You are on page 1of 39

Low-lying excited stated of natural carotenoids

viewed by ab initio methods

Daniil Khokhlov∗ and Aleksandr Belov

Department of Chemistry, Lomonosov Moscow State University, Moscow, 119991, Russia

E-mail: daniilkh@phys.chem.msu.ru

1
Abstract

Low-lying excited states of carotenoids (the optically dark 2A− +


g and bright 1Bu )

are deeply involved in energy transfer processes in photosynthetic antennas such as

light-harvesting and non-photochemical quenching. Though any ab initio modeling of

these phenomena has to rely on precise energies of the carotenoid electronic states, their

accurate evaluation remains a challenging problem due to a different nature of the states

of interest. The paper aims to study how accurate are the excitation energies of the low-

lying excited states of certain open- and closed-chain carotenoids obtained by a state-of-

the-art multireference approach for electronic structure calculation. Here, DMRGSCF

and a perturbative approach based on DSRG-MRPT2 were used for treatment of static

and dynamic correlation, respectively. Nuclear geometries of the electronic states were

optimized with DFT-based approaches. It was demonstrated that spin-flip TD-DFT

can replace multiconfigurational methods for the geometry optimization of the 2A−
g

state, but not for the calculation of the excitation energy. Adiabatic excitation energies

to the 1B+ −1 with an appropriate


u state were shown to be within a margin of 1000 cm

flow-parameter value. Adiabatic excitation energies to the 2A−


g state for the open-chain

carotenoids lie within a range of experimental values (taking into account the broad

range of experimental estimates); for the closed-chain ones the error does not exceed

2000 cm−1 . Ab initio stationary (1A− + − +


g →1Bu ) and transient (2Ag →1Bu ) absorption

spectra were modeled for violaxanthin and lycopene, and they show a good agreement

with the experimental ones both in terms of vibronic structure and transition energies.

Introduction

Carotenoids, the natural substituted polyenes, play many important roles including, but not
limited to, light-harvesting in blue region of visible light 1,2 and dissipation of excessive en-
ergy in the form of heat. 3 Such a combination of almost contradictory properties is a result
of a unique structure of their excited electronic states, which are usually classified assuming
the quasi-C2h point group symmetry of polyenes together with the “particle-hole” symmetry

2
(+ for the “ionic”, − for the “covalent” states) arising from the Pariser-Parr-Popple Hamil-

tonian. 4 The optically accessible 1B+
u state resides energywise above the optically dark 2Ag

state, at least in terms of adiabatic excitation energies (from the relaxed geometry of the
ground state to the relaxed geometry of the excited state). Vertical excitation energies can-
not be immediately extracted from the experiment (especially for the 2A−
g state), thus they

are more contradictory. However, it should be noted that there is strong theoretical evidence
that this ordering is inverted for vertical energies. 5,6 Such relative position of energy levels
together with their position with respect to energy levels of chlorophylls makes possible the
aforementioned dual function. On one hand, a carotenoid can be excited into the 1B+
u state

by blue light. Further, this excitation energy can be transferred to the pool of chlorophylls
in the photosynthetic apparatus. On the other hand, the 2A−
g state acts as an acceptor of

excitation energy from the low-energy cluster of chlorophylls. 3,7–9 The 2A−
g state undergoes

rapid radiationless decay, thus serving as a sink for excessive excitation energy under high
illumination conditions. While the latter mechanism has been questioned recently, 10 it re-
mains one of the most probable pathways for non-photochemical quenching of chlorophyll
fluorescence, although it is probably not a trivial excitation energy transfer. However, the
mechanism should involve energy transfer to the 2A−
g state in some way, since an alternative

dissipation pathway involving the formation of cation-radical chlorophyll-lutein pair has only
limited evidence based on rather weak computational arguments 11 or experiments showing
only a small fraction of this pathway. 12
The structure of the low-lying electronic states of carotenoids is not only important for
understanding of the photosynthesis process, but it is itself an interesting challenge for com-
putational chemistry. The wavefunction of the “ionic” 1B+
u state is dominated by one singly-

excited determinant (corresponding to HOMO→LUMO excitation) and, therefore, can be


accurately described by means of single-reference methods such as time-dependent density
functional theory (TD-DFT) 13–15 or equation-of-motion coupled clusters (EOM-CC), namely
EOM-CC with singles and doubles (EOM-CCSD). 16 In contrast, dark “covalent” states, such

3
as 2A− −
g and 1Bu , have a large contribution of the doubly excited determinants (table 1), thus

requiring multi-reference methods as a natural framework for their accurate description 17


since single-reference methods are not accurate enough (even the EOM-CC method with
singles, doubles, and truncated triples, EOM-CC3 18 ). The general solution is to use mul-
ticonfigurational self-consistent-field (MCSCF) approach in one or another formulation to
account for static correlation with further perturbative correction by means of multirefer-
ence perturbation theory (MRPT). This approach is reasonably accurate for the “covalent”
states, however, it is much less accurate for the “ionic” 1B+
u state. MCSCF is not accurate

enough for its accurate description even in the active space including the entire conjugated
π-system 5,6,19 which is quite unusual for a state with one dominating singly-excited determi-
nant in the wavefunction decomposition. Lack of excitations to higher lying diffuse orbitals
(commonly referred to as contribution of dynamic correlation), supported by the impor-
tance of diffuse basis sets for accurate description of the 1B+
u state
20
and good performance
of EOM-CCSD, is one of the main reasons of such behavior.
This problem can be avoided when “double” active spaces including not only the entire
π-system, but a set of the same size with more diffuse 3pz orbitals are used as a reference.
While such an approach could be very promising for smaller polyenes, 19,20 its computational
complexity remains prohibitive for natural carotenoids which require up to 52 active orbitals
(in the case of a 13-ene system) in the MCSCF active space. Density matrix renormal-
ization group (DMRG) 21–25 is capable of dealing with such large active spaces and thus is
extremely well-suited for treatment of static correlation in linear polyenes. 5,6,26–28 However,
further post-SCF perturbative treatment remains a significant computational bottleneck.
On the contrary, smaller active spaces can also be used to reduce error in the excitation
energy to the 1B+ 5
u state by reducing the perturbative correction by absolute value. While

this approach can be applied to much longer polyenes, it introduces additional bias into
the process of active space selection which is obviously undesirable. Another approach is to
replace commonly used perturbation theories such as complete active space second-order per-

4
turbation theory (CASPT2) 29 or N-electron valence state second-order perturbation theory
(NEVPT2) 6,19,30 by driven similarity renormalization group (DSRG) perturbation theory
(DSRG-MRPT2). 28 While DSRG-MRPT2 itself is not capable of completely removing a
“poor” reference problem, it produces excitation energies which are closer to the experimen-
tal data than the NEVPT2 ones. Moreover, it has a much better computational scaling
with respect to the number of active orbitals, therefore, in principle, it could pave the way
to “double” active space calculations with post-SCF MRPT treatment for long polyenes.
Nevertheless, the main goal of this paper is to explore the applicability of DSRG-MRPT2 to
natural carotenoids in “normal” active spaces, thus expanding the previous work 28 on linear
unsubstituted polyenes.
When discussing and modeling excitation energies into the excited “covalent” states, it
is crucial to distinguish between vertical and adiabatic excitation energies since the energy
gap between the states changes dramatically during geometry relaxation (by ca. 0.8 eV for
octatetraene 17 ). While relaxed geometries of the ground state (1A− +
g ) and the 1Bu state

can be optimized relatively easy by means of TD-DFT and these geometries (and excitation
energies calculated on them) are in excellent agreement with DMRG self-consistent field
(DMRGSCF) results, 28 geometries of the “covalent” states have to be obtained from MCSCF
(or DMRGSCF) optimization. This optimization requires numerical gradients from state-
averaged DMRGSCF and, despite the sufficient simplification due to the differentiation by
the state weight, 31 remains a very time-consuming process for natural polyenes. Thus, there
is an obvious desire to replace DMRGSCF by TD-DFT in the optimization process. However,
it can not be done in a straightforward manner due to the large contribution of doubly excited
determinants to the wavefunctions of the “covalent” states.
The remedy for the absence of double excitations in TD-DFT framework was proposed ca.
20 years ago, 32 namely spin-flip TD-DFT (SF-TD-DFT). In the conventional configuration
interaction (CI) singles approach (or in a very similar manner in TD-DFT) excited state
singlet wavefunction ΨSex of the excited state can be represented through spin-adapted singlet

5
Table 1: Excited state wavefunction structure for the low-lying excited states of
polyenes and carotenoids. Excitation operators Êia and T̂ia are described in the
text. H and L in subscript and superscripts denote HOMO and LUMO orbital,
respectively.
Electronic Wavefunction composition expressed through excitation operators acting on ...
state singlet HF reference |ΦS0 i triplet HF reference |ΦT0 i
L+1
2A−g c1 ÊH L
+ c2 ÊH−1 L L
+ c3 ÊH ÊH H
c1 T̂H−1 + c2 T̂LL+1 + c3 T̂HL
1B−u c1 ÊH + c2 ÊH−2 + c3 ÊH ÊH + c4 ÊH−1 ÊH−2 c1 T̂L + c2 T̂H−2 + c3 T̂HL+1 + c4 T̂H−1
L+2 L L+1 L+2 L L L+2 H L

1B+u c1 ÊHL
|ΦS0 i c1 T̂LL + c2 T̂HH

  √
excitation operators Êia = â†a,α âi,α + â†a,β âi,β / 2 (α and β denotes spin direction) acting
on a singlet Hartree-Fock (HF) determinant |ΦS0 i as follows:

X
ΨSex = cSia Êia |ΦS0 i, (1)
i,a

where indexes i and a correspond to core and virtual orbitals, respectively, cSia – decom-
position coefficients obtained by diagonalization of Hamiltonian in the basis of excited de-
terminants, ↠and â are creation and annihilation operators, respectively. In contrast, the
spin-flip counterpart uses triplet high spin (S = 1, MS = 1) HF determinant |ΦT0 i as a refer-
ence. In this formalism, the same triplet wavefunction ΨSex is represented through spin-flip
down excitation operators T̂ia = â†a,β âi,α acting on |ΦT0 i as follows:

X
ΨSex = cTia T̂ia |ΦT0 i, (2)
i,a

Further response equations used to determine decomposition coefficients cTia are quite similar
to the ones used in conventional TD-DFT. 32
Spin-flip ansatz allows to include several important double excitations (with respect to
the singlet reference |ΦS0 i) using only single excitation operators T̂ia . Thus, SF-TD-DFT
has almost the same computational complexity as TD-DFT, but adds several physically
important configurations and, in general, allows for a higher number of degrees of freedom in
the CI wavefunction decomposition. Certainly, SF-TD-DFT does not account for all double

6
excitations, but it captures the most important ones for the low-lying doubly excited states.
Inherent wavefunction structure of the 2A− − +
g , 1Bu , and 1Bu states within the SF-TD-DFT

framework is shown in the table 1. As can be seen, all “covalent” states of interest can now be
represented as a linear combination of spin-flip operators T̂ia acting of |ΦT0 i. Since a physically
correct form of the wavefunction is used (in contrast to a conventional TD-DFT), the correct
state ordering and reasonably accurate excitation energies might be expected. Indeed, SF-
TD-DFT was shown to significantly improve description of the 2A−
g state in short polyenes

(up to 5 conjugated double bonds). 33 However, it should be noted that performance of SF-
TD-DFT depends not only on the functional chosen (as its conventional analogue), but also
on the way of how exchange-correlation kernel contribution to the coupling between excited
determinants is treated. The originally proposed 32 so-called collinear approximation includes
in the coupling only the exact HF exchange, thus the SF-TD-DFT results strongly depend on
a fraction of exact HF contribution to the functional. So, it is not surprising that functionals
with a large HF fraction such as “half-and-half” BHHLYP are shown to be more accurate in
general, and for polyenes in particular. 33 A more complex non-collinear approach has been
implemented, 34 and while it performs better for the functionals with a small fraction of HF
exchange, for “half-and-half” functionals the accuracy of both approaches is comparable.
The goal of the present paper is twofold. First, it is to examine whether the SF-TD-DFT
optimization of the 2A−
g state could replace DMRGSCF optimization in terms of accuracy.

Second, it is to explore the applicability of DMRGSCF with the DSRG-MRPT2 correction


computational protocol to natural polyenes. The paper is organized as follows. First, we
describe computational details, namely geometry optimization, calculation of excitation ener-
gies, and calculation of absorption spectra. Second, we compare various functionals used for
geometry optimization of several carotenoid-like molecules. Third, we compare the obtained
results with the experimental data. Last, we examine the applicability of the approach to
carotenoids, discuss its perspectives and possible improvements. Also, we examine whether
it is possible to use TD-SF-TDDFT both for geometry optimization and calculation of the

7
excitation energies.

Computational details

Geometry optimization

Nuclear geometries of three open-chain carotenoids (neurosporene, spheroidene, and ly-


copene), three closed-chain ones (violaxanthin, lutein, and zeaxanthin) were relaxed to the
minimums of the electronic states of interest, namely 1A− − +
g , 2Ag , and 1Bu . Initially, the struc-

ture of zeaxanthin was taken from X-ray data. 35 Other closed-chain carotenoids were con-
structed from zeaxanthin. Open-chain carotenoids were built from scratch. Non-conjugated
and non-polar parts of the open-chain carotenoids were removed (and replaced by a hydro-
gen atom) to simplify calculations as shown in the fig. 1. Additionally, the 1A− −
g and 2Ag

states of several carotenoid-like molecules, namely unsubstituted icosa-10-ene (10-ene here-


inafter), 2,6,2’,6’-tetramethyl-hexadeca-7-ene (6me-7-ene), and 3,4,7,8,5,́6-́hexahydro-7’,8’-
didehydrospheroidene (tetrahydrospheroidene) were optimized.
Due to a varying structure of the electronic states of interest, different optimization
approaches were used, albeit all of them were based on the DFT framework:

• the 1A−
g state was optimized in ORCA
36,37
by means of DFT with the CAM-B3LYP 38
functional (with a standard range-separation parameter µ =0.33);

• the 1B+
u state was optimized in ORCA
36,37
by means of TD-DFT with the CAM-
B3LYP 38 functional (µ =0.33);

• the 2A−
g state was optimized in GAMESS-US
39,40
by means of spin-flip TD-DFT with
two different hybrid functionals: BHHLYP (exchange: 50% HF exchange and 50%
B88 functional 41 , correlation: 100% LYP 42 ) and PBE0 43,44 . A restricted open-shell
HF triplet high spin determinant served as a reference for spin-flip excitations; the
collinear approximation was used for an exchange-correlation kernel 32 . Additionally,

8
the 10-ene molecule was also optimized by means of spin-flip TD-HF using the same
approach.

Double-zeta basis cc-pVDZ was used throughout all calculations.

Figure 1: Structures of the studied carotenoids. Removed parts are shown in red.

Excitation energies

We mainly followed the protocol which has earlier been proposed for unsubstituted polyenes
by the same authors 28 . DMRGSCF calculations were carried out in the active space including
the entire π-system to avoid an additional bias in its selection. Therefore, the active space
for open-chain carotenoids with N double bonds had (2N , 2N ) size, where the first number
in brackets denotes the number of active electrons and the second one – the number of

9
active orbitals. The same was true for closed-chain carotenoids, except for violaxanthin
and lutein, where not only (18,18) and (20,20) active spaces were used, but also (22,22) for
both molecules for the reasons described below. In DMRG calculations matrix product state
(MPS) tensors had a size of m = 500, except for violaxanthin with an active space of (22,22),
for which the size of m = 700 was necessary to achieve an appropriate accuracy. DMRGSCF
was carried out in a state-averaged manner. The three lowest electronic states (1A− −
g , 2Ag ,

and 1B− +
u ) were included in the averaging. The 1Bu state was excluded from the averaging

for two reasons. First, it resides very high energywise if dynamic correlation is not included.
Second, its absence from the averaging does not change DSRG-MRPT2 excitation energies
significantly as was shown earlier for unsubstituted polyenes. 28 The converged DMRGSCF
wavefunctions were used to calculate the necessary reduced density matrices (RDM) for
DSRG-MRPT2 calculations, and the state-averaged RDMs were used to calculate the dressed
DSRG Hamiltonian. Further, the dressed Hamiltonian was diagonalized by means of DMRG
to obtain the excitation energies for four electronic states (the three included in DMRGSCF
and the 1B+
u ).

The MP2 natural orbitals were the starting orbitals for all carotenoids in question. The π-
orbitals from the MP2 method were localized by means of the Pipek-Mezey procedure 45 and
used further in the DMRGSCF procedure without any modifications for all molecules except
for lutein and violaxanthin. The MP2 orbitals for the lutein molecule were resorted before
DMRGSCF in order to include the orbitals from the conjugated π-system and exclude the
non-conjugated π-orbital from the active space. The orbitals for the violaxanthin molecule
were modified in order to select a proper active space. Since the accuracy-performance
balance of DMRG calculations strongly depends on the quality of localization, it is desirable
to localize orbitals in such a way that one orbital is localized predominantly on one atom.
This criterion could be easily satisfied in unsubstituted polyenes and, fortunately, in many
natural carotenoids, where the active space π-orbitals ψi are linear combinations of atomic

10
pz orbitals of carbon atoms ϕj (C):

N
X
ψi = cij ϕj (C), (3)
j

where N – is the number of carbon atoms in the conjugated system and ci j – decomposition
coefficients. However, in violaxanthin 18 carbon π-orbitals mix with 4 orbitals ϕj (O) from
oxirane rings forming a set of 22 orbitals:

18
X 4
X
ψi = cij ϕj (C) + cij ϕj (O) (4)
j=1 j=1

This prevents the effective separate localization of any of 18 orbitals from this set. In order
to provide a reasonable active space for DMRG, all 22 orbitals were localized, 18 carbon
orbitals ϕj (C) were selected and mixed again into the delocalized ones (for the sake of a
better convergence in the DMRGSCF procedure) by means of Hŭckel-like procedure while
oxirane orbitals remained localized. The orbitals obtained in this way formed an (18,18)
active space which was used for violaxanthin in all further calculations. The initial (22,22)
active space was used for comparison only (hereinafter numbers in round brackets mean
number of electrons and orbitals in the active space).
The aforementioned calculations were carried out in an ad hoc modified version of PSI4 46
with Forte module 47 . Intermediate DMRG calculations, such as evaluation of energies and
reduced density matrices, were performed in StackBlock 23 . The cc-pVDZ basis set with
an auxiliary basis set cc-pVDZ-JK 48 for SCF calculations and with an auxiliary basis set
cc-pVDZ-RI 49 for correlated calculations was used.

Vibronic spectra

The displaced harmonic oscillator model was used to calculate the vibronic spectra for the
1A− + − +
g →1Bu and 1Ag →1Bu electronic transitions together with the Frank-Condon (FC)

11
approximation. The FC approximation is well justified since both transitions are optically
allowed with large transition dipole values, thus all the terms except the first one in the
Herzberg-Teller series can be neglected. Under these approximations the one-photon ab-
sorption intensity I(ω) depending on the light frequency ω reads as follows:

X En − Em
I(ω) ∼ ω|~µe |2 ρm hχm |χn i2 δ( − ω), (5)
m,n

where the nuclear vibrational wavefunction |χm i corresponds to the ground electronic state
and the vibrational wavefunction |χn i corresponds to the excited one, µ
~ e is the transition
dipole moment between the electronic states, En and Em are the energies of the vibronic
states, ρm is the Boltzmann population of the m-th vibronic state. The overlap integral
hχm |χn i is the Frank-Condon factor. Since the number of vibrational modes contributing
to the vibronic spectrum can be relatively high, we used the time-dependent approach to
calculate the spectra following the paper of Baiardi et el 50 developed for a more general
case. The main idea is rather common 51 and it is to replace Dirac δ-function with its Fourier
transform:
Z +∞
1
δ(ω) = eiωt dt (6)
2π −∞

Inserting this expression into eq. 5 and rearranging the summation and integration, one
obtains the following:

Z +∞ X
2
I(ω) ∼ ω|~µe | dt e−εm /kB T hχm |χn ihχn |χm ieiεn t/h̄ e−iεm t/h̄ ei(ωads −ω)t , (7)
−∞ m,n

where En − Em is rewritten using the energies of vibrational levels εm and εn of the ground
and excited states, respectively, as h̄ωads + εn − εm , i.e. h̄ωads is the adiabatic excitation
energy as defined above. Taking into account the completeness of the basis set of excited
P
state vibrational wavefunctions n |χn ihχn | = 1 and the fact that vibrational wavefunctions
are eigenfunctions of vibrational Hamiltonians H and H of the ground and excited states,

12
one gets:
Z +∞ X
I(ω) ∼ ω|~µe | 2
dt hχm |e−τ H e−τ H |χm iei(ωads −ω)t , (8)
−∞ m

where characteristic times are τ = 1/kB T − it/h̄ and τ = it/h̄. Since the vibrational
wavefunctions of the ground state also form a complete basis set, the remaining sum can be
replaced with the trace in an appropriate basis set:

Z +∞   Z +∞
−τ H −τ H i(ωads −ω)t
I(ω) ∼ ω|~µe |2
Tr e e e dt = ω|~µe | 2
χF C (t)ei(ωads −ω)t dt (9)
−∞ −∞

Here the correlation function χF C (t) contains all necessary information about the lineshape
of the vibronic spectrum. Here we omit the derivation of the equation 50 and only show its
explicit form under FC approximation and in the absence of Dushinsky effect 52 (i.e. rotation
of normal modes upon electronic excitation):

s
1 a·a
χF C (t) = 2N
exp(v − k), (10)
(ih̄) C · D

where the elements of vectors a, a, C, and D are expressed in the following way:

ωi ωi
ai = ai =  (11)
sinh (h̄τ i ω i ) sinh h̄τ i ω i

   
ωi h̄τ i ω i ωi h̄τ i ω i
ci = coth ci = coth C i = ci + ci (12)
h̄ 2 h̄ 2
   
ωi h̄τ i ω i ωi h̄τ i ω i
di = tanh di = tanh D i = di + di (13)
h̄ 2 h̄ 2
Y Y 2
v= di Ki2 k= di (Di )−1 Ki2 (14)
i i

The shift vector K corresponding to the dimensionless displacement between the normal co-
ordinates of the ground and excited states was expressed through the transformation matrix
from the mass-weighted Cartesian coordinates to the normal coordinates of the ground state

13
L, the diagonal matrix of atomic masses M, and the displacement vector ∆ between the
Cartesian coordinates of the excited and ground states in the following way:

K = LM1/2 ∆ (15)

The unitary matrix L was obtained as a matrix of eigenvectors of the mass-weighted Hessian
of the ground state. The sets of vibrational frequencies {ωi } and {ωi } were considered
identical in all calculations (combined with the absense of the Dushinsky effect that meant
that the Hessians of the two electronic states were the same). The Hessian was calculated
for the 1A−
g state only by means of DFT with CAM-B3LYP functional (cc-pVDZ basis set)

in ORCA package. 36,37 Rescaling of vibrational frequencies was not used in any form.

Results and discussion

Geometry optimization

We begin with the comparison of the SF-TD-DFT geometry optimization of the 2A−
g state

with the DMRGSCF optimization results. The bond length alternation (BLA) value defined
as a difference in length between the adjacent C-C bonds is one of the most useful metrics
to compare the geometries of the electronic states of polyenes since the BLA pattern along
the carbon backbone is highly sensitive to both electronic state and optimization method.
The 10-ene molecule serves as an example of an unsubstituted highly symmetric polyene for
which the BLA behavior is relatively easy to analyze.
Three different functionals with a varying fraction of exact HF exchange (100 % HF,
BHHLYP with 50% HF, and PBE0 with 25 % HF) were used for comparison. The rationale
behind such a choice of functionals is that the fraction of HF has an important effect on
collinear SF-TD-DFT energies, 33 and we can expect the same effect for their gradients with
respect to nuclear coordinates. Indeed, the difference in the BLA patterns for different

14
Figure 2: BLA patterns for the 2A−
g state of 10-ene.

functionals is obvious (fig. 2); the pattern for 100 % HF stands out sharply while the
difference between BHHLYP and PBE0 is relatively small. However, the effect of dynamic
correlation hidden in the exchange-correlation kernel seems to be more important than the
fraction of HF itself as can be seen from comparison with the MCSCF optimized geometries.
Two polar examples are included for comparison – CASSCF in a minimal active space
(4,4) 6 and DMRGSCF (20,20) 28 in a natural active space including the entire π-system
which can be considered a reference. The former completely lacks any dynamic correlation
contribution due to a small number of determinants in the wavefunction expansion and its
BLA pattern closely resembles that of SF-TD-DFT with 100 % HF denoted as SF-TD-HF.
In contrast, the latter has a very long wavefunction expansion which partially accounts for
dynamic correlation and matches well the results of SF-TD-DFT with two other functionals.
Though these results do not provide any quantitative metric for the quality of SF-TD-DFT
optimization of 2A−
g , at least they show that the method could give a quantitatively correct

15
BLA pattern for the 2A−
g state with appropriate functionals. In order to obtain quantitative

results, adiabatic excitation energies from the ground to the 2A−


g state will be compared.

Moreover, since further complication of molecular structure by substitution leads to less


illustrative BLA patterns, the excitation energies of substituted polyenes will be used only
for comparison between SF-TD-DFT and DMRGSCF geometries.
Table 2: Adiabatic excitation energies for the 1A− − −1
g →2Ag transition (in cm ) of
10-ene as a function of geometry optimization method.

Geometry minimization method


Adiabatic excitation energy method SF-TD-DFT
DMRGSCF
HF PBE0 BHHLYP
DMRGSCF 16600 17770 17380 17460
DMRGSCF+DSRG-MRPT2 13100 14980 13230 14210

Adiabatic excitation energies from the ground to the 2A−


g state were calculated using

DMRGSCF without post-SCF correction (denoted as DMRGSCF in the tables 2-4) and
DMRGSCF with a further post-SCF correction by means of DSRG-MRPT2 (flow parameter
s = 5). While the former shows clearly how far the SF-TD-DFT minimum is from the DM-
RGSCF minimum, the latter allows to find an estimate to the (unknown) DSRG-MRPT2
minimum of 2A−
g and, moreover, it is more useful in the context of application to polyenes

since the excitation energies of DSRG-MRPT2 will be compared with the experiment. The
simplest case, 10-ene (table 2), exhibits a surprisingly comparable performance for all func-
tionals except for 100 % HF (with a slightly higher error) in comparison with DMRGSCF.
While all the geometries are at the same energy “distance” from the DMRGSCF minimum,
in no case does it mean that they reside at the same point of the potential energy surface.
However, we were unable to identify any simple combination of shifts along normal modes
responsible for this difference, so the analysis will focus directly on the energies.
So, the situation becomes more interesting when we switch to the energies obtained
with a post-SCF correction. While DMRGSCF still corresponds to the minimal energy, the
behavior of different functionals varies greately. So, 100 % HF is now much worse with an
energy difference of ca. 2000 cm−1 which is in a complete agreement with the observation

16
about the poor performance of the smallest CAS(4,4). 6 Either a large active space (such
as the entire π-system), or an inclusion of dynamic correlation by means of an exchange-
correlation functional is necessary to produce a minimum close to the MRPT2 minimum.
The two remaining functionals also perform differently: while PBE0, surprisingly, exhibits an
outstanding agreement with DMRGSCF, the BHHLYP optimization introduces a significant
deviation from the DMRGSCF minimum. While it seems very attractive at this step to
declare that PBE0 is accurate enough to replace DMRGSCF in the optimization of the
2A−
g state, we think that this is premature for several reasons. First, the general goal is to

describe carotenoids rather than their unsubstituted analogues. Second, BHHLYP is known
to perform better in terms of both state ordering and excitation energies than hybrids with
lower fractions of HF exchange. 33 Last, it is not clear why PBE0 is more close to DMRGSCF;
additional testing may shed light on the reasons of such behavior.

Table 3: Adiabatic excitation energies for the 1A− − −1


g →2Ag transition (in cm ) of
6me-7-ene as a function of geometry optimization method.

Geometry optimization method


Adiabatic excitation energy method SF-TD-DFT
DMRGSCF
PBE0 BHHLYP
E(DMRGSCF) 20340 21350 21150
E(DMRGSCF+DSRG) 17620 17470 18230

Table 4: Adiabatic excitation energies for the 1A− − −1


g →2Ag transition (in cm ) of
tetrahydrospheroidene as a function of geometry optimization method.

Geometry optimization method


Adiabatic excitation energy method SF-TD-DFT
DMRGSCF
PBE0 BHHLYP
E(DMRGSCF) 18590 20630 19850
E(DMRGSCF+DSRG) 17400 16110 17710

Therefore, the same calculations were carried out for substituted polyenes – a small
model molecule 6me-7ene (table 3 optimized also by means of DMRGSCF(14,14)) and a
larger tetrahydrospheroidene molecule (table 4) which was earlier optimized by means of
DMRGSCF and used to study the impact of substituents on DSRG-MRPT2 calculations. 28

17
Indeed, they exhibit an interesting behavior. The conclusions about the DMRGSCF en-
ergies are broadly similar – both energies are higher (ca. 1000 cm−1 ) than the ones from
the DMRGSCF optimization except for an unusually poor performance of the PBE0 func-
tional for tetrahydrospheroidene. In the case of DSRG-MRPT2 energies, both functionals
have a comparable deviation (by an absolute value) for the rigid 6me-7ene molecule, and
a completely different behavior for the flexible tetrahydrospheroidene molecule. So, while
BHHLYP produces the excitation energy close to the one of DMRGSCF, optimization by
means of PBE0 leads to an unusually low excitation energy which is 1300 cm−1 lower than
the one of DMRGSCF which means that this minimum is even closer to the MRPT2 min-
imum than the DMRGSCF one. As a first guess, the difference could be attributed to a
computational error stemming from different conformational minimums. However, careful
examination of structures shows no meaningful changes in the polyene core and no signif-
icant relative rotation of flexible groups (figures are shown in Supplementary): the RMSD
between the two SF-TD-DFT geometries is only 0.03Åand that between the DMRGSCF and
BHHLYP geometries is 0.2Å. The relatively high difference of 1300 cm−1 in excitation en-
ergies for different geometries seem to be due to a cumulative effect of small relative atomic
displacements in different optimization protocols but a more specific explanation is hard to
come up with.
All these controversies lead us to decision to keep both functionals in further calcu-
lations to compare the performance of the combined approach (SF-TD-DFT optimization
and DSRG-MRPT2 excitation energy) with the experimental data and not with DMRGSCF,
since the latter certainly lacks the important dynamic correlation contribution to the energy.
However, we should note that the goal of the paper is to check whether SF-TD-DFT is an
appropriate replacement for the DMRGSCF optimization and not the extensive benchmark-
ing of the existing DFT functionals within the SF-TD-DFT framework. So, we conclude that
the replacement of DMRGSCF optimization by the SF-TD-DFT optimization is a promising
approach for large natural polyenes where the DMRGSCF optimization remains extremely

18
demanding, but the precise choice of the functional would benefit from additional study.

Excitation energies

The accuracy of the experimental data is very different for the 1B+
u and the 2Ag states due

to their different accessibility in one-photon processes. While the former is easily observable
in one-photon absorption experiments, the latter is optically dark and could be detected
by more sophisticated methods with a low intensity of the signal. For this reason, we first

discuss the 1B+
u excitation energies, and further we will switch to the 2Ag state with all its

controversies. So, for the 1B+


u state (table 5), DSRG-MRPT2 shows satisfactory accuracy

Table 5: Experimental 0→0 and calculated DSRG-MRPT2 adiabatic excitation


energies (in cm−1 ) from the ground to the 1B+
u state.

Open-chain
Calculation
Expt (hexane 298 K 53 )
s=5 s=1
Neurosporene 21340 19110 21260
Spheroidene 20660 18050 20180
Lycopene 19810 17180 19130
Closed-chain
Calculation
Expt (EPA 293 K 54 )
s=5 s=1
Violaxanthin 21310 18470 20810
Lutein 20980 17780 20150
Zeaxanthin 20490 17160 19540

(the error varies from -2200 to -3300 cm−1 depending on the molecule) when experimental
data is compared with the converged DSRG-MRPT2 (flow parameter s = 5). DSRG-MRPT2
is considered to be converged in the sense that the flow parameter is large enough to cap-
ture almost all dynamic correlation included in the DSRG-MRPT2 model. This error is
slightly higher by absolute value than the one found previously for linear polyenes, 28 espe-
cially in the case of closed-chain polyenes. Nevertheless, the origin of this error for polyenes
in general is well known. 5,6,19 Most probably, it is unavoidable in the standard model where
DMRGSCF (or CASSCF) in the natural active space is followed by any MRPT. Two coun-

19
teracting factors should be taken into account in order to understand the worse performance
for the closed-chain carotenoids in comparison with the open-chain ones. First, substituents
in polyene chain tend, in general, to decrease excitation energy to the 1B+
u state. So, two

additional methyl groups decrease the excitation energy by 770 cm−1 55 in tetraene. Sec-
ond, closed-chain carotenoids with 10 and 11 double bonds have higher excitation energies
than their open-chain analogues due to decreased conjugation between the linear conjugated
backbone and the double bond(s) in the terminal rings. If the conjugation is stronger or the
impact of substituents is larger than it should be, the error will be larger than the one for
the open-chain carotenoids. We cannot discard both factors at this step. Conformational
flexibility is another reason which cannot be ruled out completely, but a wide selection of
objects motivates us to refrain from deep conformational analysis. Regarding the general
performance of DSRG-MRPT2, the previous findings about the accuracy of the proposed
approach 28 can be extended to carotenoids. As is the case for polyenes, if the flow parameter
is chosen to be unity, the absolute error in the excitation energy to the 1B+
u state does not

exceed 1000 cm−1 (and it is even lower for most studied carotenoids).
When discussing the excitation energy to the 2A−
g state, it is necessary to briefly review

the experimental data used. Observation of fluorescence from the 2A−


g state
54–57
and tran-
sient absorption 58–60 from the 2A− +
g state to the 1Bu state are the two main methods for

experimental study of the excitation energies in question. However, both methods are not
perfect in terms of accuracy. So, fluorescence from the 2A−
g state has a very low quantum

yield 61 and it could be easily mixed with the red wing of the fluorescence peak from the 1B+
u

state. Moreover, the FC factor is extremely small for the 0→0 transition between the 2A−
g

and 1A−
g states due to a significant displacement of PES minimums of the electronic states in

question. Regarding the transient absorption experiments, excitations to the vibrationally


excited 1B+
u could be confused with the 0→0 transition due to relatively poor resolution and

a narrow detection wavelength range. In addition, another experiment 53 in solution based on


accumulated one-photon absorption for open-chain carotenoids should be mentioned whose

20
results tell in favor of reassignment of vibrational origins for the 2A−
g states. However, this

reassignment suffers from the same problem as the fluorescence experiments – very small FC
factors for 0→0 between the 1A− −
g and 2Ag states. We intentionally do not discuss resonance

Raman data here, since it is highly likely that carotenoids have different conformations in
KBr than in nonpolar liquid solvents (such as hexane) or in vacuum.

Figure 3: Adiabatic excitation energies to the 2A− g state for open-chain carotenoids at the
DSRG-MRPT2 level. Calculated values are shown in blue and green depending on the
functional used for optimization of the 2A− g state (blue for BHHLYP, green for PBE0).
Filled circles represent calculated values, lines correspond to linear fits of energy with re-
spect to 1/(2N + 1). The experimental data from one-photon absorption experiment 53 is
shown by filled down triangles, the data from fluorescence experiments (neurosporene and
spheroidene, 56 lycopene 62 ) is shown by empty up triangles, the data from transient ab-
sorption experiments (neurosporene, 63 spheroidene, 58 lycopene 64 ) is shown by filled squares.
All experimental spectra were measured in hexane, except for transient absorption of neu-
rosporene measured in EPA.

The experimental and calculated energies are shown in fig. 3 and will be analyzed in
plots of excitation energy with respect to 1/(2N + 1), where N is the number of double
bonds in carotenoid molecules. It is well-known 65 that for linear polyenes and, in many

21
cases, for carotenoids linear dependence should hold in these coordinates. First, it should
be noted that experimental values fit this linear dependence very well within one method
(even if the data was taken from different papers of different authors 53,56,58,62,64 ). But,
there is an obvious discrepancy between different methods which cannot be explained by
the solvent effect (the 2A−
g state excitation energy is insensitive to the solvent and most

of the experiments were carried out in the same solvent - hexane) or by conformational
differences (almost all experiments were for carotenoids in non-polar solvent and not in
protein). Energy gap between the series is close to 1000 cm−1 which allows to relate it
to the vibrational quantum in the vibrational progression forming the lineshape of one-
photon spectra of carotenoids, though the latter is generally considered to be slightly higher
(ca. 1300-1500 cm−1 ). Therefore, the maximum energy difference between the different
experimental assignments is close to two vibrational quanta or, probably, one vibrational
quantum and an additional conformational contribution. All the calculated results (both
on BHHLYP and PBE0 geometries) reside in this range which could be considered a good
agreement. Although we abstain here from favoring any of the experimental assignments
in this section, we will provide some support to transient absorption experiments when the
modeled spectra will be discussed.
The comparison is less straightforward for the series of closed-chain carotenoids due to
the lack of consistent experimental data. The lowest estimates for all molecules came from
transient absorption experiment in the LHCII complex, 59 in which the molecules reside
in polar protein medium and can adopt protein-specific conformations. For this reason,
these energies are mostly shown for illustrative purpose due to the lack of experiments
featuring all carotenoids of interest in one experimental setup. As with the open-chain
carotenoids, the same trend is present – transient absorption experiments provide lower
excitation energies than the fluorescence ones. In contrast with the open-chain molecules,
a monotonous dependence of excitation energy on the reciprocal number of double bonds is
absent.

22
However, the calculated excitation energies still fit this linear dependence well although
with a smaller slope than in the case of the open-chain carotenoids. The lack of confor-
mational analysis seems to be the reason of such difference and all molecules “resemble”
zeaxanthin more than expected. The accuracy of DSRG-MPRT in general is worse for them
than for the open-chain molecules: not all excitation energies fit into the experimental range.
Nevertheless, the error does not exceed 2000 cm−1 for all molecules and depends on an ex-

Figure 4: Adiabatic excitation energies to the 2A− g state for the closed-chain carotenoids at
DSRG-MRPT2 level. The calculated values are shown in blue and green depending on the
functional used for optimization of the 2A− g state (blue for BHHLYP, green for PBE0). Filled
circles represent calculated values, lines correspond to linear fits of energy with respect to
1/(2N + 1). The experimental data from the fluorescence experiment in EPA 54 is shown by
filled up triangles, from the fluorescence experiment in hexane 66 – by empty up triangles. The
data from the transient absorption experiments in methanol 67 is shown by filled squares, and
the transient absorption data in recombinant LHCII complexes 59 is shown by empty squares.

perimental reference and optimization method. Such accuracy is comparable with the one
obtained for linear polyenes, 28 but, it is clear that this class of molecules requires a more
elaborated approach for each molecule based on their conformational stability. The problem

23
is that such an analysis should account for possible conformational changes taking place

during the internal conversion from the 1B+
u state to the 2Ag state. All experiments cited

above include this step, and the two-photon absorption experiment 68 for lutein is the only
experiment probing the ground state conformation. Interestingly, the energy estimate from
this experiment is 15100 cm−1 which differs from the values shown above. To sum up, though
the accuracy of the computational protocol can benefit from the conformational search on
the potential energy surface of the 2A−
g state, it is a non-trivial task and deserves a separate

study. Thus, the provided DSRG-MRPT2 accuracy is the worst estimate for the closed-chain
carotenoids.
Certainly, it would be interesting to evaluate the quality SF-TD-DFT calculations not
only for geometries of the 1A−
g state, but also for adiabatic excitation energies. These

energies are shown in fig. 5 both for open-chain and closed-chain polyenes. First, we should
note the dramatically different behavior of the two functionals in question – BHHLYP and
PBE0. While the distinction between them was not clear during the geometry optimization,
it cannot be overestimated when excitation energies are considered. Even the state ordering
is different – in the case of PBE0 the 1B+
u state has a lower excitation energy than BHHLYP

which contradicts the experimental data. 61 On the contrary, it is in a complete agreement


with the behavior observed for smaller polyenes 33 (although the B3LYP functional was used
instead of PBE0, both functional have a similar fraction of HF exchange). Due to this
incorrect behavior we strongly advise against using the functionals with a low fraction of
HF exchange for carotenoids, even if their energy minimum of the 2A−
g state is closer to the

MRPT2 result (as it was for tetrahydrospheroidene, table 4).


Although BHHLYP produces rather accurate results for the 1B+
u state, especially for the

open-chain carotenoids, the same cannot be said about the excitations to the 2A−
g state.

The lowest error estimate is ca. 2000 cm−1 for the open-chain carotenoids and even higher
for the closed-chain ones. The reason of such discrepancy can be not only the accuracy of
the current calculations, but also the interpretation of the experiments for the open-chain

24
carotenoids which yield the excitation energies that are greater by one vibrational quantum.
To conclude, the performance of SF-TD-DFT with BHHLYP applied to carotenoids is far
from perfect. Theoretically, a proper 50-50 functional could be constructed to provide better
energies, but that is not the aim of the current paper.

Open-chain Closed-chain
Figure 5: Adiabatic excitation energies of carotenoids at the SF-TD-DFT level. Excitation
energies corresponding to 1A− +
g → 1Bu are shown in orange, excitation energies corresponding
to 1A− −
g → 2Ag are shown in violet. Solid lines correspond to the BHHLYP functional,
dashed lines – to the PBE0 functional. Black vertical lines denote the range of experimental
excitation energies 61 to the 2A− +
g state, experimental data for the 1Bu state is shown by black
dots.

Spectral properties

Absorption spectra from the ground state and transient absorption spectra at room temper-
ature were modeled for lycopene and violaxanthin representing both open-chain and closed-
chain carotenoids. Both experimental spectra exhibit a well-resolved vibrational structure
that allows one to compare with the experiment not only the 0→0 excitation energies, but

25
also the vibrational progressions. Spectra for lycopene (fig. 6) were calculated with the

Figure 6: Experimental (shown in black 53,69 ) and calculated absorption spectra of the ly-
copene molecule corresponding to 1A− + − +
g →1Bu (blue) and 2Ag →1Bu (red).

DSRG-MRPT2 (s = 1) ab initio energies for the 0→0 transition, vibrational lineshapes were
calculated as described above. Spectral origins were shifted by 450 cm−1 in the case of 1A−
g

−1
→1B+
u transition and by 1000 cm for the 2A− +
g →1Bu in order to match the absorption

maximums of the experimental and calculated spectra for clarity. While this shift is still
necessary, it should be noted that it is relatively small even for the transient absorption
spectrum. Both spectra agree with the experimental ones reasonably well – vibrational pro-
gression is reproduced correctly and relative intensities of the peaks are qualitatively correct.
Naturally, the deviation is higher for the 2A− +
g →1Bu transition since the optimization meth-

ods for nuclear geometry are not in complete accord due to different functionals. Also, we
should note a larger distance between the vibrational peaks which is caused by two reasons:
an absence of any scaling for DFT frequencies and use of the ground state frequencies for
both electronic states. The same conclusions can be made for the spectrum of the closed-
chain carotenoid (fig. 7), though the agreement observed in terms of lineshapes is slightly
better. At the same time, a larger origin shift is necessary for the transient absorption

26
spectrum (2450 cm−1 ) due to a worse energy estimate for the 1B+
u state (table 5).

Figure 7: Experimental (shown in black 67 ) and calculated absorption spectra of the violax-
anthin molecule corresponding to 1A− + −
g →1Bu (blue) and 2Ag →1Bu (red).
+

The agreement between the experimental and calculated peak intensities in the transient
absorption spectrum allows us to theorize about an interpretation of the experimental data
which is very important in connection with the energy transfer in photosynthetic complexes.
The spectra were modeled as excitation spectra from the ground vibrational state (in fact,
vibrational states have Boltzmann distribution in the model, but at room temperature the
vibrational modes forming the vibrational progression are predominantly not excited), and
they match the experimental ones for both carotenoids. This good agreement is in accord
with the observations made during pump-probe experiments 61,63 in which the time required
for thermal equilibration of the 2A− + −
g state after the 1Bu →2Ag interconversion is shorter

than the delay between the pump and probe pulses. It is interesting that this effect has not
been detected in unsubstituted polyenes. Due to this, we abstained from any (re)assignments
of the experimental data when modeling their spectra. 6
If, therefore, we assume that the assignment by Polivka and coworkers 58,61,64,67 is correct
(which is backed not only by experimental works, but also by the presented ab initio model),

27
we can propose the following explanation for the observed differences between the experi-
mental datasets. The simplest interpretation can be done for the overestimated excitation
energies by Wang et al. - the intensity of the 0→0 transition in the absorption spectrum
corresponding to the 1A− −
g →2Ag excitation is so low that this state is virtually optically

inaccessible due to both a large displacement (in terms of normal modes) of the ground and
excited states and a low transition dipole moment. Therefore, the signal interpreted as the
0→0 transition is in fact the 0→2 transition (or the 0→1 transition to high-energy conforma-
tion of the 2A−
g state), which leads to an overestimation of the excitation energy of polyenes.

This hypothesis could be verified by modeling of the corresponding absorption spectrum,


but the low value of the transition dipole moment requires an inclusion of transition dipole
derivatives into the Herzberg-Teller series which seems prohibitive for natural carotenoids at
this moment. The interpretation of the difference with the fluorescence experiments is less
straightforward and there is an ongoing discussion about the relaxation dynamics between
the target excited states. It is in many cases carotenoid-specific, so we can only sketch out a
general concept of how the difference between the energies can originate, and this hypothesis
is mostly speculative. If the 2A−
g state has several conformational minimums which can

be accessed during relaxation from the 1B+


u state, they have different FC factors for the

transition to the ground state. The conformation “closer” (in terms of shifts along normal
modes) to the ground state will have a higher FC factor and, therefore, it will produce higher
fluorescence yield, but also a higher photon energy. In contrast, a deeper energy minimum
would correspond to a lower FC factor, and it could have a lower contribution to the fluo-
rescence spectrum, but would produce the main signal in the transient spectrum. However,
it is worth noting that such a hypothesis requires a robust computational verification.

28
Conclusion

The computational protocol earlier proposed for unsubstituted polyenes 28 has been modified
to efficiently carry out calculations for natural carotenoids following the objectives outlined
in the previous paper. The most challenging computational step, the SA-DMRGSCF ge-
ometry optimization of the 2A−
g state, was replaced by the SF-TD-DFT optimization which

was proven to give reasonable geometries. However, SF-TD-DFT itself is not capable to

accurately describe the excitation energies to both 1B+
u and 2Ag state, at least with the

tested functionals. Though there is still room for improvement, it is far from the goals of
the current paper which is focused on the multireference ab initio approach.
A combination of DMRGSCF followed by DSRG-MRPT2 perturbative correction de-

scribes well both the 1B+
u and 2Ag states, if an appropriate flow parameter value (s = 1) is

chosen. In this case, the error in the adiabatic excitation energy does not exceed 2000 cm−1
and, moreover, the energy gap between the 2A− +
g and 1Bu states crucial for modeling of their

interconversion is described reasonably well. Such an accuracy is enough to model both sta-
tionary adsorption spectra (from 1A− + −
g to 1Bu ) and transient absorption spectra (from 2Ag

−1
to 1B+
u ). While an empirical energy shift is still necessary, it does not exceed 2500 cm for
violaxanthin, and is even lower (1000 cm−1 ) for the open-chain carotenoid lycopene.
The modeled spectra demonstrated a good agreement in vibrational structure such as
the number of distinct absorption peaks and their relative intensities. Some disagreement
such as the distances between the peaks comes directly from the relatively simple model
for vibronic transition (the unscaled DFT Hessian of the ground state and the absence of
Dushinsky rotation). However, we would like to emphasize that the fundamental agreement
between the modeled transient spectrum and the experimental one was achieved in terms
of relative intensities of the peaks which was absent in linear polyenes taken as a model for
carotenoids. 6 Such an agreement allows us to criticize the assignment of excitation energies
of carotenoids proposed by Wang et al. 53 and express theoretical support for the assignment
following from transient absorption experiments.

29
Unfortunately, we have to state that the relative performance of the method is relatively
worse for the closed-chain carotenoids in comparison with the open-chain analogues. We
believe that the main reason is the absence of conformational analysis for the 2A−
g state,

and further exploration of its conformational landscape is desirable. Moreover, SF-TD-DFT


seems to be a reasonable tool to do it due to its low computational complexity. Regarding
the overall accuracy of the approach, further improvement can be made via expansion of the
active space with 3pz orbitals which has been proven to significantly improve the description
of the 1B+
u state. Though this would be significantly more computationally expensive, DSRG-

MRPT2 combined with DMRGSCF seems to be a promising tool due to a moderate scaling
of both methods with respect to the number of active orbitals. This work is currently in
progress.

Supporting Information Available

Relaxed nuclear geometries of carotenoids and trial molecules, absolute DMRGSCF and
DSRG-MRPT2 energies of the electronic states in respective nuclear geometries, and com-
parison of the 2A−
g geometries of tetrahydrospheroidene are provided in the Supplementary.

References

(1) Demmig-Adams, B. Carotenoids and photoprotection in plants: A role for the xantho-
phyll zeaxanthin. BBA - Bioenergetics 1990, 1020, 1–24.

(2) Frank, H. A.; Cogdell, R. J. Carotenoids in photosynthesis. Photochem. Photobiol.


1996, 63, 257–264.

(3) Ruban, A. V.; Berera, R.; Ilioaia, C.; van Stokkum, I. H. M.; Kennis, J. T. M.; Pas-
cal, A. A.; van Amerongen, H.; Robert, B.; Horton, P.; van Grondelle, R. Identification

30
of a mechanism of photoprotective energy dissipation in higher plants. Nature 2007,
450, 575–578.

(4) Pariser, R. Theory of the electronic spectra and structure of the polyacenes and of
alternant hydrocarbons. J. Chem. Phys. 1956, 24, 250–268.

(5) Taffet, E. J.; Lee, B. G.; Toa, Z. S. D.; Pace, N.; Rumbles, G.; Southall, J.;
Cogdell, R. J.; Scholes, G. D. Carotenoid nuclear reorganization and interplay of bright
and dark excited states. J. Phys. Chem. B 2019, 123, 8628–8643.

(6) Khokhlov, D.; Belov, A. Ab initio study of low-lying excited states of carotenoid-derived
polyenes. J. Phys. Chem. A 2020, 124, 5790–5803.

(7) Duffy, C. D.; Chmeliov, J.; Macernis, M.; Sulskus, J.; Valkunas, L.; Ruban, A. V.
Modeling of fluorescence quenching by lutein in the plant light-harvesting complex
LHCII. J. Phys. Chem. B 2013, 117, 10974–10986.

(8) Chmeliov, J.; Bricker, W. P.; Lo, C.; Jouin, E.; Valkunas, L.; Ruban, A. V.; Duffy, C.
D. P. An ‘all pigment’ model of excitation quenching in LHCII. Phys. Chem. Chem.
Phys. 2015, 17, 15857–15867.

(9) Balevičius, V.; Fox, K. F.; Bricker, W. P.; Jurinovich, S.; Prandi, I. G.; Mennucci, B.;
Duffy, C. D. Fine control of chlorophyll-carotenoid interactions defines the functionality
of light-harvesting proteins in plants. Sci. Rep. 2017, 7, 1–10.

(10) Gray, C.; Wei, T.; Polı́vka, T.; Daskalakis, V.; Duffy, C. D. P. Trivial Excitation Energy
Transfer to Carotenoids Is an Unlikely Mechanism for Non-photochemical Quenching
in LHCII. Front. Plant Sci. 2022, 12, 1–13.

(11) Dreuw, A.; Fleming, G. R.; Head-Gordon, M. Charge-transfer state as a possible sig-
nature of a zeaxanthin-chlorophyll dimer in the non-photochemical quenching process
in green plants. J. Phys. Chem. B 2003, 107, 6500–6503.

31
(12) Amarie, S.; Standfuss, J.; Barros, T.; Kühlbrandt, W.; Dreuw, A.; Wachtveitl, J.
Carotenoid Radical Cations as a Probe for the Molecular Mechanism of Nonphotochem-
ical Quenching in Oxygenic Photosynthesis. J. Phys. Chem. B 2007, 111, 3481–3487.

(13) Hsu, C. P.; Hirata, S.; Head-Gordon, M. Excitation energies from time-dependent den-
sity functional theory for linear polyene oligomers: butadiene to decapentaene. J. Phys.
Chem. A 2001, 105, 451–458.

(14) Peach, M. J.; Tellgren, E. I.; Salek, P.; Helgaker, T.; Tozer, D. J. Structural and
electronic properties of polyacetylene and polyyne from hybrid and coulomb-attenuated
density functionals. J. Phys. Chem. A 2007, 111, 11930–11935.

(15) Silva-Junior, M. R.; Schreiber, M.; Sauer, S. P. A.; Thiel, W. Benchmarks for elec-
tronically excited states: time-dependent density functional theory and density func-
tional theory based multireference configuration interaction. J. Chem. Phys. 2008, 129,
104103.

(16) Boguslawski, K. Targeting excited states in all-trans polyenes with electron-pair states.
J. Chem. Phys. 2016, 145, 234105.

(17) Serrano-Andres, L.; Lindh, R.; Roos, B. O.; Merchan, M. Theoretical study of the
electronic spectrum of all-trans-1,3,5,7-octatetraene. J. Phys. Chem. 1993, 97, 9360–
9368.

(18) Samanta, P. K.; Mukherjee, D.; Hanauer, M.; Köhn, A. Excited states with internally
contracted multireference coupled-cluster linear response theory. J. Chem. Phys. 2014,
140, 134108.

(19) Angeli, C.; Pastore, M. The lowest singlet states of octatetraene revisited. J. Chem.
Phys. 2011, 134, 184302.

32
(20) Dong, S. S.; Gagliardi, L.; Truhlar, D. G. Nature of the 11 Bu and 21 Ag excited states
of butadiene and the Goldilocks principle of basis set diffuseness. J. Chem. Theory
Comput. 2019, 15, 4591–4601.

(21) White, S. R. Density matrix formulation for quantum renormalization groups. Phys.
Rev. Lett. 1992, 69, 2863–2866.

(22) Chan, G. K.-L.; Head-Gordon, M. Highly correlated calculations with a polynomial


cost algorithm: a study of the density matrix renormalization group. J. Chem. Phys.
2002, 116, 4462–4476.

(23) Sharma, S.; Chan, G. K. L. Spin-adapted density matrix renormalization group algo-
rithms for quantum chemistry. J. Chem. Phys. 2012, 136, 124121.

(24) Keller, S. F.; Reiher, M. Determining factors for the accuracy of DMRG in chemistry.
Chimia 2014, 68, 200–203.

(25) Baiardi, A.; Reiher, M. The density matrix renormalization group in chemistry and
molecular physics: Recent developments and new challenges. J. Chem. Phys. 2020,
152.

(26) Ma, H.; Liu, C.; Jiang, Y. Theoretical study of the lowest π → π* excitation energies
for neutral and doped polyenes. J. Chem. Phys. 2005, 123, 084303.

(27) Ghosh, D.; Hachmann, J.; Yanai, T.; Chan, G. K.-L. Orbital optimization in the density
matrix renormalization group, with applications to polyenes and β-carotene. J. Chem.
Phys. 2008, 128, 144117.

(28) Khokhlov, D.; Belov, A. Toward an accurate ab initio description of low-lying singlet
excited states of polyenes. J. Chem. Theory Comput. 2021, 17, 4301–4315.

(29) Andersson, K.; Malmqvist, P.-Å.; Roos, B. O. Second-order perturbation theory with

33
a complete active space self-consistent field reference function. J. Chem. Phys. 1992,
96, 1218–1226.

(30) Angeli, C.; Cimiraglia, R.; Evangelisti, S.; Leininger, T.; Malrieu, J. P. Introduction of
n-electron valence states for multireference perturbation theory. J. Chem. Phys. 2001,
114, 10252.

(31) Granovsky, A. A. Communication: an efficient approach to compute state-specific nu-


clear gradients for a generic state-averaged multi-configuration self consistent field wave-
function. J. Chem. Phys. 2015, 143, 231101.

(32) Shao, Y.; Head-Gordon, M.; Krylov, A. I. The spin–flip approach within time-dependent
density functional theory: Theory and applications to diradicals. J. Chem. Phys. 2003,
118, 4807–4818.

(33) Rinkevicius, Z.; Vahtras, O.; Ågren, H. Spin-flip time dependent density functional the-
ory applied to excited states with single, double, or mixed electron excitation character.
J. Chem. Phys. 2010, 133.

(34) Bernard, Y. A.; Shao, Y.; Krylov, A. I. General formulation of spin-flip time-dependent
density functional theory using non-collinear kernels: Theory, implementation, and
benchmarks. J. Chem. Phys. 2012, 136.

(35) Linden, A.; Bürgi, B.; Eugster, C. H. Confirmation of the Structures of Lutein and
Zeaxanthin. Helv. Chim. Acta 2004, 87, 1254–1269.

(36) Neese, F. The ORCA program system. WIREs Comput. Mol. Sci. 2012, 2, 73–78.

(37) Neese, F. Software update: the ORCA program system, version 4.0. WIREs Comput.
Mol. Sci. 2018, 8.

(38) Yanai, T.; Tew, D. P.; Handy, N. C. A new hybrid exchange–correlation functional

34
using the Coulomb-attenuating method (CAM-B3LYP). Chem. Phys. Lett. 2004, 393,
51–57.

(39) Gordon, M. S.; Schmidt, M. W. Theory and Applications of Computational Chemistry;


Elsevier, 2005; pp 1167–1189.

(40) Barca, G. M. J.; Bertoni, C.; Carrington, L.; Datta, D.; De Silva, N.; Deustua, J. E.;
Fedorov, D. G.; Gour, J. R.; Gunina, A. O.; Guidez, E. et al. Recent developments in
the general atomic and molecular electronic structure system. J. Chem. Phys. 2020,
152, 154102.

(41) Becke, A. D. Density-functional exchange-energy approximation with correct asymp-


totic behavior. Phys. Rev. A 1988, 38, 3098–3100.

(42) Lee, C.; Yang, W.; Parr, R. G. Development of the Colle-Salvetti correlation-energy
formula into a functional of the electron density. Phys. Rev. B 1988, 37, 785–789.

(43) Adamo, C.; Barone, V. Toward reliable density functional methods without adjustable
parameters: The PBE0 model. J. Chem. Phys. 1999, 110, 6158–6170.

(44) Ernzerhof, M.; Scuseria, G. E. Assessment of the Perdew–Burke–Ernzerhof exchange-


correlation functional. J. Chem. Phys. 1999, 110, 5029–5036.

(45) Pipek, J.; Mezey, P. G. A fast intrinsic localization procedure applicable for ab initio
and semiempirical linear combination of atomic orbital wave functions. J. Chem. Phys.
1989, 90, 4916–4926.

(46) Parrish, R. M.; Burns, L. A.; Smith, D. G. A.; Simmonett, A. C.; DePrince, A. E.; Ho-
henstein, E. G.; Bozkaya, U.; Sokolov, A. Y.; Remigio, R. D.; Richard, R. M. et al. Psi4
1.1: an open-source electronic structure program emphasizing automation, advanced
libraries, and interoperability. J. Chem. Theory Comput. 2017, 13, 3185–3197.

35
(47) Forte, a suite of quantum chemistry methods for strongly correlated electrons. https:
//github.com/evangelistalab/forte, [Online; accessed 01-Oct-2021].

(48) Weigend, F. A fully direct RI-HF algorithm: implementation, optimised auxiliary basis
sets, demonstration of accuracy and efficiency. Phys. Chem. Chem. Phys. 2002, 4,
4285–4291.

(49) Weigend, F.; Köhn, A.; Hättig, C. Efficient use of the correlation consistent basis sets
in resolution of the identity MP2 calculations. J. Chem. Phys. 2002, 116, 3175–3183.

(50) Baiardi, A.; Bloino, J.; Barone, V. General Time Dependent Approach to Vibronic
Spectroscopy Including Franck–Condon, Herzberg–Teller, and Duschinsky Effects. J.
Chem. Theory Comput. 2013, 9, 4097–4115.

(51) Mukamel, S.; Abe, S.; Islampour, R. Generating function for electronic spectra of poly-
atomic molecules. J. Phys. Chem. 1985, 89, 201–204.

(52) F., D. The importance of the electron spectrum in multi atomic molecules. Concerning
the Franck-Condon principle. Acta Physicochim. USSR 1937, 551–556.

(53) Wang, P.; Nakamura, R.; Kanematsu, Y.; Koyama, Y.; Nagae, H.; Nishio, T.;
Hashimoto, H.; Zhang, J. P. Low-lying singlet states of carotenoids having 8-13 conju-
gated double bonds as determined by electronic absorption spectroscopy. Chem. Phys.
Lett. 2005, 410, 108–114.

(54) Josue, J. S.; Frank, H. A. Direct determination of the S1 excited-state energies of


Xanthophylls by low-temperature fluorescence spectroscopy. J. Phys. Chem. A 2002,
106, 4815–4824.

(55) Christensen, R. L.; Galinato, M. G. I.; Chu, E. F.; Howard, J. N.; Broene, R. D.;
Frank, H. A. Energies of low-lying excited states of linear polyenes. J. Phys. Chem. A
2008, 112, 12629–12636.

36
(56) Fujii, R.; Onaka, K.; Kuki, M.; Koyama, Y.; Watanabe, Y. The 2A−
g energies of all-

trans-neurosporene and spheroidene as determined by fluorescence spectroscopy. Chem.


Phys. Lett. 1998, 288, 847–853.

(57) Frank, H. A.; Josue, J. S.; Bautista, J. A.; van der Hoef, I.; Jansen, F. J.; Lugtenburg, J.;
Wiederrecht, G.; Christensen, R. L. Spectroscopic and photochemical properties of
open-chain carotenoids. J. Phys. Chem. B 2002, 106, 2083–2092.

(58) Polı́vka, T.; Zigmantas, D.; Frank, H. A.; Bautista, J. A.; Herek, J. L.; Koyama, Y.;
Fujii, R.; Sundström, V. Near-infrared time-resolved study of the S1 state dynamics of
the carotenoid spheroidene. J. Phys. Chem. B 2001, 105, 1072–1080.

(59) Polı́vka, T.; Zigmantas, D.; Sundström, V.; Formaggio, E.; Cinque, G.; Bassi, R.
Carotenoid S1 state in a recombinant light-harvesting complex of photosystem II.
Biochemistry-US 2002, 41, 439–450.

(60) Polı́vka, T.; Zigmantas, D.; Herek, J. L.; He, Z.; Pascher, T.; Pullerits, T.; Cogdell, R. J.;
Frank, H. A.; Sundström, V. The carotenoid S1 state in LH2 complexes from purple
bacteria Rhodobacter sphaeroides and Rhodopseudomonas acidophila: S1 energies, dy-
namics, and carotenoid radical formation. J. Phys. Chem. B 2002, 106, 11016–11025.

(61) Polı́vka, T.; Sundström, V. Ultrafast dynamics of carotenoid excited states-from solu-
tion to natural and artificial systems. Chem. Rev. 2004, 104, 2021–2072.

(62) Zhang, J.-P.; Fujii, R.; Qian, P.; Inaba, T.; Mizoguchi, T.; Koyama, Y.; Onaka, K.;
Watanabe, Y.; Nagae, H. Mechanism of the carotenoid-to-bacteriochlorophyll energy
transfer via the S1 state in the LH2 complexes from purple bacteria. J. Phys. Chem. B
2000, 104, 3683–3691.

(63) Niedzwiedzki, D. M.; Sandberg, D. J.; Cong, H.; Sandberg, M. N.; Gibson, G. N.;
Birge, R. R.; Frank, H. A. Ultrafast time-resolved absorption spectroscopy of geometric
isomers of carotenoids. Chem. Phys. 2009, 357, 4–16.

37
(64) Hörvin Billsten, H.; Herek, J. L.; Garcia-Asua, G.; Hashøj, L.; Polı́vka, T.;
Hunter, C. N.; Sundström, V. Dynamics of energy transfer from lycopene to bac-
teriochlorophyll in genetically-modified LH2 complexes of Rhodobacter sphaeroides.
Biochemistry 2002, 41, 4127–4136.

(65) Tavan, P.; Schulten, K. Electronic excitations in finite and infinite polyenes. Phys. Rev.
B 1987, 36, 4337–4358.

(66) Mechanism of nonphotochemical quenching in green plants: energies of the lowest ex-
cited singlet states of violaxanthin and zeaxanthin. Biochemistry 2000, 39, 2831–2837.

(67) Polivka, T.; Herek, J. L.; Zigmantas, D.; Akerlund, H.-E.; Sundstrom, V. Direct obser-
vation of the (forbidden) S1 state in carotenoids. Proceedings of the National Academy
of Sciences 1999, 96, 4914–4917.

(68) Walla, P. J.; Linden, P. A.; Ohta, K.; Fleming, G. R. Excited-State Kinetics of the
Carotenoid S1 State in LHC II and Two-Photon Excitation Spectra of Lutein and β-
Carotene in Solution: Efficient Car S1→Chl Electronic Energy Transfer via Hot S 1
States? J. Phys. Chem. A 2002, 106, 1909–1916.

(69) Hörvin Billsten, H.; Herek, J. L.; Garcia-Asua, G.; Hashøj, L.; Polı́vka, T.;
Hunter, C. N.; Sundström, V. Dynamics of energy transfer from lycopene to bac-
teriochlorophyll in genetically-modified LH2 Complexes of Rhodobacter sphaeroides.
Biochemistry-US 2002, 41, 4127–4136.

38
TOC Graphic

39

You might also like