SciPost Submission Page
Deriving densitymatrix functionals for excited states
by Julia Liebert, Christian Schilling
This is not the latest submitted version.
This Submission thread is now published as
Submission summary
Authors (as registered SciPost users):  Julia Liebert · Christian Schilling 
Submission information  

Preprint Link:  https://arxiv.org/abs/2210.00964v1 (pdf) 
Date submitted:  20221007 23:23 
Submitted by:  Schilling, Christian 
Submitted to:  SciPost Physics 
Ontological classification  

Academic field:  Physics 
Specialties: 

Approach:  Theoretical 
Abstract
We initiate the recently proposed $\boldsymbol{w}$ensemble oneparticle reduced density matrix functional theory ($\boldsymbol{w}$RDMFT) by deriving the first functional approximations and illustrate how excitation energies can be calculated in practice. For this endeavour, we first study the symmetric Hubbard dimer, constituting the building block of the Hubbard model, for which we execute the LevyLieb constrained search. Second, due to the particular suitability of $\boldsymbol{w}$RDMFT for describing BoseEinstein condensates, we demonstrate three conceptually different approaches for deriving the universal functional in a homogeneous Bose gas for arbitrary pair interaction in the Bogoliubov regime. Remarkably, in both systems the gradient of the functional is found to diverge repulsively at the boundary of the functional's domain, extending the recently discovered BoseEinstein condensation force to excited states. Our findings highlight the physical relevance of the generalized exclusion principle for fermionic and bosonic mixed states and the curse of universality in functional theories.
Current status:
Reports on this Submission
Anonymous Report 2 on 202312 (Invited Report)
 Cite as: Anonymous, Report on arXiv:2210.00964v1, delivered 20230102, doi: 10.21468/SciPost.Report.6426
Strengths
(1) Derivation of universal excitedstate 1RDM functional
(2) Application to two systems of bosons
(3) Illustration of the universal divergence of the functionals
(4) Review of the wensemble 1RDM theory
Report
The manuscript "Deriving densitymatrix functionals for excited states" by Julia Liebert and Christian Schilling derives the universal excitedstate 1RDM functionals explicits for two bosonic systems. The functional forms reveal the divergence of the functionals at the domain boundaries. The work is elegantly presented with a detailed discussion of the findings.
Requested changes
(1) Explain in a sentence or two why the functionals are not convex before closure. (2) Verify the third line of Eq. (3) for correct factors of N given the normalization of the 1RDM to N.
Anonymous Report 1 on 20221212 (Invited Report)
 Cite as: Anonymous, Report on arXiv:2210.00964v1, delivered 20221212, doi: 10.21468/SciPost.Report.6295
Report
The paper presents an explicit expression for the universal functionals in wensemble RDMFT in the bosonic Hubbard dimer for two particles with momentum 0 and for the homogenous BEC. The latter was derived in three different manners. They all lead to the same result. The BEC force, already obtained in previous works, was generalized to excited states. This is very interesting work indicating how such functionals could be developed. The work is therefore recommended for publication, but the current manuscript would benefit a lot from an additional round of improvements.
Typos
• Two lines after Eq. (7) "represents" should be changes to "represent".
• In Fig. 1 the legend has $\omega_i$ instead of $w_i$.
• Two lines above Eq. (45) it should be "spanned by the"
Consistency
• The symbol $\equiv$ should also be used in Eq. (4) and (7). Also in Eq. (5) the first equality should probably be \equiv instead of the second one.
• Sometimes the symbol $\text{Tr}_N$ is used (Eq. (3)) and sometime Tr is used (Eq. (2)). It would be better to stick to one.
• Either keep $\hat{\gamma}_{\hat{h}}$ for vrepresentable 1RDMs or mention that you drop the subscript and that all 1RDMs need to be vrepresentable. (For example see beginning of Section IV E 2.)
Readability
• The word "dispersion" is confusing and should be removed.
• Above Eq. (2) $\mathcal{E}^N(\mathbf{w})$ should be defined, at least in words.
• For people outside the GOKfield “$r=2$” might not be clear. Please define it, e.g. between brackets.
• In Eq. (35) $\mathcal{S}^d$ is not defined (both $\mathcal{S}$ and $d$).
• The abbreviation GOK should be explained.
• For completeness, add that the weights sum up to one.
Questions/Comments
• In Eq. (A15) why do you not divide out a factor 2 in the fraction?
• All energies from the Hubbard dimer seem to be missing a factor 1/2. For example above Eq. (13) your expression is Tr$_1[\hat{t}\hat{\gamma}]=4t(n_01)$. I would expect $2t(n_01)$. Maybe there is a normalization factor missing.
• Eq. (43) also has an additional factor 1/2 compared to Eq. (25) and Eq. (26). Please check.
• The paragraph after Eq. (10) is a bit lengthy discussion for the fact that the single translation element in the periodic setting is the same operation as the inversion operation. So they are equivalent. Perhaps you could put this when introducing the dimer. Inversion is also a more natural description of the symmetry element than translation + periodicity, so I would rather liked to have seen g(erade) and u(ngerade) designations.
• I got confused about the precise definition of your creation and annihilation operator. Do they keep the standard normalization of the states? And around Eq. (20) is the operator $\beta_0$ not only needed to compensate for the inconvenience of the standard normalization of the states.
• I would like to see the derivation of Eq. (24) from Eq. (19). Maybe in the appendix.
• Section IV C seems a bit off topic at this point in the paper, since it does not seem to be relevant for the functional for the Bogloiubov approximated homogeneous BEC. Perhaps it is better discussed in section V. Further, the state crossing is a well known issue in functional theories e.g. spinDFT [von Barth and Hedin J Phys C 5, 1629 (1972), Eschrig “The Fundamentals of Density Functional Theory (revised and extended version)” (2003), DFT on graphs [Leeuwen Penz, JCP 155, 244111 (2021)], RDMFT [Giesbertz JCP 143, 054102 (2015)]. The nonanalyticity related a lot to the possible kinks in the adiabatic connection mostly discussed in the context of DFT. It would be good to also make connections to that.
• I am not sure about Eq. (41). Eq. (39) and (40) is only for vrepresentable 1RDMs and the minimum over convex functionals is not necessarily convex. I think you also need Eq. (37) or something more.
• Why has Eq. (45) creation operators and above, the spanning orthonormal states have annihilation operators? That does not seem to be compatible.
• How did you get Eq. (47)?
• After Eq. (50) it is clear what is intended with the statement about $F_{w,q’}$ but the sentence is illogical. It should relate to the biconjugate of $F_{w,q’}$. Perhaps you could just say that you got effectively the same result.
• In Eq. (55) what is the q' referring to?
• After Eq. (55) what do you mean by "gradient force is collectively diverging? Do you mean divergence of all components?
• In Section V, 4th line, $w=0$ is also finite. You probably mean nonzero.
• It would be good to explicitly mention in the title that your work is focused on bosonic systems.
• How general do you expect the BEC force to be?
• It seems that the construction also works for attractive interactions. Is this correct?
Author: Christian Schilling on 20230122 [id 3257]
(in reply to Report 1 on 20221212)
We thank the Referee for this positive assessment of our work and the constructive feedback which helped us to further improve the manuscript.
Typos $\cdot$ Two lines after Eq. (7) "represents" should be changes to "represent". $\cdot$ In Fig. 1 the legend has $\omega_i$ instead of $w_i$ $\cdot$ Two lines above Eq. (45) it should be "spanned by the"
We thank the Referee for pointing out these three typos which we corrected.
Consistency The symbol $\equiv$ should also be used in Eq. (4) and (7). Also in Eq. (5) the first equality should probably be $\equiv$ instead of the second one.
We agree that in old Eqs. (4) and (7)/new Eqs. (5) and (8) ''$\equiv$'' should be used as suggested by the referee. The set of 1RDMs $\overline{\mathcal{E}}^1_N(\boldsymbol{w})$ if defined via the partial trace since it contains all those 1RDMs that are compatible with an $N$particle state in $\overline{\mathcal{E}}^N(\boldsymbol{w})$. Based on that, $\overline{\mathcal{E}}^1_N(\boldsymbol{w})= \mathrm{conv}(\mathcal{E}^1_N(\boldsymbol{w}))$ is a mathematical consequence of the definition $\overline{\mathcal{E}}^1_N(\boldsymbol{w})\equiv N\mathrm{Tr}_{N1}(\overline{\mathcal{E}}^N(\boldsymbol{w}))$. To clarify this we interchanged the second and third expression in old Eq. (5)/new Eq. (6).
Sometimes the symbol $\mathrm{Tr}_N$ is used (Eq. (3)) and sometime $\mathrm{Tr}$ is used (Eq. (2)). It would be better to stick to one.
We thank the referee for this comment and stick to $\mathrm{Tr}_N$ to denote the trace over $N$particles. We therefore added the subscript $N$ in the trace in the paragraph below Eq. (1) and in old Eq. (2)/new Eq. (3).
Either keep $\hat{\gamma}_{\hat{h}}$ for vrepresentable 1RDMs or mention that you drop the subscript and that all 1RDMs need to be vrepresentable. (For example see beginning of Section IV E 2.)
The set of $v$representable 1RDMs consists of all those $\hat{\gamma}$ for which there exists some oneparticle Hamiltonian $\hat{h}$ such that $\gamma$ can be obtained via the sequence of maps in old Eq. (29)/new Eq. (30). Thus, given the sequence in old Eq. (29)/new Eq. (30) we can assign a $\hat{h}$ to $\gamma$ and explicitly write $\hat{\gamma}_{\hat{h}}$. However, given the set of $v$representable 1RDMs it is a priori not known which $\hat{h}$ would correspond to which $\hat{\gamma}$ in this set. In that sense it is not meaningful to label all $v$representable 1RDMs by $\hat{\gamma}$ by a subscript $\hat{h}$ except for the cases where a fixed $\hat{h}$ is given from the very beginning.
Readability The word "dispersion" is confusing and should be removed.
In the context of optics, a dispersion relation relates the frequency (energy) of a wave to the wave number (or momentum). It is therefore also common in condensed matter physics to talk about a ''dispersion law'' when the energy of a particle and its momentum are related. For instance, $\varepsilon(\boldsymbol{p})=\boldsymbol{p}^2/2m$ is usually referred to as ''free dispersion''. In the same way it is common to refer to a ''dispersion relation'' in the context of the elementary excitations of Bogoliubov quasiparticles (e.g. see textbook [L. Pitaevskii, S. Stringari, BoseEinstein Condensation and Superfluidity, Oxford University Press (2016)]). To follow standard terminology we therefore feel that it is appropriate to refer to a ''dispersion relation'' for the Bogoliubov excitation spectrum before old Eq. (28)/new Eq. (29) and only removed the term in the context of $t_{\boldsymbol{p}}$.
Above Eq. (2) $\mathcal{E}^N(\boldsymbol{w})$ should be defined, at least in words.
The set $\mathcal{E}^N(\boldsymbol{w})$ is the subset of $N$particle density operators that satisfy the spectral constraint $\mathrm{spec}^\downarrow(\hat{\Gamma})=\boldsymbol{w}$. To clarify its definition and make it more explicit we added the new Eq. (2) in the manuscript.
For people outside the GOKfield ''$r=2$'' might not be clear. Please define it, e.g. between brackets.
We explain the definition of ''$r=2$'' in the first paragraph of Sec. IV.E. and the corresponding weight vector $\boldsymbol{w}=(w, 1w, 0, ..)$ is given by the old Eq. (32)/new Eq. (33). To clarify this aspect, we added this information in the first paragraph of Sec. IV.E and added the term ''nonvanishing weights'' also after old Eq. (36)/new Eq. (37).
In Eq. (35) $\mathcal{S}_d$ is not defined (both $\mathcal{S}$ and $d$).
The group $\mathcal{S}_d$ denotes the permutation group of a set of $d$ elements. In our case $d$ is the dimension of the oneparticle Hilbert space $\mathcal{H}_1$, i.e. $d=\mathrm{dim}(\mathcal{H}_1)$. We thank the referee for this comment and added the definition of $\mathcal{S}_d$ and $d$ accordingly.
The abbreviation GOK should be explained.
We added the explanation of the abbreviation ''GOK'' in the sentence below Eq. (1).
For completeness, add that the weights sum up to one.
We added this information in the paragraph above the new Eq. (2).
Questions/Comments In Eq. (A15) why do you not divide out a factor 2 in the fraction? All energies from the Hubbard dimer seem to be missing a factor $1/2$. For example above Eq. (13) your expression is $\mathrm{Tr}_1[\hat{t}\hat{\gamma}]=4t(n_01)$. I would expect $2t(n_01)$. Maybe there is a normalization factor missing.
We thank the referee for this comment. Indeed, the Fourier transform of the kinetic energy reads
$$\hat{t} = t \sum_{\nu=0,1} \mathrm{cos}(\pi\nu)\hat{n}_\nu$$and not
$$\hat{t} = 2t \sum_{\nu=0,1} \mathrm{cos}(\pi\nu)\hat{n}_\nu.$$This leads to $2t(n_01)$ for the kinetic energy. This then also modifies Eq. (A15) and we changed Eqs.~(A13)(A19) accordingly. We also checked the expectation value of
$$\mathrm{Tr}_2[W\Gamma_{\boldsymbol{w}}]$$which is actually correct.
Eq. (43) also has an additional factor 1/2 compared to Eq. (25) and Eq. (26). Please check.
The factor $1/2$ in old Eq. (43)/new Eq. (44) compared to old Eqs. (25) and (26)/new Eqs. (26) and (27) is due to the introduction of the new variable $\eta_{\boldsymbol{p}^\prime} = n_{\boldsymbol{p}^\prime} + n_{\boldsymbol{p}^\prime}$ in old Eq. (17)/new Eq.(18). The sum over $\boldsymbol{p}^\prime$ contains each pair $(\boldsymbol{p}, \boldsymbol{p}), \boldsymbol{p} \neq \boldsymbol{0}$ only once, whereas in the old Eq. (25)/new Eq. (26) the summation is still performed over $\boldsymbol{p}$. To remind the reader of the definition of $\boldsymbol{p}^\prime$ we recall it below the old Eq. (43)/new Eq. (44).
The paragraph after Eq. (10) is a bit lengthy discussion for the fact that the single translation element in the periodic setting is the same operation as the inversion operation. So they are equivalent. Perhaps you could put this when introducing the dimer. Inversion is also a more natural description of the symmetry element than translation + periodicity, so I would rather liked to have seen g(erade) and u(ngerade) designations.
Since both systems the symmetric BoseHubbard dimer in Sec. III and the homogeneous BEC in Sec. IV are translational invariant we would like to use Sec. III to already prepare the reader for the more technical Sec. IV. We therefore feel that it is clearer to use momentum occupation numbers in both cases instead of switching to $n_e, n_u$ for the BoseHubbard dimer. To improve the readability we followed the suggestion of the referee and moved the paragraph concerning the inversion symmetry to the beginning of Sec. III before we derive the universal functional.
I got confused about the precise definition of your creation and annihilation operator. Do they keep the standard normalization of the states? And around Eq. (20) is the operator $\beta_0$ not only needed to compensate for the inconvenience of the standard normalization of the states.
The operator $\hat{\beta}_0$ in the unitary $\hat{U}$ in old Eq. (20)/new Eq. (21) ensures that the Hamiltonian remains particlenumber conserving and that the Bogoliubov quasiparticle vacuum is a state in the $N$boson Hilbert space. This is both not the case for a standard Bogoliubov transformation violating particle number conservation. Therefore, the purpose of the $\hat{\beta}_0, \hat{\beta}_0^\dagger$ operators is to ensure particle number conservation. We would like to stress here that particle number conservation is absolutely vital in the context of the LevyLieb constrained search. Indeed, the universal functional is defined via the minimization over all $N$particle states mapping to a given 1RDM with fixed normalization $\mathrm{Tr}_1[\hat{\gamma}]=N$. Thus, if we restrict the set of all $N$particle states (or $\overline{\mathcal{E}}^N(\boldsymbol{w})$) in the LevyLieb constrained search to a smaller variational manifold of states in Sec. IV, all states in this variational manifold are $N$particle states by construction which is ensured by a particlenumber conserving Bogoliubov theory. To support the reader we added this information below old Eq. (20)/new Eq. (21).
I would like to see the derivation of Eq. (24) from Eq. (19). Maybe in the appendix.
To keep our paper selfcontained we recap in Sec. IV.A the derivation of the grounds state Bogoliubov approximated universal functional discussed in more detail in our previous paper Ref. [3]. Nevertheless, the derivation of old Eq. (24)/new Eq. (25) is already complete up to some straightforward analytical calculations. We start with the expectation value
$$\langle N\hat{U}^\dagger\hat{W}_{\mathrm{B}}\hat{U}N\rangle$$as explained below old Eq. (22)/new Eq. (23) and insert identities $\hat{U}^\dagger \hat{U}$ such that each bosonic creation/annihilation operator can be transformed individually using old Eq. (21)/new Eq. (22). This calculation can be also found in [M. Girardeau, R. Arnowitt, Theory of ManyBoson Systems: Pair Theory, Phys. Rev. \textbf{113}, 755 (1959)], where the particle number conserving Bogoliubov theory used in Sec. IV.A was introduced. The second step to derive
$$\mathcal{F}_{\boldsymbol{w}_0}$$in old Eq. (24)/new Eq. (25) is to replace the variational parameters $\phi_{\boldsymbol{p}^\prime}$ by $\eta_{\boldsymbol{p}^\prime}$ using old Eq. (23)/new Eq. (24). To explain this in more detail and to support the reader we extended the paragraph below old Eq. (22)/new Eq. (23) accordingly.
Section IV C seems a bit off topic at this point in the paper, since it does not seem to be relevant for the functional for the Bogloiubov approximated homogeneous BEC. Perhaps it is better discussed in section V. Further, the state crossing is a well known issue in functional theories e.g. spinDFT [von Barth and Hedin J Phys C 5, 1629 (1972), Eschrig “The Fundamentals of Density Functional Theory (revised and extended version)” (2003), DFT on graphs [Leeuwen Penz, JCP 155, 244111 (2021)], RDMFT [Giesbertz JCP 143, 054102 (2015)]. The nonanalyticity related a lot to the possible kinks in the adiabatic connection mostly discussed in the context of DFT. It would be good to also make connections to that.
The complexity based on state crossings as discussed in Sec. IV. C is indeed relevant for the derivation of excited state functionals in $\boldsymbol{w}$RDMFT in general and thus also for the Bogoliubov approximated functional. We therefore feel that it is better to discuss this before the derivation of the $\boldsymbol{w}$ensemble universal functional and also refer back to Sec. IV.C in Sec. IV.E.1.
We thank the referee for providing these interesting references. It is in fact known that crossings of the ground state and first excited state lead to several conceptual issues in both DFT and RDMFT. A problem, which will be even more present in $\boldsymbol{w}$RDMFT (or GOKDFT) for $\boldsymbol{w}\neq (1, 0, ...)$. Nevertheless, the references do not discuss the effect of the level crossings on the analyticity of the functional and the splitting of the functional's domain into different cells. To provide the reader with a broader perspective on the consequences of level crossings in functional theories we added in Sec. IV.C a comment with the references suggested by the referee.
I am not sure about Eq. (41). Eq. (39) and (40) is only for vrepresentable 1RDMs and the minimum over convex functionals is not necessarily convex. I think you also need Eq. (37) or something more.
By referring to the LevyLieb constrained search, the universal functional in ground state RDMFT is defined for all 1RDMs and not only pure state $v$representable 1RDMs as in Gilbert's formulation of RDMFT. The analogous reasoning applies in $\boldsymbol{w}$ensemble RDMFT. Thus, also the maximization in the LegendreFenchel transformation in old Eq. (37)/new Eq. (38) and the minimization in old Eq. (40)/new Eq. (41) over all $\hat{\gamma}\in\overline{\mathcal{E}}^1_N(\boldsymbol{w})$ and not only over $\boldsymbol{w}$ensemble $v$representable 1RDMs. This is crucial since we usually do not know a priori which 1RDMs are $\boldsymbol{w}$ensemble $v$representable. The idea in Sec. IV.E.1 is to use the knowledge of the energy $E_{\boldsymbol{w}}$ to derive the universal functional $\overline{\mathcal{F}}_{\boldsymbol{w}}$ via a LegendreFenchel transformation in old Eq. (37)/new Eq. (38) and the LegendreFenchel transform of any function is always convex since the supremum (maximum) over a family of affine functions is convex. In particular, the universal functional can be calculated via the LegendreFenchel transform for all $\hat{\gamma}\in\overline{\mathcal{E}}^1_N(\boldsymbol{w}) $ and not only $\boldsymbol{w}$ensemble $v$representable 1RDMs as explained in Sec. IV.D.
Why has Eq. (45) creation operators and above, the spanning orthonormal states have annihilation operators? That does not seem to be compatible.
We thank the referee for pointing out that typo which we corrected. The orthonormal states should of course also have creation operators since the quasiparticle operators $\hat{c}_{\boldsymbol{q}}$ annihilate the quasiparticle vacuum.
How did you get Eq. (47)?
Old Eq. (47)/new Eq. (48) is derived in the same way as the expectation value
$$\langle N\hat{U}^\dagger\hat{W}_\mathrm{B}\hat{U}N\rangle$$is expressed in terms of the variational parameters ${\phi_{\boldsymbol{p}^\prime}}$ in Sec. IV.A and reduces to it for $w=1$, i.e. in the ground state case. To clarify this in the manuscript we added an explanation below old Eq. (47)/new Eq. (48).
After Eq. (50) it is clear what is intended with the statement about $\mathcal{F}_{w, q^\prime}$ but the sentence is illogical. It should relate to the biconjugate thereof. Perhaps you could just say that you got effectively the same result.
The universal functional and its biconjugate coincide for all $\boldsymbol{w}$ensemble $v$representable 1RDMs. Therefore, the same holds for $\mathcal{F}_{\boldsymbol{w},\boldsymbol{q}^\prime}$, where $\boldsymbol{q}^\prime$ denotes the momentum corresponding to the first excitation.
In Eq. (55) what is the $\boldsymbol{q}^\prime$ referring to?
The momentum $\boldsymbol{q}^\prime$ is defined as the momentum of the lowestlying quasiparticle excitation on top of the quasiparticle vacuum as defined in old Eq. (45)/new Eq. (46). To remind the reader about this definition we added a respective explanation below old Eq. (55)/new Eq. (56).
After Eq. (55) what do you mean by "gradient force is collectively diverging? Do you mean divergence of all components?
Yes, the gradient force is collective in the sense that all individual components for each momentum $\boldsymbol{p}^\prime$ diverge. We agree that this aspect should be clarified and added this information in the paragraph below old Eq. (55)/new Eq. (56).
In Section V, 4th line, $w=0$ is also finite. You probably mean nonzero.
The referee is right and we corrected the corresponding sentence accordingly.
It would be good to explicitly mention in the title that your work is focused on bosonic systems.
It is worth noticing that the functional for the Hubbard dimer in Sec. III will be the same for the FermiHubbard dimer restricted to the singlet subspace. This is explained in the first paragraph of Sec. III. Moreover, the concepts behind the three approaches to develop universal functional in Sec. IV are equally applicable to fermionic systems. For instance, both Bogoliubov theory for bosons and BCS theory for femions have a similar foundation with respect to the corresponding set of quasifree states (e.g. see [J. Manuceau, A. Verbeure, Commun. Math. Phys.$\mathbf{9}$, 293–302 (1968)], [V. Bach, E.H. Lieb, J.P. Solovej, J. Stat. Phys. $\mathbf{76}$, 3–89 (1994)]). In addition, also the conceptual aspects discussed in Secs. IV.C and IV.D apply to fermions as well as bosons as mentioned in the last sentence before Sec. IV.C. Therefore, despite the focus on bosonic systems, the scope of this paper is not exclusively narrowed to bosons. Thus, we feel that it is indeed appropriate to not specify this further in the title. To further emphasise this point in the paper we added two respective comments in the first paragraph of Sec. IV and at the end of the second paragraph in Sec. V.
How general do you expect the BEC force to be?
This is indeed a very interesting question which still requires further investigations. We expect that for nonzero interactions a universal functional exhibits this diverging gradient at the boundary of its domain. Thus, the BEC force merely originates from the geometry of a certain class of quantum states. However, we would also like to stress that a physical system will only be affected by the BEC force if the minimizer 1RDM is close enough to the boundary of the functional's domain, e.g. when the system is close to a complete BEC. For a general Hamiltonian, in particular for inhomogeneous systems, this regime is rather narrow.
It seems that the construction also works for attractive interactions. Is this correct?
Yes, the same approach works also for universal functional in the case of attractive interactions. In fact, the minimization over the phases $\sigma_{\boldsymbol{p}^\prime}$ in old Eq. (54)/new Eq. (55) can be executed independent of the sign of the Fourier coefficients $W_{\boldsymbol{p}^\prime}$ leading to $\sigma_{\boldsymbol{p}^\prime} = \mathrm{sign}(W_{\boldsymbol{p}^\prime})$. We discussed this aspect in more detail in our previous work [J. Liebert, C. Schilling, Phys. Rev. Research $\mathbf{3}$, 013282 (2021)]. To comment on this we extended the paragraph below old Eq. (54)/new Eq. (55) accordingly.
Author: Christian Schilling on 20230122 [id 3256]
(in reply to Report 2 on 20230102)We thank the referee for the effort made and the positive assessment of our work.
For simplicity let us first discuss the ground state functional which is contained in $\boldsymbol{w}$RDMFT for $\boldsymbol{w}_0 = (1, 0, ...)$ and comment on general $\boldsymbol{w}$ afterwards. Then, the set $\mathcal{E}^N(\boldsymbol{w}_0)$ of all $N$particle pure state is not convex and it is wellknown that the pure universal functional
$$\mathcal{F}_{\boldsymbol{w}_0}(\hat{\gamma}) \equiv \min_{\mathcal{E}^N(\boldsymbol{w}_0)\ni\hat{\Gamma}\mapsto\hat{\gamma}}\mathrm{Tr}_N[\hat{\Gamma} \hat{W}]$$is in general not convex. A famous example illustrating this is the BoseHubbard dimer discussed in [C.L. BenavidesRiveros, J. Wolff, M.A.L. Marques, C. Schilling, Phys. Rev. Lett. $\mathbf{124}$, 180603 (2020)]. The same holds for the the fermionic Hubbard dimer restricted to the singlet subspace. Furthermore, we proved in [J. Liebert, F. Castillo, J.P. Labbé, C. Schilling, J. Chem. Theory Comput. $\mathbf{18}$, 124–140 (2021)] that an exact convex relaxation where the nonconvex set $\mathcal{E}^N(\boldsymbol{w})$ is replaced by $\overline{\mathcal{E}}^N(\boldsymbol{w})$ yields
$$\overline{\mathcal{F}}_{\boldsymbol{w}}=\mathrm{conv}(\mathcal{F}_{\boldsymbol{w}}).$$Thus, the universal functional is indeed convex after applying an exact convex relaxation. To support the reader we added a corresponding sentence below old Eq. (2)/new Eq. (3).
The 1RDM is defined as $\hat{\gamma} \equiv N\mathrm{Tr}_{N1}[\hat{\Gamma}]$, where $\hat{\Gamma}$ is a $N$particle density operator with normalization $\mathrm{Tr}_N[\hat{\Gamma}]=1$. Due to the normalization $\mathrm{Tr}_1[\hat{\gamma}]=N$ there is no additional factor $N$ needed in the reduction
$$\mathrm{Tr}_N[\hat{h}\hat{\Gamma}]=\mathrm{Tr}_1[\hat{h}\hat{\gamma}]$$when going from the $N$particle to the oneparticle level. Note that a mathematically rigorous treatment would require to use different symbols for the oneparticle Hamiltonian
$$\hat{h}:\mathcal{H}_N\to \mathcal{H}_N$$and its restriction to the oneparticle Hilbert space $\hat{h}_{\mathcal{H}_1}$. Since this is rather natural in the formalism of second quantization we use in the manuscript the same symbol $\hat{h}$ for simplicity. To clarify this in the manuscript we added this information below old Eq. (2)/new Eq. (3). In first quantization
$$\hat{h} = \hat{h}_1\otimes \mathbb{\hat{1}}^{N1}+ ...+\mathbb{\hat{1}}^{N1}\otimes\hat{h}_1,$$where for illustrative purposes we denote here the one particle Hamiltonian on $\mathcal{H}_1$ as $\hat{h}_1$. Then, we explicitly see that
$$\mathrm{Tr}_N[\hat{h}\hat{\Gamma}] = \mathrm{Tr}_1[N\hat{h}_1\mathrm{Tr}_{N1}[\hat{\Gamma}]]= \mathrm{Tr}_1[\hat{h}_1\hat{\gamma}]$$and thus the factors of $N$ are correct.