PHYSICAL REVIEW A 105, 047101 (2022)

Comment on “Origin of symmetry-forbidden high-order harmonic generation

in the time-dependent Kohn-Sham formulation”

Ofer Neufeld ,1,* Nicolas Tancogne-Dejean ,1 and Angel Rubio 1,2,†

1
Max Planck Institute for the Structure and Dynamics of Matter and Center for Free-Electron Laser Science, Hamburg 22761, Germany
2
Center for Computational Quantum Physics (CCQ), The Flatiron Institute, New York, New York 10010, USA

(Received 24 June 2021; accepted 22 March 2022; published 6 April 2022)

In their recent paper [Phys. Rev. A 103, 043106 (2021)], Zang et al. theoretically investigated high harmonic
generation (HHG) in benchmark two-electron systems that are inversion symmetric with time-dependent density
functional theory (TDDFT) in the Kohn-Sham formulation. They found that the theory wrongly predicted the
emission of symmetry-forbidden even harmonics and concluded that this error originates from an inherent
problem of TDDFT that unphysically populates one- and two-electron excited states. They further claimed
that this effect results in an incorrect HHG cutoff energy. We reproduced their main results, but found that
the unphysical even harmonics that they observed originated from numerical errors introduced by the boundary
conditions. We show that contrary to their claims, the HHG cutoff energy calculated within TDDFT agrees
perfectly with the standard and well-established models of HHG.

DOI: 10.1103/PhysRevA.105.047101

In their recent work [1], Zang et al. performed HHG calcu- correctly reproduces the HHG cutoff energy in agreement
lations for inversion-symmetric two-electron systems (H2 and with well-established models for HHG [5–8]. Lastly, we high-
He) with various levels of theory, including time-dependent light that any errors in excited state occupation in TDDFT (i)
configuration interaction (TCI), time-dependent density func- cannot break any inversion (or other) inherent symmetry of
tional theory (TDDFT) in the Kohn-Sham (KS) formulation, the system, and (ii) are conceptually fixable by using better
and their recently developed time-dependent natural Kohn- approximations for the XC functional (and are absent in for-
Sham formulation (TDNKS) [2]. Their numerical results for mally exact TDDFT).
HHG using standard TDDFT (with either the local density We begin by reproducing the main result of Ref. [1]—
approximation (LDA) or Perdew-Burke-Ernzerhof (PBE) [3] Fig. 1(a) shows our calculated HHG spectra from H2 using
approximation for the exchange-correlation (XC) functional) TDDFT within the LDA with an added self-interaction cor-
showed the presence of symmetry-forbidden even harmonics rection (SIC) [9]. The laser parameters are chosen identical
that should not have been observed in these conditions due to a to those in Figs. 1–3 in Ref. [1], where the laser elec-
dynamical inversion symmetry [4], and which are standardly tric field is polarized along the H2 main axis with power
not observed with other levels of theory nor experimentally. I0 = 3.36 × 1012 W/cm2 and wavelength λ = 800 nm (the
They concluded that the error in H2 and He is due to an resulting Keldysh parameter is ∼6.2). Clearly, unphysical
inherent flaw in multielectron TDDFT in the KS formulation. even harmonics are emitted in accordance with the results
Specifically, they argued that the occupation of excited states in Ref. [1] [see highlighted harmonic orders 8, 10, and 12
in TDDFT is wrong and spuriously allows for this symmetry in Fig. 1(a)]. We also note that a similar result is obtained
breaking to occur. Lastly, they claimed that the HHG cutoff for a laser field transverse to the H2 main axis, but the even
energy was wrongly reproduced within TDDFT due to similar harmonics in that case were much less pronounced (therefore
considerations. this case is not presented or considered from this point on).
Here we repeat the calculations in Ref. [1] and argue Numerically, the result in Fig. 1(a) was obtained with a
that the observation of symmetry-forbidden harmonics in fact Cartesian grid with the real-space TDDFT code, OCTOPUS
originated from convergence issues. We demonstrate that con- [10–12], where the box was a sphere of radius 45 bohr and
trary to the claims in Ref. [1], TDDFT (even within the LDA) grid spacing of 0.4 bohr. The results in Ref. [1] were obtained
with a similar approach but with a much smaller box size of
radius 20 bohr that may initially seem too small. However,
*
ofer.neufeld@gmail.com for these laser conditions the quiver length of the electron
†
angel.rubio@mpsd.mpg.de motion is 3 bohr; thus a priori both box sizes should suffice
for obtaining converged spectra. It is important to note that
Published by the American Physical Society under the terms of the counterintuitively, it can be much more difficult to numeri-
Creative Commons Attribution 4.0 International license. Further cally converge calculations for weak laser driving and high
distribution of this work must maintain attribution to the author(s) Keldysh parameters than for stronger field powers. This is
and the published article’s title, journal citation, and DOI. Open because the resulting harmonic yield is often numerically
access publication funded by the Max Planck Society. close to the noise level of the calculation. Also, for low laser

2469-9926/2022/105(4)/047101(4) 047101-1 Published by the American Physical Society

COMMENTS PHYSICAL REVIEW A 105, 047101 (2022)

FIG. 1. Origin of spurious even harmonics in HHG calculations from H2 . (a) HHG spectra from H2 calculated with KS-TDDFT with a
15 bohr wide CAP. (b) Same as (a) but with a 30 bohr wide CAP. (c) Same as (a) but for a SAE model from the TDSE. (d) Same as (b) but for
a SAE model from the TDSE. Arrows highlight positions of unphysical even harmonics for the narrow absorber, and their absence in the wide
absorber case. HHG yield is presented in logarithmic scale (with basis 10).

powers low-energy electrons are ionized that are difficult to shows spurious even harmonics just as in the TDDFT case.
absorb in the boundaries [13]; these can also lead to spurious Importantly though, Fig. 1(c) corresponds to a time-dependent
reflections or numerical instability (especially for very long Schrödinger equation (TDSE) calculation for a SAE which
laser pulses as those used in Ref. [1] and Fig. 1(a) (20 opti- is formally exact, and which cannot be accused of leading
cal cycles long)). Figure 1(a) was calculated with a 15 bohr to spurious even harmonics. Figure 1(d) shows that with the
wide complex absorbing potential (CAP), which is usually wider CAP these errors again vanish. The fact that this er-
sufficient to obtain well-converged HHG spectra. In contrast, ror is reproduced with a SAE model directly shows that it
Fig. 1(b) shows HHG spectra in the exact same conditions does not originate from any inherent problem in KS-TDDFT
as those in Fig. 1(a), but with a CAP that is 30 bohr wide. for multielectron systems. Notably, in Ref. [1] the authors
It clearly shows that the spurious even harmonic emission performed calculations for atomic hydrogen which is a one-
is almost fully suppressed. This suggests that the unphysical electron system, but did not show such errors. We believe that
even harmonics are a result of convergence issues. this might be a result of the lower Ip of atomic hydrogen (e.g.,
To further show that the unphysical emission in Fig. 1(a) even worse errors came up in Ref. [1] for He which has an
is not a fundamental consequence of TDDFT or its approx- even higher Ip ). Similarly, the errors go away for higher laser
imations, we repeat a similar calculation for a single-active- powers (also seen in Ref. [1]) because the Keldysh parameter
electron (SAE) model of H2 (such models are commonly is reduced and an apparent convergence is obtained (since the
employed in the HHG community [14–17]). For this model harmonic spectrum is dominated by faster electrons for which
we use the following potential that has the correct asymptotic the reflection error is smaller).
form and mimics some local screening effects (given in atomic We also point out that in one-dimensional (1D) systems
units): one can perform numerically exact calculations with large
spatial box sizes without resorting to absorbing boundaries
0.5 + 0.5 × e−(r−R0 ) 0.5 + 0.5 × e−(r+R0 )
2 2

V (r) = −  −  , (because the grid is sufficiently large to avoid reflections). We

(r − R0 )2 + a (r + R0 )2 + a have performed such TDDFT calculations for a 1D model of
H2 in conditions similar to those in Fig. 1, and have simi-
(1)
larly seen only odd harmonics being emitted (not presented).
where R0 = 0.6692 bohr is half the equilibrium H-H spacing Overall, we see no evidence that KS-TDDFT is breaking any
for the H2 molecule, r is the SAE radial coordinate, and inherent symmetry of the system.
a is a free parameter set to a = 0.102 such that the model Another point worth discussing is that symmetry-breaking
correctly reproduces the first ionization potential of H2 , Ip = harmonics can also appear in HHG in standard conditions
15.4 eV. Figures 1(c) and 1(d) show results of calculations due to symmetry-breaking elements in the laser-matter system
using this model potential in the exact same conditions as itself (e.g., the laser pulse envelope and its final duration, and
those in Figs. 1(a) and 1(b), respectively. Clearly Fig. 1(c) the system occupying more than one Floquet state) [4,18–20].

047101-2
COMMENTS PHYSICAL REVIEW A 105, 047101 (2022)

I0=2×1013 I0=3×1013 I0=4×1013

W/cm2 W/cm2 W/cm2
Yield

I0=5×1013 I0=6×1013 I0=7×1013

W/cm2 W/cm2 W/cm2
Yield

I0=8×1013 I0=9×1013 I0=1014

W/cm2 W/cm2 W/cm2
Yield

Harmonic order Harmonic order Harmonic order

FIG. 2. HHG cutoff energy scaling with respect to laser power in TDDFT. Plots show HHG spectra from H2 calculated with TDDFT in
the same approach as in Fig. 1(b), but for various laser powers. Red dashed lines denote the Ip , red solid lines denote the Lewenstein model
cutoff prediction, and dashed blue lines signify the numerical cutoff. Odd harmonics are highlighted by gray lines. HHG yield is presented in
logarithmic scale (with basis 10).

Moreover, TDDFT describes nonlinear equations of motion nition originates from analytical considerations arising from
(because the XC potential depends on the time-dependent the saddle-point method applied in the Lewenstein model
density); thus, harmonic selection rules might also be broken itself—energy conservation can only be obtained above the
by imperfections in the approximations for the XC functional cutoff for imaginary times, leading to an exponentially de-
[21]. Nonetheless, in this particular case we find that these caying harmonic yield [6–8]. In that sense, it is conceptually
issues are not the source of the symmetry-breaking harmonics. not reasonable to compare cutoff energies defined as the
Lastly, we address the claims in Ref. [1] that TDDFT leads last observable harmonic above noise level to the analytical
to an incorrect cutoff energy as compared to the Lewenstein formula from the Lewenstein model. We highlight that with
model of HHG [6]. Figure 2 presents calculations in H2 with the correct definition, results in Ref. [1] make much more
the same approach as in Fig. 1(b), but for laser powers ranging sense for TDDFT, while the TDNKS formulation seems to
from 1013 to 1014 W/cm2 . In each subplot the Ip is highlighted incorrectly describe the HHG cutoff and plateau (the response
in dashed red, the cutoff harmonic is highlighted in blue, and seems largely perturbative). (ii) The box size used in Ref. [1]
the cutoff prediction from the Lewenstein model (Ip + 3.17Up was too small, causing an artificial saturation of the cutoff for
where Up is the laser ponderomotive energy) is highlighted higher laser powers. (iii) The fitting procedure used in Ref. [1]
in red. The results agree extremely well, always within one was possibly incorrect (as Eq.(5) in Ref. [1] has a redundant
harmonic order, as expected for long-duration laser driving 1.3 factor in front of the Ip that should be negligible in these
pulses. Some deviations are expected due to approximations conditions).
in the Lewenstein model (e.g., Coulomb focusing effects and In summary, we could not find any evidence for an intrinsic
electron-electron interactions that are neglected), but these error introduced by TDDFT that breaks inherent symmetries
should be small. We therefore conclude that the incorrect of the light-matter Hamiltonian (and such evidence cannot
scaling discussed in Ref. [1] was a result of several errors: (i) be obtained from unconverged numerical results). Addition-
Zang et al. incorrectly defined the cutoff as the last harmonic ally, if such hypothetical errors were indeed present and
order above noise level. This definition is sometimes used non-negligible, they may be suppressed by employing bet-
experimentally, but is inappropriate when comparing theory ter approximations for the XC functional (beyond PBE) and
to the Lewenstein model where the cutoff energy is defined would not appear in formally exact TDDFT. We also showed
as the start of an exponential drop in yield. This cutoff defi- that the HHG cutoff energy agrees perfectly with standard

047101-3
COMMENTS PHYSICAL REVIEW A 105, 047101 (2022)

well-established models for HHG, and the errors discussed in The Flatiron Institute is a division of the Simons Foun-
Ref. [1] result from an incorrect analysis. Thus, in our opinion dation. O.N. gratefully acknowledges the support of the
TDDFT remains one of the best affordable approaches for Humboldt foundation, and support from a Schmidt Sci-
describing strong-field physics from first principles. ence Fellowship. This work was supported by the Cluster
of Excellence Advanced Imaging of Matter (AIM), Gru-
We thank an anonymous referee for insightful contri- pos Consolidados (Grant No. IT1249-19) and SFB925,
butions. We acknowledge financial support from the Euro- Light induced dynamics and control of correlated quantum
pean Research Council (Grant No. ERC-2015-AdG-694097). systems.

[1] X. Zang, C. Shang, P. Zhang, Y. Zhu, Z. Zhu, and U. [12] N. Tancogne-Dejean, M. J. T. Oliveira, X. Andrade, H. Appel,
Schwingenschlögl, Origin of symmetry-forbidden high-order C. H. Borca, G. Le Breton, F. Buchholz, A. Castro, S. Corni,
harmonic generation in the time-dependent Kohn-Sham formu- A. A. Correa, U. De Giovannini, A. Delgado, F. G. Eich, J.
lation, Phys. Rev. A 103, 043106 (2021). Flick, G. Gil, A. Gomez, N. Helbig, H. Hübener, R. Jestädt,
[2] X. Zang, U. Schwingenschlögl, and M. T. Lusk, Identifica- J. Jornet-Somoza et al., Octopus, a computational framework
tion and Resolution of Unphysical Multielectron Excitations for exploring light-driven phenomena and quantum dynamics
in the Real-Time Time-Dependent Kohn-Sham Formulation, in extended and finite systems, J. Chem. Phys. 152, 124119
Phys. Rev. Lett. 124, 026402 (2020). (2020).
[3] J. P. Perdew, K. Burke, and M. Ernzerhof, Generalized Gradient [13] U. De Giovannini, A. H. Larsen, and A. Rubio, Modeling
Approximation Made Simple, Phys. Rev. Lett. 77, 3865 (1996). electron dynamics coupled to continuum states in finite vol-
[4] O. Neufeld, D. Podolsky, and O. Cohen, Floquet group theory umes with absorbing boundaries, Eur. Phys. J. B 88, 56
and its application to selection rules in harmonic generation, (2015).
Nat. Commun. 10, 405 (2019). [14] L. Medišauskas, J. Wragg, H. van der Hart, and M. Y. Ivanov,
[5] P. B. Corkum, Plasma Perspective on Strong Field Multiphoton Generating Isolated Elliptically Polarized Attosecond Pulses
Ionization, Phys. Rev. Lett. 71, 1994 (1993). Using Bichromatic Counterrotating Circularly Polarized Laser
[6] M. Lewenstein, P. Balcou, M. Y. Ivanov, A. L’Huillier, and Fields, Phys. Rev. Lett. 115, 153001 (2015).
P. B. Corkum, Theory of high-harmonic generation by low- [15] L. Barreau, K. Veyrinas, V. Gruson, S. J. Weber, T. Auguste,
frequency laser fields, Phys. Rev. A 49, 2117 (1994). J.-F. Hergott, F. Lepetit, B. Carré, J.-C. Houver, D. Dowek, and
[7] T. Schultz and M. Vrakking, Attosecond and XUV Physics: P. Salières, Evidence of depolarization and ellipticity of high
Ultrafast Dynamics and Spectroscopy (Wiley, New York, 2014). harmonics driven by ultrashort bichromatic circularly polarized
[8] K. Amini, J. Biegert, F. Calegari, A. Chacón, M. F. Ciappina, fields, Nat. Commun. 9, 4727 (2018).
A. Dauphin, D. K. Efimov, C. F. de M. Faria, K. Giergiel, [16] O. Neufeld and O. Cohen, Background-Free Measurement of
P. Gniewek, A. Landsman, M. Lesiuk, M. Mandrysz, A. S. Ring Currents by Symmetry-Breaking High-Harmonic Spec-
Maxwell, R. Moszynski, L. Ortmann, J. A. Perez-Hernandez, troscopy, Phys. Rev. Lett. 123, 103202 (2019).
A. Picon, E. Pisanty, J. S. Prauzner-Bechcicki et al., Symphony [17] E. Bordo, O. Neufeld, O. Kfir, A. Fleischer, and O. Cohen,
on strong field approximation, Rep. Prog. Phys. 82, 116001 Spectroscopy of atomic orbital sizes using bi-elliptical high-
(2019). order harmonic generation, Phys. Rev. A 100, 043419
[9] J. P. Perdew and A. Zunger, Self-interaction correction to (2019).
density-functional approximations for many-electron systems, [18] R. Bavli and H. Metiu, Properties of an electron in a quantum
Phys. Rev. B 23, 5048 (1981). double well driven by a strong laser: Localization, low-
[10] A. Castro, H. Appel, M. Oliveira, C. A. Rozzi, X. Andrade, F. frequency, and even-harmonic generation, Phys. Rev. A 47,
Lorenzen, M. A. L. Marques, E. K. U. Gross, and A. Rubio, 3299 (1993).
Octopus: A tool for the application of time-dependent density [19] N. Moiseyev and M. Lein, Non-Hermitian quantum mechanics
functional theory, Phys. Status Solidi 243, 2465 (2006). for high-order harmonic generation spectra, J. Phys. Chem. A
[11] X. Andrade, D. Strubbe, U. De Giovannini, H. Larsen, M. J. 107, 7181 (2003).
T. Oliveira, J. Alberdi-Rodriguez, A. Varas, I. Theophilou, N. [20] V. Kapoor and D. Bauer, Floquet analysis of real-time wave
Helbig, M. J. Verstraete, L. Stella, F. Nogueira, A. Castro, M. functions without solving the Floquet equation, Phys. Rev. A
A. L. Marques, and A. Rubio, Real-space grids and the octopus 85, 023407 (2012).
code as tools for the development of new simulation approaches [21] V. Kapoor, M. Ruggenthaler, and D. Bauer, Periodicity of the
for electronic systems, Phys. Chem. Chem. Phys. 17, 31371 time-dependent Kohn-Sham equation and the floquet theorem,
(2015). Phys. Rev. A 87, 042521 (2013).

047101-4