Professional Documents
Culture Documents
(Download PDF) Progress in Ultrafast Intense Laser Science Xiii 1St Edition Kaoru Yamanouchi Full Chapter PDF
(Download PDF) Progress in Ultrafast Intense Laser Science Xiii 1St Edition Kaoru Yamanouchi Full Chapter PDF
(Download PDF) Progress in Ultrafast Intense Laser Science Xiii 1St Edition Kaoru Yamanouchi Full Chapter PDF
https://ebookmass.com/product/fluorine-a-paradoxical-element-
volume-5-progress-in-fluorine-science-alain-tressaud/
https://ebookmass.com/product/laser-therapy-in-veterinary-
medicine-photobiomodulation-1st-edition-ebook-pdf/
https://ebookmass.com/product/laser-based-mid-infrared-sources-
and-applications-a-wiley-science-wise-co-publication-1st-edition-
konstantin-l-vodopyanov/
Chain Mobility and Progress in Medicine,
Pharmaceuticals, and Polymer Science and Technology
George Wypych
https://ebookmass.com/product/chain-mobility-and-progress-in-
medicine-pharmaceuticals-and-polymer-science-and-technology-
george-wypych/
https://ebookmass.com/product/laser-induced-breakdown-
spectroscopy-2nd-edition-jagdish-p-singh-editor/
https://ebookmass.com/product/progress-in-medicinal-chemistry-
volume-56-1st-edition-david-r-witty/
https://ebookmass.com/product/progress-in-optics-volume-62-1st-
edition-edition-taco-d-visser-eds/
https://ebookmass.com/product/algal-green-chemistry-recent-
progress-in-biotechnology-1st-edition-edition-rajesh-prasad-
rastogi/
Springer Series in Chemical Physics 116
Kaoru Yamanouchi
Wendell T. Hill III
Gerhard G. Paulus Editors
Progress in
Ultrafast Intense
Laser Science XIII
Springer Series in Chemical Physics
Volume 116
Series editors
A.W. Castleman Jr., University Park, USA
J.P. Toennies, Göttingen, Germany
K. Yamanouchi, Tokyo, Japan
W. Zinth, München, Germany
The Springer Series in Chemical Physics consists of research monographs in basic
and applied chemical physics and related analytical methods. The volumes of this
series are written by leading researchers of their fields and communicate in a
comprehensive way both the basics and cutting-edge new developments. This series
aims to serve all research scientists, engineers and graduate students who seek
up-to-date reference books.
123
Editors
Kaoru Yamanouchi Gerhard G. Paulus
Department of Chemistry Institute of Optics and Quantum Electronics
The University of Tokyo Friedrich Schiller University Jena
Tokyo Jena, Thüringen
Japan Germany
ISSN 0172-6218
Springer Series in Chemical Physics
Progress in Ultrafast Intense Laser Science
ISBN 978-3-319-64839-2 ISBN 978-3-319-64840-8 (eBook)
https://doi.org/10.1007/978-3-319-64840-8
Library of Congress Control Number: 2017954485
v
vi Preface
for her help with the editing processes. Last but not least, our gratitude goes out to
Dr. Claus Ascheron, Physics Editor of Springer-Verlag at Heidelberg, for his kind
support.
We hope this volume will convey the excitement of ultrafast intense laser sci-
ence to the readers and stimulate interdisciplinary interactions among researchers,
thus paving the way to explorations of new frontiers.
vii
viii Contents
xi
xii Contributors
F.H.M. Faisal
Abstract Despite its long-standing usefulness for the analysis of various processes
in intense laser fields, it is well-known that the KFR or strong-field approximation
(SFA) does not account for the final-state Coulomb interaction for ionisation. Var-
ious ad hoc attempts have been made in the past to face this problem within the
SFA, however, till now no systematic S-matrix expansion accounting for it has been
found. To overcome this long standing limitation of SFA we present here a systematic
series expansion of the strong-field S-matrix that could accounts for the final-state
Coulomb interaction in all orders.
1.1 Introduction
Over the past several decades the well-known strong-field approximation in the form
of the so-called KFR or SFA ansatz [1–3] has provided much fruitful insights into
the highly non-perturbative processes in intense laser fields. However, it is also well-
known that SFA does not account for the Coulomb interaction in the final state that is
specially significant for the ubiquitous ionisation process in strong fields. Due to this
problem, many authors in the past decades have made various heuristic corrections to
the SFA. Thus, for example, attempts to account for the Coulomb effect appear within
early ionisation models [4–6]. Other approaches include WKB-like approximations
[7, 8], semi-clssical and/or “quantum trajectory” approach [9–11], semi-analytic R-
matrix approach [12], and more recently an approach employing ansätze with phase
correction plus inhomogeneous differential equation [13]. Until now, however, no
systematic strong-field S-matrix theory could be found that unlike the usual plane-
wave SFA would be able to account for the laser plus Coulomb interaction in the
final state to all orders.
Here we present a strong-field S-matrix theory that overcomes this long standing
problem and derive a systematic all order S-matrix series explicitly incorporating
the laser plus Coulomb interaction in the final state. To this end we shall use be-
low a three-interaction formalism developed earlier in connection with the so-called
intense-field S-matrix theory or IMST (see, e.g. review [14] or, original references
cited therein).
In this section we briefly indicate the three-interaction technique suitable for the
problem at hand. The Schroedinger equation of the interacting atom+ laser field is
𝜕
(iℏ − H(t))|𝛹 (t)⟩ = 0 (1.1)
𝜕t
For example, for an effective one electron atomic system interacting with a laser
field, we may take
𝐩op 2 Ze2
Ha = ( − + Vs.r. (𝐫)) (1.3)
2m r
where Z is the core charge and Vs.r. (𝐫) is a short-range potential that goes to zero for
asymptotically large r faster than the Coulomb potential.
The laser-atom interaction is assumed here in the minimal coupling gauge (in
“dipole approximation”)
e e2 A2 (t)
Vi (t) = (− 𝐀(t) ⋅ 𝐩op + ) (1.4)
mc 2mc2
where 𝐀(t) is the vector potential of the laser field, and 𝐩op ≡ −iℏ∇.
Since all information of the interacting system is contained in the full wavefunc-
tion 𝛹 (t) and in general this is not known explicitly, we shall consider a more useful
formal expression of the full wavefunction in terms of the appropriate partial in-
teractions among the sub-systems and, the associated sub-propagators (or Green’s
functions). The latter objects may be already known, or could be found, to expand
the total wavefunction in terms of them.
Thus, first, we may formally define the full propagator, G(t, t′ ), associated with
the total Hamiltonian H(t), by the inhomogeneous equation
1 Strong-Field S-Matrix Series with Coulomb Wave Final State 3
𝜕
(iℏ − H(t))G(t, t′ ) = 𝛿(t − t′ ). (1.5)
𝜕t
The solution of the Schroedinger equation (1.1) can then be expressed as
|𝛹 (t)⟩ = |𝜙i (t)⟩ + G(t, t1 )Vi (t1 )|𝜙i (t1 )⟩dt1 (1.6)
∫
where |𝜙i (t)⟩ is a given initial state. We may note here already that due to the implicit
presence of the Heaviside theta-function in all the propagators (see, for example, the
Volkov propagator given in the sequel) the time integration limits are always from
a given initial time ti to a given final time tf since the limits of the intermediate
time-integrations are automatically controlled by the propagators at the appropriate
positions by themselves. Usually the interaction time interval tf − ti is taken to be
long, e.g., from −∞ to +∞. Note, however, that there is no difficulty in using the
theory for interactions with any finite laser pulse, for during the rest of the time, from
and to the long-time limits, the pulse could be assumed to be vanishingly small.
In general, as for the full wavefunction, we do not have explicit knowledge of
the full propagator G(t, t′ ). Therefore, we intend to express it in terms of certain
most relevant sub-propagators. Clearly, the two most relevant states in any quantum
mechanical transition process are the initial state, in which the system is prepared,
and the final state, in which the system is detected. Since in any ionisation process
the final state interaction is governed by the long-range Coulomb interaction of the
outgoing electron and the residual ion-core, it is highly desirable that the final state
incorporates the long-range Coulomb interaction from the beginning.
Let us define a final reference Hamiltonian Hf (t) that incorporates the final-state
Coulomb interaction in the presence of the laser field. Hf , and the corresponding
final-state interaction Vf (t) are related to each other by the total Hamiltonian, H(t),
𝜕
(iℏ − Hf (t))Gf (t, t′ )(t) = 𝛿(t − t′ ) (1.8)
𝜕t
Assuming for a moment that a suitable Hf (t) and Gf (t, t′ ) for the present purpose
could be found, the total G(t, t′ ) can be re-expressed in terms of Gf (t, t′ ) as
Substituting this in |𝛹 (t)⟩ above we get a closed form expression of the full state
vector in the form
4 F.H.M. Faisal
+ dt2 dt1 Gf (t, t2 )Vf (t2 )G(t2 , t1 )Vi (t1 )|𝜙i (t1 )⟩ (1.10)
∫
This formally closed form of the wavefunction of the interacting system has been
originally derived and discussed in connection with non-sequential double ionization
processes (see, review [14]). Here we shall make use of it for the problem at hand.
In fact, the transition amplitude (or the S-Matrix element Sfi ) from an initial state,
|𝜙i (t)⟩, to a final state |𝜓f (t)⟩ of the system is given, by definition, by the projection
of the final state on to the total wavefunction evolving from the initial state. Thus,
using the above form of |𝛹 (t)⟩, we get
= ⟨𝜓f (t)|𝜙i (t)⟩ + dt1 ⟨𝜓f (t1 )|Vi (t1 )⟩|𝜙i (t1 )⟩ +
∫
+ dt2 dt1 ⟨𝜓f (t2 )|Vf (t2 )G(t2 , t1 )Vi (t1 )|𝜙i (t1 )⟩ + ... (1.11)
∫
𝐩2op e e2 A2
HVol (t) = ( − 𝐀(t) ⋅ 𝐩 + ) (1.12)
2m mc 2mc2
The solutions of the corresponding Schroedinger equation are easily found
2
t pt′′
i
dt′′
𝜓𝐩 (𝐫, t) = ⟨𝐫|𝐩⟩e− ℏ ∫t′ 2m (1.13)
i
where 𝐩t ≡ (𝐩 − ec 𝐀(t)) and ⟨𝐫|𝐩⟩ = e ℏ 𝐩⋅𝐫 is a plane wave of momentum 𝐩.
The Volkov propagator GVol (t, t′ ) is the solution of the inhomogeneous equation
𝜕 𝐩2op e e2 A2
(iℏ − ( − 𝐀(t) ⋅ 𝐩 + ))GVol (t, t′ ) = 𝛿(t − t′ ). (1.14)
𝜕t 2m mc 2mc2
which is therefore given explicitly by:
i ∑ i t p2′′
dt′′
GVol (t, t′ ) = − 𝜃(t − t′ ) |𝐩⟩e− ℏ ∫t′
t
2m ⟨𝐩| (1.15)
ℏ 𝐩
G(t, t′ ) = GVol (t, t′ ) + GVol (t, t1 )V0 (t1 )GVol (t1 , t′ )dt1 + ⋯ . (1.16)
∫
Vi (t) = H(t) − Ha
e e2 A2
= (− 𝐀(t) ⋅ 𝐩op + ) (1.18)
mc 2mc2
For the final state, we intend to take account of the long-range Coulomb interac-
tion explicitly. One such state is the so-called “Coulomb-Volkov” state. It has been
introduced a long time ago [15, 16] by taking the usual stationary Coulomb-wave
and augmenting it heuristically by the time-dependent Volkov-phase:
2
t pt ′
−ℏ ∫i
dt′
𝛷𝐩 (𝐫, t) = 𝜙(−)
𝐩 (𝐫) × e
2m (1.19)
1 𝜋 i i
𝜙(−)
𝐩 (𝐫) = e 2 𝜂p 𝛤 (1 + i𝜂p )e ℏ 𝐩⋅𝐫 1 F1 (−i𝜂p , 1, − (pr + 𝐩 ⋅ 𝐫)) (1.21)
L
3
2
ℏ
To determine the Hamiltonian HCV (t) to which the Coulomb-Volkov state belongs,
we introduce a vector operator defined by
1 Strong-Field S-Matrix Series with Coulomb Wave Final State 7
∑
𝜋c ≡ |𝜙𝐬 ⟩𝐬⟨𝜙𝐬 | (1.22)
𝐬
where |𝜙𝐬 ⟩ stands for the Coulomb wave (“+” or “-”) with momentum 𝐬, cf. (1.21).
Consider next the exponential operator
We can write the Coulomb-Volkov Hamiltonian HCV (t) with the help of the op-
erator 𝜋c ,
where j ≡ 𝐩, stands for the momentum 𝐩 of the Coulomb wave state |𝜙(−) 𝐩 ⟩ and j ≡ D
stands for the discrete indices of the bound states |𝜙D ⟩ of the attractive Coulomb
potential. To establish that (1.27) indeed satisfies (1.26), let us first consider the case
{j ≡ 𝐩} and use (1.24) to calculate,
i
e ℏ 𝛼(t)⋅𝜋c |𝜙𝐩 ⟩ = T(𝜋c )|𝜙𝐩 ⟩
∑ i
= |𝜙𝐩 ⟩ − |𝜙𝐬 ⟩(1 − e ℏ 𝛼(t)⋅𝐬 )⟨𝜙𝐬 |𝜙𝐩 ⟩
𝐬
i
= |𝜙𝐩 ⟩ − |𝜙𝐩 ⟩(1 − e ℏ 𝛼(t)⋅𝐩 )
i
= e ℏ 𝛼(t)⋅𝐩 |𝜙𝐩 ⟩ (1.28)
Also we have
8 F.H.M. Faisal
e e
− 𝐀(t) ⋅ 𝜋c |𝜙𝐩 ⟩ = − 𝐀(t) ⋅ 𝐩|𝜙𝐩 ⟩ (1.29)
mc mc
Thus, substituting (1.27) in (1.26) for the continuum case, we get on the left hand
side
i t
(Ep + e
2 A2 (t′ )
)dt′ −𝛼(t)⋅𝐩) e2 A2 (t)
l.h.s. = e− ℏ (∫ 2mc2 (Ep + − 𝛼(t)
̇ ⋅ 𝐩)|𝜙𝐩 ⟩ (1.30)
2mc2
and on the right hand side,
i t
(Ep + e
2 A2 (t′ )
)dt′ −𝛼(t)⋅𝐩)
𝐩op 2 Ze2 e2 A2 (t′ ) e
r.h.s. = e− ℏ (∫ 2mc2 (( − )+ − 𝐀(t) ⋅ 𝐩)|𝜙𝐩 ⟩
2m r 2mc 2 mc
(1.31)
p 2
e
Noting that 𝛼(t)
̇ = mc 𝐀(t) and HCou |𝜙𝐩 ⟩ = Ep |𝜙𝐩 ⟩, where, Ep = 2m , one easily finds
from above that the l.h.s = r.h.s and hence the given solution is exactly fulfilled.
In a similar way it is easily seen that
∑ i
T(𝜋c )|𝜙D ⟩ = |𝜙D ⟩ − |𝜙𝐬 ⟩(1 − e ℏ 𝛼(t)⋅𝐬 )⟨𝜙𝐬 |𝜙D ⟩
𝐬
= |𝜙D ⟩ + 0
(1.32)
since, the overlap integral between the discrete and the continuum eigenstates of the
Coulomb Hamiltonian vanish by orthogonality, ⟨𝜙𝐬 |𝜙D ⟩ = 0. Hence, on substituting
(1.27) in (1.26) in the discrete case we get
i t
(ED + e
2 A2 (t′ )
dt′ +0) e2 A2 (t)
l.h.s. = e− ℏ ∫ 2mc2 (ED + + 0)|𝜙D ⟩ (1.33)
2mc2
and
i t
(ED + e
2 A2 (t′ )
)dt′ +0)
𝐩op 2 Ze2 e2 A2 (t′ )
r.h.s. = e− ℏ (∫ 2mc2 (( − )+ + 0)|𝜙D ⟩ (1.34)
2m r 2mc2
𝐩2 2
Moreover, ( 2m
op
− Zer )|𝜙D ⟩ = ED |𝜙D ⟩ and, therefore, the l.h.s = r.h.s and the verifi-
cation is complete.
To summarise, the complete set of solutions of the CV-Schroedinger equation
defined by (1.26) is given explicitly for the continuum case by
t p2
−ℏ ∫ i
( 2m + 12 A(t′ )2 − ec 𝐀(t′ )⋅𝐩)dt′
𝛷𝐩(−) (𝐫, t) = 𝜙(−)
𝐩 (𝐫)e
2c (1.35)
t p2
i
( 2m + 12 A(t′ )2 )dt′
𝛷D (𝐫, t) = 𝜙D (𝐫)e− ℏ ∫ 2c (1.36)
1 𝜋𝜂p i i
⟨𝐫|𝜙(−)
𝐩 ⟩ = e 2 𝛤 (1 + i𝜂p )e ℏ 𝐩⋅𝐫 1 F1 (−i𝜂p , 1, − (pr + 𝐩 ⋅ 𝐫)) (1.37)
L3∕2 ℏ
Zℏ
with 𝜂p = pa0
, and the well known hydrogenic bound states,
i
GCV (t, t′ ) = − 𝜃(t − t′ )
ℏ
∑ i t (𝐩− ec 𝐀(t′′ )2
dt′′
× { |𝜙𝐩 ⟩e− ℏ ∫t′ 2m ⟨𝜙𝐩 |
𝐩
∑ i t e2 A2 (t′′ )
)dt′′
+ |𝜙nlm ⟩e− ℏ ∫t′ (Enl + 2mc2 𝜙nlm |} (1.39)
nlm
We are now in a position to obtain the desired S-matrix amplitude. From the knowl-
edge of HCV (t) obtained above the rest-interaction in the final-state turns out to be,
Thus, finally, we have arrived at the desired systematic S-matrix series for the strong-
field ionisation amplitude, which systematically accounts for the final state long
range Coulomb interaction through the Coulomb-Volkov state in all orders. We quote
the first three terms more explicitly and, give the rule of construction for all the higher
order terms of the series:
∞
∑
Sfi = Sfi(n) (1.42)
n=0
where,
Sfi(0) = ⟨𝛷𝐩 (𝐫, t)|𝜙i (𝐫, t)⟩ (1.43)
i e e2 A2 (t1 )
Sfi(1) = − dt1 ⟨𝛷𝐩 |(𝐫1 , t1 )(− 𝐀(t1 ) ⋅ 𝐩op + )|𝜙i (𝐫1 , t1 )⟩ (1.44)
ℏ∫ mc 2mc2
i e
Sfi(2) = − dt dt ⟨𝛷 (𝐫 , t )|(− 𝐀(t2 ) ⋅ (𝐩op − 𝜋c ) + Vs.r. (𝐫2 ))GVol (𝐫2 , t2 ; 𝐫1 , t1 )
ℏ∫ 2 1 𝐩 2 2 mc
e e2 A2 (t1 )
× (− 𝐀(t1 ) ⋅ 𝐩op + )|𝜙i (𝐫1 , t1 )⟩ (1.45)
c 2mc2
i e
Sfi(3) = − dt dt dt ⟨𝛷 (𝐫 , t )|(− 𝐀(t3 ) ⋅ (𝐩op − 𝜋c ) + Vs.r. (𝐫3 ))
ℏ∫ 3 2 1 𝐩 3 3 mc
Ze2
×GVol (𝐫3 , t3 ; 𝐫2 , t2 )(− + Vs.r. (𝐫2 ))GVol (𝐫2 , t2 ; 𝐫1 , t1 )
r2
e e2 A2 (t1 )
×(− 𝐀(t1 ) ⋅ 𝐩op + )|𝜙i (𝐫1 , t1 )⟩
mc 2mc2
… (1.46)
where the angle brackets stand for the integration with respect to the space coordi-
nates and “⋯” stands for the higher orders terms. The higher order terms can be
written down easily, if required, for they follow the same pattern as the third order
term but are to simply extended by an extra intermediate factor GVol V0 and an extra
time integration for each successive order, to all orders.
1 Strong-Field S-Matrix Series with Coulomb Wave Final State 11
Before ending this report it is interesting to consider the S-matrix expansion of the
strong-field amplitude in the presence of an asymptotically short range potential.
This can be obtained simply by taking the limit Z = 0 in the result derived above. In
i
this limit the Coulomb waves 𝜙𝐩 (𝐫) reduce to the plane waves e ℏ 𝐩⋅𝐫 and the Coulomb-
Volkov state 𝛷𝐩 (𝐫, t) (1.19) reduces to the Volkov state (1.13). This implies that
the factor with the final state interaction in all terms, beginning with the second
order term, reduces to the short range potential Vs.r. (𝐫) only, due to the following
simplification
e e
⟨𝐩|(− 𝐀(t) ⋅ (𝐩op − 𝜋c ) + Vs.r. (𝐫)) = (− 𝐀(t) ⋅ (𝐩 − 𝐩)⟨𝐩| + ⟨𝐩|Vs.r. (𝐫))
mc mc
= ⟨𝐩|Vs.r. (𝐫) (1.47)
Also, the intermediate interaction V0 in all terms (from the second order term on-
wards) for Z = 0 reduces to the short-range potential Vr.s. (𝐫) only. Hence, in general,
for Z ≡ 0, the Coulomb-Volkov S-matrix series, (1.41), goes over to the simpler se-
ries
(a) For the sake of concreteness we have presented the result starting with the
Schroedinger equation of the interacting system in the minimal coupling gauge (so-
called velocity gauge). A similar result can be derived in the same way (or by a
gauge transformation) starting from the Schroedinger equation in the so-called length
gauge. This and the issue of gauge invariance of the theory will be presented and
discussed elsewhere.
(b) It is expected that the present theory would be helpful in clarifying a number
of issues of much current interest in strong-field physics involving (i) the shape of the
so-called “low energy structure” (LES) [18], (ii) the number of peaks associated with
the “very low energy structures” (VLES) [19, 20], (iii) origin of the “zero energy
structure” (ZES) [21], and (iv) possible existence of an as yet unknown “threshold
law” for the energy dependence of the strong-field ionization probability. Most or
all of these issues possibly depend crucially on the role of the long-range final state
Coulomb interaction specially in the low energy regime (e.g. c.f. [22]).
(c) The explicit expression of the Coulomb-Volkov propagator (or Green’s func-
tion) GCV (t, t′ ) given here suggests that the theory also would be useful for strong-
field processes involving excitation of the discrete states, either as a final state, or as
intermediate mediating states, or both, for example, in connection with the so-called
“frustrated ionization” (e.g. [23]) in strong fields.
(d) We may point out for that the terms of the S-matrix series (1.42), for example,
the amplitudes Sif(1) and Sif(2) , could be evaluated by a combination of stationary phase
method and numerical evaluation, provided the coordinates dependent Coulomb in-
tegrals can be evaluated analytically, for example by Norsieck’s method [24]. The
Coulomb integral of the first order amplitude (and of the first factor of the second
order amplitude) are of the form
(1) e
M𝐩,i = 𝜙(−)∗
𝐩 (𝐫)(− 𝐀(t) ⋅ 𝐩op )e−𝜅r d3 r (1.49)
∫ mc
Zℏ
where 𝜂(p) ≡ pa0
. The second Coulomb integral of the 2nd order amplitude is of the
form
(2) e
M𝐩,𝐤 = 𝜙(−)∗
𝐩 (𝐫)(− 𝐀(t) ⋅ (𝐩op − 𝜋c ))ei𝐤⋅𝐫 d3 r
∫ mc
e
= (− 𝐀(t) ⋅ (𝐤 − 𝐩)) 𝜙(−)∗ (𝐫)ei𝐤⋅𝐫 d3 r
mc ∫ 𝐩
(1.50)
They have the same form as of the following two prototypical integrals which we
give explicitly below:
References
2.1 Introduction
where H is the Hamiltonian, E is the eigenenergy, and 𝛹 is the wave function. The
non-relativistic ground state energy levels E0 of many-electron molecules (where the
number of electrons N ≤ 15) can be obtained to about 0.01 eV accuracy using the CI
method [2] or the CC method [3]. For small systems like He [4, 5] and H2 [6], the
exact eigenenergies of the bound states can be calculated. It remains a challenge to
obtain reliable estimates of the energies of highly excited states [7–9].
Much less effort has been spent on developing methods for obtaining solutions to
the time-dependent Schrödinger equation (TDSE),
𝜕𝛹 (t)
iℏ = H(t)𝛹 (t), (2.2)
𝜕t
where the Hamiltonian H(t) depends on time. The time-dependence of the Hamil-
tonian usually arises from the coupling with a laser field. Equation (2.2) describes
an initial-value problem, where we provide the initial wave function 𝛹 (t = 0) = 𝛹0 ,
which is the lowest energy solution to (2.1), and we seek an approximation to the
wave function 𝛹 (t) for 0 ≤ t ≤ T for some finite time T. To compute 𝛹 (t) could be a
significantly more difficult problem than solving the time-independent SE (2.1). The
reason is that we require a method that describes well the electron correlation not
only in the initial state 𝛹 (0) but also at later time t. Particularly in the case of intense
laser-molecule and laser-atom interactions, the wave function is strongly perturbed
by the laser field so that 𝛹 (t) becomes in general very different compared to the ini-
tial state 𝛹 (0). After the interaction with the laser field, the wave function becomes
a superposition of the initial state and more than a few excited states, which may
include singly excited states, doubly excited states, Rydberg states, and continuum
states representing ionization. Ideally, a theoretical method we develop can describe
all of these components equally well.
On the experimental side, a variety of phenomena have been observed, the mech-
anisms of which need to be explored theoretically. The best studied example is the
correlated motion of the two electrons in helium, leading to non-sequential double
ionization [10, 11] and the creation of wave packets consisting of doubly-excited
states [12]. Signatures of correlated electron dynamics can also be seen in high-
harmonic spectra of molecules [13, 14] and atoms [15], and in molecular dissocia-
tion [16, 17].
In the first part of this chapter, Sect. 2.2, we introduce the basics of multicon-
figuration-based wave function approaches to the approximate solution of the TDSE
(2.2). A multiconfiguration wave function is a wave function which consists of
several or many Slater determinants. The simplest wave function of this kind, in
which only one Slater determinant is included, is called the Hartree-Fock (HF)
wave function. Adding more Slater determinants gives a better description of the
dynamics of the system. Since the inclusion of all possible Slater determinants
is computationally unfeasible for any system having more than two electrons, the
essential point of a multiconfiguration method is to find a way of effectively reduc-
ing the number of Slater determinants. In Sect. 2.3, we review three approaches for
reducing the number of Slater determinants: the time-dependent configuration inter-
2 Multiconfiguration Methods for Time-Dependent Many-Electron Dynamics 17
action with single excitations method (Sect. 2.3.1), the time-dependent restricted-
active-space configuration-interaction method (Sect. 2.3.2), and the time-dependent
R-matrix theory (Sect. 2.3.3). These three methods all have in common that the Slater
determinants are independent on time. In Sect. 2.4, we review another two meth-
ods: the multiconfiguration time-dependent Hartree-Fock method (Sect. 2.4.1) and
the time-dependent complete active-space self-consistent field method (Sect. 2.4.2).
In these two methods a different approach is introduced: the Slater determinants
are made time-dependent. This means in general that, compared to the case where
time-independent Slater determinants are used, much fewer Slater determinants are
required to achieve a high level of accuracy. The drawback is that the dynamical
equations become nonlinear, and therefore, they become harder to solve.
In Sect. 2.5, we introduce our recent proposal of reducing the computational effort
in time-dependent multiconfiguration methods. In this method, called Factorized CI,
all Slater determinants are included in the expansion of the wave function, but a
product approximation is made for the CI coefficients.
A summary is given in Sect. 2.6. The reader may also refer to the recent articles
[18, 19] for a complementary review of time-dependent many-electron methods.
The problem that we have to solve is defined by (2.2), with the time-dependent
Hamiltonian
∑N
e2 ∑
N
1
H(t) = h(𝐫j , t) + . (2.3)
j=1
4𝜋𝜀0 j,k=1 |𝐫j − 𝐫k |
j<k
ℏ2 2
h(𝐫, t) = h0 (𝐫) + f (𝐫, t), h0 (𝐫) = − ∇ + V(𝐫), (2.4)
2me
1 ∑N
𝛷I (𝐫1 , s1 , … , 𝐫N , sN ) = √ 𝜀k1 ,⋯,kN 𝜒I1 (𝐫k1 , sk1 ) ⋯ 𝜒IN (𝐫kN , skN ), (2.8)
N! k1 ,…,kN =1
where 𝜀k1 ,⋯,kN is the totally antisymmetric Levi-Civita symbol. The coefficient CI (t)
in front of each Slater determinant is called a CI coefficient. A wave function defined
as in (2.7) automatically satisfies the Pauli principle: 𝛹 is antisymmetric under the
exchange of the coordinates of two electrons. An important consequence is that a
determinant vanishes if Ij = Ik holds for a set of j and k. In order to have each Slater
determinant appear only once in the expansion (2.7), we require that the the (index ) I
satisfies I1 < I2 < ⋯ < IN . This means that we can construct a total of Itot = 2M N
=
(2M)!
(2M−N)!N!
different determinants from the orbital set 𝜒j .
For simplicity, we usually assume a slightly less general form of (2.7), where we
have a set of M spatial orbitals 𝜙j (𝐫), j = 1, … M, and define 𝜒2j−1 (𝐫, s) = 𝜙j (𝐫)𝛼(s)
and 𝜒2j (𝐫, s) = 𝜙j (𝐫)𝛽(s). This means that we use the same set of spatial orbitals for
both the 𝛼 (up-spin) and 𝛽 (down-spin) electrons. If we restrict the wave function
expansion to determinants with N𝛼 𝛼 electrons and N𝛽 𝛽 electrons (N𝛼 + N𝛽 = N),
we obtain the expansion
∑ ∑
𝛹 (t) = CIJ (t)𝛷IJ = CIJ (t)‖𝜙I1 𝛼 ⋯ 𝜙IN 𝛼 𝜙J1 𝛽 ⋯ 𝜙JN 𝛽‖,
𝛼 𝛽
IJ IJ
∑
= CIJ (t)‖𝜙I1 ⋯ 𝜙IN 𝜙J1 ⋯ 𝜙JN ‖. (2.9)
𝛼 𝛽
IJ
The last line of (2.9) defines a convenient abbreviation where the 𝛽 orbitals are
distinguished from the 𝛼 orbitals by an overline. The orbitals 𝜙j in the last line
of (2.9) denote spatial orbitals, and it is implied that a spatial orbital with no
overline is always multiplied by an 𝛼 spin function to form a spin-orbital, and a
spatial orbital with an overline is multiplied with a 𝛽 spin function. Each deter-
minant is here labeled by two indexes, I = (I1 , … , IN𝛼 ) specifying the 𝛼-orbitals,
2 Multiconfiguration Methods for Time-Dependent Many-Electron Dynamics 19
and J = (J1 , … , JN𝛽 ) specifying the 𝛽-orbitals. In this case, there are Itot = L𝛼 L𝛽
( )
determinants, where Lx = NM and x = 𝛼 or 𝛽.
x
In order to apply a numerical approximation to the TDSE based on the ansatz
(2.9), we expand the orbitals 𝜙j in a finite number of basis functions
∑
nmax
𝜙j (𝐫) = bnj 𝜉n (𝐫), (2.10)
n=1
where 𝜉n (𝐫), n = 1, … , nmax is a set of linearly independent basis functions. Using the
basis set 𝜉n , we can construct M = nmax orthonormal orbitals 𝜙j . For example, assum-
∑
ing that 𝜙j is time-independent, we can take the bnj to satisfy m ⟨𝜉n |h0 |𝜉m ⟩bmj =
∑
𝜀j m ⟨𝜉n |𝜉m ⟩bmj , representing that the orbitals 𝜙j are eigenfunctions of the single-
electron Hamiltonian h0 in the space spanned by the set 𝜉n . We may also take 𝜙j to be
the Hartree-Fock orbitals, which are eigenfunctions of the Fock operator. The Fock
operator is described in more detail later in Sect. 2.3.1. Another option is to take
simply 𝜙j = 𝜉j , which would require an orthonormalization step by a method such
as the Gram-Schmidt procedure when 𝜉n is not an orthonormal set at the beginning.
A direct and straightforward way to solve the TDSE would be in the following
manner. First, starting from the time-independent orbitals 𝜙j , we construct all pos-
sible Slater determinants 𝛷IJ . Then, we compute the matrix representation of the
Hamiltonian using the basis of 𝛷IJ , HIJI ′ J ′ (t) = ⟨𝛷IJ |H(t)|𝛷I ′ J ′ ⟩, and integrate the
ordinary differential equation
L
dC (t) ∑ 𝛼 ∑ L 𝛽
This approach is referred to as the time-dependent full CI. The problem with time-
dependent full CI is that the number of Slater determinants becomes too large. To
see this, let us estimate the number of determinants Itot of the form (2.9) that can
be constructed from the orbital set 𝜙j . For a simple estimate for an atomic system
represented by the spherical coordinates, if a total of nrad
max = 100 basis function cov-
ang
ering the radial direction and nmax = 10 basis function covering the angular part, we
ang
need to have a total of nmax = nrad n
max max
= 103 basis functions covering the spher-
ical coordinates. We remark that this is still a conservative estimate. In previous
investigations, larger values were employed for the number of radial basis functions,
for example nrad rad
max = 250 [20] for two-photon ionization of He, nmax = 240 [21] for
rad
single-photon ionization of Be, and nmax ≈ 1600 [22] for strong-field ionization of
Be and Ne. With nmax = M = 103 , and assuming that N𝛼 = N𝛽 = N∕2 for even N,
and N𝛼 = N𝛽 + 1 = (N + 1)∕2 for odd N, the total number of determinants becomes
( M )2 ( 103 )2 ( M )( M ) ( 103 )( 103 )
Itot = N∕2 = N∕2 for even N, and Itot = (N+1)∕2 (N−1)∕2
= (N+1)∕2 (N−1)∕2
for odd N.
20 E. Lötstedt et al.
10
10
5
10
0
10
2 4 6 8 10 12 14
Number of electrons
where ∑
𝜓(𝐫1 , 𝐫2 , t) = Cjk (t)𝜙j (𝐫1 )𝜙k (𝐫2 ), (2.13)
jk
and Cjk is a symmetric matrix. For He, the spatial part 𝜓(𝐫1 , 𝐫2 , t) is usually expanded
in the spherical coordinates as [19, 23, 26]
2 Multiconfiguration Methods for Time-Dependent Many-Electron Dynamics 21
∑ m m
𝜓(𝐫1 , 𝐫2 , t) = ⟨l1 m1 l2 m2 |LM⟩𝛬LM
l l
(r1 , r2 , t)Yl 1 (𝜃1 , 𝜑1 )Yl 2 (𝜃2 , 𝜑2 ), (2.14)
1 2 1 2
LMl1 l2 m1 m2
In this section, we review several attempts to solve the TDSE using a wave function
of the form shown in (2.9), where the set of determinants included in the expansion
is restricted. In other words, we assume that the total wave function is expanded as
∑
𝛹 (t) = CIJ (t)𝛷IJ , (2.16)
IJ∈S
where the sum runs over a subset S of all possible determinants. Once the reduced
set S is defined, we can in principle solve the TDSE straightforwardly, that is, we
construct the matrix representation of the Hamiltonian HIJI ′ J ′ = ⟨𝛷IJ |H|𝛷I ′ J ′ ⟩ and
solve the initial value problem of
dCIJ (t) ∑
iℏ = HIJI ′ J ′ (t)CI ′ J ′ (t) (2.17)
dt I ′ J ′ ∈S
e2 ∑ 𝜙∗ (𝐫 ′ ) [
N∕2
]
F𝜙j (𝐫) = h0 (𝐫)𝜙j (𝐫) + d3 r′ k ′ 2𝜙k (𝐫 ′ )𝜙j (𝐫) − 𝜙j (𝐫 ′ )𝜙k (𝐫)
4𝜋𝜀0 k=1 ∫ |𝐫 − 𝐫 |
(2.20)
using the occupied orbitals 𝜙j , j ≤ N∕2, but apart from these orbitals, there are also
unoccupied orbitals 𝜙q , q = N∕2 + 1, … , nmax , which are eigenfunctions of F. As
seen in (2.10), nmax is the number of basis functions used to expand the orbitals.
Then, we define a singly excited configuration state function 𝛷j→q as the sum of the
two determinants where the orbital j is replaced by the orbital q (q > N∕2),
(
1
𝛷j→q = √ ‖𝜙1 ⋯ 𝜙j−1 𝜙q 𝜙j+1 ⋯ 𝜙N∕2 𝜙1 ⋯ 𝜙N∕2 ‖
2
)
+ ‖𝜙1 ⋯ 𝜙N∕2 𝜙1 ⋯ 𝜙j−1 𝜙q 𝜙j+1 ⋯ 𝜙N∕2 ‖ . (2.21)
Since an excited configuration state function 𝛷j→q is defined as a sum of two deter-
minants, the spin of 𝛷j→q vanishes, that is, Ŝ 2 𝛷j→q = 0. In general, a configura-
tion state function is defined as an eigenfunction of the total spin operator Ŝ 2 con-
structed from a linear combination of a few Slater determinants [1]. The reduced set
of determinants S in the TDCIS method is now defined as the reference determi-
nant 𝛷ref plus all singly excited configuration state functions 𝛷j→q , j = 1, … , N∕2,
q = N∕2 + 1, … , nmax . The number nCIS of singly excited configurations 𝛷j→q is
nCIS = (N∕2)(nmax − N∕2) ≈ Nnmax ∕2, which is a manageable number for systems
with N ≲ 100. Due to the property ⟨𝛷HF |H(0)|𝛷j→q ⟩ = 0 known as the Brillouin’s
theorem [1], the ground state wave function in the TDCIS model is the HF determi-
nant, provided that 𝛷ref = 𝛷HF .
2 Multiconfiguration Methods for Time-Dependent Many-Electron Dynamics 23
In [43], it was shown how to reformulate the equation of motion (2.17) for the
TDCIS method as an equation of motion for a set of time-dependent orbitals. By
∑nmax
defining 𝜁j (𝐫, t) = q=N∕2+1 cjq (t)𝜙q (𝐫) as a time-dependent orbital that describes
the excitations from the occupied orbital j, it is possible to formulate the TDCIS
equations of motion (2.17) as a linear coupled Schrödinger-type equation for the
∑ ̃ ̃
orbitals 𝜁j , iℏ𝜕𝜁j (𝐫, t)∕𝜕t = k H jk (t)𝜁k (𝐫, t) + C0 (t)Z(t), where the Hamiltonian Hjk
and the factor Z(t) were those derived in [43], and C0 (t) is the the CI coefficient of
the reference determinant 𝛷ref . It should be noted that H ̃ jk and Z(t) do not depend
on 𝜁j , which means that the TDCIS equations of motion are linear. By solving the
equation for 𝜁j , we can avoid calculating explicitly all the unoccupied orbitals 𝜙q ,
q > N∕2 by diagonalizing the Fock Hamiltonian.
The TDCIS method was applied to strong-field ionization of atoms in [43–46], in
particular to clarify the origin of the enhancement of the high-harmonic spectrum
in Xe at a high-harmonic energy of around 100 eV [45]. In [47–50], the TDCIS
method was used to study strong-field ionization of a variety of small molecules
such as CO2 , C2 H4 and NH3 . A large gaussian basis set was employed to expand the
spatial orbitals.
P1 = {𝜒1 , … , 𝜒p1 },
Pi = {𝜒pi−1 +1 , … , 𝜒pi } for 1 < i < K, and
PK = {𝜒pK−1 +1 , … , 𝜒2nmax }. (2.22)
We proceed by defining the total number of allowed electrons in each set, by giving
i
a list Nocc of occupation numbers, specifying the minimum number of electrons Nmin
i
and the maximum number of electrons Nmax in each set i,
24 E. Lötstedt et al.
[ 1 ]
Nocc = (Nmin , Nmax
1
), … , (Nmin
K
, Nmax
K
) . (2.23)
The determinants included in the reduced set S are now defined by these determi-
nants where the number of orbitals Ni included from the set Pi satisfies Nmin i
≤
Ni ≤ Nmax
i
. The total number of orbitals should equal the number of electrons,
∑K
N = N. As an example, let us consider a system with three electrons, ten spin-
i=1 i
orbitals, and the division P = (4, 8). This choice of P implies that the three spin-
orbital sets are P1 = {𝜒1 , 𝜒2 , 𝜒3 , 𝜒4 }, P2 = {𝜒5 , 𝜒6 , 𝜒7 , 𝜒8 }, and P3 = {𝜒9 , 𝜒10 }.
If we select the occupation numbers as Nocc = [(1, 2), (1, 2), (0, 1)], then determi-
nants like ‖𝜒1 𝜒2 𝜒7 ‖ and ‖𝜒2 𝜒6 𝜒10 ‖ are included in S since at most only two spin-
orbitals from the sets P1 and P2 are included, and at most only one from the set
P3 . On the other hand, a determinant like ‖𝜒1 𝜒9 𝜒10 ‖ is not included in S , since
there are two spin-orbitals from the set P3 .
In practice, the division of the spin-orbitals into the different subsets Pi is per-
formed based on the orbital energy, or based on the spatial distribution of the orbitals.
It is also possible to make the division of the spin-orbitals based on criteria that
depend on both the orbital energies and the spatial distributions [51]. In the applica-
tions of TDRASCI so far, the focus has been placed on a single photoionization. In
this case, the orbitals are partitioned in two sets, Pcore , which describes the region
close to the atomic core, and Pion , which contains orbitals describing the motion of
an electron far from the atomic core. The occupation numbers Nocc are chosen so that
there is at most one electron in the set Pion . It was suggested [51, 52] that we can
use HF orbitals for the set Pcore , and localized orbitals for the set Pion . See Fig. 2.2
for an illustration. The use of localized orbitals leads to a grid-type discretization of
the wave function in the outer region, which facilitates the treatment of the ejected
electron.
1
φ1
radial function (Å -1/2)
-0.5
0 10 20 30 40
radial distance (Å)
Fig. 2.2 Illustration of an orbital partition in the TDRASCI method. The region close to the atomic
core is represented by HF orbitals (𝜙1 , 𝜙2 , and 𝜙3 ), while the region far from the core is described
by localized orbitals (here exemplified by tent functions [53]). Note that in practical calculations
[51, 52] a much larger number of radial orbitals are used, compared to the number actually shown
in the figure
2 Multiconfiguration Methods for Time-Dependent Many-Electron Dynamics 25
where 𝐫N = (rN , 𝜃N , 𝜑N ) is the coordinate vector of the ejected electron. Note that
the wave function is not antisymmetrized with respect to rN , and that the time-
dependence of 𝛹outer originates exclusively from 𝜓q (rN , t). This results in a set of
single-particle Schrödinger equations for the electron in the outer region, interacting
with the core via local potentials [64]. The time-independent functions 𝛺q depend
on the coordinates of all the N − 1 bound electrons, and on the angles and spin coor-
dinates of the ejected Nth electron. The function 𝛺q is labeled by the channel index
q and is obtained by coupling (N − 1)-electron states of the residual ion with angular
and spin functions of the ejected electron, so that a total wave function of a specific
total angular momentum and spin is obtained. If we consider photoionization of He
26 E. Lötstedt et al.
as an example and assume a final state where the system as a whole (He+ + e− ) is
in a singlet P state, the 𝛺q describing a 1s state of He+ and an ejected electron with
angular momentum l = 1 would be
This kind of expansion was first conceived in the context of time-dependent vibra-
tional wave functions for the motion of nuclei within a molecule [75, 76]. This
method is called the multiconfiguration time-dependent Hartree method. The idea
behind using time-dependent electronic orbitals is to obtain a compact representa-
tion of the time-dependent total wave function 𝛹 (t). A time-dependent orbital is able
to vary in response to the change in the time-dependent Hamiltonian, and we there-
fore expect that a relatively small number of orbitals is needed to obtain a convergent
wave function. Even if a time-dependent orbital 𝜙j starts at t = 0 as a 2s orbital for
example, 𝜙j (t = 0) = 𝜙2s , at later times the same orbital may contain contributions
2 Multiconfiguration Methods for Time-Dependent Many-Electron Dynamics 27
from excited and continuum states, 𝜙j (t) = b2s (t)𝜙2s + b3p (t)𝜙3p + b𝜀p (t)𝜙𝜀p + …. If
we expand each orbital in terms of a basis set 𝜉n (𝐫),
∑
nmax
𝜙j (𝐫, t) = bnj (t)𝜉n (𝐫), (2.28)
n=1
where the coefficients bnj (t) are now time-dependent, the expansion (2.27) becomes
∑
Itot
𝛹 (t) = CIJ (t)‖𝜙I1 (t) ⋯ 𝜙IN (t) 𝜙J1 (t) ⋯ 𝜙JN (t)‖.
𝛼 𝛽
IJ
∑ ∑
= CIJ (t)bn1 I1 (t) ⋯ bnN I (t)bm1 J1 (t) ⋯ bmN J (t)
𝛼 N𝛼 𝛽 N𝛽
IJ n1 ,…,nN𝛼
m1 ,…,mN𝛽
This expansion should be compared with the expansion using the time-independent
basis functions as time-independent orbitals,
̃
∑
Itot
𝛹 (t) = ̃ IJ (t)‖𝜉I ⋯ 𝜉I 𝜉 J ⋯ 𝜉 J ‖.
C (2.30)
1 N 1 N 𝛼 𝛽
IJ
If we identify the indexes n and m in (2.29) with the indexes I and J in (2.30), we
see that the two equations are equal to each other because
∑
̃ IJ (t) =
C Cnm (t)bI1 n1 (t) ⋯ bIN nN (t)bJ1 m1 (t) ⋯ bIN mN (t). (2.31)
𝛼 𝛼 𝛽 𝛽
nm
Even though the two expressions (2.29) and (2.30) represent the same wave func-
tion if Itot = Ĩtot , it is hoped that the expression (2.29) is more efficient in the sense
that a small Itot < Ĩtot is sufficient to obtain convergent results. In fact, thanks to the
factorized approximation of the CI coefficients in (2.31), the number of determi-
nants Itot required in( the)( expansion
) (2.29) can be much smaller than the number of
determinants Ĩtot = nNmax nNmax included in the sum in (2.30).
𝛼 𝛽
An example of a system with two electrons is helpful to understand the general
idea. Let us assume a two-electron system, and a single particle basis set 𝜉n (𝐫) con-
sisting of nmax = 100 basis functions. In the approach in which time-independent
orbitals are used, we would write the wave function as
∑
nmax
𝛹 (t) = ̃ nm (t)‖𝜉n 𝜉 m ‖,
C (2.32)
n,m=1
28 E. Lötstedt et al.
where n2max = 104 parameters are required to specify the wave function. Now, we
assume that 𝛹 (t) can in fact be represented well by only two time-dependent orbitals
as
𝛹 (t) = C1 (t)‖𝜙1 (t)𝜙1 (t)‖ + C2 (t)‖𝜙2 (t)𝜙2 (t)‖, (2.33)
L𝛼 L𝛽
∑ ∑
𝛹 (t) = CIJ (t)‖𝜙I1 (t) ⋯ 𝜙IN (t)𝜙J1 (t) ⋯ 𝜙JN (t)‖. (2.34)
𝛼 𝛽
I=1 J=1
We assume that there are M spatial orbitals 𝜙j ( j = 1, … , M), and that each orbital
is expanded in terms of a basis set 𝜉n (𝐫), n = 1, … , nmax as in (2.28). The orbitals
are required to be orthonormal at all times, ⟨𝜙j (t)|𝜙k (t)⟩ = 𝛿jk . The sum in (2.34)
( )( )
goes over all the L𝛼 L𝛽 = NM NM determinants that can be constructed by the M
𝛼 𝛽
spatial orbitals 𝜙j . Note that the number of time-dependent orbitals M is an inde-
pendent parameter, which is not related to the size nmax of the basis set. Typically,
M ≈ N spatial orbitals are used in a MCTDHF calculation. An important property
of the MCTDHF wave function is that it is invariant under orbital transformations
∑
𝜙j → 𝜙′j = k Ujk 𝜙k , where Ujk is a unitary matrix. Given the new set of orbitals 𝜙′j ,
′
it is always possible to find a new set of CI coefficients CIJ such that the wave function
∑ ′ ′
can be written as 𝛹 = IJ CIJ 𝛷IJ , where 𝛷IJ are the determinants constructed from
′
the transformed orbitals 𝜙′j [79, 94]. This invariance under unitary orbital transfor-
mations is a property of full-CI type wave functions, where all possible determinants
are included in the expansion of the wave function. For example, the time-dependent
full CI wave function described in Sect. 2.2 is invariant under unitary orbital trans-
formations, but the TDCIS wave function (see Sect. 2.3.1) is not.
When M takes the minimum value, M = N∕2, the MCTDHF method becomes
the time-dependent restricted Hartree-Fock (TDHF) method. The TDHF method, in
which the wave function consists of one single determinant,
is one of the first methods introduced for going beyond the single active electron
approximation in laser-atom interaction [80]. Because of its relative simplicity, the
TDHF method has been applied to non-adiabatic electron dynamics of a variety of
molecular systems [81–87]. The major disadvantage of the TDHF method is that
ionization is not correctly described [80, 88, 89]. The ansatz (2.35) implies that in
the case of a two-electron system, both electrons are described by the same spa-
tial orbital 𝜙1 (𝐫). However, to model an ionization process in a qualitatively correct
way, we have to provide at least two orbitals, one orbital for the electron to be ejected,
and the other orbital for the bound electron. In order to solve this problem, a time-
dependent unrestricted Hartree-Fock (UHF) approximation [90, 91] was proposed,
but the more general MCTDHF method is preferable in most cases, since the MCT-
DHF method can describe electron correlation beyond the mean-field level, and it
includes UHF as a limiting case [89].
The MCTDHF method has been used to investigate mainly the laser-driven
dynamics of small, few-electron systems such as H2 [92–95], He [20, 96–98], Be
[18, 21], and LiH [99–102]. In order to reduce the numerical burden, the dimen-
sions of the systems were in many cases reduced [88, 89, 103–114]. Thanks to the
advances in computational methods, it has become possible to apply the MCTDHF
method to real, three-dimensional many-electron atoms and molecules in intense
laser fields, such as Ne [22, 115], HF [21], H2 O [116] and NO [117]. The calcu-
lation on NO [117] included M = 12 time-dependent orbitals, by which a total of
Itot = 392040 Slater determinants were constructed. Very recently, it was shown [22]
that the spectrum of the high-order harmonics generated using He as a non-linear
medium by a laser pulse with a central wavelength of 400 nm and a peak intensity of
4 × 1014 W/cm2 calculated with the MCTDHF method with M = 14 agreed perfectly
with the high-harmonic spectrum calculated by solving the TDSE using a method
similar to the time-dependent full CI method introduced in the last part of Sect. 2.2.
Before delving into some of the details of the theory behind MCTDHF, we point
out that a wave function ansatz like (2.34) has also been applied to describe bosonic
particles [118–122], where the total wave function is required to be symmetric under
particle exchange. The main application of this variant of MCTDHF is the time-
dependent dynamics of Bose-Einstein condensates.
Since the orbitals are time-dependent in MCTDHF, we must specify a dynamical
equation that determines the time-dependence. This is accomplished by the applica-
tion of the Dirac-Frenkel time-dependent variational principle [123, 124]
𝜕
⟨𝛿𝛹 (t)|H(t) − iℏ |𝛹 (t)⟩ = 0. (2.36)
𝜕t
The Dirac-Frenkel time-dependent variational principle (2.36) guarantees that the
orbitals and the CI coefficients in the wave function expansion (2.34) change with
time in such a way that the difference between H(t)𝛹 (t) and iℏ𝜕𝛹 (t)∕𝜕t is kept to be
as small as possible. Taking variations 𝛿𝛹 (t) with respect to the orbitals and impos-
ing the condition that ⟨𝜙j |𝜕𝜙k ∕𝜕t⟩ = 0 lead to the orbital equation of motion [20,
78, 106]
30 E. Lötstedt et al.
[ ]
𝜕𝜙j (𝐫, t) ∑
iℏ = Q h(𝐫, t)𝜙j (𝐫, t) + D−1
jk (t)dklmn (t)gmn (𝐫, t)𝜙l (𝐫, t) . (2.37)
𝜕t klmn
∑
M
Qf (𝐫) = f (𝐫) − 𝜙j (𝐫, t)⟨𝜙j (t)|f ⟩, (2.38)
j=1
is a projection operator that makes the time derivative 𝜕𝜙j ∕𝜕t of an orbital orthogonal
to 𝜙k (t). The definitions of the first- and second order spin-summed reduced density
matrices Djk (t) and djklm (t) can be found in [1] or [20]. The mean-field potential
gmn (𝐫, t) in (2.37) reads
e2 𝜙∗ (𝐫 ′ , t)𝜙n (𝐫 ′ , t)
gmn (𝐫, t) = d3 r′ m . (2.39)
4𝜋𝜀0 ∫ |𝐫 − 𝐫 ′ |
The equation of motion for the CI coefficients CIJ (t) is the same as that for time-
independent orbitals, and is given by
L
dC (t) ∑ 𝛼 ∑ L 𝛽
The MCTDHF equations (2.37) and (2.40) together guarantee that the norm
⟨𝛹 (t)|𝛹 (t)⟩ of the wave function is conserved throughout the time propagation. If
the Hamiltonian is time-independent, then the total energy ⟨𝛹 (t)|H|𝛹 (t)⟩ is con-
served as well.
It should be noted that the equations of motion (2.37) are nonlinear with respect to
the orbitals since both the orbitals and the CI coefficients are dependent on time in the
MCTDHF theory. The right hand side of (2.37) contains higher order terms of 𝜙j , i.e.,
terms proportional to 𝜙3 and 𝜙5 . In addition, the right hand side of (2.37) depends on
the CI coefficients through the density matrices Djk and djklm . The nonlinearity of the
equations means that they are significantly more difficult to be solved than a standard,
linear Schrödinger-type equation. The most straightforward way of performing the
time propagation is to use a general-purpose Runge-Kutta solver [125], but more
sophisticated approaches such as predictor-corrector schemes have been investigated
[95].
2 Multiconfiguration Methods for Time-Dependent Many-Electron Dynamics 31
As we have seen in the previous Sect. 2.4.1, the MCTDHF ansatz (2.34) includes
all possible Slater determinants, that is, a full CI expansion. For the treatment of
systems with many electrons N > 10, where N is the number of electrons, the
number of determinants becomes prohibitively large. If we assume that we need
M = N spatial orbitals, the number of determinants Itot scales exponentially with N
( M )2 1
as Itot = N∕2 ≈ 4N+ 2 ∕(N𝜋). Just like the case with time-independent orbitals as
described in Sect. 2.3, we may be able to obtain a more efficient method if a way
is found to reduce the number of determinants. In particular for the interaction of
long-wavelength (around 1 µm) intense laser pulses with atoms and molecules, we
expect that only the valence electrons are strongly perturbed by the laser field. This
means that determinants which correspond to the excitation of core electrons are not
important, and can be excluded from the expansion.
The above idea of excluding core-excited configurations in a time-dependent con-
text was suggested in [126]. The method is called the time-dependent complete
active-space self-consistent field (TD-CASSCF) method. The set of spatial orbitals
is divided into three groups in this method: frozen-core orbitals 𝜙fc j
, j = 1, … , Mfc ,
dynamical-core orbitals 𝜙dc j
, j = 1, … , Mdc , and active orbitals 𝜙 a
j
, j = 1, … , Ma ,
with a total of M = Mfc + Mdc + Ma orbitals. As in the MCTDHF method, it is
y
assumed that the orbitals form an orthonormal set, ⟨𝜙xj (t)|𝜙k (t)⟩ = 𝛿jk 𝛿xy at all t, for
all j and k and for all x, y ∈ {f c, dc, a}. The wave function is constructed by taking a
linear combination of determinants,
∑
𝛹 (t) = CIJ (t)𝛷IJ (t), (2.41)
IJ∈SCASSCF
where Na𝛼 and Na𝛽 are the number of active 𝛼 and 𝛽 electrons, respectively. The
total number of electrons is N = Na𝛼 + Na𝛽 + 2(Mfc + Mdc ). The core orbitals are
always double occupied, and are therefore used to construct all the determinants.
The indexes I, J specify the active orbitals 𝜙aj included in the determinant 𝛷IJ . The
frozen-core orbitals 𝜙fc are, as suggested by their name, time-independent. Thus,
j
( )( )
the number of possible determinants is reduced to Itot = NMa NMa . The restriction
a𝛼 a𝛽
to determinants having doubly occupied core orbitals can reduce the total number of
determinants Itot significantly. If we consider a Ne atom described by M = 10 spatial
32 E. Lötstedt et al.
( )2
orbitals as an example, we would have Itot = 10 5
= 63504 in MCTDHF, but only
(9)2
Itot = 4 = 15876 in TD-CASSCF with the 1s orbital taken as a doubly occupied
( )2
core orbital, and Itot = 83 = 3136 in the case when both the 1s and the 2s orbital are
considered to be core orbitals. Even though the number of determinants is reduced,
the TD-CASSCF wave function can still be a good approximation. If we consider the
excitation by near-infrared radiation, the 2p electrons are active almost exclusively,
and most of the determinants in the MCTDHF expansion are in fact not necessary
[22].
The equations of motion in the TD-CASSCF method are very similar to those in
the MCTDHF method,
dCIJ ∑
iℏ = ⟨𝛷IJ (t)|H(t)|𝛷I ′ J ′ (t)⟩CI ′ J ′ (t) (2.43)
dt I ′ J ′ ∈S CASSCF
for the orbitals. The spin-summed density matrices Djk (t) and djklm (t) in (2.44) are
evaluated using the TD-CASSCF wave function in (2.41). The only difference of
the TD-CASSCF orbital equation (2.44) from the orbital equation (2.37) in MCT-
∑
DHF is the presence of the term k Rjk (t)𝜙k (𝐫, t) on the right hand side of (2.44).
In the TD-CASSCF case, we cannot assume that ⟨𝜙j |𝜕𝜙k ∕𝜕t⟩ = 0, as was done in
the MCTDHF method. The reason is that the core orbitals and the active orbitals are
not equivalent. The TD-CASSCF wave function is not invariant under unitary trans-
formations, which mix the core and active orbitals. The rotation matrix Rjk must
therefore be calculated consistently at each time step, as described in [126]. See also
[127, 128] for a general discussion on the orbital rotation matrix Rjk in the context
of general multiconfiguration theories with time-dependent orbitals.
The TD-CASSCF method has successfully been applied to one-dimensional
model molecules (LiH and LiH dimer) [126] and atoms (Be and Ne) [22] exposed
to intense laser pulses.
2.5 Factorized CI
In this section, we describe our recent proposal on how to reduce the computational
cost of the MCTDHF method [129]. In the factorized CI method, we start from the
MCTDHF expansion of the total wave function,
Another random document with
no related content on Scribd:
PLEASE READ THIS BEFORE YOU DISTRIBUTE OR USE THIS WORK
1.D. The copyright laws of the place where you are located also
govern what you can do with this work. Copyright laws in most
countries are in a constant state of change. If you are outside the
United States, check the laws of your country in addition to the terms
of this agreement before downloading, copying, displaying,
performing, distributing or creating derivative works based on this
work or any other Project Gutenberg™ work. The Foundation makes
no representations concerning the copyright status of any work in
any country other than the United States.
• You pay a royalty fee of 20% of the gross profits you derive from
the use of Project Gutenberg™ works calculated using the
method you already use to calculate your applicable taxes. The
fee is owed to the owner of the Project Gutenberg™ trademark,
but he has agreed to donate royalties under this paragraph to
the Project Gutenberg Literary Archive Foundation. Royalty
payments must be paid within 60 days following each date on
which you prepare (or are legally required to prepare) your
periodic tax returns. Royalty payments should be clearly marked
as such and sent to the Project Gutenberg Literary Archive
Foundation at the address specified in Section 4, “Information
about donations to the Project Gutenberg Literary Archive
Foundation.”
• You comply with all other terms of this agreement for free
distribution of Project Gutenberg™ works.
1.F.
1.F.4. Except for the limited right of replacement or refund set forth in
paragraph 1.F.3, this work is provided to you ‘AS-IS’, WITH NO
OTHER WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR ANY PURPOSE.
Please check the Project Gutenberg web pages for current donation
methods and addresses. Donations are accepted in a number of
other ways including checks, online payments and credit card
donations. To donate, please visit: www.gutenberg.org/donate.
Most people start at our website which has the main PG search
facility: www.gutenberg.org.