SciPost Submission Page
Photoemission spectroscopy from the threebody Green's function
by Gabriele Riva, Timothée Audinet, Matthieu Vladaj, Pina Romaniello and J. Arjan Berger
This is not the current version.
Submission summary
As Contributors:  Arjan Berger 
Preprint link:  scipost_202110_00015v2 
Date submitted:  20211127 11:34 
Submitted by:  Berger, Arjan 
Submitted to:  SciPost Physics 
Academic field:  Physics 
Specialties: 

Approach:  Theoretical 
Abstract
We present an original approach for the calculation of direct and inverse photoemission spectra from first principles. The main goal is to go beyond the standard Green's function approaches, such as the $GW$ method, in order to find a good description not only of the quasiparticles but also of the satellite structures, which are of particular importance in strongly correlated materials. Our method uses as a key quantity the threebody Green's function, or, more precisely, its holeholeelectron and electronelectronhole parts. We show that, contrary to the onebody Green's function, satellites are already present in the corresponding noninteracting Green's function. Therefore, simple approximations to the threebody selfenergy, which is defined by the Dyson equation for the threebody Green's function and which contains manybody effects, can still yield accurate spectral functions. In particular, the selfenergy can be chosen to be static which could simplify a selfconsistent solution of the Dyson equation. We also show how the onebody Green's function can be retrieved from the threebody Green's function. We illustrate our approach by applying it to the symmetric Hubbard dimer.
Current status:
Author comments upon resubmission
Thank you for sending us the reports of the referees.
We thank the referees for their careful reading of the manuscript and for their questions and comments.
We are pleased to read that the referees consider the manuscript as "well written and potentially of strong interest for the community" and "an important addition to the literature".
Both referees ask for some points to be clarified.
In the following we do so and we provide a list of changes.
We also address the comments of the editorincharge.
We hope that our revised manuscript will be suitable for publication in SciPost Physics.
Sincerely, the authors.
REVIEWER 1:
Reviewer 1 considers our manuscript as ``well written and potentially of strong interest for the community".
Reviewer 1 would like us to address the following two points.
Reviewer's comment:
"a satellite in photoemission arise from the interaction between electrons. In the noninteracting limit, only singleparticle poles exist. It is ok that the G3 contains extra poles also in the noninteracting case, but such poles should not contribute to the ARPES spectral function, i.e. they should have zero intensity. Only interaction, i.e. a static selfenergy, could give finite intensity to these poles. The authors find these poles in the analytical description (eq. 24), and they seem to suggest that they contribute to ARPES also in the noninteracting limit. Indeed there is a pole, which they call satellite, in Fig. 1. This point should be clarified."
Our reply:
We agree with the referee that in the ARPES spectral function the satellite amplitudes are nonzero only when the interaction is switched on. This spectral function is defined as the imaginary part of the onebody Green's function (1GF), for which satellite amplitudes are zero when the interaction is switched off. The confusion stems from the fact that in figures 13 we had also reported the imaginary part of the threebody Green's function (3GF), i.e., without the contraction to get the 1GF. The imaginary part of the 3GF is not equal to the ARPES spectral function. The 3GF contains more information than the 1GF. Therefore the imaginary part of the 3GF has nonvanishing satellite amplitudes also in the noninteracting case. The 1GF are obtained using the contractions in equations (34) and (35) and from the 1GF the ARPES spectral function can be obtained. Nevertheless, for analysis purposes, it is convenient to introduce a 3body spectral function as the imaginary part of the 3GF.
To clarify these points in the revised manuscript we have modified section II.D by adding the following two paragraphs:
Since the spectral representation of $G_3^{e+h}(\omega)$ given in Eq. (12) is similar to the one of $G_1$
it is convenient to introduce a 3body spectral function for $G_3^{e+h}(\omega)$ that is similar to the spectral function corresponding to $G_1$.
The latter is defined as
\begin{equation}
A(x_1,x_{1'};\omega)=\frac{1}{\pi}\text{sign}(\mu\omega)\text{Im} G_1(x_1,x_{1'};\omega).
\end{equation}
We can thus define the spectral function $A_3(\omega)$ corresponding to $G_3^{e+h}(\omega)$ according to
\begin{equation}
A_3(\omega)=\frac{1}{\pi}\text{sign}(\mu\omega)\text{Im} G_3^{e+h}(\omega),
\end{equation}
where, for notational convenience, the spinposition arguments are omitted.
and
We note that the 3body spectral function is not the spectral function that corresponds to photoemission spectroscopy. Both spectral functions have the same poles but the corresponding amplitudes are different.
In particular, in the noninteracting case the amplitudes of satellites can be nonzero in the 3body spectral function.
To retrieve the spectral function that corresponds to photoemission spectra Eq. (17) has to be used.
Moreover, figures 13 have been modified to emphasize more clearly the differences between the 1 and 3body spectral functions. These figures are now divided into two panels; in the upper panel we report the 1body spectral function (the ARPES spectral function), while the 3body spectral function is reported in the bottom panel.
We included this latter case to show that the 3GF has the poles at the same position as the 1GF and that only the amplitudes differ.
Reviewer's comment:
"An important approximation of standard approaches based on G1 (especially in the ab inito community) is to take the diagonal only component of the selfenergy, i.e. only the poles are corrected with respect to the zeroorder (usually DFT) simulation. G3 has six indexes, and it is not easy to understand where such approximation would enter. This is also related to how demanding would be the inversion of G3 in the QP basis set (appendix C), and which kind of satellites the approach could give. I would expect that, the contraction of the external indexes (indexes im in eqs. 3334) could give the QP approximation. However, for correlated satellites (plasmons, excitons, magnons, etc .. ) the internal indexes should be allowed to mix. Side comment, something is wrong in the indexes of eq. 30. The Hubbard dimer does not help much here. At least a general discussion in this direction would be useful."
Our reply:
We thank the referee for raising this very interesting point. Indeed the diagonal approximation to the selfenergy is an important practical tool to calculate the poles of the onebody Green's function. If the selfenergy is dynamical both quasiparticle energies and satellites can be obtained with this approximation, although in practice mainly quasiparticle energies are calculated. Instead, if the selfenergy is static only quasiparticles can be calculated.
A diagonal approximation can also be made for the threebody selfenergy and this could be very interesting because it would reduce the numerical cost of the calculations significantly. Moreover, from a static threebody selfenergy both quasiparticles and satellites can be obtained. Of course, the quality of the quasiparticle energies and satellites not only depend on the diagonal approximation but also on the approximation to the selfenergy. The latter approximation is probably the most crucial. As alluded to by the referee we can not test the diagonal approximation on the Hubbard dimer since the threebody selfenergy is already diagonal in the basis that diagonalizes the noninteracting 3GF. The referee mentions the contraction of the external indices in eqs. 3334 (eqs. 3435 in the revised manuscript), i.e., $m=i$. However, the equivalent of the diagonal approximation in the threebody case would be to set $m=i$, $o=j$ and $k=l$.
We have now clarified this point in the revised manuscript by adding the following sentences to the outlook given in section 4.
Moreover, we can reduce the numerical cost of the calculations by applying a diagonal approximation to the threebody selfenergy, in similar manner as is often done for the onebody selfenergy, to calculate only the poles of $G_3^{e+h}$. While a static onebody selfenergy can only yield poles that correspond to quasiparticles, a diagonal static threebody selfenergy would yield the poles corresponding to both quasiparticles and satellites.
We have also corrected the indices in Eq. (30) (Eq. (31) of the revised manuscript).
REVIEWER 2:
Referee 2 considers our manuscript ``an important addition to the literature" and referee 2 recommends publication after two points are clarified.
Reviewer's comment:
"In Figs. 15 can it be clarified whether a curve labeled G3, or Sigma3, is the 3particle spectral function or the 1particle spectral function obtained from a calculation of G3? Hopefully the latter."
Our reply:
We thank the referee for this question. This point was indeed not clear.
In the revised manuscript we have modified figures 13 to emphasize more clearly the differences between the 1 and 3body spectral functions. These figures are now divided into two panels; in the upper panel we report the 1body spectral function (the one that corresponds to photoemission spectra), while the 3body spectral function is reported in the bottom panel. We included this latter case to show that the 3GF has the poles at the same position as the 1GF and that only the amplitudes differ. In figures 4 and 5 only spectral functions corresponding to the 1GF are shown.
Reviewer's comment:
"Can the actual steps in the calculation of the spectral function using their simplified G3 (static Sigma3) be set out in the language of a general manyelectron system and of manybody perturbation theory? E.g. as a flowchart. This would help the reader assess how realistic the Hubbard model is here as a prototype manyelectron system. Is Sigma3 spatially nonlocal? What is the extent of selfconsistency imposed by the calculation? To what extent can the recommended procedure be regarded as a true perturbation theory, and if it can, what is the small quantity?"
Our reply:
The exact 3body selfenergy is defined by the Dyson equation given in Eq. (18) and it is indeed a nonlocal quantity. We do not yet have an expression for the 3body selfenergy in which it is given as an explicit functional of $G_3$ and the interaction. Therefore, in this work we are not doing manybody perturbation theory. For our application to the Hubbard dimer we have derived the exact $G_3^{e+h}$ and $G_{03}^{e+h}$. Therefore, we know the exact $\Sigma_3$ because we can solve the Dyson equation (Eq. (18)) and we can take the static approximation by setting $\omega=0$. To obtain the corresponding $G_3^{e+h}$ we solve once more the Dyson equation in Eq. (18). As a consequence selfconsistency is not an issue here. The main goal of this manuscript is to show that, with a static selfenergy, the 3GF has information about satellites, contrary to the 1GF. We are now trying to make the theory applicable to real systems, by looking for approximations to $\Sigma_3$ as explicit functionals of $G_3^{e+h}$ and the interaction. This interaction, which could be, for example, the bare Coulomb interaction or the screened Coulomb interaction, will then be the small parameter. This is, however, beyond the scope of this work.
We have now clarified these points in the conclusions by adding the following paragraph,
For the specific case of the Hubbard dimer we were able to obtain the exact $G_3^{e+h}$. Therefore we could obtain an exact threebody selfenergy by solving a Dyson equation. However, in general, the exact threebody selfenergy is unknown. Therefore, our next goal is to derive a general static approximation for the threebody selfenergy. This could be achieved, for example, by using the equation of motion for $G^{e+h}_3$ along the same lines as has been done for $G_1$ or by using a similar strategy as in Ref. [27], where a practical scheme to calculate $G_3$ for the description of Auger spectra is proposed.
We also added the following sentences to the outlook given in section 4 briefly discussing a possible strategy to reduce the numerical cost of the calculations, namely by using a diagonal approximation to the threebody selfenergy, as was mentioned by Reviewer 1.
Moreover, we can reduce the numerical cost of the calculations by applying a diagonal approximation to the threebody selfenergy, in similar manner as is often done for the onebody selfenergy, to calculate only the poles of $G_3^{e+h}$. While a static onebody selfenergy can only yield poles that correspond to quasiparticles, a diagonal static threebody selfenergy would yield the poles corresponding to both quasiparticles and satellites.
EDITORINCHARGE:
We now address the comments and questions of the editorincharge.
Editor's comment:
``The solution of the threebody green's function was already used in the literature: for the Auger problem by A. Marini and M. Cini in Journal of Electron Spectroscopy and Related Phenomena 127 (2002) 17â28 and by C. Calandra and F. Manghi in Phys. Rev. B 50, 2061 to study satellite structures and the occurrence of the metalinsulator transition. I think the authors should mention these two works in their manuscript".
Our reply:
We thank the editor for these two important references, which we have now added to the manuscript in the Introduction, as:
We note that the threebody Green's function has been employed to describe Auger spectra [27] and to study satellite structures and the occurrence of the metalinsulator transition. [28]
Editor's comment:
"Another physical phenomenon that could be studied with the present approach is probably "trions". May the author comment on this possibility? Do they expect trions will be well described by solving the G3 problem?"
Our reply:
We agree with the editor that trions would be another interesting applications of our approach. We had briefly mentioned this possibility in the conclusions. We have now added more references there. We cannot foresee the performance of our approach to describe these excitations because it will depend on the quality on the approximations to the 3body selfenergy. It will be interesting to explore this problem in the future.
Editor's comment:
"I think in Eq. 54 you wrote "$G_2(\omega)$" instead of "$G_1(\omega)$"."
Our reply:
We thank the editor for noticing this typo. We have now corrected it.
List of changes
1.
We have modified section II.D by adding the following two paragraphs:
Since the spectral representation of $G_3^{e+h}(\omega)$ given in Eq. (12) is similar to the one of $G_1$
it is convenient to introduce a 3body spectral function for $G_3^{e+h}(\omega)$ that is similar to the spectral function corresponding to $G_1$.
The latter is defined as
\begin{equation}
A(x_1,x_{1'};\omega)=\frac{1}{\pi}\text{sign}(\mu\omega)\text{Im} G_1(x_1,x_{1'};\omega).
\end{equation}
We can thus define the spectral function $A_3(\omega)$ corresponding to $G_3^{e+h}(\omega)$ according to
\begin{equation}
A_3(\omega)=\frac{1}{\pi}\text{sign}(\mu\omega)\text{Im} G_3^{e+h}(\omega),
\end{equation}
where, for notational convenience, the spinposition arguments are omitted.
and
We note that the 3body spectral function is not the spectral function that corresponds to photoemission spectroscopy. Both spectral functions have the same poles but the corresponding amplitudes are different.
In particular, in the noninteracting case the amplitudes of satellites can be nonzero in the 3body spectral function.
To retrieve the spectral function that corresponds to photoemission spectra Eq. (17) has to be used.
Moreover, figures 13 have been modified to emphasize more clearly the differences between the 1 and 3body spectral functions. These figures are now divided into two panels; in the upper panel we report the 1body spectral function (the ARPES spectral function), while the 3body spectral function is reported in the bottom panel.
We included this latter case to show that the 3GF has the poles at the same position as the 1GF and that only the amplitudes differ.
2.
We have added the following sentences to the outlook given in section 4.
Moreover, we can reduce the numerical cost of the calculations by applying a diagonal approximation to the threebody selfenergy, in similar manner as is often done for the onebody selfenergy, to calculate only the poles of $G_3^{e+h}$. While a static onebody selfenergy can only yield poles that correspond to quasiparticles, a diagonal static threebody selfenergy would yield the poles corresponding to both quasiparticles and satellites.
We have also corrected the indices in Eq. (30) (Eq. (31) of the revised manuscript).
3.
We have added the following paragraph to the conclusions,
For the specific case of the Hubbard dimer we were able to obtain the exact $G_3^{e+h}$. Therefore we could obtain an exact threebody selfenergy by solving a Dyson equation. However, in general, the exact threebody selfenergy is unknown. Therefore, our next goal is to derive a general static approximation for the threebody selfenergy. This could be achieved, for example, by using the equation of motion for $G^{e+h}_3$ along the same lines as has been done for $G_1$ or by using a similar strategy as in Ref. [27], where a practical scheme to calculate $G_3$ for the description of Auger spectra is proposed.
4.
We have added the following sentence to the Introduction,
We note that the threebody Green's function has been employed to describe Auger spectra [27] and to study satellite structures and the occurrence of the metalinsulator transition. [28]
5.
We have added some references of works in the literature that involve the 3GF
6.
in Eq. 54 we modified "$G_2(\omega)$" to "$G_1(\omega)$"."
Submission & Refereeing History
You are currently on this page
Reports on this Submission
Anonymous Report 4 on 2022112 (Invited Report)
Strengths
1) The manuscript presents a systematic discussion of the threeparticle
Green's function  i.e. the time ordered product of six fermion operators 
for an interacting manyelectron system. The calculation is shown in some
detail and can be followed relatively easily.
2) The manuscript contains a detailed comparison between various approximations
and exact results for a solvable system, the Hubbard dimer. For strongly
correlated electron systems that is a very important thing to do in my
opinion.
Weaknesses
1) It is somewhat unclear to me inhowfar the method which is presented
really is helpful for more complicated systems than a Hubbard dimer.
In partiular i wonder if in the case of the inverse photoemission spectrum in
the quarter filled ground state  i.e. with one electron in the dimer  taking
all states with one added electron and a 'particle hole excitation' of the
electron present initially is not equivalent to an exact solution?
2) I somewhat resent the use of the term 'satellites' in this manuscript.
For example in the noninteracting threeparticle Green's function I would
expect that these 'satellites' really are structureless continua and have
little to do with the features called satellites in photoemission spectra
of correlated electron systems, which are more something like
Hubbard bands.
3) I do not understand why the authors are using a Green's function of
6 Fermion operators. Would the most natural extension not be a Green's function
that has three fermions at time t_1 and one Fermion at time t_2,
i.e. the type of Green's function which shows up in the equation
of motion of the single particle Green's function?
Report
Report on
'Photomissionspectroscopy from the threebody Green's function'
by G. Riva et al.
The main point of the manuscript is the discussion of the threeparticle
Green's function  i.e. the time ordered product of six fermion operators 
for an interacting manyelectron system. The authors derive a spectral
representation which is roughly equivalent to the Lehmann representation
and introduce a special time ordering (Eq. (11) ) to apply this to
photoemssion and inverse photoemission, which is then used to derive the
singleparticle Green's function. The authors also introduce a threeparticle
selfenergy and a Dysonequation by which the interacting threeparticle
Green's function is expressed in terms of the one for noninteracting
particles. As a benchmark the authors then compute various Green's functions
in different approximations for the exactly solvable system of a Hubbard dimer
and compare the results.
The manuscript presents a remarkable amount of work but is somewhat hard to
read. It is not really clear to me inhowfar the method which is presented
really is helpful for more complicated systems than a Hubbard dimer.
Still, I think the manuscript meeets the acceptance criteria once a few minor
corrections have been made as detailed below.
Requested changes
1) Threeparticle Green's functions are being studied for a very long time.
For example the wellknown Hubbardoperators are nothing but products of
three Fermion operators. More generally, composite operators have been used
for a long time, see PHYSICAL REVIEW B104, 155128 (2021) for a recent example.
It would appear to me that these works are more physically motivated than
the rather technical approach of the authors and in any way should be mentioned.
2) The derivation of the spectral representation is rather unpleasant to
read because some equations are only in section 2.1., others only in Appendix A,
so that a lot of backandforth scrolling is necessary if one wants to follow
the calculation. I would suggest to change this.
The authors should comment on the three points in the 'weak points' section
Anonymous Report 3 on 20211217 (Invited Report)
Report
The direct and inverse photoemission intensity is given, within the sudden approximation, by the oneelectron removal and addition spectra that are proportional to the imaginary part of the retarded single particle Green function (e.g. Rev. Mod. Phys. 75, 473 2003, Phys. Rev. B 94, 115119 2016). This is the case for both (strongly) interacting and noninteracting/weakly interacting systems. This results stems from a direct calculation of the photoelectron current using scattering theory. Beyond the sudden approximation corrections to the photocurrent appear. Hedin and coworkers (Phys. Rev. B 58 15565 1998) have shown that whereas the sudden approximation includes "intrinsic losses" or satellite structure, adiabatic corrections provide further the "extrinsic losses".
The authors state that the threebody Green function contribute to photoemission intensity, but do not derive this statement from a calculation of the photocurrent. Instead the existing literature (some of which referred to above) seems to agree that within the sudden approximation (and apart from matrixelement effects), the oneparticle Green function contains all spectral information relevant for photoemission.
In this context it is not clear how the authors challenge the present status quo and can justify their statement that the threebody Green function is a fundamental quantity to the calculation photoemission spectra. Possibly the implication is that the threebody Green function embodies corrections beyond the sudden approximation. If so, these corrections in terms of the threebody Green function needs to be derived in a mathematically consistent fashion from fundamental considerations on the photocurrent.
Author: Arjan Berger on 20211217 [id 2034]
(in reply to Report 3 on 20211217)
There seems to be a misunderstanding.
In this work we are always working within the sudden approximation which is the standard in our field. When we write that we use the 3body Green’s function (3GF) to calculate photoemission spectra it is implied that we mean photoemission spectra within the sudden approximation. It is not mentioned explicitly since it is standard practice in our field. We will make this point explicit in the second revision of our work in order to avoid any possible misunderstanding by rewriting the following sentence in the introduction
"The main reason is that the onebody Green’s function (1GF) can be easily linked to photoemission spectra since its poles are the electron removal and addition energies.”
as
"The main reason is that the onebody Green’s function (1GF) can be easily linked to photoemission spectra (within the sudden approximation) since its poles are the electron removal and addition energies.”
We note that the referee claims that we state that “the threebody Green function contribute to photoemission intensity”.
We want to make clear that nowhere in the paper do we make this statement.
The purpose of this work is completely different.
The final goal is still to calculate the onebody Green’s function (1GF) since, within the sudden approximation, it indeed has all the required information about photoemission spectra.
However, that is if one has the exact 1GF.
The main idea of this work is to use the 3GF to improve the approximations to the 1GF and, in particular, to capture satellites. This point is explained in detail in the Introduction and Results sections of the paper. In a nutshell, to capture satellites using the standard 1body approach one requires a dynamical selfenergy since the noninteracting 1GF only contains information about quasiparticles. It is wellknown that it is difficult to obtain good dynamical approximations for the selfenergy. For example, the GW approximation does not yield very accurate satellites. Instead, when using a 3body approach, information about satellites is already contained in the noninteracting 3GF and, therefore, a simpler static 3body selfenergy is sufficient to capture satellites. Once the 3GF is obtained we contract (according to Eq. (17)) to obtain the 1GF and therefore the photoemission spectrum including the satellites.
To make this point clearer in the second revision we will modify the following sentence in the Introduction,
"Therefore, we will study here the threebody Green’s function (3GF) as the fundamental quantity from which to calculate photoemission spectra”
to
“Therefore, we will study here the threebody Green’s function (3GF) as the fundamental quantity from which to calculate the 1GF and, hence, photoemission spectra"
Now that this misunderstanding has been cleared up and in view of the two recommendations to publish our work in SciPost Physics already given by the other two referees, we hope that our work can now finally be published in SciPost Physics.
Report 2 by Davide Sangalli on 2021123 (Invited Report)
Report
The authors have addressed the points raised in my review.
In particular for the definition of the ARPES spectral function from G3. The role of the basis set and the related diagonal approximation remains to be further explored. Future applications on real materials could possibly clarify this.
I think the manuscript can be now accepted for publication.
Anonymous Report 1 on 20211130 (Invited Report)
Report
The authors have made helpful modifications to their manuscript that provide the clarification that I (and, in my opinion, also the other referee and the Editor) requested.
Davide Sangalli on 20211203 [id 2006]
The authors have addressed the points raised in my review.
In particular for the definition of the ARPES spectral function from G3. The role of the basis set and the related diagonal approximation remains to be further explored. Future applications on real materials could possibly clarify this.
I think the manuscript can be now accepted for publication.