RES EARCH

SPECTROSCOPY response. Notably, the delays expected from

that travel time of a photon across molecular
Zeptosecond birth time delay in orbitals are one to two orders of magnitude
shorter than the Wigner delay—i.e., they man-
molecular photoionization ifest in the zeptosecond (1 zs = 10−21 s) domain.
In the description of light-matter interactions,
Sven Grundmann1*, Daniel Trabert1, Kilian Fehre1, Nico Strenger1, Andreas Pier1, Leon Kaiser1, such ultrashort time differences are often ig-
Max Kircher1, Miriam Weller1, Sebastian Eckart1, Lothar Ph. H. Schmidt1, Florian Trinter1,2,3, nored, and the dipole approximation is invoked.
Till Jahnke1*, Markus S. Schöffler1, Reinhard Dörner1* Doing so corresponds to neglecting the spatial
dependence of the light wave, and the light’s
Photoionization is one of the fundamental light-matter interaction processes in which the absorption electric field is approximated to be present
of a photon launches the escape of an electron. The time scale of this process poses many open instantaneously with the same phase over
questions. Experiments have found time delays in the attosecond (10−18 seconds) domain between the whole relevant region of space. Beyond
electron ejection from different orbitals, from different electronic bands, or in different directions. this approximation, however, the birth time
Here, we demonstrate that, across a molecular orbital, the electron is not launched at the same time. delay leads to a phase shift between the corre-
Rather, the birth time depends on the travel time of the photon across the molecule, which is sponding contributions to the overall photo-
247 zeptoseconds (1 zeptosecond = 10−21 seconds) for the average bond length of molecular hydrogen. electron wave. Such relative phases—and
Using an electron interferometric technique, we resolve this birth time delay between electron hence the birth time delays across a molecular
emission from the two centers of the hydrogen molecule. orbital—are accessible by experiments that ex-
ploit the interference between the different

Downloaded from http://science.sciencemag.org/ on October 16, 2020

P
parts of the wave function. Figure 1 outlines
hotoionization is a fundamental quan- ingly, the birth time delay quantifies to what our metrology to measure tb.
tum process that has become a powerful extent a delocalized molecular orbital reacts Our approach builds on the close analogy
tool to study atoms, molecules, liquids, simultaneously as a single unit on being hit by between a plane wave behind a double slit
and solids. Facilitated by the advent of a photon. For example, it shows whether the (Fig. 1, A and B) and the photoelectron wave
attosecond technology, it can now even part of the orbital facing toward an approach- emitted from a homonuclear diatomic molecule
be addressed in the time domain. Timing in ing photon reacts first and whether the part (Fig. 1, C and D). This analogy was proposed
photoionization usually refers to the time it downstream of the photon beam has a retarded by Cohen and Fano (8) and is well established
takes for an electron to escape to the con-
tinuum after absorption of the photon. This
time depends, for example, on the electronic
orbital (1, 2), on the energy band in solids (3, 4),
or on the orientation (5) and handedness (6)
of the target molecule. The escape time differ-
ence manifests as a phase shift of the electron
wave in the far field. This relation is based on
the concept of the Wigner delay, which is the
energy derivative of the phase of a photoelec-
tron wave at an asymptotic distance from the
source (7). Typical numbers are, for example,
20 attoseconds (1 as = 10−18 s) for the Wigner
delay between emission from the 2s and 2p
shells in neon (1).
However, the Wigner delay does not cover
another notable question about the timing in
photoionization of extended systems—namely
the temporal buildup of the photoelectron wave
across its spatially extended source. Hereafter,
we refer to these variations in the temporal
structure of the electron wave as the birth time
delay tb. Although the Wigner delay is caused
during the travel of the electron to the contin-
uum after its birth, tb relies on different birth
times of the contributions to the total photo-
electron wave along a molecular orbital. Accord- Fig. 1. Concept of birth time delay measurement. (A) Intensity distribution on a screen in the far field
behind the double slit in (B). (B) A plane wave impinges on a double slit. The phase shift ½Df in the
right slit causes a tilt of the interference pattern. (C and D) Emission of a photoelectron wave from
1
Institut für Kernphysik, Goethe-Universität, Max-von-Laue- two indistinguishable atoms of a homonuclear diatomic molecule mimics the double-slit setup in (B).
Strasse 1, 60438 Frankfurt, Germany. 2Photon Science,
Here, the angle a is enclosed by the electron momentum vector and the molecular axis. A time delay (Dt)
Deutsches Elektronen-Synchrotron (DESY), Notkestrasse 85,
22607 Hamburg, Germany. 3Molecular Physics, Fritz-Haber- between the emission from one of the two centers—e.g., originating from the travel time of the photon
Institut der Max-Planck-Gesellschaft, Faradayweg 4-6, 14195 impinging from the left side in (D)—leads to a shift of the interference fringes in (C). The ratio of slit distance
Berlin, Germany. (molecular bond length R, respectively) to wavelength is 1.65 in both cases [(B) and (D)]. In (B) the
*Corresponding author. Email: [email protected]
(S.G.); [email protected] (T.J.); doerner@atom. right-hand part of the wave is delayed by Df ¼ p=2, whereas in (D) a birth time delay of 247 zs causes
uni-frankfurt.de (R.D.) Df ≈ p=11 for R = 0.74 Å. light dir., light direction; mol. dir., molecular direction.

Grundmann et al., Science 370, 339–341 (2020) 16 October 2020 1 of 3

RES EARCH | R E P O R T

today (9, 10). It has been used, for example, to

Fig. 2. Electron emission from H2 mimics
study the onset of decoherence of a quantum
the double-slit experiment. (A and
system (11) and entanglement in an electron
B) Interference pattern of fast electrons
pair (12). The angular emission pattern from a
(Ee = 735 ± 15 eV) from one-photon
gerade orbital has a maximum that is perpen-
double ionization of H2 by 800 eV circularly
dicular to the molecular axis, which corresponds
polarized photons for the average inter-
to the zeroth-order interference maximum be-
nuclear distance of R = 0.74 ± 0.02 Å
hind the double slit. If a constant phase shift
[purple line in (A)] and as function of R (B).
Df is introduced to one of the slits, the inter-
The blue line in (A) models a double-slit
ference pattern behind the double slit be-
interference pattern for a slit distance
comes asymmetric, and the angular position
R = 0.74 Å and l = 0.45 Å, which is the
of the interference fringes moves as a function
average de Broglie wavelength of the fast
of Df on an intensity screen in the far field.
electron. The subset S of the data
In Fig. 1, A and B, we illustrate this relation.
are used for (A) and for the subsequent
A corresponding shift may also occur when
analysis of the birth time delay.
two interfering electron waves are emerging
from the two indistinguishable centers of a
homonuclear diatomic molecule upon photon
absorption (Fig. 1, C and D). If the contribu-
tions to the photoelectron wave are launched

Downloaded from http://science.sciencemag.org/ on October 16, 2020

simultaneously across an orbital of a diatomic
molecule, the electron emission pattern in the
molecular frame of reference is symmetric
with respect to the normal of the bond axis.
However, any initial phase shift between the
waves emerging from one or the other center
leads to an angular shift of the diffraction
pattern—just as in the double-slit case. Hence,
inspecting the angular emission distribu-
tion of photoelectrons emitted from a homo-
nuclear diatomic molecule for such angular
shifts offers a way to measure the birth time
delay tb (13).
For an electron of kinetic energy Ee and mo-
mentum pe, tb can be obtained from the mea-
sured angle a0—the angle to which the central Fig. 3. Results of birth time delay measurement. Birth time delay of fast electrons (Ee = 735 ± 15 eV) from
interference maximum is shifted—in the follow- one-photon double ionization of H2 by 800 eV circularly polarized photons for the average internuclear
ing way: An electron wave is emitted and prop- distance of R = 0.74 ± 0.02 Å (selected subset S as shown in Fig. 2B). (A) Electron angular distribution with
agates with the phase velocity vph = Ee/pe for respect to the molecular axis, which is aligned parallel to the light propagation direction [cos(b) > 0.87,
the time tb before a second electron wave is corresponding to the top row of bins in (B)]. Red curve is the Gaussian fit used to obtain the angular
born at a distance R. The zeroth-order inter- position of the zeroth-order maximum cos(a0). (B) Electron angular distribution in the molecular frame of
ference maximum occurs under the emission reference as a function of cos(b). Dashed line is perpendicular to the molecular axis—i.e., the location
angle a0 for which the path length difference of the zeroth-order maximum in the absence of birth time delays. (C) Location of the maxima of the
between both electron waves vanishes zeroth-order interference fringe as a function of cos(b). The maxima are obtained with Gaussian fits, as
indicated by the red line in (A). The error bars include statistical and systematic errors, and the purple-
 
cosða0 Þ  R tb  vph shaded error range indicates the systematical error (see supplementary materials for further details).
2p  ¼0 ð1Þ
l l Left axis: cos(a0); right axis: birth time delay calculated with Eq. 2. The blue line resembles a birth time delay
given by the travel time of light across the molecule (Eq. 3). The red line is a prediction combining atomic
nondipole effects and the travel time of the photon (see text for details). elec., electron.
where l is the electron’s de Broglie wave-
length. Accordingly, the birth time delay tb can
be inferred from the angular shift a0, and an
angle of a0 = 90° corresponds to zero birth time dimensional momenta of both protons in (KER) from the relative momentum of the
delay (simultaneous emission) coincidence with one electron. From the sum protons (15, 16). In the reflection approxima-
momentum of the three measured particles, tion, the internuclear distance R at the moment
R we inferred the missing electron momentum of photoabsorption is related to the kinetic
tb ¼ cosða0 Þ ð2Þ vector by means of momentum conservation energy release through KER = e2/(4pe0R),
vph
including the photon momentum. The mol- where e is the elementary charge and e0 is the
ecules in the target gas jet were randomly vacuum permittivity.
We implemented this scheme by studying oriented with respect to the light propagation. One-photon double ionization typically pro-
one-photon double ionization of H2 using left- After ejection of the two electrons, the two ceeds by means of one of two sequential pro-
handed circularly polarized photons with an protons were driven apart by their Coulomb cesses. A primary photoelectron is set free by
energy of 800 eV. Using a COLTRIMS reac- repulsion. We obtained the orientation of the the absorption of the photon, and the second
tion microscope (14), we measured the three- molecular axis and the kinetic energy release electron is either shaken off or knocked out to

Grundmann et al., Science 370, 339–341 (2020) 16 October 2020 2 of 3

RES EARCH | R E P O R T

the continuum (17). The fraction of the photon far—would likely depend on a and b, but they The analogy between electron emission from
energy that exceeds the sum of the adiabatic are equal for the different pathways that inter- the H2 molecule and a classical double-slit ex-
double ionization energy of H2 (31.03 eV) and fere under a certain emission angle a and do periment suggests that the birth time delay
the KER is shared between the two electrons. not influence our measurement. could be interpreted as the travel time of the
The symmetry of this energy sharing is a We compared our experimental findings with photon from one molecular center to the other,
measure for the strength of the Coulomb two simple models. First, we assumed that t′ b is which is up to 247 zs for the average bond
interaction among the two electrons, which given by the time difference with which a point length of H2. Our experimental results support
has the potential to destroy the single-particle of constant phase of the photon wave hits the this picture, but studies targeting more-complex
quantum interference pattern (12). Therefore, two centers molecules and applying more-sophisticated the-
we restricted our investigation to fast elec- oretical models are necessary to further unveil
trons that carried >96% of the excess energy R the scope of birth time delay. This work can
t′ b ¼ cosðbÞ ð3Þ
(fig. S2). For such fast electrons, double-slit c function as a benchmark for such studies.
interference effects are well established on
the single-particle level (11, 18). Corresponding where c is the speed of light. In the case of REFERENCES AND NOTES
slow electrons of the double ionization pro- cos(b) = 0, the photon wave hits both centers 1. M. Schultze et al., Science 328, 1658–1662 (2010).
cess had <4% of the excess energy and are not of the molecule simultaneously, and there can- 2. M. Isinger et al., Science 358, 893–896 (2017).
3. A. L. Cavalieri et al., Nature 449, 1029–1032 (2007).
shown here (19). not be any birth time delay between electrons 4. Z. Tao et al., Science 353, 62–67 (2016).
In Fig. 2A, we display the electron angular emitted from one or the other center because 5. J. Vos et al., Science 360, 1326–1330 (2018).
distribution of those fast electrons for the aver- the outgoing waves are exactly in phase. On 6. S. Beaulieu et al., Science 358, 1288–1294 (2017).
7. J. M. Dahlström, A. L’Huillier, A. Maquet, J. Phys. B 45, 183001
age internuclear distance of 0.74 Å (purple line). the other hand, cos(b) = ±1 resembles the max- (2012).

Downloaded from http://science.sciencemag.org/ on October 16, 2020

The results show a rich structure, which—as imum possible travel time of the photon from 8. H. D. Cohen, U. Fano, Phys. Rev. 150, 30–33 (1966).
expected—resembles the interference pattern one molecular center to the other. For this case, 9. J. Fernández, O. Fojón, A. Palacios, F. Martín, Phys. Rev. Lett.
98, 043005 (2007).
of electrons emerging from a double slit. Figure the expected birth time delay was ±247 zs for 10. S. E. Canton et al., Proc. Natl. Acad. Sci. U.S.A. 108,
2B displays this measured interference pattern R = 0.74 Å. Between these extreme cases, the 7302–7306 (2011).
of the fast electron as a function of the inter- birth time delay showed a linear dependence 11. D. Akoury et al., Science 318, 949–952 (2007).
12. M. Waitz et al., Phys. Rev. Lett. 117, 083002 (2016).
nuclear distance. The results show how the on cos(b). For comparison to the experimental 13. This interferometric technique is insensitive to the length of
number of interference fringes increased and data, the blue line in Fig. 3C resembles this the used synchrotron light pulses up to 100 ps.
how the angular separation of the maxima simple model. Note that the model agreed with 14. J. Ullrich et al., Rep. Prog. Phys. 66, 1463–1545 (2003).
15. T. Weber et al., Nature 431, 437–440 (2004).
decreased with increasing internuclear dis- the prediction from equation 12 of (20), if one
16. M. S. Schöffler et al., Phys. Rev. A 78, 013414 (2008).
tance R, which affirms the double-slit nature neglects the ionization potential. 17. C. Siedschlag, T. Pattard, J. Phys. B 38, 2297–2310
of the electron emission. Second, the red line in Fig. 3C shows the (2005).
The data shown in Fig. 2 were averaged over result from a more refined model. This model 18. D. A. Horner et al., Phys. Rev. Lett. 101, 183002 (2008).
19. The emission of the second electron is mediated through
all orientations of the molecular axis with re- accounted for the fact that—other than in the the Coulomb interaction and is therefore subject to the
spect to the light propagation and the light’s optical double slit—in photoionization, the two retardation of the electromagnetic field. This possible
polarization plane. Thus, the results must be interfering waves are not simply spherical. At additional delay of the double ionization process is
symmetric with respect to the inversion of the two centers
symmetric. To search for possible shifts of the the high photon energy used here, the atomic of the molecule. Therefore, our interferometric technique
interference fringes resulting from birth time nondipole effect tilts the electron angular is blind to this absolute delay.
delays, we inspected the interference pattern distributions from each center slightly in the 20. G. L. Yudin, S. Chelkowski, A. D. Bandrauk, J. Phys. B 39,
L17–L24 (2006).
of the fast electron for different angles b be- forward direction with respect to the photon 21. S. Grundmann et al., Phys. Rev. Lett. 124, 233201 (2020).
tween the photon propagation direction and propagation (21). This fact led to an additional 22. S. Chelkowski, A. D. Bandrauk, Phys. Rev. A 97, 053401
the molecular axis in Fig. 3 for the subset S angular shift of the interference pattern that (2018).
(see Fig. 2B). Figure 3A shows the measured slightly increased the slope of the red line com- 23. S. Grundmann, Zeptosecond Birth Time Delay in
Molecular Photoionization, version v1, Zenodo (2020);
fringes for molecules aligned parallel to the pared with the blue one. The red line obtained http://doi.org/10.5281/zenodo.3899920.
light propagation direction (see fig. S3 for a from considering molecular photoionization in
corresponding polar plot). The zeroth-order the time domain was in line with the prediction AC KNOWLED GME NTS

maximum of the distribution was displaced to of calculations of molecular photoionization in We acknowledge DESY (Hamburg, Germany), a member of
the Helmholtz Association HGF, for the provision of experimental
the right, which suggested the existence of a the frequency domain if nondipole effects are facilities. Parts of this research were carried out at PETRA III,
birth time delay. To confirm this assumption, included in full (22). This model is in reason- and we thank J. Seltmann and K. Bagschik for excellent support
we depicted the interference fringes as func- able agreement with the experimental results, during the beam time. S.G. is grateful for discussions on the
topic with A. Kheifets and H. Kremer. Funding: We acknowledge
tion of cos(b) in Fig. 3B. The histogram shows but more theoretical work including electron- support by BMBF and by DFG. K.F. acknowledges support
a clear dependence of the central fringe on the electron correlation is needed to clarify the by the German National Merit Foundation. Author contributions:
photon direction. For a quantitative analysis, deviation. M.S.S., T.J., K.F., N.S., A.P., L.K., S.E., M.W., S.G., D.T., L.Ph.H.S.,
R.D., and F.T. designed, prepared, and performed the experiment.
we determined cos(a0)—i.e., the angular posi- We have shown that the birth of a photo- S.G. performed the data analysis. S.G. and M.K. created the figures.
tion of the central maximum—for each row of electron wave from a molecular orbital did not All authors contributed to the manuscript. Competing interests:
the histogram in Fig. 3B through a Gaussian fit occur simultaneously across the molecule. With None declared. Data and materials availability: Data presented in
this study are available on Zenodo (23).
(red curve in Fig. 3A). Figure 3C shows the re- an electron interferometric technique, we ob-
sults of these fits and the corresponding birth served the resulting birth time delay, which
SUPPLEMENTARY MATERIALS
time delays (according to Eq. 2) on the right was imprinted as a phase difference between
science.sciencemag.org/content/370/6514/339/suppl/DC1
vertical scale. The birth time delay might be the parts of the wave launching from the two Materials and Methods
interpreted as a nondipole Wigner delay be- sides of the H2 molecule. The observed effect Supplementary Text
tween photoionization from different locations is general and does not only alter molecular Figs. S1 to S3
References (24–26)
of the spatially extended molecular orbital. photoionization, but it is also expected to be
However, the usual Wigner times—treated relevant for electron emission from solids and 28 April 2020; accepted 26 August 2020
entirely within the dipole approximation so liquids. 10.1126/science.abb9318

Grundmann et al., Science 370, 339–341 (2020) 16 October 2020 3 of 3

Zeptosecond birth time delay in molecular photoionization
Sven Grundmann, Daniel Trabert, Kilian Fehre, Nico Strenger, Andreas Pier, Leon Kaiser, Max Kircher, Miriam Weller,
Sebastian Eckart, Lothar Ph. H. Schmidt, Florian Trinter, Till Jahnke, Markus S. Schöffler and Reinhard Dörner

Science 370 (6514), 339-341.

DOI: 10.1126/science.abb9318

The travel time of light in a molecule

There is currently considerable interest in experimental studies of various ultrafast processes. Of particular
interest are the real-time dynamics of photoionization, one of the most fundamental processes caused by the light-matter
interaction, in which the absorption of a photon leads to the ejection of an electron and the formation of anion. Using an

Downloaded from http://science.sciencemag.org/ on October 16, 2020

electron interferometric technique, Grundmann et al. report a birth time delay on the order of a few hundred
zeptoseconds between two electron emissions from the two sides of molecular hydrogen, which is interpreted as the
travel time of the photon across the molecule. The proposed technique is generally applicable to more complex systems,
and further studies are necessary to support this interpretation.
Science, this issue p. 339

ARTICLE TOOLS http://science.sciencemag.org/content/370/6514/339

SUPPLEMENTARY http://science.sciencemag.org/content/suppl/2020/10/14/370.6514.339.DC1
MATERIALS

REFERENCES This article cites 26 articles, 7 of which you can access for free
http://science.sciencemag.org/content/370/6514/339#BIBL

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Use of this article is subject to the Terms of Service

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement of
Science, 1200 New York Avenue NW, Washington, DC 20005. The title Science is a registered trademark of AAAS.
Copyright © 2020 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of
Science. No claim to original U.S. Government Works