week ending

PRL 105, 233002 (2010) PHYSICAL REVIEW LETTERS 3 DECEMBER 2010

Delay in Atomic Photoionization

A. S. Kheifets1,2,* and I. A. Ivanov1
1
Research School of Physical Sciences, The Australian National University, Canberra ACT 0200, Australia
2
The Kavli Institute for Theoretical Physics, University of California, Santa Barbara, California 93106-4030, USA
(Received 13 October 2010; published 1 December 2010)
We analyze the time delay between emission of photoelectrons from the outer valence ns and np
subshells in noble gas atoms following absorption of an attosecond extreme ultraviolet pulse. Various
processes such as elastic scattering of the photoelectron on the parent ion and many-electron correlation
affect the apparent ‘‘time zero’’ when the photoelectron leaves the atom. This qualitatively explains the
time delay between photoemission from the 2s and 2p subshells of Ne as determined experimentally by
attosecond streaking [Science 328, 1658 (2010)]. However, with our extensive numerical modeling, we
were only able to account for less than half of the measured time delay of 21
5 as. We argue that the
extreme ultraviolet pulse alone cannot produce such a large time delay and it is the streaking IR field that
is most likely responsible for this effect.

DOI: 10.1103/PhysRevLett.105.233002 PACS numbers: 32.30.Rj, 31.15.ve, 32.70.n, 32.80.Fb

Among other spectacular applications of the attosecond photoionization in noble gas atoms. We employ both the
streaking technique, it has become possible to determine explicit time-dependent and stationary treatments of the
the time delay between subjecting an atom to a short laser photoionization process. To this end, we solve the time-
pulse and subsequent emission of the photoelectron. In a dependent Schrödinger equation (TDSE) in the single
recent work by Eckle et al. [1], the helium atom was active electron approximation. By carefully examining
subjected to a near-infrared laser pulse with an intensity the time evolution of the photoelectron wave packet, we
of several units of 1014 W=cm2 . Such a strong field ioniza- establish the apparent ‘‘time zero’’ when the photoelectron
tion regime could be characterized by a fairly small leaves the atom. To account for electron correlation, we
Keldysh parameter
’ 1. The time delay in such a photo- solve a set of coupled integral equations in the random
emission process can be conveniently analyzed in terms of phase approximation with exchange (RPAE) [4].
nonadiabatic tunneling [2]. In a subsequent experiment by Within an independent electron approximation, the time
Schultze et al. [3], the time delay was measured in neon in delay is caused by the energy dependence of the elastic
the extreme ultraviolet (XUV) photon energy range by scattering phase shifts of the photoelectron moving in the
high-order harmonic conversion of the driving near-infrared Hartree-Fock (HF) potential of the Neþ ion. The many-
laser pulse. In this regime, which is characterized by a electron correlation, which is due to intershell 2s-2p cou-
moderate intensity, short wavelength, and
 1, it is pling, depends on the energy of the photon and thus adds an
believed that the formation of the outgoing wave packet additional component to the quantum phase of the dipole
follows instantaneously temporal variation of the incident matrix element. Both these effects account for the time
electromagnetic field. Nevertheless, a sizable time delay of delay not exceeding 10 as. This recovers only about one-
21
5 as was reported between photoemission from the 2s half of the experimental value of 21
5 as. We carefully
and 2p valence subshells of Ne. Schultze et al. [3] argued examine other correlation and polarization corrections but
that a comprehensive temporal characterization of photo- find them unable to produce any sizable contribution to the
emission on the attosecond time scale could provide a new measured time delay. We also analyze these effects in other
insight into intra-atomic electron correlations. Indeed, the noble gas atoms.
best theoretical treatment within an independent electron The time-dependent calculation of photoionization in
model could only account for 4.0 as time delay. When the Ne was performed by radial grid integration of the TDSE
theoretical model was corrected for electron correlations using the matrix iteration method [5]. We employed a
before and after photoionization, a relative delay of 6.4 as one-electron basis in a parametrized optimized effective
was obtained. This unresolved difference between the potential [6]. We used the linearly polarized XUV pulse
measured and calculated time delays puts many-electron EðtÞ ¼ E0 gðtÞ cos!t with the envelope gðtÞ represented by
models of atomic photoionization under significant strain. the Nutall window function and centered at t ¼ 0. The
If substantiated, this difference could potentially point to following field parameters were chosen: E0 ¼ 0:119 a:u.
new physical mechanisms underpinning electromagnetic (corresponding to the peak intensity of 5  1014 W=cm2 ),
interaction in atoms on the attosecond time scale. ! ¼ 106 eV, T ¼ 2=! ¼ 39 as, and FWHM ¼ 182 as.
In this Letter, we perform an extensive study of the time Experimental field intensity was not reported by Schultze
delay between the ns2 and np6 outer valence subshell et al. [3]. Given a typical high-order harmonic conversion

0031-9007=10=105(23)=233002(4) 233002-1 The American Physical Society

 week ending
PRL 105, 233002 (2010) PHYSICAL REVIEW LETTERS 3 DECEMBER 2010

efficiency of 106 [7], the presently chosen XUV field asymptotic Rkl / sin½kr þ l ðkÞ þ ð1=kÞ lnð2krÞ  l=2
strength is most certainly larger than the one used experi- and the spherical harmonic YL ðr=rÞ with L  l; m. The
mentally. With this choice, our calculation is guaranteed projection coefficients akl ðtÞ cease to depend on time for
to account for nonperturbative, with respect to the field, jtj > T1 when the driving XUV pulse is off.
ionization effects if these effects were to be sizable. Other There are two convenient indicators of the evolution
XUV field parameters used in the present work were of the wave packetP R(1). One is the norm given by the
identical to those used in the experiment. The XUV pulse integral NðtÞ ¼ L dkjakL ðtÞj2 . This norm is plotted
described above is shown in the top panel of Fig. 1 (dotted in the top panel of Fig. 1 with the red solid and green
black line). The pulse is truly off outside the interval
T1 , dashed lines for the wave packets that originated from
where T1 ’ 5T, which is about twice the FWHM. the 2s and 2p subshells, respectively. For better clarity,
The solution of the TDSE satisfies the initial condition these curves are scaled and overplotted on the electromag-

ðr; t ¼ T1 Þ ¼ i ðrÞ, which corresponds to a bound elec- netic pulse.
tron state on the atomic shell i to be ionized. So the shell The figure shows clearly that the evolution of the 2s and
index i is implicit in the following but omitted for brevity. 2p wave packets starts and ends at the same time without
The wave packet representing the photoelectron ejected any noticeable delay. This is further visualized in the inset,
from a given shell is defined as where the variation of the norm ½NðtÞ  NðT1 Þ=NðT1 Þ is
plotted on an expanded time scale near the driving pulse
XZ
ðr; tÞ ¼ akL ðtÞkL ðrÞeiEk t dk; (1) end. Indeed, the norm starts deviating from zero with the
L rise of the XUV pulse and reaches its asymptotic value
once the interaction with the XUV pulse is over.
where akL ðtÞ ¼ eiEk t hkL j
ðtÞi are the projection coeffi- Another marker of the wave packet dynamics is the crest
cients of the solution of the TDSE on the continuum position, defined as a location of the global maximum of
spectrum of the atom. The continuum state kL ðrÞ ¼ the electron density. The latter quantity is truly informative
Rkl ðrÞYL ðr=rÞ is the product of the radial orbital with the only when the electron is outside the atom and the wave
packet is fully formed, having one well-defined global
maximum. In the bottom panel of Fig. 1, we show the crest
0.2
position of the 2s and 2p wave packets propagating in
Electromagnetic field E (au)

0.1
time. This figure can be viewed as a more realistic version
of a somewhat idealized and simplified graph presented in
0 Fig. 1 of Schultze et al. [3]. We see that evolution of the
norm and the movement of the crest commence and cease
-0.1 at about the same time.
The movement of the crest becomes almost linear when
-0.2 E(t)
N2s(t) the norm reaches its asymptotic value and the wave packet
N2p(t) 80 100 120
is fully formed. Once fitted with the linear time depen-
25
-200 -100 0 100 200
dence r ¼ kðt  t0 Þ þ r0 for large times t > T1 (shown as a
dotted straight line) and backpropagated inside the atom,
2
20 the 2s wave packet seems to have an earlier start time t0
Crest position r (au)

1 than that of the 2p wave packet. This difference is magni-

15
fied in the inset. It is about 6 as at the origin r0 ¼ 0 and
10
0
-5 0 5 about 4 as at the distance r0  1 au, which corresponds to
r2s(t)
the size of the valence shell of the Ne atom. We see that at
2p
5 r2p(t) the origin t2s
0 < 0 and t0 > 0 are shifted to the opposite
direction with respect to the peak of the driving XUV pulse
0
-200 -100 0 100 200 which sets the start time of the photoionization process.
Time t (as) Thus the seeming (or apparent according to Ref. [3]) time
zero of the wave packet, which is inferred by the backward
FIG. 1 (color online). Top: The norm of the wave packets NðtÞ time propagation, is different from the physical (or real)
(scaled arbitrarily) emitted from the 2s and 2p subshells is time zero t ¼ 0.
plotted as a function of time with the red solid and green dashed
The origin of this shift is most clearly elucidated within
lines, respectively. The XUV pulse is overplotted with the black
dotted line. In the inset, the norm variation ½NðtÞ  the perturbation theory framework, which should be appli-
NðT1 Þ=NðT1 Þ is shown on an expanded time scale near the cable under the present field conditions of a single photon
pulse end. Bottom: The crest position of the 2s and 2p wave transition with
 1 [8]. Under these conditions,
packets is shown with the same line styles. The crest position
after the pulse end is fitted with the straight line, which corre- Z1
sponds to the free propagation. In the inset, extrapolation of the akL ðt > T1 Þ ¼ i hkL jzji ieiðEk i Þ EðÞd: (2)
free propagation inside the atom is shown. 1

233002-2
 week ending
PRL 105, 233002 (2010) PHYSICAL REVIEW LETTERS 3 DECEMBER 2010

Wave packet coefficients |akl|2

Here we extended the integration limits outside the pulse l=0
0.06 1
duration and wrote the dipole matrix element hkL jzji i in 2
the length gauge. By separating the angular and radial
0.04
integration, we can present this matrix element in the
reduced form
0.02
hkL jzji i / Clm
10li mi d ðkÞ; (3)

10li mi is the Clebsch-Gordan coefficient,   l; i,

0
where Clm
1
and d ðkÞ is real. With this definition, we can write akL / l=0 (-π)
1 (-π/2)
~ k  i Þ, where the Fourier transform of the
id ðkÞEðE

HF phases δl(k) (rad)

2
~ R 0.5
XUV field Eð!Þ ¼ 1 1 e
i! EðÞd is real for a symmetric

pulse that we presently consider. We note that Eqs. (1)–(3) 0

are equivalent to Eqs. (S5)–(S8) of Schultze et al. [3] given
in their supporting online material. -0.5
To describe the motion of the wave packet (1), we apply
the usual saddle-point method. For each l, the crest of the -1

wave packet is moving at large times t > T1 quasiclassi- 0.5

RPAE phases arg[Dλ(k)] (rad)

l=0
cally along the trajectory which is given by the equation 0.4
1
2

  
d 1
r¼k t  ðkÞ þ lnð2krÞ pffiffiffiffiffiffi : (4) 0.3
dE l k k¼ 2E0
0.2
Since the logarithm is a slowly varying function which can
0.1
be absorbed into a constant, Eq. (4) describes a straight
line: r ¼ kðt  t0 Þ þ r0 , with t0 ¼ dl ðkÞ=dEjk¼pffiffiffiffiffiffi
2E
. 0
0 20 40 60 80 100 120 140
Thus the relative time delay between various photoioniza- Photoelectron energy Ek (eV)
tion channels is determined primarily by the derivatives of
the corresponding elastic scattering phases [9]. FIG. 2 (color online). Top: Expansion coefficients jakl j2 plot-
The scattering phases l ðkÞ of the photoelectron moving ted versus the photoelectron energy Ek ¼ k2 =2, which is ex-
in the field of a singly charged Neþ ion are shown in the pressed in eV. Middle: The HF scattering phases. Bottom: Phases
middle panel of Fig. 2. The photoelectron ejected from the of the RPAE dipole matrix elements arg½D ðkÞ.
2s shell has only one value of the angular momentum
l ¼ 0, whereas the 2p photoelectron can acquire two an- So far, we confined ourselves with an independent elec-
gular momenta l ¼ 0 and 2. The phases in the s and p tron approximation and calculated the dipole matrix
waves are shifted downwards by  and =2, respectively, elements d ðkÞ and the scattering phases l ðkÞ in the HF
for better clarity. In the top panel of the same figure, we approximation [11,12]. It is well known, however, that
display the asymptotic projection coefficients aL ðkÞ for many-electron correlation modifies strongly the dipole
l ¼ 0, 1, and 2 and m ¼ 0. The centers of the energy matrix elements. The full account for this effect can be
distribution of the l-projected coefficients (indicated by taken within the RPAE model [4] by solving a set of
the vertical dotted lines in the top panel) define the position coupled integral equations:
of the energy derivative of the corresponding scattering XZ
phases dl ðkÞ=dEjk¼pffiffiffiffiffiffi
2E
(indicated by the straight lines in D ðkÞ ¼ d ðkÞ þ dpD ðpÞ ðpÞU  ðp; kÞ: (5)
0

the middle panel). We see that the energy derivatives of the

s and p phases are negative, whereas that of the d phase is Here  ðpÞ ¼ ð!  Ep   þ iÞ1 is the Green’s func-
positive. This is so due to the presence of the occupied s tion, and U  ðp; kÞ is the Coulomb interaction matrix. The
and p states in the Neþ ion which disturbs the otherwise one-electron HF basis corresponding to the field of the
monotonic increase with energy of the Coulomb phase (the singly charged Neþ ion accounts for direct photoelectron
Levinson-Seaton theorem [10]). Because the 2p ! kd interaction with its parent shell. It is therefore the intershell
transition is strongly dominant over the 2p ! ks one, as Coulomb interaction with Þ  that should only be in-
is seen from the corresponding projection coefficients in cluded into Eq. (5). Since the Green’s function is complex,
the top panel of Fig. 2, it is the d phase that determines the the dipole matrix elements D ðkÞ acquire an additional
shift of the apparent time zero of the 2p wave packet phase, which is plotted in the bottom panel of Fig. 2.
relative the physical time zero t ¼ 0. This shift is positive The HF phase derivatives alone account for the apparent
for the 2p wave packet and negative for the 2s one, in time zero shift between the 2s and 2p ionization t02s-2p ¼
accordance with our observation displayed in the inset of 6:2 as. The RPAE correction adds an extra 2.2 as. In
the bottom panel of Fig. 1. total, this accounts for the apparent time zero shift
233002-3
 week ending
PRL 105, 233002 (2010) PHYSICAL REVIEW LETTERS 3 DECEMBER 2010

t02s-2p ¼ 8:4 as. Both the HF and RPAE phases are photoionization begins, which is fully determined by the
smooth functions of the photoelectron energy, and their driving XUV pulse alone. In this sense, the attosecond
averaging over the bandwidth of the XUV pulse does not streaking is not informative on the early stages of the
change these numbers in a noticeable way. The analogous photoionization process. However, this technique allows
values reported by Schultze et al. [3] for the independent one to determine the energy derivative of the quantum
electron model and the correlation correction are 4.0 and phase of the dipole matrix element [15], thus facilitating
2.4 as, respectively. Both sets of calculations are quite close the so-called complete photoionization experiment [16].
and well below the experimental value of 21
5 as. This is particularly important in those targets where the
One could argue that the complete account for many- many-electron correlation is significant.
electron correlation within the TDSE, rather than adding The full potential of the attosecond streaking technique
this correlation ad hoc, could modify the present result. and its successful application in atomic collision physics
This is, however, unlikely given the nature of the RPAE can be realized only if the current strong disagreement
which is a direct generalization of the HF method in the between theory and experiment in Ne is resolved. The
presence of an oscillatory external electromagnetic field present study was not able to do so. Our simulations and
[13]. The only approximation taken when deriving Eq. (5) analytic arguments indicate that the XUV pulse alone
is that at any instant of time the atomic wave function is cannot produce such a large time delay and it is the streak-
an antisymmetric product of one-electron functions. It is ing IR field that is most likely responsible for this effect.
quite a robust approximation under the field parameters The authors acknowledge support of the Australian
considered above. Research Council in the form of Discovery Grant
We also evaluated the time delay of the wave packet No. DP0771312. Resources of the National Computa-
relative to the XUV pulse in other noble gases. In He, the tional Infrastructure (NCI) Facility were employed. One
wave packet emitted from the 1s shell is delayed by 2 as of the authors (A. S. K.) thanks the Kavli Institute for
relative to the center of the XUV pulse. This follows from Theoretical Physics for hospitality. This work was sup-
the independent electron HF calculation, which returns ported in part by the NSF Grant No. PHY05-51164
a positive derivative of the p phase shift as there is no
occupied p orbital in the Heþ ion. It is also confirmed by
the correlated convergent close-coupling model, which is
known to produce benchmark photoionization results for *Corresponding author.
He in the XUV range [14]. It is to be compared with 5 as [email protected]
delay reported for He by Schultze et al. [3]. In heavier [1] P. Eckle et al., Science 322, 1525 (2008).
noble gases, Ar and Kr, the difference of the HF p- and [2] G. L. Yudin and M. Y. Ivanov, Phys. Rev. A 63, 033404
d-phase derivatives becomes smaller as occupation of the (2001).
ionic orbitals increases in line with the Levinson-Seaton [3] M. Schultze et al., Science 328, 1658 (2010).
theorem. In Kr, the d-phase derivative becomes negative as [4] M. Y. Amusia, Atomic Photoeffect (Plenum, New York,
the 3d orbital is occupied. Accordingly, the time delay 1990).
between the wave packets emitted from the ns and np [5] M. Nurhuda and F. H. M. Faisal, Phys. Rev. A 60, 3125
valence subshells is getting smaller. When the HF and (1999).
[6] A. Sarsa, F. J. Gálvez, and E. Buendia, At. Data Nucl. Data
RPAE phase derivatives are combined, it results in 5.8 as
Tables 88, 163 (2004).
delay in Ar and nearly zero delay in Kr around the 100 eV [7] E. Goulielmakis et al., Science 320, 1614 (2008).
photon energy mark. [8] N. B. Delone and V. P. Krainov, Atoms in Strong Light
In conclusion, we examined various effects leading to Fields (Springer-Verlag, Berlin, 1985).
the shift between the apparent time zero of the photoelec- [9] C. A. A. de Carvalho and H. M. Nussenzveig, Phys. Rep.
tron wave packets emitted from the 2s and 2p shells in 364, 83 (2002).
neon relative to the center of the XUV pulse which sets the [10] L. Rosenberg, Phys. Rev. A 52, 3824 (1995).
timing of the photoionization process. We found that this [11] L. V. Chernysheva, N. A. Cherepkov, and V. Radojevic,
shift is primarily due to the energy derivative of the HF Comput. Phys. Commun. 11, 57 (1976).
elastic scattering phase shifts which differs significantly [12] L. V. Chernysheva, N. A. Cherepkov, and V. Radojevic,
for various partial waves. The RPAE correction, which Comput. Phys. Commun. 18, 87 (1979).
[13] D. J. Thouless, The Quantum Mechanics of Many-Body
accounts for many-electron correlation, is rather small
Systems (Academic, New York, 1972).
and cannot explain the profound difference between the [14] I. Bray, D. V. Fursa, A. S. Kheifets, and A. T. Stelbovics,
theoretical and experimental time delay. J. Phys. B 35, R117 (2002).
The apparent time zero is meaningful only when the [15] V. S. Yakovlev, J. Gagnon, N. Karpowicz, and F. Krausz,
wave packet is detected at large distances from the atom as Phys. Rev. Lett. 105, 073001 (2010).
in attosecond streaking experiments. This apparent time [16] N. A. Cherepkov and S. K. Semenov, J. Phys. B 37, 1267
zero has little to do with the real time when the atomic (2004).

233002-4