Efficient calculation of electronic absorption spectra by means of

Efficient calculation of electronic absorption spectra
by means of intensity selected TD-DFTB
1,2
1
1
2
Robert R¨
uger , You Lu , Erik van Lenthe and Lucas Visscher
1
Scientific
2
Computing & Modelling NV
Division of Theoretical Chemistry, VU University Amsterdam
Summary of ground state SCC-DFTB
i
1
hφi|H |φii +
2
ˆ0
N
atom
X
AB
1
∆qAγAB ∆qB +
2
N
atom
X
I
UAB
AB
Orbital energies:
P P
I LCAO basis: φi(~
r) = A µ∈A cµiχµ(~r)
I Hamiltonian matrix elements are parametrized from DFT:

free atom

ε

Dµ E for µ = ν
ˆ
0
0
0 ˆ
hχµ|H |χν i =
χµ T + vKS [ρA + ρB ] χν
for µ ∈ A, ν ∈ B, A 6= B


0
otherwise
Self-consistent charge correction:
P 0 P
0
I Total density: ρ = ρ + δρ =
A ρA +
A δρA
~ A|)
I Density fluctuations: δρA(~
r) ≈ ∆qAξA(|~r − R
free atom
I ∆qA = qA − q
are calculated by Mulliken charge
A
analysis from the molecular orbitals:
Nel X X X
1
qA =
cµiSµν cνi + cνiSνµcµi
2 i
ν
γAB =
d3~r
2
1
δ
Exc
3 0
d ~r ξA(~r)
+
0
|~r − ~r | δρ(~r)δρ(~r 0)
Z
3
ρ
Example: Fullerene C60
[2]
= δij δab (εa − εi) + 4 εa −
Z
Z
S
Kia,jb
= d3~r d3~r 0φi(~r)φa(~r)
6
I large
number of low-lying single orbital transitions
with almost no oscillator strength:
Number of single orbital transitions
all
18% (fia > 0.005)
Absorption [arb. units]
{
800
all
15% (fia > 0.01)
500
400
300
3
4
200
all
fia > 0.001
fia > 0.02
fia > 0.1
1000
100
10
1
1
2
5
6
(εa − εi ) [eV]
I intensity
all
7% (fia > 0.05)
S/T p
εiKia,jb εb
− εj
2
δ Exc 1
0
0
+
φ
(~
r
)φ
(~
r
)
j
b
0
0
GS
|~r − ~r | δρ(~r)δρ(~r ) ρ
Eigenvalue equation that gives excitation energies and excited
states in the basis of the single orbital transitions
I Ω too large to be diagonalized directly ⇒ iterative methods
selection reduces the number of
excitations in the target window and makes the
calculation of interesting quantities possible,
e.g. spectral shift upon embedding of the
chromophore into the protein environment:
I
2
3
4
5
6
∆I [eV]
Tyrosine
Ubiquitin
Goal: Find an approximation for the coupling matrices K
that is easy to calculate within the SCC-DFTB framework!
lot of single orbital transitions have zero
oscillator strength due to symmetry
I essentially the same absorption spectrum can be
obtained with only 25% of the basis
I quality of the approximation decreases as more
transitions are removed
Original derivation by Niehaus:
I Atomic contributions to the transition density:
X
pij (~r) = φi(~r)φj (~r) =
pij,A(~r)
Results: Carefully applied intensity selection preserves the resulting
absorption spectra within the general accuracy of the TD-DFTB method,
while the reduced basis size together with the smaller number of excitations
per energy interval reduce the computational effort considerably.
Linear response TD-DFTB
[3]
S/T
Ia
7
Absorption [arb. units]
√
2
5
λ [nm]
Linear response excitations from Casida’s eq.:
Ω F~I = ∆I F~I
S/T
Ωia,jb
4
Energy[eV]
Example: Ubiquitin (1231 atoms)
all
25% (fia > 0.001)
{
Single orbital transitions:
I excited states and excitation energies
are straightforward to calculate
I in itself not a very good approximation
for the true excitations
I can be used as the basis of a
Nocc × Nvirt dimensional space in
which other approximations of the
excited states can be expressed
DFTB (Quasinano)
Question: Can one remove the single orbital transitions
with small oscillator strengths from the basis without changing
the resulting linear response excitations too much?
0 ξB (~r 0)
Summary of linear response TD-DFT
DFT (PBE-D3(BJ)+TZP)
Observations for non-intensity selected TD-DFTB:
I The dimension of the single orbital transition space grows quadratically with system size.
I Only a limited number of excitations can be calculated when using iterative eigensolvers.
~ia contribute little to the
I Single orbital transitions with small transition dipole moments d
q
P
2(εa−εi)
~
~
transition dipole moments of the linear response excitations: dI = ia
F
d
ia,I ia
∆I
Precalculate γ for all elements and distances using DFT:
Z
experimental [5]
Intensity selected TD-DFTB
µ∈A
I
TD-DFTB is an approximation of TD-DFT
⇒ same problems and limitations
I TD-DFTB results are on average not much
worse than TD-DFT, see [3] and [4]
I TD-DFTB is several orders of magnitude
faster than TD-DFT; possible applications:
I structures which are too large to be
treated with regular TD-DFT
I pre-screening of a large number of
different structures
Example: fac-Ir(ppy)3
Absorption [arb. units]
EDFTB =
Nel
X
The pros and cons of TD-DFTB
[1]
240
250
260
270
280
290
300
λ [nm]
I easier
to use than QM/MM with model systems!
A
I
Multipole expansion and monopolar approximation:
~ A|)
pij,A(~r) ≈ qij,AξA(|~r − R
X
X
1
qij,A =
cµiSµν cνj + cνiSνµcµj
2
ν
µ∈A
I
Insert expansion into the (singlet) coupling matrix:
X
S
Kia,jb =
qia,Aγ˜AB qjb,B
Z
γ˜AB =
I
3
d ~r
Z
AB
2
1
δ Exc
d ~r ξA(~r)
+
0
|~r − ~r | δρ(~r)δρ(~r 0)
3 0
ρGS
Ignore charge fluctuations and assume γ˜AB ≈ γAB .
Conclusions & Outlook
TD-DFTB is a useful tool to quickly calculate the absorption spectrum of large structures.
I Intensity selection can be used to further reduce the computational cost of TD-DFTB.
I An implementation of TD-DFTB with intensity selection will be available in ADF2014.
I
References
ξB (~r 0)
⇒ Straightforward way to build the coupling
matrix from ground state results and parameters!
[1] Augusto F. Oliveira et al., J. Braz. Chem. Soc. 20, 1193-1205 (2009)
[2] M. E. Casida and D. P. Chong, Recent Advances in Density Functional Methods, 1, p. 155, 1995
[3] Thomas Niehaus et al., Phys. Rev. B 63, 085108 (2001)
[4] Fabio Trani et al., J. Chem. Theory Comput. 2011, 7, 3304–3313
[5] J. Fine et al., Molecular Physics 110:15-16, 1849-1862 (2012)
mail: [email protected]