −A(N) N < M ≤ N + 1
∆ = −A(N) + I(N) = µ(N + η)− µ(N − η)
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Band gap for finite systems with exact functional
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Band ap for LiH [EPL 77, 67003 (2007)]
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Band ap for LiH [EPL 77, 67003 (2007)]
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Band gap for solids
Sharma et al. cond-mat/0801.3787
µ(η) has discontinuity at η = 0 withEg = limη→0+(µ(η)− µ(−η)) being identical to exactfundamental gap.
In the vicinity of η = 0 one finds a linear behavior
µ(η) = µ(η = 0−) +{
clη for η < 0Eg + crη for η > 0
with cl = 2∫ n−(r)
|r−r′|d3rd3r′ and cr = 2
∫ n+(r)|r−r′|d
3rd3r′
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Band gap for finite systems with exact functional
S. Sharma Reduced Density Matrix Functional Theory
RDMFTNumerics
Chemical potential for solids
S. Sharma Reduced Density Matrix Functional Theory