The Free Electron Gas in Cavity Quantum Electrodynamics

Vasil Rokaj,1, ∗ Michael Ruggenthaler,1, † Florian G. Eich,1 and Angel Rubio1, 2, ‡ 1Max Planck Institute for the Structure and Dynamics of Matter, Center for Free Electron Laser Science, 22761 Hamburg, Germany 2Center for Computational Quantum Physics (CCQ), Flatiron Institute, 162 Fifth Avenue, New York NY 10010 (Dated: May 31, 2021) Cavity modification of material properties and phenomena is a novel research field largely mo- tivated by the advances in strong light-matter interactions. Despite this progress, exact solutions for extended systems strongly coupled to the photon field are not available, and both theory and experiments rely mainly on finite-system models. Therefore a paradigmatic example of an exactly solvable extended system in a cavity becomes highly desireable. To fill this gap we revisit Som- merfeld’s theory of the free electron gas in cavity quantum electrodynamics (QED). We solve this system analytically in the long-wavelength limit for an arbitrary number of non-interacting elec- trons, and we demonstrate that the electron-photon ground state is a Fermi liquid which contains virtual photons. In contrast to models of finite systems, no ground state exists if the diamagentic A2 term is omitted. Further, by performing linear response we show that the cavity field induces plasmon-polariton excitations and modifies the optical and the DC conductivity of the electron gas. Our exact solution allows us to consider the thermodynamic limit for both electrons and photons by constructing an effective quantum field theory. The continuum of modes leads to a many-body renormalization of the electron mass, which modifies the fermionic quasiparticle excitations of the Fermi liquid and the Wigner-Seitz radius of the interacting electron gas. Lastly, we show how the matter-modified photon field leads to a repulsive Casimir force and how the continuum of modes introduces dissipation into the light-matter system. Several of the presented findings should be experimentally accessible.

I. INTRODUCTION light and matter has led to a number of great funda- mental discoveries, like the laser cooling [15–17], the first The free electron gas introduced by Sommerfeld in realization of Bose-Einsten condensation in dilute gases 1928 [1] is a paradigmatic model for solid state and con- and the atom laser [18, 19], the theory of optical coher- densed matter physics. It was originally developed for ence [20] and laser-based precision spectroscopy [21, 22], the description of thermal and conduction properties of and the manipulation of individual quantum systems metals, and has served since then as one of the funda- with photons [23, 24]. mental models for understanding and describing mate- In most cases simplifications of QED are employed for rials. The free electron gas with the inclusion of the the practical use of the theory (due to its complexity) in electron-electron interactions, was transformed into the which matter is described by a few states. This leads to homogeneous electron gas [2, 3], known also as the jellium the well-known models of quantum optics, like the Rabi, model, and with the advent of density functional theory Jaynes-Cummings or Dicke models [25–27]. Although, (DFT) and the local density approximation (LDA) [4] has these models have served well and have been proven very become one of the most useful computational tools and succesful [28], recently they are being challenged by novel methods in physics, chemistry and materials science [5]. developments in the field of cavity QED materials [12]. Also within the Fermi liquid theory, developed by Lan- For this, first-principle approaches have already been put dau [6], the free electron gas model was used as the funda- forward using Green’s functions methods [29], the ex- mental building block [7]. In addition, the free electron act density-functional reformulation of QED, known as arXiv:2006.09236v4 [quant-ph] 27 May 2021 gas in the presence of strong magnetic fields has also QEDFT [30–32], hybrid-orbital approaches [33, 34], or been proven extremely important for the description of generalized coupled cluster theory for electron-photon the quantum Hall effect [8, 9]. systems [35, 36]. On the other hand, the cornerstone of the modern de- Cavity QED materials [12, 37, 38] is an emerging scription of the interaction between light and matter, in field, combining many different platforms for manipulat- which both constituents are treated on equal quantum ing and engineering quantum materials with electromag- mechanical footing, and both enter as dynamical entities, netic fields, ranging from quantum optics [14], polari- is quantum electrodynamics [10–14]. This description of tonic chemistry [12, 39–47], and light-induced states of matter using either classical fields [48, 49] or quantum fields originating from a cavity [50–53]. A plethora of pathways have been explored recently from both theo- ∗ [email protected] rists and experimenters. Quantum Hall systems under † [email protected] cavity confinement, in both the integer [54–58] and the ‡ [email protected] fractional [59, 60] regime, have demonstrated ultrastrong 2 coupling to the light field and modifications of trans- Performing linear response [3, 91, 92] for the interact- port [61]. Light-matter interactions have been suggested ing electron-photon system in the cavity, we compute the to modify electron-phonon coupling and superconductiv- optical conductivity σ(w) in which we identify diamag- ity [62–65] with the first experimental evidence already netic modifications to the standard conductivity of the having appeared [66]. Cavity control of excitons has been free electrons gas, coming from the cavity field. Fur- investigated [67–69] and exciton-polariton condensation ther, in the static limit we find that the cavity field sup- has been achieved [70, 71]. Further, the implications presses the DC conductivity and the Drude peak of the of coupling to chiral electromagnetic fields has also at- 2DEG. This shows that a cavity can alter the conduc- tracted interest and is currently investigated [72–75]. tion properties of 2D materials as suggested also experi- Much of our understanding and theoretical descrip- mentally [61, 66]. Our linear response formalism demon- tion of light-matter interactions and of these novel ex- strates that plasmon-polariton resonances exist for this periments, is based on finite-system models from quan- interacting electron-photon system [93, 94] and provides tum optics. However, extended systems like solids be- a microscopic quantum electrodynamical description of have very much differently than finite systems and it plasmon-polaritons. is questionable whether the finite-system models can be To overcome the discrepancy between the electronic straightforwardly extended to describe macroscopic sys- sector, in which the energy density of the electrons is tems, like materials, strongly coupled to a cavity. It is finite, and the photonic sector, whose energy density in therefore highly desirable, in analogy to the Rabi and the the thermodynamic limit vanishes, we promote the single Dicke model [25–27], to have a paradigmatic example of mode theory into an effective quantum field theory in the an extended system strongly coupled to the quantized 2D continuum by integrating over the in-plane modes of cavity field. the photon field. The area of integration in the photonic The aim of this work is to fill this gap, by revisiting momentum space is directly connected to the effective Sommerfeld’s theory [1] of the free electron gas in the cavity volume and the upper cutoff in the photon mo- framework of QED and providing a new paradigm for menta defines the effective coupling of the theory. More- many-body physics in the emerging field of cavity QED over, in the effective field theory the energy density of the materials. photon field becomes finite and renormalizes the electron In this article we introduce and study in full general- mass [95–97]. The renormalized mass depends on the full ity the 2D free electron gas (2DEG) coupled to a cav- electron density in the cavity which means that we have ity. We show that this system in the long-wavelength a many-body contribution to the renormalized mass due limit and for a finite amount of cavity modes is ana- to the collective coupling of the electrons to the cavity lytically solvable and we find the full set of eigenstates field. In addition, the renormalized electron mass shows and the eigenspectrum of the system. Specializing to the up in the expression for the chemical potential and mod- paradigmatic case of just one effective mode (with both ifies the fermionic quasiparticle excitations of the Fermi polarizations included) we highlight that in the large N liquid. Upon the inclusion of the Coulomb interaction, or thermodynamic limit the ground state of the elec- the mass renormalization leads also to a shrinking of the trons is a Slater determinant of plane waves with the Wigner-Seitz radius, which implies a localization effect momenta of the electrons distributed on the 2D Fermi for the electrons. From the energy density of the photon sphere, thus it is a Fermi liquid. On the other hand, the field in the cavity we compute the corresponding Casimir photon field gets strongly renormalized by the full elec- force [98, 99] (pressure) and we find that due to the in- tron density and the combined light-matter ground state teraction of the cavity field with the 2DEG, the Casimir exhibits quantum fluctuation effects and contains virtual force is repulsive [100]. Furthermore, we are able to de- photons. Moreover, we study the full phase diagram of scribe consistently and from first principles dissipation the system (see Fig. 4) and we find that when the cou- and absorption processes without the need of any artifi- pling approaches its maximum value (critical coupling) cial damping parameter [3, 92]. a critical situation appears with the ground state being Outline of the Paper—In section II we introduce the infinitely degenerate. Above the critical coupling (which 2DEG in cavity QED and we solve the system exactly. in principle is forbidden) the system is unstable and has In section III we find the ground state of the system in no ground state. The lack of a ground state shows up the large N (or thermodynamic) limit. In section IV we also when the diamagnetic A2 term is neglected in the provide the phase diagram of the system for any value of Hamiltonian. This is in stark contrast to the standard the coupling constant and we discuss under which condi- quantum optics models, like the Rabi or the Dicke model, tions the system is stable and has a ground state. In sec- which have a ground state even without the diamagnetic tion V we perform linear response, we introduce the four A2 term. This highlights that the A2 term is necessary fundamental responses (matter-matter, photon-photon, for the stability of extended systems like the 2DEG. This photon-matter and matter-photon) and we compute the result we believe sheds light on the ongoing discussion optical and the DC conductivity of the 2DEG in the cav- about whether the A2 term can be eliminated or not [76– ity. In section VI out of the single mode theory we con- 79] which is related to the existence of the superradiant struct an effective quantum field theory in the contin- phase transition [80–90]. uum. Finally, in section VII we conclude and highlight 3 the experimental implications of this work and give an overview of the future perspectives.

II. ELECTRON GAS IN CAVITY QED

Our starting point is the Pauli-Fierz Hamiltonian which describes slowly moving electrons in the non- relativistic limit, minimally coupled to the photon field [11, 14, 101]

N 2 N 2 ˆ 1 X  ˆ  1 X e H = i~∇j + eA(rj) + 2me 4π0 |rj − rk| j=1 jfree electron model introduced by Som- in QED, the assumption of spatially non-varying fields it merfeld [1]. Since we restrict our considerations in two is necessary because otherwise homogeneity and transla- dimensions1, the momentum operator has only two com- tional invariance would not be respected. These assump- ponents ∇ = (∂x, ∂y). We thus assume the 2DEG re- tions are fundamental for the electrons in the Sommer- stricted on the (x, y) plane, in which the system is con- feld model [1, 2] and it is necessary to enforce them to sidered macroscopic. Then the electrons can be described the photon field as well. We note that the description of with the use of periodic boundary conditions, as in the solids in QED beyond the dipole approximation remains original Sommerfeld model [1]. We would like to mention an open research question. that for macroscopic systems the choice of the boundary As a starting point, we consider the case where the conditions does not affect the bulk properties [102]. electromagnetic field consists of a single mode of fre- For the mode functions Sκ,λ(r) to satisfy the bound- quency ω but with both polarization vectors ελ kept. ary conditions of the cavity, the momenta of the Although, as shown in appendix E, we can solve this photon field take the values κ = (κx, κy, κz) = problem even for arbitrarily many discrete modes ana- 3 (2πnx/L, 2πny/L, πnz/Lz) with n = (nx, ny, nz) ∈ Z . lytically, the one-mode case serves as a stepping stone In the long-wavelength limit [101, 103], which has been to construct an effective quantum field theory that takes into account the continuum of modes. In this way the fact that the cavity is open is also taken into account. In section VI we then show, with the help of the exact 1 We would like to mention that all the derivations we present here analytic solution for the many-mode case, that the pre- do not depend on the choice of dimensions and can be performed sented effective field theory is a good approximation for also in the 3D case. most current experimental situations. 4

The polarization vectors are chosen to be in the (x, y) interpreted as a plasmon-polariton frequency, and as we plane such that the mode to interact with the 2DEG. will show in section V it corresponds to a plasmon- The polarization vectors have to be orthogonal and we polariton excitation (or resonance) of the system. The choose ε = e and ε = e . Under these assumptions ˆ ˆ† 1 x 2 y operators {bλ, bλ} satisfy bosonic commutation relations the Pauli-Fierz Hamiltonian of Eq. (1), after expanding ˆ ˆ† 0 [bλ, b 0 ] = δλ,λ0 for λ, λ = 1, 2. In terms of this new set the covariant kinetic energy, is λ of operators the photonic part Hˆp of our Hamiltonian, is N equal to the sum of two non-interacting harmonic oscil- X  2 ie  Hˆ = − ~ ∇2 + ~Aˆ · ∇ lators 2m j m j j=1 e e 2  1 2 2   ˆ X ˆ† ˆ Ne ˆ 2 X † 1 Hp = ~ωe bλbλ + (9) + A + ~ω aˆλaˆλ + , (3) 2 2me 2 λ=1 λ=1 | {z } ˆ Hˆp and the quantized vector potential A is

1 2 and the quantized vector potential of Eq. (2) is   2 ˆ ~ X ελ ˆ ˆ†  A = √ bλ + bλ . (10) 1 2 0V 2ω   2 λ=1 e ˆ ~ X ελ  †  A = √ aˆλ +a ˆλ . (4) 0V λ=1 2ω From this expression we see that the vector potential Aˆ got renormalized and depends on the dressed frequency In the Hamiltonian of Eq. (3) we have a purely photonic ωe [55]. Substituting back into the Hamiltonian of Eq. (3) part Hˆp which depends only on the annihilation and cre- the expressions for the photonic part Hˆp and the vector ation operators of the photon field {aˆ† , aˆ }. Substituting λ λ potential Aˆ given by Eqs. (9) and (10) respectively, and ˆ the expression for the vector potential A given by Eq. (4) introducing the parameter g and introducing the diamagnetic shift ωp 1   2 s s e~ ~ e2n e2n g = , (11) e 2D me 0V 2ω ωp = = , (5) e me0 me0Lz the Hamiltonian of Eq. (3) looks as the photonic part Hˆp takes the form 2 N 2   ~ X 2 X † 1 2 " # Hˆ = − ∇ + ω ˆb ˆb + .   2 2 j ~e λ λ X † 1 ~ωp  †  2me 2 Hˆ = ω aˆ aˆ + + aˆ +a ˆ . (6) j=1 λ=1 p ~ λ λ 2 4ω λ λ λ=1 2 N X ˆ ˆ†  X + ig bλ + bλ ελ · ∇j. (12) The diamagnetic shift ωp is induced due to the collective λ=1 j=1 coupling of the full electron density ne = N/V to the transversal quantized field [55, 94, 101, 103–105]. This The parameter g in Eq. (11) can be interpreted as p 2 means that ωp = e ne/me0 is the plasma frequency in the single-particle light-matter coupling constant. The the cavity. We note that the electron density ne = N/V Hamiltonian is invariant under translations in the elec- is defined via the 2D electron density of the material tronic configuration space, since it only includes the mo- inside the cavity n2D = N/S and the distance between mentum operator of the electrons. This implies that Hˆ the mirrors of the cavity Lz as ne = n2D/Lz. commutes with the momentum operator ∇,[H,ˆ ∇]=0, The photonic part Hˆp can be brought into diago- and they share eigenfunctions. As we already stated, nal form by introducing a new set of bosonic operators for the electrons we employ periodic boundary condi- ˆ† ˆ tions [1, 2]. Thus, the eigenfunctions of the momentum {bλ, bλ} operator ∇ and the Hamiltonian are plane waves of the ˆ 1 h † i form bλ = √ aˆλ (ω + ω) +a ˆ (ω − ω) (7) 2 ωω e λ e e eikj ·rj † 1 h † i φ (r ) = √ with 1 ≤ j ≤ N. (13) ˆ √ kj j bλ = aˆλ (ωe − ω) +a ˆλ (ωe + ω) . S 2 ωωe x y where the frequency where kj = 2π(nj /L, nj /L) are the momenta of the elec- trons, with n = nx, ny
∈ 2, and S = L2 is the ar- q j j j Z 2 2 eas of the material inside the cavity depicted in Fig. 1. ωe = ω + ωp (8) The wavefunctions of Eq. (13) are the single-particle is a dressed frequency which depends on the cavity fre- eigenfunctions. But the electrons are fermions and the quency ω and the diamagnetic shift (or plasma fre- many-body wavefunction must be antisymmetric under quency) ωp. Thus, the dressed frequency ωe should be exchange of any two electrons. To satisfy the fermionic 5 statistics we use a Slater determinant Φ(r1σ1, .., rN σN ) The operators defined in Eq. (16) also satisfy bosonic built out of the single-particle eigenfunctions of Eq. (13). † 0 commutation relations [ˆc, cˆλ0 ] = δλλ0 for λ, λ = 1, 2. For For convenience we denote this Slater determinant as ˆ  †  the quadratic operator Hλ = ~ωe cˆλcˆλ + 1/2 which is ΦK ≡ Φ(r1σ1, .., rN σN ), (14) of the form of a harmonic oscillator we know that the full P set of eigenstates is given by the expression[106] where K = j kj is the collective momentum of the elec- trons. This makes the notation shorter but also indicates the fact that the ground state and the excited states of the system depend on the distribution of the electrons in k-space and particularly on the collective momentum K. Applying Hˆ of Eq. (12) on the wavefunction ΦK we obtain ( 2     X 1   † ˆ ˆ† ˆ ˆ ˆ† (ˆc )nλ HΦK = ~ωe bλbλ + − g bλ + bλ ελ · K λ 2 |nλ, ελ · Kiλ = √ |0, ελ · Kiλ with nλ ∈ Z, λ = 1, 2 λ=1 nλ! N ) N 2 X X (18) + ~ k2 Φ where K = k . (15) 2m j K j e j=1 j=1 Defining now another set of annihilation and creation † operators {cˆλ, cˆλ}

ˆ gελ · K † ˆ† gελ · K cˆλ = bλ − andc ˆλ = bλ − , (16) ~ωe ~ωe ˆ the operator HΦK given by Eq. (15) simplifies as follows where |0, ελ · Kiλ is the ground state of Hˆλ, which gets annihilated byc ˆλ [106], and the eigenergies of Hˆλ are ( 2    2  X † 1 g 2 ˆ ˆ ~ω(nλ + 1/2). The HΦK given by Eq. (17) in terms of HΦK = ~ωe cˆλcˆλ + − (ελ · K) + e 2 ~ω † ˆ λ=1 e the operators {cˆλ, cˆλ} contains only the sum over Hλ and N ) consequently applying Hˆ Φ on the states Q |n , ε · 2 X K λ λ λ + ~ k2 Φ . (17) Ki we obtain 2m j K λ e j=1

" 2 #  2 " # N  " 2 #   2 2 2k2 ˆ Y X 1 g (ελ · K) X ~ j Y H ΦK |nλ, ελ · Kiλ =  ~ωe nλ + − +  ΦK |nλ, ελ · Kiλ . (19) 2 ω 2me λ=1 λ=1 ~e j=1 λ=1

From the above equation we conclude that the full set the expression of the eigenspectrum we see that there is of eigenstates of the electron-photon hybrid system de- a negative term which is proportional to the square of 2 scribed by the Hamiltonian of Eq. (3) is the collective momentum of the electrons ∼ (ελ · ~K) . This is an all-to-all photon-mediated interaction between 2 N Y X the electrons in which the momentum of each electron Φ |n , ε · Ki with λ = 1, 2 and K = k K λ λ λ j couples to the momenta of all the others. This photon- j=1 λ=1 mediated interaction as we will see in section VI has im- (20) plications for the effective electron mass and the quasi- and its eigenspectrum is. particle excitations of this Fermi liquid. To obtain the expression of Eq. (21) we substituted in 2 "   2 # N 2 2 X 1 γ (ελ · K) X ~ kj Eq. (19), the definition of the the single-particle coupling E = ω n + − ~ + nλ,k ~e λ g given by Eq. (11), and we introduced the parameter γ 2 N 2me 2me λ=1 j=1 2m N g2 ω2 ω2 (21) γ = e = p = p ≤ 1. (22) 2 ω ω2 ω2 + ω2 It is important to mention that the electron-photon ~ ~e e p eigenstates constitute a correlated eigenbasis, because The parameter γ can be viewed as the collective coupling the bosonic eigenstates |nλ, ελ · Kiλ depend on the col- of the electron gas to the cavity mode and depends on lective momentum of the electrons K. Moreover, from the cavity frequency and the full electron density ne via 6 the frequency ωp defined in Eq. (5). This implies that whose origin q is also arbitrary (see Fig. 2) is the more charges in the system the stronger the coupling +∞ +∞ between light and matter in the cavity. Further, we note ZZ ZZZ 2 0 2 0 that the collective coupling parameter γ is dimensionless ΩD = D(k − q)d k = D(k )d k , (23) and most importantly γ has an upper bound and cannot −∞ −∞ be larger than one. As we will see in section IV this up- 0 per bound guarantees the stability of the system. Lastly, where we performed the shift k = k − q. The num- we highlight that also in the case of a multi-mode quan- ber of electrons N we can accommodate in the vol- tized field, with the mode-mode interactions included, ume ΩD is 2 times (due to spin degeneracy) the num- the structure of the many-body spectrum stays the same ber of allowed states. Thus, the 2D electron density is 2 with the one in Eq. (21), but with a different coupling n2D = 2ΩD/(2π) . constant, frequencies and polarizations (due to the mode- The energy of Eq. (21) minimizes for nλ = 0 for both mode interactions) and a sum over all the modes [103]. λ = 1, 2. Thus, the photonic contribution to the ground This is shown in detail in appendix E. state energy is constant Ep = ~ωe and does not influ- ence the electrons in k-space. Then, the ground state energy Egs(≡ E0,k) = Ep + Ek is the sum of the pho- tonic contribution Ep and the part which depends on the electronic momenta Ek, which includes two terms: a positive one, which is the sum over the kinetic energies III. GROUND STATE IN THE LARGE N LIMIT of all the electrons and we denote by T , and a negative one which is minus the square of the collective momen- P tum K = j kj. To find the ground state we need to Having diagonalized the Hamiltonian of Eq. (3) we minimize the energy density Ek/S with respect to the want now to find the ground state of this many-body distribution D(k − q). In the large N,S limit the sums system in the large N limit. For this we need to min- in the expression for the energy density Ek/S turn into imize the energy of the many-body spectrum given by integrals. Thus, the kinetic energy density T/S (with Eq. (21) in the limit where the number of electrons N doubly occupied momenta) is [2] and the area S become arbitrarily large and approach +∞ the thermodynamic limit, but in such a way that the 2D 2 2 ZZ T ~ 2 X 2 ~ 2 2 2 electron density n2D = N/S stays fixed. The electron = lim k = 2 d kD(k − q)k S 2me S→∞ S 2me (2π) density can be defined by the number of allowed states k −∞ in a region of k-space, of volume ΩD with respect to a (24) distribution D in k-space [2]. The number of states in 2 0 the volume ΩD is: #states = ΩDS/(2π) . The volume and after performing the transformation k = k − q we ΩD with respect to an arbitrary distribution D(k − q) obtain

+∞ +∞ +∞ 2 " ZZ ZZ ZZ # T ~ 2 2 0 0 0 2 2 2 0 0 0 2 2 2 0 0 = 2 d k D(k )(k ) +2q 2 d k D(k )k +q 2 d k D(k ) = S 2me (2π) (2π) (2π) −∞ −∞ −∞ | {z } | {z } | {z } tD KD n2D 2 ~ 2
= tD + 2q · KD + q n2D . (25) 2me

The term tD is the kinetic energy of free electrons with large N limit is respect to a distribution centered at zero D(k0) [2]. The  +∞ 2  2 ZZ term KD is the collective momentum of the electrons ελ · K 2 0 2 = d2kD(k − q)ε · k .(26) with respect to D(k ), and q n2D is the kinetic energy S (2π)2 λ  due to the arbitrary origin of the distribution (see Fig. 2). −∞ This last term depends on the 2D density n2D and the Performing the transformation k0 = k − q and multiply- origin q, but not on the shape of the distribution D. ing by the area S we find Let us compute now the negative term appearing in 2 Eq. (21). The square of the collective momentum per (ελ · K) 2 2 = S (ελ · KD + ελ · qn2D) . (27) area (ελ · K/S) (for doubly occupied momenta) in the S Summing the two contributions which we computed in 7

Eqs. (25) and (27) we find the energy density as function of the shape of the distribution D and the origin q

FIG. 2. Schematic depiction of a generic distribution D(k−q) in k-space. The shape D of the distribution as well as its origin q are arbitrary. To find the ground state distribution in k-space one needs to minimize the energy density of the system with respect to both the shape D and the origin q.

2 " 2 # Ek ~ 2 γ X 2 E[D] ≡ = tD + 2q · KD + q n2D − (ελ · KD + ελ · qn2D) . (28) S 2me n2D λ=1

The energy density has to be minimized with respect to the origin of the distribution q = (qx, qy). For that we compute the derivative of the energy density E[D] with respect to q

∂E[D] KD = (1 − γ)(KD + qn2D) = 0 =⇒ q0 = − . ∂q n2D (29)

The optimal origin q0 is independent of the coupling γ, and substituting q0 into Eq. (28) we find 2  2  ~ KD E[D]|q0 = tD − . (30) 2me n2D FIG. 3. Graphic representation of the ground state distribu- The remaining task now is to optimize the energy den- tion of the 2DEG not coupled to a cavity. The ground state sity with respect to the shape D of the distribution. In distribution is the 2D Fermi sphere S(k) (circle) with radius general to perform such a minimization it is not an easy |kF| (Fermi wavevector). For the 2DEG coupled to the cavity task. Thus, to find the optimal k-space distribution we we find that the ground state distribution in k-space is also will use some physical intuition. the 2D Fermi sphere S(k) with radius |kF|. The energy density E[D] (as well as q0) given by Eq. (30) is independent of the coupling constant γ. This indicates that the ground state and the ground state en- a 2D sphere (circle) as shown in Fig. 3. But let us forget ergy in the thermodynamic limit are independent of the for a moment the fact that we know the ground state coupling to the cavity. Driven by this observation let us distribution of the electrons, and let us consider again a compare the energy density in Eq. (30) with the energy generic distribution in k-space D(k−q) as the one shown density of the original free electrons gas [2] without any in Fig. 2. For such a distribution the ground state energy coupling to a cavity mode. density, as we found in Eq. (25), is In the original free electron model the energy of the 2 nc ~ 2
system is the sum over the kinetic energies of all the elec- E [D] = tD + 2q · KD + q n2D . (31) nc P 2 2me trons Ek = j ~ kj/2me [1, 2], and due to rotational symmetry the ground state momentum distribution is Minimizing Enc[D] with respect to the origin q we find the standard Fermi sphere S(k) [2], which in our case is that the optimal origin of the distribution is q0 = 8

−KD/n2D. This is the same with the one we found in the ground state energy, as we can deduce from Eq. (21) Eq. (29) for the 2DEG coupled to the cavity mode. Sub- is Ep/S = ~ω/Se . In the large N,S (or thermodynamic) stituting q0 = −KD/n2D into the expression for the en- limit this contribution is miniscule and strictly speaking ergy density of the uncoupled electron gas in Eq. (31) we goes to zero. On the other hand the electrons have a nc find E [D]|q0 to be equal to the energy density E[D]|q0 finite energy density E[S]. This implies that only the of the coupled system 2DEG contributes to the ground state energy density of the interacting electron-photon hybrid system in the cav- Enc[D]| = E[D]| . (32) q0 q0 ity. This energy mismatch shows up because in the elec- This means that both energy functionals, the coupled and tronic sector we have N electrons in the thermodynamic the uncoupled, get minimized by the same k-space dis- limit, while in the photonic sector we have only one mode. tribution D. For the uncoupled 2DEG, the shape of the This discrepancy between the two sectors hints towards the fact that for both sectors to contribute on the same distribution in k-space is the 2D Fermi sphere S(k − q0). For a sphere the collective momentum is zero, K = 0, level, a continuum of modes of the photon field have to be S taken into account such that the photon field to acquire and consequently the optimal origin is also zero q0 = 0. Thus, for the coupled system the ground state momen- a finite energy density in its ground state. We explore tum distribution is the 2D Fermi sphere S(k) centered this direction further in section VI. Before we continue at zero, as depicted in Fig. 3. Most importantly since we note that the photon field in its highly excited states the collective momentum is zero the ground state of the can still contain arbitrarily large amounts of energy. Yet 2DEG coupled to the cavity is for the considerations of the ground state these highly- excited photon-states do not play a role. 2 Y From the fact that the ground state of the electrons is |Ψgsi = |Φ0i ⊗ |0, 0iλ, (33) the standard Fermi sphere and that the energy density λ=1 of the photon field in the thermodynamic limit is negligi- ble, one might conclude that the electron-photon ground where Φ is the Slater determinant given by Eq. (14) with 0 state of the system is trivial and there are no quantum zero collective momentum K = 0. It is important to men- fluctuation effects due to the electron-photon coupling. tion that since in the ground state the collective momen- However, this is not the case. To classify completely the tum is zero, the ground state is a tensor-product state electron-photon ground state one needs to look also at between the electrons and the photons. The fact that the ground state photon occupation. the ground state distribution of the electrons in k-space is the Fermi sphere implies that the electronic system is a Fermi liquid [6]. Further, having found the ground state of the electrons, we can compute also the ground state A. Ground State Photon Occupation energy density of the electrons as a function of the Fermi wavevector and we find The photon number operator is +∞ 2 ZZ 2 4 2 ~ 2 2 2 ~ kF ˆ X † E[S] = d kS(k)k = . (34) Nph = aˆ aˆλ. (35) 2m (2π)2 16πm λ e e λ=1 −∞ Mismatch of Energies.—Moreover, we would like to To compute the ground state photon occupation we need point out a fundamental discrepancy which appears be- to write the number operator in terms of the bosonic † tween the electronic and photonic sector, with respect to operators {cˆλ, cˆλ} defined in Eq. (16). Using Eqs. (7) † their contributions in the ground state energy density. and (16) we find that the number operator in terms ofc ˆλ The contribution of the (single-mode) photon field, to andc ˆλ is

2 "  2     X 1 gελ · K 2 gελ · K gελ · K Nˆ = ω2 − ω2
cˆ + + (ω − ω) cˆ + cˆ† + ph 4ωω e λ ω e λ ω λ ω λ=1 e ~e ~e ~e      2 # 2 gελ · K † gελ · K 2 2
† gελ · K + (ω + ωe) cˆλ + cˆλ + + ω − ωe cˆλ + . (36) ~ωe ~ωe ~ωe

† In the ground state the collective momentum is zero, K = cˆλcˆλ gives a non-zero contribution. Thus, we find for the 0, and out of all the terms appearing above only the term that first creates and then destroys a bosonic excitation 9 ground state photon occupation A. Critical Coupling and Infinite Degeneracy

2 ˆ ˆ (ωe − ω) At the critical coupling γc = 1 the energy density E[D] hNphigs ≡ hΨgs|Nph|Ψgsi = . (37) given by Eq. (28) becomes independent of the origin q 2ωωe 2  K2  From the result above we see that the ground state pho- ~ D E[D]|γc = tD − . (39) ton occupation is non-zero. This means that there are 2me n2D virtual photons in the ground state of the interacting The fact that the energy density becomes degenerate electron-photon system. This phenomenon has also been with respect to the origin q means that the ground state reported for dissipative systems [107]. From the fact that of the system is not unique. Moreover, Eq. (29) from the ground state of the 2DEG in the cavity contains pho- which we determined the optimal value for the vector q, tons we conclude that there are quantum fluctuations of gets trivially zero. the photon field in the ground state due to the electron- The energy density of Eq. (39), as we explained in the photon coupling. Thus, our system is not a trivial Fermi previous section, minimizes for a sphere S(k − q). But liquid, but rather it is a Fermi liquid dressed with pho- since the energy density E[D]|γc is degenerate with re- tons. spect to the origin q and the optimal q cannot be deter- Further, the ground state photon occupation shows mined from Eq. (29), all spheres of the form S(k−q) are an interesting dependence on the electron density. For degenerate and have exactly the same ground state en- electron densities small enough for the plasma frequency ergy. This means that the optimal ground state k-space p 2 ωp = e ne/me0 to be much smaller than the cavity distribution it is not unique but rather the ground state q frequency, ω  ω, the dressed frequency ω = ω2 + ω2 of the system at the critical coupling γc = 1 is infinitely p e p degenerate with respect to origin of the k-space distribu- is approximately equal to the cavity frequency, ωe ≈ ω. tion of the electrons. In this case the ground state photon occupation is zero, Such an infinite degeneracy appears also for a 2D elec- ˆ hNphigs = 0. However, for large electronic densities such tron gas in the presence of perpendicular, homogeneous that ωp  ω, the dressed frequency is ωe ≈ ωp and the magnetic field where we have the Landau levels demon- ˆ numerator in the expression for hNphigs is approximately strating exactly this behavior [108]. The infinite degen- 2 ωp. Thus, we find that for large electron densities the eracy is also directly connected to the quantum Hall ef- ground state photon occupation has a square root de- fect [8]. The connection between quantum electrodynam- pendence on the electron density ics and the quantum Hall effect has also been explored re- √ cently in the context of quantum electrodynamical Bloch hNˆphigs ∼ ne. (38) theory [55]. Lastly, we note that the fact that all spheres S(k−q) of This implies that the amount of photons in the ground arbitrary origin q are degenerate means that the ground state increases with the number of electrons. This be- state energy of our system, at the critical coupling γc, is havior might be related to the superradiant phase transi- invariant under shifts in k-space, which implies that is tion [80] and could potentially provide some insights on invariant under Galilean boosts. how to achieve this phase transition, which remains still elusive. B. No Ground State Beyond the Critical Coupling

For completeness we would also like to consider the IV. CRITICAL COUPLING, INSTABILITY & case where the coupling constant goes beyond the crit- THE DIAMAGNETIC A2 TERM ical coupling γc and becomes larger than one, γ > 1. In principle from its definition in Eq. (22) the coupling So far we have examined rigorously and in full gen- constant γ is not allowed to take such values, but investi- erality the behavior of the 2DEG coupled to the cavity, gating this scenario will provide further physical insight in the regime where the cavity mode ω is finite and the why this should not happen. collective coupling parameter γ, defined in Eq. (22), is For simplicity and without loss of generality, we sim- less than one. But now the following questions arises: plify our consideration to the case where the cavity field what happens in the limit where the frequency of the has only one polarization vector ε1 = ex and ε2 = 0. In quantized field goes to zero, ω → 0, and the collective this case the energy density E[D] given by Eq. (28) as a coupling parameter takes its maximum value γ → 1? function of the x-component of the vector q = (qx, qy) is We will refer to the maximum value of the coupling 2 ~ x 2
constant γ as critical coupling, γc = 1, because as we will E(qx) = (1 − γ) 2qxKD + qxn2D , (40) see at this point an interesting transition happens for the 2me system, from a stable phase to an unstable phase, as it where we neglected all terms in Eq. (28) independent of is also summarized by the phase diagram in Fig. 4. qx. For γ > 1 the energy density above has no minimum 10 and it is unbounded from below because 1 − γ < 0 is elimination of the A2 is responsible for the notorious su- negative and taking the limit for qx to infinity the energy perradiant phase transition of the Dicke model [26]. The density goes to minus infinity superradiant phase transition was firstly predicted by

2 Hepp and Lieb [80] for the Dicke model in the thermody- ~ x 2
namic limit and soon after derived in an alternative way lim E(qx) = lim (1 − γ) 2qxKD + qxne → −∞. qx→∞ 2me qx→∞ by Wang and Hioe [81]. The existence though of the su- (41) perradiant phase was challenged by a no-go theorem [82] which showed that the superradiant phase transition in This proves that the free electron gas coupled to the cav- atomic systems appeared completely due to the fact that ity mode for γ > 1 has no ground state and the system in the A2 term was not taken into account. More recently, this case is unstable, because shifting further and further another demonstration of a superradiant phase transi- the distribution in k-space, by moving its center q, we tion was predicted in the framework of circuit QED [83], can lower indefinitely the energy density2. Thus, we con- which again was challenged by another no-go theorem clude that the upper bound for the collective coupling γ which applied also to circuit QED systems [84]. Also the given by Eq. (22) guarantees the stability of the coupled inclusion of qubit-qubit interactions was shown to be im- electron-photon system. Lastly, we would like to men- portant for such circuit QED systems [87]. Further, the tion that due to the lack of ground state, equilibrium is application of these no-go theorems it is argued that it de- not well-defined in the unstable phase and equilibrium pends on the gauge choice [78, 90, 109]. Nevertheless, the phenomena cannot be described properly. debate over the existence of the superradiant phase tran- sition is still ongoing, with new demonstrations coming from the field of cavity QED materials [85, 110] accompa- nied though by the respective no-go theorems [86, 111]. Lastly, the possibility of a superradiant phase transition beyond the dipole approximation has also been investi- gated [88, 89]. For our system to examine the importance of the dia- magnetic A2 term here, we study the free electron gas coupled to the cavity in the absence of the A2 term. From the Hamiltonian Hˆ in Eq. (3) it is straightforward to derive the Hamiltonian Hˆ 0 for the electron gas cou- pled to the cavity mode when the A2 term is neglected 0 2 2 Hˆ = Hˆ − Ne Aˆ /2me

N " # 2 2∇2   ˆ 0 X ~ j ie~ ˆ X † 1 H = − + A · ∇j + ~ω aˆλaˆλ + . 2me me 2 FIG. 4. Phase diagram for the free electron model in cavity j=1 λ=1 QED. The system has a stable ground state for coupling con- (42) stant γ < 1. At the critical coupling γc = 1 the ground state is infinitely degenerate. Beyond the critical coupling γc = 1 As we explained in section II in the electronic config- the system is unstable and the system has no ground state. uration space we have translational symmetry, and the electronic eigenfunction is the Slater determinant given by Eq. (14). Introducing now the parameter

1/2 2 e   C. No-Go Theorem and the A Term g0 = ~ ~ , (43) me 20ωV In what follows we are interested in the importance ˆ 0 of the often neglected [77] diamagnetic A2 term for the applying the Hamiltonian H on the Slater determinant ˆ 2DEG coupled to the single mode quantized field. The in- ΦK and substituting the definition for quantized field A fluence of this quadratic term has been studied theoreti- given by Eq. (4) we obtain cally in multiple publications [76–79] and its influence has ( 2 X   1    also been experimentally measured [58]. Moreover, the Hˆ 0Φ = ω aˆ† aˆ + − g0 aˆ +a ˆ† ε · K K ~ λ λ 2 λ λ λ λ=1 2 N ) N ~ X 2 X 2 We would like to point out that this argument is similar to the + kj ΦK where K = kj. (44) 2me one for the lack of ground state in the length gauge when the j=1 j=1 dipole self-energy is omitted. In the length gauge the energy can be lowered indefinitely by shifting further and further in real The Hamiltonian Hˆ 0 is of exactly the same form as Hˆ space the charge distribution [101]. of Eq. (15). Following exactly the same procedure for 11 diagonalizing Hˆ , which we showed in section II, we can and show that without the A2 term the coupling constant diagonalize also Hˆ 0 and we find that its eigenspectrum is has no upper bound and the center of mass can obtain an arbitrarily large momentum which subsequently leads to 2 "   0 2 2 # N 2 2 X 1 γ ~ (ελ · K) X ~ kj an arbitrarily negative energy. This implies that energy Enλ,k = ~ω nλ + − + 2 N 2me 2me of the system is unbounded from below and the system λ=1 j=1 has no ground state. (45) where we substituted the parameter g0 of Eq. (43) and we introduced the parameter γ0 V. CAVITY MODIFIED RESPONSES 2m N (g0)2 ω2 γ0 = e = p (46) 2 ω ω2 So far we have considered the electron gas inside the ~ ~ cavity to be in equilibrium without any external pertur- in complete analogy to the coupling constant γ given by bations, like fields, potentials, forces or other kind of Eq. (22). The dressed frequency ω does not show up any- sources being applied to it. The aim of this section is 0 e more neither in the coupling γ nor in the energy spec- exactly to go in this direction and apply external pertur- trum (45), because the quantized field and the energy of bations to our interacting electron-photon system, and 2 the cavity mode do not get renormalized by the A , since compute how particular observables of the system re- it is absent. Comparing now the spectrum of Eq. (45) for spond to the external perturbations. ˆ 0 the Hamiltonian H , with the spectrum given by Eq. (21) In standard quantum mechanics and solid state physics derived for the Hamiltonian Hˆ of Eq. (3) which included usually one applies to the system an external field, the A2 term, we see that they are exactly the same, up force or potential and then focuses on how the elec- 0 to replacing ωe with ω and γ with γ . The last one is a trons respond to the perturbation by computing matter- very important difference, because the coupling constant matter response functions, like the current-current re- γ0 has no upper bound and can be arbitrarily large, as ω J p sponse function χj , which is related to the conductive can be larger than ω. In section IV B we proved that the properties of the electrons [2, 3]. On the other side in spectrum of the form given by Eq. (45), has no ground quantum optics one focuses on the responses of the elec- state if the coupling constant gets larger than one. For tromagnetic field by computing photon-photon response 0 large densities ωp can become larger than ω and γ will be functions, like the A-field response function χA. 0 A larger than one, γ > 1. Consequently, the Hamiltonian Quantum electrodynamics combines both perspectives ˆ 0 H will be unstable and will not have a ground state. under a common unified framework and except of per- 2 This highlights that eliminating the diamagnetic A turbing by external fields, forces and potentials offers term, is a no-go situation for the free electron gas cou- the possibility of coupling to external currents. This pled to the cavity, and that for a sound description of implies that QED gives us the opportunity to access such a macroscopic solid state system the diamagnetic novel observables and response functions which might 2 A term is absolutely necessary. For finite-system mod- provide new insights in the emerging field of cavity 2 els like the Rabi or the Dicke model the A term is of QED [12]. In addition to the matter-matter and photon- course important, but these models have a stable ground photon responses QED allows to access also cross- 2 state even without the A term. This is in stark con- correlated response functions, like matter-photon and trast to the 2DEG (which is macroscopic) coupled to the photon-matter. As we will see in what follows, all four cavity mode which has no ground state without the dia- sectors (matter-matter, photon-photon, matter-photon magnetic term. This demonstrates explicitly that finite- and photon-matter) have the same pole structure but system models should be applied to extended systems with different strengths. More specifically we will show with extra care. Our demonstration strongly suggests that all sectors exhibit plasmon-polariton excitations or that the quadratic term should be included for the de- resonances, which modify the radiation and conductive scription of extended systems, like 2D materials, coupled properties of the electron gas in the cavity. to a cavity and we believe contributes substantially to the ongoing discussion about the proper description of light- matter interactions [76–79, 101, 112] and particularly for the emerging field of cavity QED materials. A. Linear Response Formalism Finally, we emphasize that our proof can be ex- tended also for interacting electrons. This is because Our considerations throughout this section will remain the Coulomb interaction involves only the relative dis- within the framework of linear response, in which a sys- tances of the electrons and preserves translational sym- tem originally assumed to be at rest and described by a metry. Thus, one can go to the center of mass and relative Hamiltonian Hˆ , is perturbed by a time-dependent ex- ˆ ˆ distances frame in which the relative distances decouple ternal perturbation of the form Hext(t) = fext(t)P. The from the quantized vector potential A and from the cen- external perturbation fext(t) couples to some observable ter of mass. The center of mass though stays still coupled of the system represented by an operator Pˆ. The strength to A. Then one can follow the proof we presented here of the perturbation is considered to be small, such that 12 the response of the system to be of first order in per- The external current influences the hybrid system in the turbation theory. This is how linear response formal- cavity, and induces electromagnetic fields, as depicted ism (also known as Kubo formalism) is usually formu- in Fig. 5. The influence of the external current on the lated [3, 91, 92]. Then, by going into the interaction photonic observables is exactly what we are interested in picture the response of any observable Oˆ to the external here. perturbation is defined as [3, 91, 92] Z t i 0 0 0 δhOˆ(t)i = − dt h[OˆI (t), PˆI (t )]ifext(t ). (47) ~ t0 The correlator above is computed with respect to the ground state |Ψgsi of the unperturbed Hamiltonian Hˆ , 0 0 h[OˆI (t), PˆI (t )]i ≡ hΨgs|[OˆI (t), PˆI (t )]|Ψgsi, and OˆI (t) = ˆ ˆ eiHt/~Oˆe−iHt/~ is the operator Oˆ in the interaction pic- ture. From Eq. (47) by introducing the theta function Θ(t − t0) we can re-write the response δhOˆ(t)i with the O 0 help of a function χP (t − t ) as Z ∞ ˆ 0 O 0 δhO(t)i = dt χP (t − t )fext(t), (48) t0 FIG. 5. Material confined inside a cavity, perturbed by exter- O 0 nal time dependent current J (t). The external current per- with the function χP (t − t ) defined as ext turbs the interacting light-matter system, a time-dependent 0 O 0 iΘ(t − t ) ˆ ˆ 0 electric field is induced, and the cavity radiates. We note χP (t − t ) = − h[OI (t), PI (t )]i. (49) ~ that in an experiment the emitted radiation can be accessed Functions of this form are known as response functions through the openness of the cavity. and are of great importance because they give us infor- mation about how observables of the system respond to an external perturbation [3, 91, 92]. From Eq. (48) by 1. A-Field Response & Absorption performing a Laplace transform we can also obtain the response of the observable Oˆ in the frequency domain The first thing we would like to compute is the re- ˆ O sponse of the A-field δhAˆ (t)i due to the external time- δhO(w)i = χP (w)fext(w), (50) dependent current Jext(t). The response of the vector O where χP (w) and fext(w) are the response function and potential δhAˆ (t)i is defined via Eq. (47) and is given by the external perturbation respectively in the frequency A 0 the A-field response function χA(t−t ). From Eq. (49) we domain [3, 91, 92]. A 0 can define the response function χA(t − t ), and perform- ing the computation for the A-field response function, which we show in detail in appendix B, and we find B. Radiation & Absorption Properties in Linear Response 0 0 A 0 Θ(t − t ) sin(ωe(t − t )) χA(t − t ) = − . (52) 0ωVe Let us start by applying linear response to the pho- Performing a Laplace transform on the response function tonic sector by computing response functions related to χA(t−t0) we can find the response function χA(w) in the the electromagnetic field. From such responses we obtain A A frequency domain, which is given in appendix B, and we information about the radiation and absorption proper- deduce the real <[χA(w)] and the imaginary =[χA(w)] ties of the electron gas coupled to the cavity. To compute A A parts of χA(w) these properties, we apply an external time dependent A current J (t) as shown in Fig. 5. We would like to   ext A 1 w − ωe w + ωe emphasize that in standard quantum mechanics the pos- <[χA(w)] = 2 2 − 2 2 , 20ωV (w − ω) + η (w + ω) + η sibility of perturbing with an external current does not e e e η  1 1  exist and only QED makes this available. A =[χA(w)] = 2 2 − 2 2 To couple the external current to our system we need 20ωVe (w + ωe) + η (w − ωe) + η to add to the Hamiltonian Hˆ of Eq. (3) an external (53) time dependent term Hˆ (t) = −J (t) · Aˆ as it is ext ext which are depicted in Fig. 6. From this expression we done in quantum electrodynamics [10, 11, 92]. The ex- see that the pole of the response function is at frequency ternal current is chosen to be in only in the x-direction w = ±ω. The frequency ω defined in Eq. (8) depends on J (t) = e |J (t)|. Adding the external perturbation e e ext x ext the cavity frequency ω and the plasma frequency ω in the full time-dependent Hamiltonian is p the cavity. This means that the electron gas in the cavity Hˆ (t) = Hˆ − Jext(t) · Aˆ . (51) has a plasmon-polariton resonance. 13

E 0 For a self-adjoint operator the real and the imaginary definition of Eq. (49). The computation of χA(t − t ) is part of any response function have to be respectively even presented in appendix C and we find and odd [3]. In our case the A-field is self-adjoint and we see that the real and imaginary parts of χA(w) shown in 0 0 A E 0 Θ(t − t ) cos(ωe(t − t )) Fig. 6 satisfy these properties. χA(t − t ) = . (56) 0V Before we continue let us comment on how of these parts of the response function should be interpreted. The The response function above describes the generation of A real part <[χA(w)] is the component of the response func- a time dependent electric field due to the external time tion which is in-phase with the external current that dependent current Jext(t). This means that the external drives the system. The real part describes a polariza- current makes the coupled light-matter system radiate. tion process in which the wavefunction is modified peri- From Eq. (56) we see the radiation is at the plasmon- odically without any energy being absorbed or released polariton frequency ω since the response function in time on average by the external driving [3]. On the other e is a cosine of ωe. This fact can also be understood from hand, the imaginary part =[χA(w)] is the out-of-phase E A the response function in the frequency domain χA(w), component of χA(w), with respect to the external driv- E E A whose real <[χA(w)] and imaginary parts =[χA(w)] are ing current. The imaginary part is responsible for the appearance of energy absorption in the system, with the   E η 1 1 absorption rate W given by the expression [3] <[χA(w)] = 2 2 − 2 2 , 20V (w + ωe) + η (w − ωe) + η W = −w=[χA(w)]|J (w)|2. (54)   A ext E 1 w + ωe w − ωe =[χA(w)] = 2 2 − 2 2 20V (w + ωe) + η (w − ωe) + η (57)

from which we see that the poles are at the plasmon- polariton resonance w = ±ωe as shown also in Fig. 7. Lastly, we would like to mention that the response E 0 function of the electric field in time χA(t−t ) of Eq. (56), A 0 and the response function of the A-field χA(t − t ) E 0 of Eq. (52) satisfy Maxwell’s equation χA(t − t ) = A 0 −∂tχA(t − t ) [113]. This is a beautiful consistency check for our computations and of the whole linear response formalism in QED [92], because it shows that linear re- sponse theory even for coupled electron-photon systems respects the classical Maxwell equations.

A A FIG. 6. Real <[χA(w)] and imaginary =[χA(w)] parts of A the A-field response function χA(w) in the frequency domain, plotted with a finite η. The resonances for both parts appear at the plasmon-polariton frequency w = ±ωe.

2. Electric Field Response & Current Induced Radiation

Having computed the response of the A-field we would also like to compute the response of the electric field E due to the external current. The electric field operator in dipole approximation and polarized in the x-direction is [101] E E FIG. 7. Real <[χA(w)] and imaginary =[χA(w)] parts of E 1 the E-field response function χA(w) in the frequency domain   2 E ˆ ~ω  † with a finite broadening parameter η. The poles of <[χA(w)] E = i aˆ1 − aˆ1 ex. (55) and =[χE (w)] both appear at the frequency w = ±ω and 20V A e signify the frequency at which an time-dependent electric field With the definition of the electric field we can compute is oscillating. Radiation should come out of the cavity at this E 0 frequency. the electric field response function χA(t − t ) using the 14

C. Cavity Modified Conductivity & Drude Peak is Suppression N X ie e2  Hˆ (t) = Hˆ + ~∇ + Aˆ · A (t) m j m ext j=1 e e In what follows we are interested in the conduction   properties of the 2DEG inside the cavity and more specif- = Hˆ − Jˆp + Jˆd · Aext(t) (58) ically on whether the cavity field modifies the conductiv- ity of the 2DEG. This is a question of current theoretical where Hˆ is the Hamiltonian of Eq. (3). The external and experimental interest, because recently cavity mod- field couples to the internal parts of the current operator, ifications of transport and conduction properties have ˆ P which are the paramagnetic part Jp = (−ie~/me) j ∇j, been observed for 2D systems of Landau polaritons [61], 2 and the diamagnetic part Jˆd = −e NAˆ /me. The full as well as modifications of the critical temperature of su- physical current includes also the contribution due to the perconductors due to cavity confinement [64, 66]. external vector potential Aext(t) [30, 108]

To describe such processes we will follow what is usu- N ie~ X e2N e2N ally done in condensed matter physics, namely perturb Jˆ = − ∇j − Aˆ − Aext(t). (59) the system with an external, uniform, time-dependent me me me j=1 | {z } electric field Eext(t), as depicted in Fig. 8, and then | {z } Jˆ ˆ d compute how much current flows due to the pertur- Jp bation. Here, the electric field is chosen to be polar- Following the standard linear response formalism the ex- ized in the x-direction Eext(t) = |Eext(t)|ex and can be pectation value for the full physical Jˆ current is [3, 91] represented as the time derivative of a vector potential Z ∞ Eext(t) = −∂tAext(t). We note that to have a causal ˆ ˆ ˆ ˆ 0 J 0 0 hJ(t)i = hJi + δhJ(t)i = hJi − dt χJ (t − t )Aext(t ) external perturbation the electric field needs to be zero t for all times prior to an instant of time t . This implies 0 0 (60) that in the frequency domain the electric field and vector potential are related via Eext(w) = i(w + iη)A(w) with where δhJˆ(t)i is the response of the current Jˆ, which + η → 0 . can be computed from the the current-current response function 0 J 0 −iΘ(t − t ) ˆ ˆ 0 χJ (t − t ) = h[JI (t), JI (t )]i. (61) ~

Neglecting all contributions coming from Aext(t), such that the current response δhJˆi stays in first order to Aext, we find for the commutator of Eq. (61) the following four terms

0 0 0 [JˆI (t), JˆI (t )] = [Jˆp,I (t), Jˆp,I (t )] + [Jˆd,I (t), Jˆp,I (t )] 0 0 + [Jˆd,I (t), Jˆp,I (t )] + [Jˆd,I (t), Jˆd,I (t )] (62) For the paramagnetic contribution using the self- adjointness of the paramagnetic current operator we have 0 0 0 ∗ FIG. 8. An external time dependent electric field Eext(t) h[Jˆp,I (t), Jˆp,I (t )]i = hJˆp,I (t)Jˆp,I (t )i − hJˆp,I (t)Jˆp,I (t )i . perturbs the combined light-matter system, electrons start to Using the definition for the paramagnetic current op- flow, and a current is generated in the material. erator in the interaction picture and the fact that the expectation value is computed in the ground state 0 which has energy E0,k we find hJˆp,I (t)Jˆp,I (t )i = 0 ˆ 0 iE0,k(t−t )/ ˆ iH(t −t)/ ˆ To couple the external field we need to add the ex- e ~hJpe ~Jpi. Because the momentum ˆ ternal vector potential Aext(t) in the covariant kinetic operator commutes with the Hamiltonian H, the ground energy of the Pauli-Fierz Hamiltonian of Eq. (1), which state |Ψgsi = |Φ0i ⊗ |0, 0i1|0, 0i2 is also an eigenstate of ˆ 2 ˆ P becomes then (i~∇j + eA + eAext(t)) [11, 30, 108]. In the paramagnetic current operator Jp ∼ j ∇j. Acting linear response the current is computed to first order in with the paramagnetic current operator on the ground perturbation theory and the conductivity is defined as state we get the full paramagnetic current Jˆp|Ψgsi = P the function relating the induced current to the exter- j kj|Ψgsi, and because in the thermodynamic limit the nal electric field [3, 91, 92]. The Pauli-Fierz Hamiltonian ground state distribution of the momenta is the Fermi with the electrons coupled to a single mode, in dipole sphere, as we showed in section III, the total paramag- approximation, to first order in the external field Aext(t) netic current is zero and we have Jˆp|Ψgsi = 0. This 15 means that all expectation values and correlators which hJˆi involve Jˆ are zero. This argument applies also to p 2 ˆ ˆ 0 ˆ ˆ 0 e N the mixed terms [Jd,I (t), Jp,I (t )] and [Jp,I (t), Jd,I (t )]. hJˆi = − Aext(t). (66) J 0 me Thus, the response function χJ (t−t ) in Eq. (61) is given purely by the diamagnetic terms. Substituting the defi- The latter is the contribution of the the full background ˆ nition for the diamagnetic current Jd of Eq. (59) we find charge of the N electrons in our system. From the equa- J 0 the current-current response function χJ (t−t ) to be pro- tion for the full physical current in time hJˆ(t)i given by A 0 portional to the A-field response function χA(t − t ) Eq. (60) we can derive the relation between the current hJˆ(w)i and the external vector potential A (w) in the  2 2 ext J 0 e N A 0 frequency domain by performing a Laplace transforma- χJ (t − t ) = χA(t − t ). (63) me tion A 0 J 0  e2N  with χA(t − t ) given by Eq. (52). Since χJ (t − t ) is J A 0 hJ(w)i = − − χJ (w) Aext(w). (67) proportional to χA(t − t ) the same will also hold in the me frequency domain As we already explained, the vector potential and the e2N 2 electric field in the frequency domain are related via the χJ (w) = χA(w) (64) relation A (w) = E (w)/i(w + iη). Using this and J m A ext ext e dividing Eq. (67) by the volume V in order to introduce where χA(w) is computed in appendix B. Last, we need the current density hj(w)i = hJ(w)i/V we can define the A frequency dependent (or optical) conductivity σ(w) as to compute the expectation value of the current hJˆi which the ratio between the external electric field E(w) and is the current density hj(w)i [2, 3] 2 ˆ ˆ ˆ e N  2 J  hJi = hJpi + hJdi − hAext(t)i. (65) e ne χJ (w) Eext(w) me hj(w)i = − − = σ(w)Eext(w). me V i(w + iη) As we already explained the contribution of the param- (68) ˆ agnetic Jp current is zero in the ground state |Ψgsi. The The equation above is the Kubo formula for the electri- diamagnetic part Jˆd is proportional to the quantized field cal conductivity [3, 91]. Using the result for the current- ˆ ˆ J Jd ∼ A. The quantized vector potential is the sum of current response function χJ (w) given by Eq. (64), and 2 2 an annihilation and a creation operator and the expecta- introducing ωp = e ne/me0 which is the plasma fre- tion values of these operators in the ground state is zero. quency in the cavity, we obtain the expression for the Thus, we find that only the external field contributes to frequency dependent (or optical) conductivity σ(w)

2 J 2 4 i e n χ (w) i0ω i0ω  1 1  σ(w) = e + J = p − p − with η → 0+. (69) w + iη me V w + iη (w + iη)2ωe w + ωe + iη w − ωe + iη

The real <[σ(w)] and imaginary =[σ(w)] parts of the op- tical conductivity are given respectively by the expres- sions

 ηω2 η ω4   0 p 0 p 2w + ωe 2w − ωe <[σ(w)] = 2 2 − 2 2 2 2 − 2 2 , (70) w + η 2ωe(w + η ) (w + ωe) + η (w − ωe) + η  wω2  ω4  2 2 2 2  0 p 0 p w − η + wωe w − η − wωe =[σ(w)] = 2 2 − 2 2 2 2 − 2 2 . w + η 2ωe(w + η ) (w + ωe) + η (w − ωe) + η

In the optical conductivity σ(w) there are two contri- bution comes from the current-current response function J butions. The first contribution comes from the full elec- χJ (w). This one is purely due to the photon field in 2 2 J tron density ne via the plasma frequency ωp = nee /me0 the cavity because χJ (w) is proportional to the A-field A and is of second order to ωp. This is the standard con- response function χA(w). The current-current response tribution of the free electron gas [3]. The second contri- function is of fourth order in the plasma frequency ωp 16

1. Cavity Suppression of the Drude Peak

The Drude peak is defined as the w → 0 limit of the real part of the optical conductivity and gives the (static) DC electrical conductivity of a material σdc = limw→0 <[σ(w)] [2, 3, 114]. In the case of an electron gas 0 2 outside a cavity the DC conductivity is σdc = 0ωp/η, which is the first term of <[σ(w)] in Eq. (70) for w → 0. However, for our system we have the extra diamagnetic contributions due to the electron-photon coupling and we find that the DC conductivity σdc(γ) of the 2DEG in the cavity is a function of the collective coupling γ (defined in Eq. (22)) FIG. 9. Real part <[σ(w)] of the conductivity σ(w) of the  γ  2DEG outside (blue dashed) and inside (orange solid line) the 0 + σdc(γ) = σdc 1 − 2 2 with η → 0 . (71) cavity for collective coupling γ = 0.2. Inside the cavity the 1 + η /ωe real part exhibits poles at the plasmon-polariton resonance To zeroth order in the infinitesimal parameter η we find w = ±ωe. At frequency w = 0 the Drude peak of the 2DEG gets suppressed due to the cavity field. that the DC conductivity in the cavity, i.e., the Drude peak, decreases linearly as function of the collective cou- pling constant γ 0 σdc(γ) = σdc (1 − γ) . (72) This is a significant result because it shows that coupling materials to a cavity does not only modify the optical properties of the system, like the optical conductivity, but also the cavity can alter the static DC electrical con- ductivity. This phenomenon, of the decrease of the DC conductivity has also been reported for Landau polari- ton systems in [116]. To be more specific, in the region of zero magnetic field (in which our theory is also ap- plicable) an increase of the longitudinal resistivity was obtained, due to the cavity confinement [116]. This im- plies that the DC conductivity due to the strong coupling to the cavity decreases, in accordance to our prediction. Most importantly, we would like to mention that this effect is also in agreement with magneto-transport mea- FIG. 10. Imaginary part =[σ(w)] of the conductivity σ(w) surements performed for such Landau polariton systems of the 2DEG outside (blue dashed) and inside (orange solid in [61]. This is a firm confirmation of our work. We hope line) the cavity for collective coupling γ = 0.2. Inside the cavity the imaginary part has poles at the plasmon-polariton that further experimental measurements, focusing solely on the behavior of the Drude peak, under strong coupling resonance w = ±ωe. The peak at w = 0 gets suppressed by the cavity. to the photon field, will further explore this phenomenon and allow for a further quantitative test of our prediction about the modification of the Drude peak. The fact that the photon field has the effect to decrease the conduction of electrons implies that the cavity field can be understood as viscous medium which slows down and is a diamagnetic modification to the standard free the motion of the charged particles. In such a picture the electron gas conductivity. To be more specific, both the suppression of the Drude peak can be also understood as real and the imaginary part of the optical conductivity an increase in the effective mass of the electrons due to shown in Figs. 9 and 10 respectively, exhibit resonances the coupling to the cavity field. From the expression for at the plasmon-polariton frequency w = ±ωe, which mod- the DC conductivity in Eq. (72) we find that the effective ify the optical conductivity of the 2DEG. In addition, in (or renormalized) electron mass is me(γ) = me/(1 − γ). the real part of the conductivity we see that at w = 0 Such an increase of the effective electron mass we will the Drude peak [114, 115] of the 2DEG is suppressed also encounter later in section VI when we will couple the by the cavity field due to the higher-order diamagnetic 2DEG to the full continuum of electromagnetic modes. contributions. As the Drude peak is very important for Lastly, we would like to mention that due to the condensed matter systems and materials let us have a fact that the collective coupling parameter has an up- closer look at it. per bound γ < 1 (see Eq. 22) the Drude peak remains 17

A 0 always larger than zero and the 2DEG is a conductor. photon-matter response function χJ (t−t ), which is pro- 0 However, if the coupling could reach the critical value portional to the correlator h[Aˆ I (t), JˆI (t )]i according to γc = 1 (which is forbidden) then the DC conductivity Eq. (49). To remain within linear response we neglect the would be zero, which would imply that the cavity can contribution of Aext(t) to the current operator Jˆ which turn the 2DEG into an insulator. For γ > 1 the DC con- would result into higher order corrections. ductivity turns negative which implies that the system The paramagnetic contribution as we already ex- becomes unstable. This explains from a different point plained is zero. Substituting the definition for the dia- of view why the collective coupling must not exceed the magnetic current Jˆd we find that the mixed response upper bound 1. A 0 function χJ (t−t ) is proportional to the A-field response A 0 2
A 0 function χJ (t − t ) = −e N/me χA(t − t ). Since this relation holds in time, it will also be true in the frequency D. Mixed Responses: Matter-Photon & domain, Photon-Matter  2  A −e N A χJ (w) = χA(w). (74) In the beginning of this section we emphasized the fact me that QED gives us the opportunity to access new mixed, cross-correlated responses. So let us now present how From the result above we see that the response function A ˆ such mixed matter-photon and photon-matter response χJ (w) is the dimensionless ratio between the induced A- functions arise in QED and compute them. field and the external field Aext.

1. Matter-Photon Response E. Linear Response Equivalence Between the Electronic and the Photonic Sector The response of the current δhJˆ(t)i is defined via Eq. (47) and can be computed directly from the mixed In this section we would like to compare the four fun- J 0 response function χA(t − t ) which is proportional to damental response sectors we introduced and discussed 0 the correlator h[JˆI (t), Aˆ I (t )]i as we can deduce from above, and most importantly demonstrate how these sec- Eq. (49). The full physical current Jˆ given by Eq. (59), tors are connected and that actually are all equivalent for Aext = 0, includes two contributions. One from the with respect to their pole structure. From all the re- paramagnetic current Jˆp and one coming from the dia- sponse functions we computed in the different sectors we can can construct the following response table magnetic current Jˆd. The paramagnetic contribution as we explained in the previous subsection is zero because  ˆ   J J    the ground state has zero paramagnetic current, and δhJ(w)i χJ (w) χA(w) Aext(w) ˆ = A A consequently only the diamagnetic current contributes. δhA(w)i χJ (w) χA(w) Jext(w) Substituting the definition for the diamagnetic current (75) ˆ J 0 Jd we find that the mixed response χA(t − t ) is pro- J 0 portional to the A-field response function χA(t − t ) = which summarizes all the different responses of the sys- 2
A 0 A 0 −e N/me χA(t−t ) where χA(t−t ) given by Eq. (52). tem. Looking back now into the Eqs. (64), (74) and (73) J J The same relation between the two response functions which give the response functions χJ (w), χA(w) and A also holds in the frequency domain χJ (w) respectively, we see that all response functions are proportional to the A-field response function χA(w).  2  A J −e N A Thus, all elements of the response table can be written χA(w) = χA(w). (73) A me in terms of χA(w)

Lastly, we would like to emphasize that the mixed re-   2 2 2  J e N −e N sponse function χA(w) is dimensionless and describes the  ˆ   me me    ratio between the induced current δhJˆ(w)i and the exter- δhJ(w)i A   Aext(w) = χA(w)   ˆ J δhAˆ (w)i   Jext(w) nal current Jext(w), δhJ(w)i = χA(w)Jext(w).  −e2N  1 me (76) 2. Photon-Matter Response The fact that all response functions are proportional to Having computed the matter-photon response func- A-field response function χA(w) means that all response J A tion χA we want to compute also the photon-matter functions have exactly the same pole structure. This A response function χJ which corresponds to the inverse shows a deep and fundamental relation between the two J physical process with respect to χA. Now we look into sectors of the theory, namely that the photonic and the the response of the vector potential δhAˆ (t)i given by the electronic sectors have exactly the same excitations and 18 resonances. This implies that in an experiment, per- VI. EFFECTIVE QUANTUM FIELD THEORY turbing an interacting light-matter system with an ex- IN THE CONTINUUM ternal time dependent current, which couples to the pho- ton field, and perturbing with an external electric field, Up to here we have investigated in full generality the which couples to the current, would give exactly the same behavior of the free electron gas in the large N or ther- information about the excitations of the system. modynamic limit for the electronic sector, coupled to a Furthermore, from the response table in Eq. (76) we single quantized mode. The single mode approximation see that the current-current response function scales has been proven very fruitful and successful for quantum J quadratically with the number of electrons χJ (w) ∼ optics and cavity QED [14, 103], but as it is known from 2 A N χA(w), while the mixed response functions linearly the early times of the quantum theory of radiation and J A A χA(w) = χJ (w) ∼ NχA(w). The photon-photon re- the seminal work of Einstein [117] to describe even one A sponse function χA(w) given by Eq. (53) also scales with of the most fundamental processes of light-matter inter- respect to the area of the 2DEG as 1/S. This implies action like spontaneous emission the full continuum of that in the large N,S limit only the responses involving modes of the electromagnetic field have to be taken into J J A matter (χJ , χA, χJ ) are finite, due to the dependence on account. Moreover, we should always keep in mind that A N, while χA goes to zero. This is the same feature that in a cavity set-up of course a particular set of modes of appears also for the energy densities of the two sectors the electromagnetic field are selected by the cavity, but as we mentioned in section III. Again, this hints towards it is never the case that only a single mode of the cav- the fact that in order to have a finite photon-photon re- ity contributes to the light-matter coupling. The single sponse, we need to include a continuum of modes for the mode models like the Rabi, Jaynes-Cummings or Dicke photon field because we are a considering a macroscopic model, describe effectively (with the use of an effective 2D system. For a finite system such a problem would not coupling) the exchange of energy between matter and the arise and this shows another point in which coupling the photon field as if there were only a single mode coupled photon field to a macroscopic system is different that to to matter [118]. a finite system. In our case the situation becomes even more severe Moreover, the light-matter coupling γ of Eq. (22) is because we consider a macroscopic system like the 2DEG, proportional to the number of particles3. This implies where the propagation of the in-plane modes becomes that the strength of the responses actually depends on important. This implies that the 2D continuum of modes the coupling constant. This suggests that light and mat- of the electromagnetic field has to be taken into account. ter in quantum electrodynamics are not only equivalent Before we proceed with the construction of the theory with respect to their excitations and resonances, but also for the photon field in the continuum let us give some the strengths of the their respective responses are related more arguments why such a theory is needed and what through the light-matter coupling constant (or number of particular observables and physical processes can only be particles). described in such a theory. Lastly, we highlight that the response functions we Why a Quantum Field Theory? computed throughout this section depend on the arbi- trarily small yet finite auxiliary parameter η, which is From the point of view of observables and physical pro- standard to introduce in linear response, in order to have cesses the main reasons are: (i) As we saw in section III a well-defined Laplace transform [3, 92]. In the limit the contribution of the single mode cavity field to the η → 0 the response functions go to zero (see for example ground state energy density Ep/S in the thermodynamic limit, where the number of electrons N and the area S be- Eq. (53)) except of the frequencies w = ±ωe where they diverge. This implies that η works like a regulator which come arbitrary large, becomes arbitrary small and tends spreads the resonance over a finite range and describes to zero. This implies that in the single mode case no sig- the coupling of the system to an artificial environment nificant contribution to the ground state of the system and how energy is dissipated to this environment [3]. To comes from the photon field, because of the discrepancy remove this arbitrary broadening parameter η, one can between the amount of the electrons and the amount of treat the matter and the photon sectors on equal foot- modes. (ii) As we mentioned in the end of the previous ing and perform the continuum-limit also for the photon section, absorption processes and dissipation can be de- field. This as we will see in the next section allows for scribed consistently and from first principles only when the description of absorption and dissipation without the a continuum of modes is considered [3]. (iii) Since the need of η. contribution of the cavity field to the energy density is zero, compared to the energy density of the electrons, no real contribution to the renormalized or effective mass of the electron can occur. This again is due to the fact that we consider a single mode of the photon field, and as it 3 This fact can be understood more easily from the coupling con- known from QED, mass renormalization shows up when stant in the effective theory g(Λ) in Eq. (79) but it is also true electrons are coupled to the full continuum of the elec- for the Dicke model [26] tromagnetic field [13, 95, 96, 119, 120]. (iv) Lastly, no 19 macroscopic forces can appear between the cavity mir- will be able to construct an analytically solvable effective rors, like the well-known Casimir-Polder forces [99], in field theory, in the thermodynamic limit for both light the single mode limit. As it well known from the litera- and matter. Before we continue we would like to men- ture such forces show up only when the full continuum of tion that the validity of the approximation to neglect the modes is considered [121, 122]. For all these reasons we mode-mode interactions depends on the how large the proceed with the construction of the effective field theory diamagnetic shift ωp [103] is. We will investigate and for a continuum of modes. test this approximation in more detail later in subsec- tions VI B and VI C.

A. Effective Field Theory, Coupling and Cutoff To construct this effective quantum field theory, first we need to introduce back the dependence to the mo- menta κ = (2πn /L, 2πn /L, πn /L ) of the all the To promote the single mode theory to a field theory we x y z z parameters of the theory. The bare modes ω of the need to perform the “thermodynamic limit” for the pho- quantized electromagnetic field in terms of the momenta ton field (in analogy to the electrons), and integrate over κ are ω(κ) = c|κ|. Furthermore, for the dressed fre- all the in-plane modes of the electromagnetic field. Such q 2 2 a procedure can be performed for an arbitrary amount of quency ωe = ω + ωp we also need to introduce the photon modes, with the mode-mode interactions included q κ-dependence by promoting it to ω(κ) = ω2(κ) + ω2. (see appendix E). However, such a treatment would make e p the theory non-analytically solvable, and particularly in As a consequence, also the single-mode (many-body) cou- 2 2 the thermodynamic limit. pling constant γ = ωp/ωe becomes κ-dependent γ(κ) = 2 2 For the latter reason, we will follow an alternative ωp/ωe (κ). With these substitutions and summing the approach. We will perform the integration in an effec- eigenspectrum of Eq. (21) over all the momenta in the tive way, where we will neglect the mode-mode interac- (x, y) plane, we find the expression for the ground state tions and we will integrate the single mode spectrum of energy (where nλ = 0 for both λ = 1, 2) for the effective Eq. (21) over all the in-plane modes. In this way we theory

    2 N Λ 2 Λ ~ X 2 X 1 X 2 X Ek(Λ) =  kj −  γ(κ) (ελ · K)  + ~ωe(κ). (77) 2me N j=1 κx,κy λ=1 κx,κy

In the energy expression above we introduced the cut- where we introduced the parameters off Λ which defines the highest allowed frequency that we e2 can consider in this effective field theory. Such a cutoff α = and ω2(κ ) = c2κ2 + ω2, (80) 2 e z z p is necessary for effective field theories and it is standard 4πc 0meLz to introduce it also for QED [10, 11]. The sum over the and the momentum κz = π/Lz (for nz = 1) depends on single mode coupling constant γ(κ) defines the effective the distance between the cavity mirrors Lz (see Fig. 1). coupling g(Λ) of the effective field theory. For the effec- Here comes a crucial point, the effective coupling g(Λ) tive coupling g(Λ) we have in Eq. (79) depends on the number of particles N. We would like to emphasize that the number of particles ap-

Λ Λ pears explicitly due to dipolar coupling, i.e. because in X e2N 1 X 1 g(Λ) = γ(κ) = . this effective field theory we couple all modes to all parti-  m L S ω2(κ) + ω2 cles in the same way. However, in QED beyond the dipole κ ,κ 0 e z κ ,κ p x y x y approximation, each mode has a spatial profile which di- (78) rectly implies that the coupling is local, in the sense that each mode couples to the local charge density and not In the limit where the area of the cavity becomes macro- to the full amount of electrons in the system. This is a scopic S → ∞, the momenta (κx, κy) of the photon field second point in which the effectiveness of our field theory become continuous variables and the sum gets replaced becomes manifest. This has implications because in the by an integral thermodynamic limit N → ∞ the effective coupling g(Λ) becomes arbitrarily large. Nevertheless, for the effective

Λ coupling g(Λ) we can derive rigorously conditions under e2N 1 ZZ dκ dκ  Λ  which the effective theory is stable and well defined. g(Λ) = x y = Nα ln 2 2 2 2 2 In section III we found the ground state of the electron- 0meLz 4π c κ + ωp ωe (κz) 0 photon system in the thermodynamic limit (with this (79) limit performed only for the electrons) for all values of 20

is known to exist for both relativistic and non-relativistic QED [10, 11, 13, 119, 123]. On the other hand the ef- fective coupling g(Λ) of our theory has no IR divergence because for arbitrarily small momenta κz = π/Lz the coupling goes to zero, g(Λ) → 0 due to the parameter α. The reason for which we have an IR divergent-free theory is the appearance of the diamagnetic shift ωp in Eq. (79) which defines the natural lower cutoff of our the- ory [55]. The diamagnetic shift appears due to the A2 term in the Pauli-Fierz Hamiltonian. Thus, we see that the diamagnetic term A2 makes non-relativistic QED IR divergent-free, while relativistic QED suffers from both UV and IR divergences. This is another fundamental reason for which the diamagnetic term A2 is of major importance. FIG. 11. Representation of the frequency domain of the pho- ton field in which the effective field theory is defined. The 2 natural lower cutoff of the theory is ωe (κz), while the highest allowed mode is Λ. B. Mode-Mode Interactions

For the sake of constructing an analytical effective field the single-mode coupling γ. Specifically we proved that if theory in the continuum, the mode-mode interactions in the coupling γ exceeds the critical coupling γ = 1 then c our treatment were neglected. The mode-mode inter- the system is unstable and has no ground state. Now actions are an important element of QED because they that we have promoted the single mode theory into an are responsible for non-linear effects for the electromag- effective field theory we need to guarantee the stability of netic field beyond the classical regime. However, as we the theory by forbidding the effective coupling to exceed can understand from the extensive treatment presented 1, 0 ≤ g(Λ) ≤ 1. From this condition and given the in appendix E, the mode-mode interactions do not alter definition of the effective coupling g(Λ) in Eq. (79) we fundamentally the energy spectrum for the 2DEG cou- find the allowed range for the cutoff Λ pled to the photon field. The mode-mode interactions 2 2 1/Nα ωe (κz) ≤ Λ ≤ ωe (κz)e . (81) shift the bare frequencies of the electromagnetic field and rotate the polarization vectors of the photon field From the expression above the highest allowed momen- 2 1/Nα (see also [103]). In a few-mode scenario these changes tum for the photon field is ωe (κz)e . Beyond this would be substantial modifications because the new nor- value the effective coupling g(Λ) becomes larger than 1 mal modes would be at different points in the photonic and the system gets unstable and the energy diverges. frequency space and would probe different parts of the In QED the finite momentum (or finite energy scale) electronic spectrum. for which the theory diverges is known as the Landau pole [119], and for that reason we will also refer here to 14 the highest allowed momentum as the Landau pole 2 modes 12 10 modes 2 1/Nα Λpole = ωe (κz)e . (82) 10 100 modes [%] l Moreover, from Eq. (81) it is clear that the cutoff Λ is 8 500 modes )|/ Ω 2 z 1000 modes a multiple of the dressed frequency ωe (κz) which means ( κ 6 that we can actually define Λ in terms of a dimensionless ˜ - ω

l 4

parameter Λ0 as | Ω 2 2 1/Nα Λ = ω (κz)Λ0 with 1 ≤ Λ0 ≤ e . (83) e 0 With this range chosen for Λ0 the effective coupling is 0.0 0.2 0.4 0.6 0.8 1.0 0 ≤ g(Λ) ≤ 1 and the system is stable and has a ground ω /ω state. p To complete this discussion on the construction of the FIG. 12. Relative percentage difference of the effective lower effective field theory, we would like to see what is the in- cutoff ωe(κz) and the lowest normal mode Ωl of the exact frared (IR) and the ultraviolet (UV) behavior of the field Hamiltonian (including the mode-mode interactions) plotted theory. From the expression for the effective coupling as function of the dimensionless ratio ωp/ω. The computa- g(Λ) in Eq. (79) it is clear that the effective coupling di- tion is performed with different amount of modes. Above 100 verges if we allow the cutoff to go to infinity, g(Λ) → ∞ modes the result is already converged. We assume all polar- for Λ → ∞, which means that our theory has a UV diver- izations to be parallel to maximize the effect of the mode- gence. This is the logarithmic divergence of QED which mode interactions. 21

However, in the continuum these shifts in the frequen- from the exact diagonalization performed in appendix E cies are not of great importance, because upon integrat- the exact coupling constant as a function of the amount ing over all the photon frequencies a continuous domain of modes M is of frequencies is spanned and all the modes within this M 2 X ωp(εα) domain are included (see Fig.11). The only difference g1D(M) = e , (86) will be on how far the frequency domain extends. Con- ex Ω2 α=1 α sequently, the effect coming from the mode-mode inter- actions is to modify the lower and upper cutoff of the where Ωα and εeα are the new normal modes and the new photon field. This means that the effective quantum polarization vectors including also the mode-mode inter- field theory is not fundamentally different to a quantum actions. The exact coupling constant can be computed field theory including the mode-mode interactions except using the exact diagonalization presented in appendix E. of a re-definition of the cutoffs. In addition, the upper In figure 13 we show the running of the exact coupling cutoff Λ in the effective theory is left open and can be constant as a function of the amount of photon modes freely adopted depending on how far we aim to probe M. the photon field energetically. As a consequence the only approximation in the effective field theory is that the 1.6 lower cutoff is assumed to be given by the expression q 2 2 2 ω(κz) = c κz + ωp. To test quantitatively the validity ωp /ω=0.005 e 1.4 of this approximation we compare the lower cutoff of the / ω ) ^2 ωp /ω= 0.05 p effective theory to the lowest normal mode Ωl resulting ω /ω= 0.5 from the exact diagonalization including the mode-mode 1.2 p 2

interactions due to the A term, which is shown in ap- ( M )/( ω pendix E. ex

1 D 1.0 In Fig. 12 we plot the relative percentage difference g between the effective lower cutoff ωe(κz) and exact lowest frequency Ωl, as a function of the dimensionless ratio 0.8 ωp/ω. As it is shown the effective cutoff and the exact 0 20 40 60 80 100 one differ by less than 10% from 0 until ωp/ω < 0.9. The relative difference exceeds 10% only for ωp > 0.9 ω. M(number of photon modes) This result shows that the effective field theory is valid 1D in the regime 0 ≤ ωp/ω ≤ 1. Further, we would like to FIG. 13. Exact many-mode coupling constant gex (M) nor- 2 mention that the regime 0 ≤ ωp/ω ≤ 1 is relevant for malized by (ωp/ω) in 1D, as a function of the number of experiments performed even in the ultrastrong coupling photon modes M. The exact coupling is plotted for three regime [56, 58]. We note that for ωp > ω the effective different values of the ratio ωp/ω corresponding to different theory can be easily corrected by replacing the effective regimes of light-matter interaction. In all cases we see that the exact coupling has arctangent dependence on the amount of cutoff ωe(κz) with the exact lowest normal mode Ωl. 2 Running of the Coupling in 1D.—To further test the photon modes. The exact coupling is normalized by (ωp/ω) such that all graphs to be visible in one plot. validity of our effective field theory and of the approxima- tion to neglect the mode-mode interactions we proceed by comparing the running of the coupling (in one dimen- As it is depicted in Fig. 13 the exact many-mode cou- sion) in the effective theory, to the exact coupling as a pling has an arctangent dependence on the amount of function of the upper cutoff of the photon field. The ef- photon modes M. This implies that the exact coupling fective coupling in one dimension is defined similarly to and the effective coupling exhibit the same running as a the effective coupling in the case of two dimensions (79) function of the upper cutoff of the photon field. This is very important because it demonstrates that the effec- Λ X tive field theory that we constructed in the continuum g1D(Λ) = γ(κ ) (84) z not only describes accurately the lower and upper ener- κz getic (or frequency) cutoffs of the photon field but also which upon introducing the fundamental cavity fre- captures the correct behavior of the coupling constant, quency ω = cπ/Lz and summing over all the photon which is of fundamental importance. This is a crucial momenta from 0 to Λ in the thermodynamic limit, is benchmark for our effective field theory. found to be 2 Z Λ   1D cωp dκz ωp cΛ g (Λ) = 2 2 2 = arctan . C. Renormalized & Effective Mass 2ω 0 c κz + ωp 2ω ωp (85) As it is known from relativistic QED when electrons From the above result we see that the effective coupling is interact with the full continuum of modes of the electro- an arctangent function of the upper cutoff Λ. Moreover, magnetic field the mass and charge of the electron get 22 renormalized. Such renormalizations are known to lead we find for the renormalized electron mass to observable radiative corrections like vacuum polariza-   −1 tion, the anomalous magnetic moment and the Lamb Λ me(Λ) = me 1 − α ln 2 . (88) shift [13, 120]. In non-relativistic QED there is no need ωe (κz) for charge renormalization, due to the elimination of From the expression above we see that the renormalized positrons from the theory [97]. However, mass renor- electron mass m (Λ) is larger than the electron mass in malization effects show up. Here, we are interested in e free space m and increases as function of the coupling the renormalization of the electron mass due the interac- e Λ. This behavior is in accordance with results coming tion of the electron with the continuum of modes of the from both relativistic and non-relativistic QED [13, 95– cavity. 97, 119]. Within the range of the cutoff Λ given by Eq. (81) the renormalized mass is always positive and the effective theory is well-defined (see Fig. 15). If the cutoff though goes beyond the Landau pole Λpole (which actually is a forbidden regime) the renormalized mass can become even negative and signifies that the theory becomes unstable, similarly to the single mode theory when the coupling coupling γ goes beyond the critical coupling γc. In the limit where the cutoff Λ takes its 2 minimum value ωe (κz) the renormalized mass me(Λ) is equal to me (see Fig. 15). This explains also why in the single mode theory the electron mass does not get renor- malized. Moreover, from Eq. (88) we see that the renormal- ized mass me(Λ) depends also on the distance between the cavity mirrors Lz (via α) and most importantly on the full electron density in the cavity ne via the FIG. 14. Schematic depiction of an electron in free space with dressed frequency ωe(κz) given by Eq. (80). The fact that m (Λ) depends on the full electron density n means that mass me and an electron coupled to the vacuum of the elec- e e tromagnetic field. The virtual photons of the cavity vacuum we can observe a many-body effect in the renormalized “dress” the electron and provide a radiative correction to the mass me(Λ). This many-body effect shows up because electron mass me(Λ). we consider here the many-body electron system of N free electrons coupled to the electromagnetic field and Generally computing the renormalized mass is a our treatment is non-perturbative. We emphasize that rather difficult task, in most cases performed perturba- such a many-body mass renormalization effect does not tively with methods ranging from dimensional regular- show up for the usual single particle mass renormaliza- ization [120], renormalization group techniques [13, 119, tion [120, 126] and is potentially very small for any finite 124] or causal perturbation theory [125]. system, but clearly not for extended systems. To the In non-relativistic QED the renormalized electron mass best of our knowledge such a many-body mass renormal- for free electrons is defined via the energy dispersion ization has not been demonstrated before. We note that of the electrons around k = 0 and is given by the for- the inclusion of the Coulomb interaction would result into mula [95, 96] further mass renormalization effects [127]. The renormalization of the electron mass due to the  2 −1 1 ∂ Ek(Λ) cavity field has experimental implications and can be me(Λ) = 2 2 (87) measured experimentally by comparing the effective mass ~ ∂ki of the electrons outside the cavity, to the effective mass where Ek(Λ) is the energy dispersion of the electron- inside the cavity (see also Fig. 16). The relation be- photon system, which depends on the momenta of the tween the two is given by the formula which we derived electrons and the cutoff of the theory. for the renormalized mass me(Λ) in Eq. (88). Having ob- In our case, we have diagonalized analytically the sin- tained experimentally the ratio me(Λ)/me the formula of gle mode Hamiltonian of Eq. (3) and then we promoted Eq. (88) allows us to deduce directly what is the high- the single mode energy spectrum given by Eq. (21) into est momentum (the cutoff) Λ to which the electrons are the energy spectrum of Eq. (77) which describes the ef- coupled to, and using Eq. (79) what is the coupling g(Λ) fective field theory we constructed. Since we have an to the cavity photons. analytic expression for the energy spectrum Ek(Λ) of the We believe this provides a novel direct way to deter- effective theory given by Eq. (77), for the computation of mine the light-matter coupling strength for extended sys- the renormalized mass we do not need to use any of the tems in cavity QED and the effective volume of the cavity. techniques we mentioned before, but we can straightfor- We elaborate on this further in section VII. wardly use the definition for me(Λ) of Eq. (87). Thus, Single-Particle Mass Renormalization in 3D.—As an 23

We note that the prefactor 2/3 is due to the fact that in 3D we have three spatial dimensions but only two pos- sible polarizations for the photon field. In 2D, the re- spective prefactor is equal to 1. In the thermodynamic limit the sum above turns into an integral. Moreover, in the single particle case N = 1 and the diamagnetic plasma frequency which depends on the electron density is negligible, ωp ≈ 0. Thus, for the effective coupling we find Λ 2 ZZZ 3 3D 2 e d κ 4αfs ~Λ g (Λ) = 3 2 2 = . 3 me0(2π) c κ 3π mec 0 FIG. 15. Renormalized mass me(Λ) as a function of the cut- (90) off Λ. For the cutoff Λ being equal to the natural lower cut- 2 off ωe (κz) the renormalized mass is equal to the free-space To obtain the above result we also introduced the fine 2 2 electron mass me(ωe (κz)) = me. As the cutoff increases the structure constant αfs = e /4π~c0. Having the expres- renormalized mass me(Λ) gets larger than the free-space mass sion for the effective coupling, we can straightforwardly me and eventually goes to infinity at the Landau pole Λpole. compute the renormalized electron mass according to the definition given in Eq. (87), and we find m Energy Dispersion m3D(Λ) = e (91) e 1 − g3D(Λ) which to the lowest order in the fine structure constant Metal in Free Space gives the following result for the renormalized mass Metal in Cavity k E 4α Λ m3D(Λ) = m + fs ~ . (92) e e 3π c The above result reproduces exactly the single-particle 0 0 mass renormalization for a free electron, which diverges k linearly to the upper cutoff of the photon field, as dis- cussed and obtained by several authors, for example FIG. 16. Energy dispersion for electrons in a metal outside Bethe [126], Hainzl and Seiringer [128] and Mandl and a cavity (black parabola) and for electrons in a metal cou- Shaw [120]. This results provides another importation pled to a cavity (yellow parabola). From the curvature of the validation of our effective field theory. parabolas the effective mass of the electrons in the respective environment can be obtained. The dispersion for the electrons Before we continue we would like to highlight another in the cavity is less steep, because the electron mass is larger important point which emerges from our effective field in the cavity due to the “dressing” by the cavity photons. theory in three dimensions. As we already discussed for our theory to be stable the effective coupling shall not exceed the value of one. Imposing this stability condition on g3D(Λ) given by Eq. (90) we find that the maximum additional test of our effective quantum field theory, and 3D our prediction for the many-body renormalization of the value for the upper cutoff Λpole is electron mass in the 2DEG, we will consider our effec-  −1 3D 4αfs ~ 15 −1 1 tive theory in 3D and compute the singe-particle mass Λpole = = 0.84 × 10 m ∼ renormalization, for which several analytic results ex- 3π mec lQCD ist [120, 126, 128]. (93) The solution for the free electron gas coupled to a cav- where l = /m c = 2.1×10−16m is the length scale of ity mode is the same also in three dimensions. Conse- QCD ~ p quantum chromodynamics (QCD), defined with respect quently, the energy spectrum in the effective quantum to the proton mass m , at which phenomena related to field theory will have the same form as the one in 2D p strong nuclear forces become important [129, 130]. Equa- given by Eq. (77). The only differences will be that the tion (93) shows that our effective quantum field theory is spectrum will depend on the 3D momenta of the elec- applicable for most regimes of light-matter interaction, trons, and that the effective coupling g3D(Λ) will be given and breaks down only at a scale inverse to the QCD by the sum of the single-particle coupling constants γ(κ) length scale l . This means, that our effective quan- over all 3D photonic momenta κ = (κ , κ , κ ) QCD x y z tum field theory is also able to predict the scale at which new phenomena take place, at the nucleus of the pro- 2 X e2N X 1 g3D(Λ) = γ(κ) = . (89) ton. Further, we would like to mention that since our 3 m  V c2κ2 + ω2 κ e 0 κ p theory is non-relativistic, we obtain a very much lower 24 value for our pole than the relativistic value of the Lan- chemical potential [131] dau pole [13, 119]. This is a beautiful consistency check 2 2 x ~ kF of our non-relativistic theory. k = µ + ~vF(k − kF) = + ~vF(k − kF), 2me(Λ) (97)

D. Modified Fermi Liquid Quasiparticle Excitations where vF is the Fermi momentum at the Fermi surface. Thus, the quasiparticle excitations in the Fermi liquid get Let us proceed by showing some further consequences also modified and depend on the renormalized electron of the effective field theory and more precisely implica- mass me(Λ). This demonstrates that the effective field tions for Fermi liquid theory. In section III we showed theory we constructed has direct implications for Fermi that the electronic ground state is one in which all single liquid theory. Lastly, we highlight that in the limit where 2 particle states with momenta less than the Fermi mo- the upper cutoff goes to the lower cutoff, Λ → ωe (κz), the renormalized mass goes to the bare electron mass, mentum pF = ~kF are occupied. All other single par- ticle states are empty. This is the starting assumption me(Λ) → me, and in this case no modification of the in Fermi liquid theory [6, 7]. The fundamental fermionic Fermi liquid shows up. This explains from another point quasiparticle excitations of the Fermi liquid theory are of view why in the single-mode theory the Fermi liquid generated by adding electrons with momentum greater does not get modified and why a field theory in the con- than the Fermi momentum pF [7, 131]. The energy of tinuum for the photon field is necessary to see a modifi- the quasiparticle at the Fermi surface is cation of the Fermi liquid.

µ = Ek(Λ,N + 1) − Ek(Λ,N), (94) E. Jellium Model & Coulomb Interaction where Ek(Λ,N) is the ground state energy of the system for N electrons distributed on the Fermi sphere with their To further illustrate the applicability of the our effec- wavevectors in the region 0 ≤ k < kF and Ek(Λ,N + 1) tive field theory and discuss its implications for electronic is the energy of the system containing one more particle systems interacting also via the Coulomb interaction we with k = kF. In the ground state where the electrons will consider the jellium model, which provides a first ap- are distributed on the Fermi sphere the collective mo- proximation to a metal or a plasma [3, 132, 133]. The mentum is zero, K = 0. Thus, in the energy Ek(Λ,N) jellium model is an interacting electron gas placed in a the negative term appearing in the effective spectrum in uniformly distributed positive background representing Eq. (77) does not contribute. However, in the N +1 state the ions, chosen to ensure the neutrality of the full sys- the last electron added on the Fermi surface with k = kF tem. Due to the positive background, the Hamiltonian of introduces a non-zero momentum to the system. Conse- the jellium model can then be written as the sum of the quently, the photon-mediated negative term now gives a kinetic energy of all the electrons plus the regularized, by contribution. Thus, we find that the quasiparticle exci- the positive background, Coulomb interaction. We note tation energy at the Fermi surface is that this regularization is very important as it eliminates a particular divergent contribution of the Coulomb inter- 2 "   2 # Λ X 2 action [3, 132, 133]. µ = ~ k2 − α ln (ε · k ) . (95) F 2 λ F However, in the effective field theory that we con- 2me ωe (κz) λ=1 structed we do not have only the homogeneous back- To obtain the above result we used that the effective cou- ground of the ions, but we also have the neutral homo- 2
geneous background of the photons in which the elec- pling per particle is g(Λ)/(N + 1) = α ln Λ/ωe (κz) as given by Eq. (79). Further, using that the polarization trons are embedded. As we already saw in the previous vectors are orthogonal and introducing the renormalized subsections, the photonic background renormalizes the electrons and introduces fermionic quasi-particles with mass me(Λ) the quasiparticle excitation at the Fermi sur- face takes the compact form an effective electronic mass me(Λ) given by Eq. (88). This makes clear that the jellium model in our effec- 2 2 tive field theory shall be one consisting of the kinetic ~ kF µ = . (96) energy of these fermionic quasi-particles with effective 2me(Λ) mass me(Λ), interacting via the regularized Coulomb in- The quasiparticle excitation at the Fermi surface µ is teraction. Thus, the jellium model (in 2D) in our effective also known as the chemical potential. From Eq. (96) we field theory is given by the following Hamiltonian see that the chemical potential depends on the renormal- 2 N h i ized (by the photon field) electron mass m (Λ) given by ˆ ~ X 2 1 X ˆ e Hjell(Λ) = − ∇j + vq nˆ−qnˆq − N 2me(Λ) 8π0S Eq. (88). This shows that in the effective field theory j=1 q6=0 the photon field modifies the chemical potential. More- | {z } | {z } ˆ ˆ over, in Fermi liquid theory the quasiparticle excitations T (Λ) We-e in the neighborhood of the Fermi surface depend on the (98) 25 which depends parametrically on the upper cutoff Λ of density [3] the photon field, via the effective mass. We note that 1 in the jellium Hamiltonian S = L2 is the area of the 2D 1  1  2 2 rs = . (99) electron gas, vq = 2πe /|q| is the Fourier transform of the a0 πn2D P −iq·rj regularized Coulomb interaction in 2D,n ˆq = j e is the Fourier transform of the electronic density operator We note that the parameter rs is known as the Wigner- Seitz radius [2] and that the length r a is the radius and Nˆ =n ˆ0 is the number operator [3]. s 0 enclosing on average one electron. With respect to the natural length rsa0 we can define the following scaled The crucial step to treat the Coulomb interaction, variables [3, 132, 133] comes from the observation that in this system there ex- r S 2 2 r = , q = rsa0q and Se = . (100) ists a natural length scale rsa0, where a0 = 4π0~ /mee e r a e (r a )2 is the Bohr radius, with respect to which the kinetic s 0 s 0 energy and the Coulomb energy scale differently. The With these definitions the jellium Hamiltonian takes the parameter rs is defined with respect to the 2D electron form

  2 e me 1 X 1 X 2π h i Hˆ (Λ) = ∇2 + nˆ nˆ − Nˆ . (101) jell  2 e j −qe qe  2a0 me(Λ) rs rsSe |qe| j qe6=0

We note that in the last step we introduced also the stan- parameter. The total energy per particles is of course dard convention for the jellium model in which 4π0 = 1 the sum of these two contributions E(rs, Λ) = τ(rs, Λ) + which implies that the energy is measured now in Ryd- x(rs) and is a function of the Wigner-Seitz radius rs. 2 berg units, 1Ry = e /2a0 [3, 132]. From the expression of Minimizing the energy with respect to rs, we find that the the jellium Hamiltonian in terms of the scaled variables minimum Wigner-Seitz radius for the interacting electron it becomes evident that with respect to rs the kinetic gas is energy and Coulomb energy scale differently. As a con- 3π me sequence, in the regime of small rs, or equivalently large (rs)min(Λ) = √ (104) 4 2 me(Λ) densities, the Coulomb interaction can be considered as perturbation to the kinetic contribution. Thus, in the which is a function of the upper cutoff Λ. Since me(Λ) large density regime the solution of the non-interacting is larger than me (see Eq. (88) or Fig. 15) the Wigner- electron gas is perfectly valid and we can use the many- Seitz radius for the interacting electrons coupled to the body eigenfunctions we obtained in section II, to com- photon field is smaller than the uncoupled. This implies pute the Coulomb contribution perturbatively. We note that when the interacting electrons are coupled to the that in our photon-modified jellium model, the kinetic photon field the radius containing one electron on aver- contribution depends on the ratio between the electron age becomes smaller. This means that the photon field mass inside and outside the cavity me/me(Λ). This ratio localizes the electrons, in the sense that (on average) an is what determines the electron-photon coupling in the electron occupies a smaller volume in space. This is a effective field theory. significant result because most electronic-structure prop- First, we compute the kinetic energy per particle erties depend on the Wigner-Seitz radius and the average with respect to the ground state of our system given electron density. We note that such localization effects on by Eq. (33). Due to the fact that the electrons are dis- the electronic density have also been reported for atomic tributed on the 2D Fermi sphere we find [3] and molecular systems under strong coupling to a cavity, with first-principle calculations [37, 38]. ˆ 2 hΨgs|T (Λ)|Ψgsi e me 1 We would like to mention that for the upper cutoff τ(rs, Λ) = = . (102) N 2a m (Λ) r2 2 0 e s being equal to the lower cutoff, Λ = ωe (κz), me(Λ) be- As we see the kinetic energy per particle is modified by comes equal to me and the photon modified jellium goes the photons because it depends on the ratio me/me(Λ). to the standard jellium Hamiltonian and we recover all Then, to first order in perturbation theory the contribu- the respective known results of the jellium model [3]. tion of the Coulomb energy per particle is [3] Beyond the First Order Coulomb Contribution.—It is √ 2 important to emphasize that in second or higher order hΨgs|Wˆ e-e|Ψgsi 8 2 e 1 x(rs) = = − . (103) contributions, the excited states of the electron-photon N 3π 2a0 rs system will contribute to the Coulomb energy. The ex- The above contribution is also known as the exchange cited states of our system, as can be easily seen from energy and as we see it does not depend on any photonic Eq. (20), are correlated states between the electrons and 26 the photons and as a consequence the photonic states will field. The zero point energy of the electromagnetic field it contribute to the higher orders and modify the correla- is known to be responsible for forces like the interatomic tion energy. This computation of the correlation energy van der Waals forces, the Casimir-Polder forces between is non-trivial and needs a separate treatment. an atom and a body 4 [99, 121], and the Casimir force be- Beyond the Large Density Regime.—It is worth to tween parallel conducting plates [98]. Since we consider a mention that the perturbative treatment we performed 2D material in a cavity we fall in the third category and here has certain limitations and it is not straightfor- the macroscopic forces in the system should be Casimir wardly applicable in the intermediate and the low den- forces. To find the Casimir force between the mirrors sity regimes [133]. The intermediate density regime is of the cavity we need to compute the zero point energy the realm of Landau’s Fermi liquid theory [6, 131] which of the electromagnetic field Ep per area S of the cavity is based on the concept of adiabatic continuity. Under mirrors. From the energy expression of the effective the- this assumption, the non-interacting ground state evolves ory in Eq. (77) we deduce that the ground state energy smoothly into the interacting one, without ruining the ex- (nλ = 0) per area is istence of a well-defined Fermi surface and well-defined Λ quasi-particle excitations. The interaction between these ZZ Ep 1 X q quasi-particles is then usually treated with perturbative = ω(κ) = ~ dκ dκ c2κ2 + ω2, S S ~e 4π2 x y p Green’s functions methods. In the low density regime, κx,κy 0 the many-body ground state is no longer the one of the (105) non-interacting (or weakly interacting) electrons, as the system it is believed to undergo a phase transition in where we also took the limit S → ∞ in which the sum which a Wigner crystal is formed [3, 132, 133]. This is gets promoted into an integral. Going now to polar coor- the regime in which the free electron gas does not provide dinates and performing the integral we obtain the result a good starting point for the many-body ground state. for the photon energy per area

E (Λ3/2 − 1) p = ~ 0 ω3(κ ). (106) F. Repulsive Casimir Force for a Non-Empty 2 e z Cavity S 6πc Using the expression for ωe(κz) given by Eq. (80) and Having defined and constructed the effective field the- taking the derivative of the photon energy per area Ep/S ory for the continuum of modes, we want to proceed by with respect to the distance of the cavity mirrors Lz we computing the zero point energy of the electromagnetic find the force per area (the pressure)

s 3/2  2 2 2  2 2 2 Fc ∂(Ep/S) ~(Λ0 − 1) 2π c e n2D π c e n2D = − = 2 3 + 2 2 + . (107) S ∂Lz 4πc Lz me0Lz Lz me0Lz

We note that to obtain the above result we also took into G. Absorption and Dissipation in the Effective account the dependence of ωp to the distance between Field Theory the cavity mirrors Lz, as given by Eq. (5). The force (or pressure) above describes the force that the paral- In section V we performed linear response for the elec- lel plates of the cavity feel due to the zero point energy tronic and the photonic sectors of the theory, in the single of the photon field of the interacting hybrid system in mode case. Our goal now is to study the linear response the cavity. The force given by Eq. (107) is positive be- behavior of the effective theory we constructed, and to see 3/2 cause Λ0 ≥ 1. This indicates that the Casimir force how the response functions get modified by the infinite is repulsive. The possibility of repulsive Casimir forces amount of in-plane modes. Here, we focus on the linear has been discussed in many different settings [134–138] response in the photonic sector, which as we showed in and has even been experimentally observed for interact- subsection V E is adequate for the description of all the ing materials immersed in a fluid [100]. In our case we do resonances of the system. not have a fluid between the cavity mirrors but a 2DEG To perturb the photon field we apply an external time- which interacts with the cavity field. dependent current Jext(t) which couples to the quantized cavity field, as shown in Fig. 5. Thus, we consider the external perturbation Hˆext(t) = −Jext(t) · Aˆ as we did in section V B. The external current is chosen to be in the x-direction Jext(t) = exJext(t). The vector potential 4 By body here we mean a macroscopic object. in the effective theory is the sum over all the in-plane 27

† modes satisfy bosonic commutation operators [ˆcκ, cˆκ0 ] = δκκ0 0 1 ∀ κ, κ . With the definition of the effective Hamiltonian   2 ˆ ~ X ex †
the full time dependent Hamiltonian under the external A = p aˆκ +a ˆκ , (108) 0V 2ω(κ) perturbation is Hˆ (t) = Hˆeff − Jext(t) · Aˆ . The vector κx,κy potential in terms of the renormalized annihilation and where we only kept the polarization in the x-direction, creation operators of Eq. (16) is because it is the only one that couples to the external perturbation. To perform linear response we need to in- 1   2 troduce and define the Hamiltonian of the effective the- ˆ ~ X ex †
ˆ A = p cˆκ +c ˆκ . (111) ory Heff. For the effective Hamiltonian it is not necessary 0V 2ω(κ) κx,κy e to give a particular expression in terms of electronic and photonic operators. The effective Hamiltonian can be defined also by giving a definition of the ground state With these definitions we can define all operators ˆ of Hˆeff and its excited states. The ground state of the of the theory in the interaction picture as OI (t) = itHˆ / −itHˆ / effective Hamiltonian Hˆeff we define it as e eff ~Oˆe eff ~ and the wavefunctions respectively as itHˆ / Y ΨI (t) = e eff ~Ψ(t). Here, we are interested in the A- |Ψgsi = |Φ0i ⊗ |0, 0iκx,κy (109) A 0 field response function χA(t−t ) which is defined through κ ,κ x y Eq. (49). Substituting the expression for Aˆ given by where |Φ0i is ground state of the electronic sector given Eq. (111) and using the fact that the effective Hamil- by the Slater determinant in Eq. (14), with the electrons tonian is a sum of non-interacting modes and that the distributed on the 2D Fermi sphere (see Fig. 3), which ground state |Ψgsi of the effective theory in the thermo- P dynamic limit is a tensor product of the photonic states consequently have zero total momentum K = j kj = 0. of all the modes, we find the response function to be the Furthermore, the set of states |0, 0iκ ,κ get annihilated x y sum of all the single mode response functions given by by the operatorc ˆκ,c ˆκ|0, 0iκx,κy = 0, ∀ κ. Having the ground state we can define the excited states of the Eq. (52) † system by applying the creation operatorsc ˆκ on it. Thus, ˆ 0 0 we find that the excited states of Heff satisfy the equation X Θ(t − t ) sin(ω(κ)(t − t )) χA(t − t0) = − e .(112) A  V ω(κ) † m    † m κ ,κ 0 e ˆ (ˆcκ) 1 (ˆcκ) x y Heff √ |Ψgsi = Ek + ~ω(κ) m + √ |Ψgsi m! e 2 m! (110) Since the response function in time is the sum of the single mode responses, also the response function in the P 2 2 where Ek = kj /2me is the kinetic energy of the A j ~ frequency domain χA(w) is the sum of all the single mode † electrons. We also note that the operators {cˆκ, cˆκ0 } response functions given in appendix B

  A X −1 1 1 χA(w) = lim − . (113) 20ω(κ)V η→0 w + ω(κ) + iη w − ω(κ) + iη κx,κy e e e

In the thermodynamic limit the above sum turns into pendix D we find the analytic expressions for the real an integral and following the derivation shown in ap- and the imaginary parts of the A-field response function A χA(w) for the effective field theory

1  (w − ω(κ ))2 + η2  (w + ω(κ ))2 + η2  A e√ z e√ z <[χA(w)] = 2 ln + ln and (114) 8πc 0Lz (w − Λ)2 + η2 (w + Λ)2 + η2 √ √ " !     !# A 1 −1 Λ + w −1 ωe(κz) + w −1 ωe(κz) − w −1 Λ − w =[χA(w)] = 2 tan − tan + tan − tan . 4πc 0Lz η η η η

If we take now the limit η → 0 for the artificial broaden- ing η, we find for the imaginary part  1 √ , for − Λ < w < −ω(κ )  2 e z 4c 0Lz  A  =[χA(w)] = 1 √ (115) − , for ω(κ ) < w < Λ  2 e z  4c 0Lz 0 , elsewhere. 28

From the expression above we see that the imaginary there are now two scales in the system, the natural lower A part =[χA(w)] is well-defined in the limit η → 0 for all cutoff ωe(κz) and upper cutoff of the effective field theory w without any divergences appearing. This is in con- as also shown in Fig. 11. A trast to =[χA(w)] of Eq. (53) in the single-mode theory Lastly, we would like to highlight that in the large which was divergent for w = ±ωe. Further, the imaginary N,S limit the imaginary and the real parts of the re- part in Eq. (115) takes a constant value in the region A √ sponse function χA(w) of Eq. (114) have a well-defined ωe(κz) < |w| < Λ and is zero everywhere else, as shown finite value and do not vanish. This is in contrast to the also in Fig. 17. This means that our system can absorb single mode response function given by Eq. (53). This energy continuously with the same√ strength in the fre- shows again that by going to the continuum of modes quency window ωe(κz) < |w| < Λ. This is because it is and constructing this effective field theory, the photon exactly this frequency range in which the effective field observables become well-defined and have a substantial theory is defined, see Fig. 11, and all modes are excited contribution to the macroscopic 2DEG in the cavity. by the external current with the same strength. The fact that the imaginary part is well-defined and does not diverge means that absorption can be consis- VII. SUMMARY, EXPERIMENTAL tently described in the effective field theory and the ab- IMPLICATIONS & FUTURE DIRECTIONS sorption rate W of Eq. (54) is well-defined and can be computed properly. This proves our claim that by con- Summary.—In this article we investigated Sommer- structing a theory of infinitely many modes in the con- feld’s theory of the free electron gas [1] in the framework tinuum we can indeed describe absorption processes and of cavity QED. In the long-wavelength limit (or dipole dissipation from first-principles, without the need of the approximation) and in case where the quantized cavity artificial broadening η, and without having to introduce field consists of a single mode we showed that the sys- some kind of environment for our system. This demon- tem is analytically solvable. This allowed us to perform strates that a system with its photon field works like its the thermodynamic limit for the electrons, in which the own heat bath [3], and more precisely in our case the ground state k-space distribution of the electrons is the continuum of modes describes the full photon bath. 2D Fermi sphere. This means that our system is a Fermi liquid. In addition, we showed that the hybrid electron- photon ground state contains virtual photons. Moreover, we provided the full phase diagram of the interacting [ A(w)] A electron-photon system for all possible couplings, and we A [ A(w)] found that when the coupling reaches its maximum value (critical coupling) the ground state becomes infinitely de- generate. Such an infinite degeneracy appears also for Landau levels [108] in the integer quantum Hall effect [8]. 0 This fact hints towards a novel connection between QED and the semiclassical theory. Beyond the critical coupling the system has no ground state and equilibrium is not well-defined. The non- existence of a ground state also occurs if the diamagnetic A2 term is neglected. This is in stark contrast to the ( z) 0 ( z) finite-system models of quantum optics, like the Rabi w and the Dicke model, in which a ground state always A A exists. This demonstrates that extended systems indeed FIG. 17. Real <[χA(w)] and imaginary =[χA(w)] parts of the A behave very much differently than finite systems and that A-field response function χA(w) in the effective field theory the well-known (finite-system) models of quantum optics with η = 0. The imaginary part has√ a finite value within the might not be straightforwardly applicable for materials frequency window ωe(κz) < |w| < Λ which indicates that in this frequency range the system can continuously absorb in cavity QED. We believe this result clarifies further the 2 energy. The real part though diverges at the√ natural lower ongoing discussion about whether the diamagnetic A cutoff w = ±ωe(κz) and the upper cutoff w = ± Λ and shows term can be neglected or not [76–79]. The elimination of that the system in the effective field theory has two scales. the diamagnetic A2 term it is known to be responsible for the notorious superradiant phase transition [80–82]. A The real part <[χA(w)] though for η =√ 0 diverges at Then, we performed linear response for the 2DEG the frequencies w = ±ωe(κz) and w = ± Λ and gives in the cavity and we introduced the four fundamen- us information about the resonances of the system. In tal response sectors: the matter-matter, photon-photon, the single mode case, in section V B, there was only one matter-photon and photon-matter. In addition, we resonance appearing at frequency w = ±ωe, while now demonstrated that all response sectors are equivalent we have two resonances at the frequency√ of the plasmon- with respect to their pole structure and that their polariton ωe(κz) and the cutoff Λ. This indicates that strengths are related via the electron-photon coupling (or 29 the number of particles). All responses show plasmon- already been observed [61, 66]. We believe our work polariton resonances which modify the conductive and can provide further insights into these experiments radiation properties of the 2DEG. and motivate new directions to be explored. To bridge the discrepancy between the electronic sec- tor, in which the energy density is finite, and the pho- • Measurement of the Effective Mass, Coupling and tonic sector, in which the single mode energy density is Cutoff.—Metals in solid state theory are described zero, we promoted the single mode theory into an effec- in most cases using the free electron model [2] in tive field theory in the continuum, by integrating over all which the energy dispersion of the electrons is de- 2 2 ∗ the in-plane modes of the photon field. In this way the scribed by a parabola Ek = ~ k /me with some ∗ energy density of the photon field becomes macroscopic effective mass me for the electrons. Measuring the and induces a radiative correction [13, 119, 120] to the dispersion of the electrons by angle-resolved pho- electron mass and renormalizes it [95–97]. The renormal- toemission spectroscopy (ARPES) [140] one can ∗ ized mass depends on the full electron density ne in the obtain the effective mass me . The effective mass cavity. To the best of our knowledge such a many-body appears because in a metal electrons are not com- effect has not been reported so far. This is a special fea- pletely non-interacting, but there are Coulomb in- ture of the 2DEG due to its macroscopicity. Further, the teractions and the potential of the ions which mod- renormalized electron mass modifies the chemical poten- ify the free electron behavior [127, 141, 142]. This tial of the system and the fermionic quasiparticle excita- pictures indicates that the mass of the electron is tions of the Fermi liquid. The concept of the fermionic not purely inherent but depends on its environ- quasi-particles of renormalized mass allowed us to intro- ment. duce within our effective quantum field theory a jellium In our case the 2DEG is coupled to a cavity. Thus, model for these quasi-particles and to include perturba- the environment of the electrons includes also the tively the Coulomb interaction. In this model the photon interactions with the photon field. Consequently, field shrinks the Wigner-Seitz radius which implies a lo- if one measures the energy dispersion of the elec- calization effect on the electrons. Moreover, the energy trons in a metal confined in a cavity, one should density of the photon field makes itself manifest by pro- find a parabolic dispersion but with a different ef- ducing a Casimir force [98, 99, 121, 122] between the fective mass (see also Fig. 16). The contribution mirrors of the cavity, which is repulsive due to the light- of the cavity photons to the effective mass of the matter coupling. Then, we performed linear response in electrons in the metal is given by the expression of the effective field theory and we showed that due to the the renormalized electron mass me(Λ) of Eq. (88). continuum of photon modes we are able to describe dissi- pation and absorption processes without the need of any We propose that the renormalized electron mass artificial damping parameter or having to introduce an due to the cavity photons can be measured by com- environment for the system. paring the effective masses outside and inside the Experimental Implications.—To a large extent this cavity. Furthermore, from such an experimental work is motivated by the great experimental progress in measurement of the effective electron mass and the cavity QED materials [56, 58, 59, 61, 139] and polari- formula in Eq. (88) one can deduce directly the tonic chemistry [39–43]. We believe that several of the cutoff Λ. The cutoff Λ gives the highest frequency results presented throughout the article are measurable (or momentum) with which the electrons interact and have experimental implications. So let us elaborate inside the cavity. Having the expression for the a bit further on the main ones. effective coupling g(Λ) given by Eq. (79) we can obtain also the coupling strength between the elec- • Cavity Modified Conductivity.—In section V C we trons and the photons in the cavity. In most cases computed the optical conductivity σ(w) of the for finite systems the light-matter coupling strength 2DEG in the cavity, given by Eq. (69). The is defined via the Rabi (or Dicke) model [25] and standard optical conductivity of the free electron the corresponding Rabi split. Our theory provides gas gets modified by the appearance of plasmon- a novel way to measure the electron-photon cou- polariton resonances which show up in both the real pling for extended systems in cavity QED, which and the imaginary part of σ(w) (see Figs. 9 and 10). goes beyond standard quantum optics models, via Since σ(w) gets modified by the cavity field this im- the effective mass. plies that also the dielectric function (w) will be modified as well (as (w) = 0+iσ(w)/w [2]). These • Modified Casimir Forces.—As it is known since the modifications can be observed with optical trans- seminal works of Casimir and Polder [98, 99] macro- mission measurements. In addition, we showed scopic vacuum forces can emerge due to the vac- that in the static limit the DC conductivity and cuum energy of the electromagnetic field between the Drude peak get suppressed due the coupling to perfectly conducting plates, like for example two the cavity field (see Eq. (72)). We would like to cavity mirrors. In section VI F we computed the mention that such modifications of the conduction Casimir force due to the vaccuum energy of the properties of 2D materials confined in cavities have 2DEG in the cavity, and we found that due to the 30

light-matter coupling the Casimir force gets mod- veloping local-density-type approximations for the ified by the electron density ne and turns out to light-matter interactions. be repulsive. Such repulsive vacuum forces have been reported in the case of a cavity immersed in • Superradiance.—Superradiance as predicted by a fluid [100]. Our theory provides an example of Dicke [26] is the enhancement of spontaneous emis- a such a repulsive force and a first-principles ex- sion due to the collective coupling of emitters. In planation on how such repulsive vacuum forces can addition, an equilibrium superradiant phase transi- emerge, due to strong light-matter interaction, and tion was also predicted for the Dicke model [80, 81], opens new pathways for manipulating and engi- which since then has triggered an ongoing de- neering Casimir forces in cavity QED. bate [82–86]. In both cases these phenomena emerge due to the collective coupling of many par- Future Directions.—The presented theory has many ticles or dipoles to the quantized photon field. The implications. Besides the ones we have pointed out so theory we presented here involves a large number of far, we would like to mention a few further research direc- electrons coupled to the quantized field of a cavity tions which are potentially interesting and to our opinion and we believe will serve as a new playground for worthwhile to pursue. the investigation of superradiance and other collec- tive phenomena in cavity QED. • Coulomb Interaction and Fermi Liquid Theory in QED.—In this work the Coulomb interaction was only treated perturbatively, as a first-order pertur- ACKNOWLEDGMENTS bation within the jellium model, and a particu- lar modification of the jellium model due to the We would like to thank M. A. Sentef, D. Welakuh, photon field was presented. The perturbative ap- M. Penz, I. Theophilou, J. Faist, M. Altarelli and proach is valid in the large density regime and can S. Y. Buhmann for useful discussions. This work was sup- be further extended to include the correlation en- ported by the European Research Council (ERC-2015- ergy of the electron-photon system in which the AdG694097), the Cluster of Excellence “Advanced Imag- photonic states will introduce non-trivial effects. ing of Matter” (AIM), Grupos Consolidados (IT1249-19) Moreover, for the regime of intermediate densities and SFB925 “Light induced dynamics and control of cor- the paradigmatic theory is the Fermi liquid the- related quantum systems”. The Flatiron Institute is a ory [6] in which the interacting electronic system is division of the Simons Foundation. described using fermionic quasi-particles, and the Coulomb interaction is treated using Green’s func- tion techniques [7]. What defines a Fermi liquid is Appendix A: Linear Response in the Photonic the fact that electrons in k-space are distributed Sector on the Fermi sphere. In section III we showed that this is the case also for the 2DEG coupled to a The aim of this appendix is to give the details of the cavity. Further, we showed how the photon modi- linear response computations, in the photonic sector, of fies the quasiparticle excitations of the Fermi liquid. the subsection V B. In subsection V B we perturbed our Thus, we believe that a theory for materials in cav- system with external time-dependent current by adding ˆ ity QED can be constructed along the lines of the to the Hamiltonian of Eq. (3) the perturbation Hext = Fermi liquid theory. −Jext·Aˆ . With this perturbation the full time-dependent Hamiltonian is given by Eq. (51). The external current is LDA functional in QEDFT.—The local density ap- • chosen to be in the x-direction, Jext(t) = exJext(t), and proximation within DFT [4] is one of the most in this case only the x-component of the vector potential successful methods for the computation of prop- Aˆ is of interest. Following the standard linear response erties of materials [5]. Recently, a generalization formalism [91, 92], which we introduced in the subsec- of DFT in the framework of QED has been intro- tion V A, the response of any observable Oˆ due to this duced [30, 31] and has already been applied [32, 92]. perturbation is The original LDA was constructed from the ana- Z t lytic solution of the free electron gas. In this work i 0 0 δhOˆ(t)i = − h[OˆI (t), Aˆ I (t )]i(−Jext(t )). (A1) we solved exactly the free electron gas in cavity ~ 0 QED and this gives the opportunity to construct ˆ 0 an LDA functional in the framework of QEDFT. If where AI (t ) is the quantized vector potential in the in- ˆ ˆ 0 such an LDA-type functional in QEDFT shares the teraction picture and the correlator h[OI (t), AI (t )]i is success of the original LDA, then we would have a defined with respect to the ground state |Ψgsi of the un- really powerful new tool for the description of ma- perturbed Hamiltonian. From the previous expression terials in the emerging field of cavity QED. Our re- the respective response function is sult on the shrinking of the Wigner-Seitz radius in 0 O 0 iΘ(t − t ) ˆ ˆ 0 the jellium model, can be potentially helpful in de- χA (t − t ) = − h[OI (t), AI (t )]i. (A2) ~ 31

† In section III we found the ground state |Ψgsi of the the annihilation and creation operators {cˆλ, cˆλ}. From unpurturbed Hamiltonian Hˆ in the thermodynamic limit Eqs. (10) and (16), and for K = 0 (which is true in the to be given by Eq. (33). Having |Ψgsi we can compute ground state) we find for the quantized vector potential the response function for any observable Oˆ. To obtain 1   2 the response function χO(t − t0) of Eq. (A2) we need to ˆ ~  † A A = cˆ1 +c ˆ1 ex. (A5) compute the commutator 20ωV˜

0 0 0 ∗ ˆ h[OˆI (t), Aˆ I (t )]i = hOˆI (t)Aˆ I (t )i − hOˆI (t)Aˆ I (t )i . Applying now A to the ground state |Ψgsi we have (A3) 1   2 iE (t−t0) ˆ ˆ 0 ~ 0,k hOI (t)AI (t )i = e ~ × where we used also the hermiticity of the opera- 20ωV˜ ˆ ˆ 0 ˆ 0 ˆ ˆ 0 tor OI (t)AI (t ) which implies that hAI (t )OI (t)i = ˆ −iH(t−t ) 0 ∗ ×hΨgs|Oe ~ |Φ0i ⊗ |1, 0i1|0, 0i2. (A6) hOˆI (t)Aˆ I (t )i . Thus, we only need to compute the cor- ˆ ˆ 0 relator hOI (t)AI (t )i. Using the definition for the oper- From the expression above we see that quantized field Aˆ ators in the interaction picture and the fact that |Ψgsi gets the ground state to the first excited state for n1 = 1. ˆ is the ground state of the Hamiltonian H, which means The state |Φ0i ⊗ |1, 0i1|0, 0i2 is the first excited state of ˆ 0 0 −iHt / −iE0,kt / ˆ that e ~|Ψgsi = e ~|Ψgsi, we find H with eigenenergy E1,k = Ek + 2~ωe. Using this we find

0 1 iE0,k(t−t ) −iHˆ (t−t0) 0   2 0 hOˆ (t)Aˆ (t )i = e ~ hOˆe ~ Aˆ i. (A4) 0 ~ −iω(t−t ) I I hOˆI (t)Aˆ I (t )i = e e × 20ωV˜ Where E0,k = Ek+~ω is the ground state energy given by e ×hΨgs|O|ˆ Φ0i ⊗ |1, 0i1|0, 0i2. (A7) Eq. (21), with nλ = 0 for both λ = 1, 2. To continue we need to apply the vector potential Aˆ to the ground state From the previous result we obtain the following expres- |Ψgsi. For that we need the expression of Aˆ in terms of sion for the commutator of Eq. (A3)

1   2 ∗ 0 ~ h −iω(t−t0) iω(t−t0)   i h[OˆI (t), Aˆ I (t )]i = e e hΨgs|O|ˆ Φ0i ⊗ |1, 0i1|0, 0i2 − e e hΨgs|O|ˆ Φ0i ⊗ |1, 0i1|0, 0i2 . 20ωV˜ (A8)

A 0 The formula above is very important because it applies to and performing a Laplace transform for χA(t − t ) we any observable Oˆ and we will use it for the computation find the response of the A-field in the frequency domain A of several different response functions. χA(w)   A −1 1 1 χA(w) = lim − . Appendix B: A-Field Response Function 20ωVe η→0+ w + ωe + iη w − ωe + iη (B3) Having derived Eq. (A8) we will use this formula to compute the response function χA(t − t0) for the quan- A Appendix C: E-Field Response Function tized vector potential Aˆ . From Eq. (A8) it is clear that all we have to compute is hΨgs|Aˆ |Φ0i ⊗ |1, 0i1|0, 0i2. Using ˆ Now we would also like to compute the response of the Eq. (A5) which gives the A-field in terms of the operators electric field due to the external time-dependent current † {cˆ1, cˆ1} we find Jext(t). The electric field in dipole approximation, in the 1 x-direction, is [101]   2 ~ hΨ |Aˆ |Φ i ⊗ |1, 0i |0, 0i = . (B1) 1 gs 0 1 2   2 20ωVe ˆ ~ω  † E = i aˆ1 − aˆ1 ex. (C1) Combining the result above with Eqs. (A8) and Eq. (A2) 20V A 0 we find the response function in time χA(t − t ) To make use of Eq. (A8) we need to write the electric † 0 0 field in terms of the operators {cˆ1, cˆ }. Using Eqs. (7) A 0 Θ(t − t ) sin(ωe(t − t )) 1 χA(t − t ) = − . (B2) and (16) we find for the electric field 0ωV e 1   2 The response function above is also the propagator of the ˆ ~ωe  † E = i cˆ1 − cˆ1 ex. (C2) A-field. Making use of the integral form of the Θ-function 20V 32

Substituting the expression for the electric field opera- Moreover, we can also deduce the real and the imaginary E tor into Eq. (A8) and then using the definition of the parts of χA(w) response function in time given by Eq. (A2) we find the   E 0 E η 1 1 response function χA(t − t ) <[χA(w)] = 2 2 − 2 2 , 20V (w + ωe) + η (w − ωe) + η 1  w + ω w − ω  =[χE(w)] = e − e A 2 V (w + ω)2 + η2 (w − ω)2 + η2 0 0 0 e e E 0 Θ(t − t ) cos(ωe(t − t )) χA(t − t ) = . (C3) (C5) 0V

Appendix D: Computation of the Response Functions in the Effective Field Theory From the response function in time by performing a Laplace transform we can obtain the response function Here we would like to give the details of the compu- in the frequency domain tation of the real and the imaginary part of the A-field A response function χA(w) in the effective field theory. In subsection VI G we showed that the response of the A-   field in the frequency domain, for the effective theory, is E i 1 1 χA(w) = lim + (C4). given by the expression (see Eq. (113)) 20V η→0+ w + ωe + iη w − ωe + iη

  A X −1 1 1 χA(w) = lim − . (D1) 20ω(κ)V η→0+ w + ω(κ) + iη w − ω(κ) + iη κx,κy e e e

In the thermodynamic limit the above sum turns into an integral and we find the following expression for the real A and the imaginary part of the response function χA(w)

Λ ZZ    A  1 w − ωe(κ) w + ωe(κ) < χA(w) = 2 2 2 − 2 2 dκxdκy and 8π 0Lz ω(κ)[(w − ω(κ)) + η ] ω(κ)[(w + ω(κ)) + η ] 0 e e e e Λ ZZ    A  η 1 1 = χA(w) = 2 2 2 − 2 2 dκxdκy. (D2) 8π 0Lz ω(κ)[(w + ω(κ)) + η ] ω(κ)[(w − ω(κ)) + η ] 0 e e e e

A A In the definition of the real and the imaginary parts <[χA(w)] and =[χA(w)] appear the following four inte- grals

Λ Λ ZZ dκ dκ ZZ dκ dκ A = x y , B = x y , (D3) ω(κ)[(w − ω(κ))2 + η2] (w − ω(κ))2 + η2 0 e e 0 e Λ Λ ZZ dκ dκ ZZ dκ dκ C = x y , D = x y . ω(κ)[(w + ω(κ))2 + η2] (w + ω(κ))2 + η2 0 e e 0 e

To simplify these integrals we go to polar coordinates tion measure is dκxdκy = κrdκrdκθ, and the dressed (κx, κy) → (κr, κθ). In polar coordinates the integra- 33

q 2 2 2 2 2 we define the frequency ω2(κ ) = c2κ2 +ω2 and we make frequency is ω(κ) = c κr + c κz + ωp. Furthermore, e z z p e 2 2 2 the variable substitution u = c κr + ωe (κz) to all the integrals in Eq. (D3) and we obtain

Λ π Z du π Z Λ du √ √ √ A = 2 2 2 , B = 2 2 2 , (D4) 2 c u[(w − u) + η ] c ω (κz ) (w − u) + η 2 e ωe (κz ) Λ π Z du π Z Λ du √ √ √ C = 2 2 2 , D = 2 2 2 . 2 c u[(w + u) + η ] c ω (κz ) (w + u) + η 2 e ωe (κz )

Performing the integration now over the variable u we obtain the following expressions for the previous integrals

 √ Λ  √  Λ 2π −1 u − w π 2w −1 u − w √ 2 2
A = 2 tan , B = 2 tan + ln (w − u) + η , (D5) c η η 2 c η η 2 ωe (κz ) ωe (κz )  √ Λ  √ Λ 2π −1 u + w π √ 2 2
2w −1 u + w C = 2 tan , D = 2 ln (w + u) + η − tan . c η η 2 c η η 2 ωe (κz ) ωe (κz )

A A The real part <[χA(w)] and the imaginary part =[χA(w)] pled to the full photon field is A of the response function χA(w), given by Eq. (D2), in N 2   terms of the integrals A, B, C and D are ˆ 1 X  ˆ  X † 1 H = i~∇j + eA + ~ω(κ) aˆκ,λaˆκ,λ + 2me 2 j=1 κ,λ A 1 (E1) <[χA(w)] = 2 (wA − B − wC − D) and 8π 0Lz A η where the quantized Aˆ -field is in dipole approximation. =[χA(w)] = 2 (C − A) . (D6) 8π 0Lz To treat the many-mode case and the mode-mode in- teraction it is convenient to introduce for the descrip- tion of the annihilation and creation operators the dis- Using the expressions we found for A, B, C and D in A placement coordinates qκ,λ and the conjugate momenta Eq. (D5) we obtain the real part <[χA(w)] and the imag- A A ∂/∂qκ,λ [101] inary part =[χA(w)] of χA(w) given in Eq. (114). 1  †  ∂ 1  †  qκ,λ = √ aˆκ,λ +a ˆκ,λ & = √ aˆk,λ − aˆκ,λ . 2 ∂qκ,λ 2 (E2) Appendix E: Exact Diagonalization with Mode-Mode Interactions The vector potential in terms of the displacement coor- dinates is [101] The aim of this appendix is to show that within the r ˆ ~ X ελ(κ) presented framework the interactions between the differ- A = p qκ,λ, (E3) 0V ω(κ) ent modes of the electromagnetic field can be treated ex- κ,λ actly without introducing any fundamental changes with and the Hamiltonian upon expanding the covariant ki- respect to the effective field theory in section VI. netic term and writing the diamagnetic Aˆ 2 explicitly The Pauli-Fierz Hamiltonian for N free electrons cou- takes the form 34

N 2 2 0 2 !   ω 0 ˆ X ~ 2 ie~ ˆ ~ p X ελ(κ) · ελ (κ ) X ~ω(κ) ∂ 2 H = − ∇j + A · ∇j + p qκ,λqκ0,λ0 + − 2 + qκ,λ . (E4) 2me me 2 0 2 ∂q j=1 κ,κ0,λ,λ0 ω(κ)ω(κ ) κ,λ κ,λ | {z } 2 Ne Aˆ 2 2me

2 The purely photonic part can be separated into a part being quadratic in the displacement coordinates qκ,λ and a part being bilinear qκ,λqκ0,λ0

N  2  2 0 2 2 !! X ~ 2 ie~ ~ωp X ελ(κ) · ελ0 (κ ) X ~ω(κ) ∂ 2 ωp Hˆ = − ∇ + Aˆ · ∇ + q q 0 0 + − + q 1 + . j j p κ,λ κ ,λ 2 κ,λ 2 2me me 2 0 2 ∂q ω (κ) j=1 κ6=κ0,λ,λ0 ω(κ)ω(κ ) κ,λ κ,λ (E5)

Further, we introduce a new scaled set of coordi- ω2(κ) = ω2(κ) + ω2. Then, the Hamiltonian takes the p e p nates uκ,λ = qκ,λ ~/ω(κ) and the dressed frequencies form

N !  2  2 2 2 ω2 ˆ X ~ 2 ie~ ˆ X ~ ∂ ωe (κ) 2 p X 0 H = − ∇j + A · ∇j + − 2 + uκ,λ + ελ(κ) · ελ0 (κ )uκ,λuκ0,λ0 . 2me me 2 ∂u 2 2 j=1 κ,λ κ,λ κ6=κ0,λ,λ0 (E6)

The vector potential in terms of the scaled coordinates is venience we introduce the enlarged “4-tuple” variable ˆ p P A = 1/0V κ,λ ελ(κ)uκ,λ. For simplicity and con- α ≡ (κ, λ) = (κx, κy, κz, λ) and everything can be writ- ten in a more compact form

N M M X  2 ie  2 X ∂2 1 X Hˆ = − ~ ∇2 + ~Aˆ · ∇ − ~ + W u u , where W = ω2 δ + ω2E , (E7) j j 2 αβ α β αβ eα αβ p αβ 2me me 2 ∂u 2 j=1 α=1 α α,β=1

−1 with the matrix Eαβ being zero for α = β, Eαα = 0, while also invertible and for its inverse U holds that is equal for α 6= β this matrix is defined as the inner product to its transpose U T . Using the orthogonal matrix U we of the polarization vectors Eαβ = εα · εβ. The matrix can define the normal coordinates zγ and the canonical W is real and symmetric and as a consequence can be momenta ∂/∂zγ [103] diagonalized by an orthogonal matrix U. This means that it can be brought into a diagonal form X ∂ X ∂ z = U u & = U . (E9) X −1 2 γ αγ α ∂z αγ ∂u Uαγ WγδUδβ = Ωαδαβ (E8) α γ α α γ,δ

2 where Ωα are the eigen-values of the matrix Wαβ. Fur- In terms of these coordinates and momenta the Hamilto- ther, because the matrix U is an orthogonal matrix it is nian takes the form [103]

N  2  M 2 2 2 ! r X ie X ∂ Ωγ 1 X Hˆ = − ~ ∇2 + ~Aˆ · ∇ + −~ + z2 with Aˆ = ε z . (E10) 2m j m j 2 ∂z2 2 γ  V eγ γ j=1 e e γ=1 γ 0 γ=1

P The new polarization vectors εeγ are defined as εeγ = α=1 εαUαγ . The Hamiltonian is translationally invari- 35 ant and thus the wavefunctions of the electronic part are given by the Slater determinant ΦK defined in Eq. (14). Applying Hˆ on ΦK we obtain

 N M ! 2 2 2 Ω2 ˆ ~ X 2 X ~ ∂ γ 2 e~ HΦK =  kj + − + zγ − gzγ εγ · K  ΦK, where g = √ . (E11) 2m 2 ∂z2 2 e e e m  V e j=1 γ=1 γ e 0

We perform a square completion and the part of the Hamiltonian depending on zγ can be written as a sum of displaced harmonic oscillators with frequencies Ωγ

  2 N M 2 2 2  2 2 ! X X ∂ Ωγ gεγ · K (gεγ · K) Hˆ Φ = ~ k2 + −~ + z − ee − ee Φ . (E12) K 2m j 2 ∂z2 2 γ Ω2 2Ω2  K e j=1 γ=1 γ γ γ

The eigenspectrum of each displaced harmonic oscilla- energy spectrum of the free electron gas coupled to an tor is Enγ = ~Ωγ (nγ + 1/2) and thus we find that the arbitrary amount of photon modes with the mode-mode interactions included is.

  2 N 2 M 2 M   X ωp X (εγ · K) X 1 E = ~ k2 − e + Ω n + (E13) k 2m  j N Ω2  ~ γ γ 2 e j=1 γ=1 γ γ=1

Thus, we see that the structure of the energy spectrum tion to the effective energy spectrum defined in Eq. (77) even with the inclusion of the mode-mode interactions is and the coupling constant in the effective field theory, the the same with the one in the effective quantum field the- exact coupling as a function of the number of the photon ory in Eq. (77), and that the mode-mode interactions do modes on each particular direction is not introduce any fundamental modification. To obtain M 2 i
2 the expression for the eigenspectrum we substituted the X ωp εγ gi (M) = e where i = x, y, z. (E14) parameter g. Also we would like mention that in connec- ex Ω2 e γ=1 γ

[1] A. Sommerfeld, “Zur elektronentheorie der metalle auf [7] P. Nozi´eres, Theory of Interacting Fermi Systems (Adi- grund der fermischen statistik,” Zeitschrift f¨urPhysik son Wesley Longman, Inc., 1964). 47, 1–32 (1928). [8] K. v. Klitzing, G. Dorda, and M. Pepper, “New method [2] N. W. Aschroft and N. Mermin, Solid State Physics for high-accuracy determination of the fine-structure (Harcourt College Publishers, 1976). constant based on quantized hall resistance,” Phys. Rev. [3] G. Giulliani and G. Vignale, Quantum Theory of the Lett. 45, 494–497 (1980). Electron Liquid (Cambridge University Press, 2012). [9] R. B. Laughlin, “Quantized hall conductivity in two di- [4] P. Hohenberg and W. Kohn, “Inhomogeneous electron mensions,” Phys. Rev. B 23, 5632–5633 (1981). gas,” Phys. Rev. 136, B864–B871 (1964). [10] W. Greiner and J. Reinhardt, Field quantization [5] G. Onida, L. Reining, and A. Rubio, “Electronic exci- (Springer, 1996). tations: density-functional versus many-body green’s- [11] H. Spohn, Dynamics of charged particles and their ra- function approaches,” Rev. Mod. Phys. 74, 601–659 diation field (Cambridge university press, 2004). (2002). [12] M. Ruggenthaler, N. Tancogne-Dejean, J. Flick, H. Ap- [6] L. D. Landau, “The theory of a fermi liquid,” pel, and A. Rubio, “From a quantum-electrodynamical Sov. Phys. JETP 30, 1058 (1956). light–matter description to novel spectroscopies,” Na- 36

ture Reviews Chemistry 2, 0118 (2018). K. J. H. and Ruggenthaler, and A. Rubio, “Light- [13] S. Weinberg, The Quantum Theory of Fields (Cam- matter hybrid-orbital-based first-principles methods: bridge University Press, 2005). The influence of polariton statistics,” Journal of [14] C. Cohen-Tannoudji, J. Dupont-Roc, and G. Grynberg, Chemical Theory and Computation 16, 5601–5620 Photons and Atoms-Introduction to Quantum Electrody- (2020). namics (Wiley-VCH, 1997). [34] S. E. B. Nielsen, C. Scha¨afer,M. Ruggenthaler, and [15] W. D. Phillips, “Nobel lecture: Laser cooling and trap- A. Rubio, “Dressed-orbital approach to cavity quantum ping of neutral atoms,” Rev. Mod. Phys. 70, 721–741 electrodynamics and beyond,” arXiv:1812.00388 [quant- (1998). ph] (2019). [16] C. N. Cohen-Tannoudji, “Nobel lecture: Manipulating [35] T. S. Haugland, E. Ronca, E. F. Kjønstad, A. Rubio, atoms with photons,” Rev. Mod. Phys. 70, 707–719 and H. Koch, “Coupled cluster theory for molecular po- (1998). laritons: Changing ground and excited states,” Phys. [17] S. Chu, “Nobel lecture: The manipulation of neutral Rev. X 10, 041043 (2020). particles,” Rev. Mod. Phys. 70, 685–706 (1998). [36] U. Mordovina, C. Bungey, H. Appel, P. J. Knowles, [18] W. Ketterle, “Nobel lecture: When atoms behave as A. Rubio, and F. R. Manby, “Polaritonic coupled- waves: Bose-einstein condensation and the atom laser,” cluster theory,” Phys. Rev. Research 2, 023262 (2020). Rev. Mod. Phys. 74, 1131–1151 (2002). [37] J. Flick, M. Ruggenthaler, H. Appel, and A. Rubio, [19] E. A. Cornell and C. E. Wieman, “Nobel lecture: Bose- “Atoms and molecules in cavities, from weak to strong einstein condensation in a dilute gas, the first 70 years coupling in quantum-electrodynamics (qed) chemistry,” and some recent experiments,” Rev. Mod. Phys. 74, Proceedings of the National Academy of Sciences 114, 875–893 (2002). 3026–3034 (2017). [20] R. J. Glauber, “Nobel lecture: One hundred years of [38] J. Flick, M. Ruggenthaler, H. Appel, and A. Ru- light quanta,” Rev. Mod. Phys. 78, 1267–1278 (2006). bio, “Kohn-sham approach to quantum electrodynami- [21] T. W. H¨ansch, “Nobel lecture: Passion for precision,” cal density-functional theory: Exact time-dependent ef- Rev. Mod. Phys. 78, 1297–1309 (2006). fective potentials in real space,” Proc. Natl. Acad. Sci. [22] J. L. Hall, “Nobel lecture: Defining and measuring opti- U. S. A. 112, 15285–15290 (2015). cal frequencies,” Rev. Mod. Phys. 78, 1279–1295 (2006). [39] T. W. Ebbesen, “Hybrid light–matter states in a molec- [23] S. Haroche, “Nobel lecture: Controlling photons in a ular and material science perspective,” Accounts of box and exploring the quantum to classical boundary,” Chemical Research 49, 2403–2412 (2016). Rev. Mod. Phys. 85, 1083–1102 (2013). [40] J. George, T. Chervy, A. Shalabney, E. Devaux, [24] D. J. Wineland, “Nobel lecture: Superposition, en- H. Hiura, C. Genet, and T. W. Ebbesen, “Multiple tanglement, and raising schr¨odinger’scat,” Rev. Mod. rabi splittings under ultrastrong vibrational coupling,” Phys. 85, 1103–1114 (2013). Phys. Rev. Lett. 117, 153601 (2016). [25] B. W. Shore and P. L. Knight, “The jaynes-cummings [41] J.A. Hutchison, A. Liscio, T. Schwartz, A. Canaguier- model,” Journal of Modern Optics 40, 1195–1238 Durand, C. Genet, V. Palermo, P. Samor`ı, and T. W. (1993). Ebbesen, “Tuning the work-function via strong cou- [26] R. H. Dicke, “Coherence in spontaneous radiation pro- pling,” Advanced Materials 25, 2481–2485 (2013). cesses,” Phys. Rev. 93, 99–110 (1954). [42] J. A. Hutchison, T. Schwartz, C. Genet, E. Devaux, [27] P Kirton, M. M. Roses, J. Keeling, and E. G. and T. W. Ebbesen, “Modifying chemical landscapes Dalla Torre, “Introduction to the dicke model: From by coupling to vacuum fields,” Angewandte Chemie In- equilibrium to nonequilibrium, and vice versa,” Ad- ternational Edition 51, 1592–1596 (2012). vanced Quantum Technologies 2, 1800043 (2019). [43] E. Orgiu, J. George, J. A. Hutchison, E. Devaux, J. F. [28] A. F. Kockum, A. Miranowicz, S. De Liberato, Dayen, B. Doudin, F. Stellacci, C. Genet, J. Schachen- S. Savasta, and F. Nori, “Ultrastrong coupling between mayer, C. Genes, G. Pupillo, P. Samor`ı, and T. W. light and matter,” Nature Reviews Physics 1, 19–40 Ebbesen, “Conductivity in organic semiconductors hy- (2019). bridized with the vacuum field,” Nat. Mater. 14, 1123– [29] P. M. M. C. de Melo and A. Marini, “Unified theory of 1129 (2015). quantized electrons, phonons, and photons out of equi- [44] J. Feist, J. Galego, and F. J. Garcia-Vidal, “Polaritonic librium: A simplified ab initio approach based on the chemistry with organic molecules,” ACS Photonics 5, generalized baym-kadanoff ansatz,” Phys. Rev. B 93, 205–216 (2017). 155102 (2016). [45] J. Galego, F. J. Garcia-Vidal, and J. Feist, “Suppress- [30] I. V. Tokatly, “Time-dependent density functional the- ing photochemical reactions with quantized light fields,” ory for many-electron systems interacting with cavity Nature Communications 7, 13841 (2016). photons,” Phys. Rev. Lett. 110, 233001 (2013). [46] C. Sch¨afer,M. Ruggenthaler, H. Appel, and A. Rubio, [31] M. Ruggenthaler, J. Flick, C. Pellegrini, H. Appel, I. V. “Modification of excitation and charge transfer in cav- Tokatly, and A. Rubio, “Quantum-electrodynamical ity quantum-electrodynamical chemistry,” Proceedings density-functional theory: Bridging quantum optics and of the National Academy of Sciences 116, 4883–4892 electronic-structure theory,” Phys. Rev. A 90, 012508 (2019). (2014). [47] T. E. Li, A. Nitzan, and J. E. Subotnik, “On the [32] C. Pellegrini, J. Flick, I. V. Tokatly, H. Appel, and origin of ground-state vacuum-field catalysis: Equilib- A. Rubio, “Optimized effective potential for quan- rium consideration,” arXiv:2002.09977 [physics.chem- tum electrodynamical time-dependent density func- ph] (2020). tional theory,” Phys. Rev. Lett. 115, 093001 (2015). [48] Basov, D. N., Averitt, R. D., and Hsieh, D., “Towards [33] F. Buchholz, I. Theophilou, M. Giesbertz, properties on demand in quantum materials,” Nature 37

Materials 16, 1077–1088 (2017). a polariton condensate and an electron system in two [49] M. Buzzi, M. F¨orst,M. Sigrist, R. Mankowsky, and dimensions,” Phys. Rev. B 93, 054510 (2016). A. Cavalleri, “Probing dynamics in quantum materials [64] M. A. Sentef, M. Ruggenthaler, and A. Rubio, “Cav- with femtosecond x-rays,” Nature Reviews Materials 3, ity quantum-electrodynamical polaritonically enhanced 299 (2018). electron-phonon coupling and its influence on supercon- [50] X. Wang, E. Ronca, and M. A. Sentef, “Cavity quan- ductivity,” Science advances 4, eaau6969 (2018). tum electrodynamical chern insulator: Towards light- [65] J. B. Curtis, Z. M. Raines, A. A. Allocca, M. Hafezi, induced quantized anomalous hall effect in graphene,” and V. M. Galitski, “Cavity quantum eliashberg en- Phys. Rev. B 99, 235156 (2019). hancement of superconductivity,” Phys. Rev. Lett. 122, [51] M. Kiffner, J. R. Coulthard, F. Schlawin, A. Ardavan, 167002 (2019). and D. Jaksch, “Manipulating quantum materials with [66] A. Thomas et al., “Exploring superconductivity under quantum light,” Phys. Rev. B 99, 085116 (2019). strong coupling with the vacuum electromagnetic field,” [52] J. Li, D. Golez, G. Mazza, A. J. Millis, A. Georges, and arXiv:1911.01459 [cond-mat.supr-con] (2019). M. Eckstein, “Electromagnetic coupling in tight-binding [67] S. Latini, E. Ronca, U. De Giovannini, H. H¨ubener, models for strongly correlated light and matter,” Phys. and A. Rubio, “Cavity control of excitons in two- Rev. B 101, 205140 (2020). dimensional materials,” Nano Letters 19, 3473 (2019). [53] Y. Ashida, A. Imamo˘glu,and˙ E. Demler, “Cavity quan- [68] M. F¨org,L. Colombier, R. K. Patel, J. Lindlau, A. D. tum electrodynamics at arbitrary light-matter coupling Mohite, H. Yamaguchi, M. M. Glazov, D. Hunger, and strengths,” Phys. Rev. Lett. 126, 153603 (2021). A. H¨ogele,“Cavity-control of interlayer excitons in van [54] D. Hagenm¨uller,S. De Liberato, and C. Ciuti, “Ul- der waals heterostructures,” Nature Communications trastrong coupling between a cavity resonator and the 10, 3697 (2019). cyclotron transition of a two-dimensional electron gas in [69] J. Levinsen, G. Li, and M. M. Parish, “Microscopic de- the case of an integer filling factor,” Phys. Rev. B 81, scription of exciton-polaritons in microcavities,” Phys. 235303 (2010). Rev. Research 1, 033120 (2019). [55] V. Rokaj, M. Penz, M. A. Sentef, M. Ruggenthaler, [70] J. Kasprzak, M. Richard, S. Kundermann, A. Baas, and A. Rubio, “Quantum electrodynamical bloch theory P. Jeambrun, J. M. J. Keeling, F. M. Marchetti, M. H. with homogeneous magnetic fields,” Phys. Rev. Lett. Szyma´nska, R. Andr´e,J. L. Staehli, V. Savona, P. B. 123, 047202 (2019). Littlewood, B. Deveaud, and L. S. Dang, “Bose– [56] J. Keller, G. Scalari, F. Appugliese, S. Rajabali, einstein condensation of exciton polaritons,” Nature 2, M. Beck, J. Haase, C. A. Lehner, W. Wegscheider, 409–414 (2006). M. Failla, M. Myronov, D. R. Leadley, J. Lloyd-Hughes, [71] J. Keeling and S. Kena-Cohen, “Bose–einstein con- P. Nataf, and J. Faist, “Landau polaritons in highly densation of exciton-polaritons in organic microcavi- nonparabolic two-dimensional gases in the ultrastrong ties,” Annual Review of Physical Chemistry 71 (2020), coupling regime,” Phys. Rev. B 101, 075301 (2020). 10.1146/annurev-physchem-010920-102509. [57] G. Scalari, C. Maissen, D. Turˇcinkov´a,D. Hagenm¨uller, [72] H. H¨ubener, U. De Giovannini, C. Sch¨afer, J. And- S. De Liberato, C. Ciuti, C. Reichl, D. Schuh, berger, M. Ruggenthaler, J. Faist, and A. Rubio, W. Wegscheider, M. Beck, and J. Faist, “Ultrastrong “Quantum cavities and floquet materials engineering: coupling of the cyclotron transition of a 2d electron gas the power of chirality,” (Under review in Nature Mate- to a thz metamaterial,” Science 335, 1323–1326 (2012). rials) (2020). [58] X. Li, M. Bamba, Q. Zhang, S. Fallahi, G. C. Gard- [73] J. Petersen, J. Volz, and A Rauschenbeutel, “Chiral ner, W. Gao, M. Lou, K. Yoshioka, M. J. Manfra, and nanophotonic waveguide interface based on spin-orbit J. Kono, “Vacuum bloch–siegert shift in landau polari- interaction of light,” Science 346, 67–71 (2014). tons with ultra-high cooperativity,” Nature Photonics [74] F. Zhang, J. Ren, L. Shan, X. Duan, Y. Li, T. Zhang, 12, 324–329 (2018). Q. Gong, and Y. Gu, “Chiral cavity quantum electro- [59] S. Ravets, P. Kn¨uppel, S. Faelt, O. Cotlet, M. Kroner, dynamics with coupled nanophotonic structures,” Phys. W. Wegscheider, and A. Imamoglu, “Polaron polaritons Rev. A 100, 053841 (2019). in the integer and fractional quantum hall regimes,” [75] P. Lodahl, S. Mahmoodian, S. Stobbe, A. Rauschenbeu- Phys. Rev. Lett. 120, 057401 (2018). tel, P. Schneeweiss, J. Volz, H. Pichler, and P. Zoller, [60] S. Smolka, W. Wuester, F. Haupt, S. Faelt, “Chiral quantum optics,” Nature 541, 473–480 (2017). W. Wegscheider, and A. Imamoglu, “Cavity quan- [76] C. Sch¨afer,M. Ruggenthaler, V. Rokaj, and A. Ru- tum electrodynamics with many-body states of a two- bio, “Relevance of the quadratic diamagnetic and self- dimensional electron gas,” Science 346, 332–335 (2014). polarization terms in cavity quantum electrodynamics,” [61] G. L. Paravicini-Bagliani, F. Appugliese, E. Richter, arXiv:1911.08427 [quant-ph] (2019). S. Fallahi, F. Valmorra, J. Keller, M. Beck, N. Bartolo, [77] A. Vukics, T. Grießer, and P. Domokos, “Elimination C. R¨ossler,T. Ihn, K. Ensslin, C. Ciuti, G. Scalari, and of the a-square problem from cavity qed,” Phys. Rev. J. Faist, “Magneto-transport controlled by landau po- Lett. 112, 073601 (2014). lariton states,” Nature Physics 15, 186–190 (2019). [78] D. De Bernardis, P. Pilar, T. Jaako, S. De Liberato, and [62] F. Schlawin, A. Cavalleri, and D. Jaksch, “Cavity- P. Rabl, “Breakdown of gauge invariance in ultrastrong- mediated electron-photon superconductivity,” Phys. coupling cavity qed,” Physical Review A 98, 053819 Rev. Lett. 122, 133602 (2019). (2018). [63] O. Cotlet¸, S. Zeytinoˇglu,M. Sigrist, E. Demler, and [79] O. Di Stefano, A. Settineri, V. Macr`ı, L. Garziano, A. Imamoˇglu,“Superconductivity and other collective R. Stassi, S. Savasta, and F. Nori, “Resolution of phenomena in a hybrid bose-fermi mixture formed by gauge ambiguities in ultrastrong-coupling cavity quan- tum electrodynamics,” Nature Physics (2019). 38

[80] K. Hepp and L. E. H. Lieb, “On the superradiant phase Proc.100N3-4,61(1997)]. transition for molecules in a quantized radiation field: [99] H.B.G. Casimir and D. Polder, “The influence of re- the dicke maser model,” Annals of Physics 76, 360–404 tardation on the london-van der waals forces,” Physical (1973). Review 73, 360 (1948). [81] Y. K. Wang and F. T. Hioe, “Phase transition in the [100] J. N. Munday, F. Capasso, and V. A. Parsegian, “Mea- dicke model of superradiance,” Phys. Rev. A 7, 831– sured long-range repulsive casimir–lifshitz forces,” Na- 836 (1973). ture 457, 2078–2084 (2009). [82] I. Bialynicki-Birula and K. Rza¸˙zewski,“No-go theorem [101] V. Rokaj, D. M. Welakuh, M. Ruggenthaler, and A. Ru- concerning the superradiant phase transition in atomic bio, “Light–matter interaction in the long-wavelength systems,” Phys. Rev. A 19, 301–303 (1979). limit: no ground-state without dipole self-energy,” Jour- [83] P. Nataf and C. Ciuti, “No-go theorem for superradiant nal of Physics B: Atomic, Molecular and Optical Physics quantum phase transitions in cavity qed and counter- 51, 034005 (2018). example in circuit qed,” Nature Communications 1 [102] J. L. Lebowitz and E. H. Lieb, “Existence of thermo- (2010), 10.1038/ncomms1069. dynamics for real matter with coulomb forces,” Phys. [84] O. Viehmann, J. von Delft, and F. Marquardt, “Super- Rev. Lett. 22, 631–634 (1969). radiant phase transitions and the standard description [103] F. H. Faisal, Theory of Multiphoton Processes (Springer, of circuit qed,” Phys. Rev. Lett. 107, 113602 (2011). Berlin, 1987). [85] G. Mazza and A. Georges, “Superradiant quantum ma- [104] Y. Todorov, A. M. Andrews, R. Colombelli, S. De Lib- terials,” Phys. Rev. Lett. 122, 017401 (2019). erato, C. Ciuti, P. Klang, G. Strasser, and C. Sirtori, [86] G. M. Andolina, F. M. D. Pellegrino, V. Giovannetti, “Ultrastrong light-matter coupling regime with polari- A. H. MacDonald, and M. Polini, “Cavity quantum ton dots,” Phys. Rev. Lett. 105, 196402 (2010). electrodynamics of strongly correlated electron systems: [105] Y. Todorov and C. Sirtori, “Intersubband polaritons in A no-go theorem for photon condensation,” Phys. Rev. the electrical dipole gauge,” Phys. Rev. B 85, 045304 B 100, 121109 (2019). (2012). [87] T. Jaako, Z-L. Xiang, J. J. Garcia-Ripoll, and P. Rabl, [106] D. J. Griffiths, Introduction to Quantum Mechanics “Ultrastrong-coupling phenomena beyond the dicke (Prentice Hall, 1995). model,” Phys. Rev. A 94, 033850 (2016). [107] S. De Liberato, “Virtual photons in the ground state of [88] G. M. Andolina, F. M. D. Pellegrino, V. Giovannetti, a dissipative system,” arXiv:1812.00388 [quant-ph] 8, A. H. MacDonald, and M. Polini, “Theory of pho- 1465 (2019). ton condensation in a spatially-varying electromagnetic [108] L. D. Landau and E. M. Lifshitz, Quantum Mechan- field,” arXiv:2005.09088 [cond-mat.mes-hall] (2020). ics, Third Edition: Non-relativistic Theory (Pergamon [89] D. Guerci, P. Simon, and C. Mora, “Superradiant Press, 1997). phase transition in electronic systems and emergent [109] D. De Bernardis, T. Jaako, and P. Rabl, “Cav- topological phases,” arXiv:2005.09088 [cond-mat.mes- ity quantum electrodynamics in the nonperturbative hall] (2020). regime,” Physical Review A 97 (2018), 10.1103/phys- [90] A. Stokes and A. Nazir, “Uniqueness of the phase tran- reva.97.043820. sition in many-dipole cavity quantum electrodynamical [110] D. Hagenm¨uller and C. Ciuti, “Cavity qed of the systems,” Phys. Rev. Lett. 125, 143603 (2020). graphene cyclotron transition,” Phys. Rev. Lett. 109, [91] R. Kubo, “Statistical-mechanical theory of irreversible 267403 (2012). processes. i. general theory and simple applications to [111] L. Chirolli, M. Polini, V. Giovannetti, and A. H. Mac- magnetic and conduction problems,” J. Phys. Soc. Jpn. Donald, “Drude weight, cyclotron resonance, and the 12 (1957). dicke model of graphene cavity qed,” Phys. Rev. Lett. [92] J. Flick, D. M. Welakuh, M. Ruggenthaler, H. Appel, 109, 267404 (2012). and A. Rubio, “Light–matter response in nonrelativistic [112] J. Galego, C. Climent, F. J. Garcia-Vidal, and quantum electrodynamics,” ACS Photonics 6 (2019). J. Feist, “Cavity casimir-polder forces and their effects [93] Y. Todorov, “Dipolar quantum electrodynamics theory in ground-state chemical reactivity,” Phys. Rev. X 9, of the three-dimensional electron gas,” Phys. Rev. B 89, 021057 (2019). 075115 (2014). [113] J. D. Jackson, Classical Electrodynamics (John Wiley [94] Y. Todorov, “Dipolar quantum electrodynamics of the & Sons Inc, 1998). two-dimensional electron gas,” Phys. Rev. B 91, 125409 [114] D. N. Basov, Richard D. Averitt, Dirk van der Marel, (2015). Martin Dressel, and Kristjan Haule, “Electrodynamics [95] T. Chen, “Infrared renormalization in non-relativistic of correlated electron materials,” Rev. Mod. Phys. 83, qed and scaling criticality,” Journal of Functional Anal- 471–541 (2011). ysis 254, 2555 – 2647 (2008). [115] D. N. Basov and T. Timusk, “Electrodynamics of high- [96] J. Fr¨ohlich and A. Pizzo, “Renormalized electron mass Tc superconductors,” Rev. Mod. Phys. 77, 721–779 in nonrelativistic qed,” Communications in Mathemat- (2005). ical Physics 294, 439–470 (2010). [116] N. Bartolo and C. Ciuti, “Vacuum-dressed cavity mag- [97] F. Hiroshima and H. Spohn, “Mass renormalization in netotransport of a two-dimensional electron gas,” Phys. nonrelativistic quantum electrodynamics,” Journal of Rev. B 98, 205301 (2018). Mathematical Physics 46, 042302 (2005). [117] A. Einstein, “Zur Quantentheorie der Strahlung,” Phys. [98] H. B. G. Casimir, “On the Attraction Between Two Per- Z. 18, 121–128 (1917). fectly Conducting Plates,” Indag. Math. 10, 261–263 [118] S. Haroche and D. Kleppner, “Cavity quantum electro- (1948), [Kon. Ned. Akad. Wetensch. Proc.51,793(1948); dynamics,” Physics Today 42, 24 (1999). Front. Phys.65,342(1987); Kon. Ned. Akad. Wetensch. [119] M. Srednicki, Quantum Field Theory (Cambridge Uni- 39

versity Press, 2007). [131] G. Baym and C. Pethick, Landau Fermi-Liquid Theory: [120] F. Mandl and G. Shaw, Quantum Field Theory (John Concepts and Applications (Wiley-VCH, 1991). Wiley & Sons, 1984). [132] A. L. Fetter and J. D. Walecka, Quantum Theory of [121] S. Y. Buhmann, Dispersion forces I: Macroscopic Many-Particle Systems (McGraw-Hill, Inc., 1971). quantum electrodynamics and ground-state Casimir, [133] A. Altland and B. Simons, Condensed Matter Field The- Casimir–Polder and van der Waals Forces, Vol. 247 ory (Cambridge University Press, 2010). (Springer, 2013). [134] J. S. Høye and I. Brevik, “Repulsive casimir force,” [122] S. Y. Buhmann, Dispersion Forces II: Many-Body Ef- Phys. Rev. A 98, 022503 (2018). fects, Excited Atoms, Finite Temperature and Quantum [135] T. H. Boyer, “Van der waals forces and zero-point en- Friction, Vol. 248 (Springer, 2013). ergy for dielectric and permeable materials,” Phys. Rev. [123] F. Hiroshima and H. Spohn, “Mass renormalization in A 9, 2078–2084 (1974). nonrelativistic quantum electrodynamics,” Journal of [136] F. S. S. Rosa, D. A. R. Dalvit, and P. W. Milonni, Mathematical Physics 46, 042302 (2005). “Casimir-lifshitz theory and metamaterials,” Phys. Rev. [124] K. G. Wilson, “The renormalization group: Critical Lett. 100, 183602 (2008). phenomena and the kondo problem,” Rev. Mod. Phys. [137] D. T. Butcher, S. Y. Buhmann, and S. Scheel, 47, 773–840 (1975). “Casimir–polder forces between chiral objects,” New [125] H. Epstein and V. Glaser, “The role of locality in pertur- Journal of Physics 14, 113013 (2012). bation theory,” Annales de l’I.H.P. Physique th´eorique [138] Henkel, C. and Joulain, K., “Casimir force between de- 19, 211–295 (1973). signed materials: What is possible and what not,” Eu- [126] H. A. Bethe, “The electromagnetic shift of energy lev- rophys. Lett. 72, 929–935 (2005). els,” Phys. Rev. 72, 339–341 (1947). [139] T. Chervy, P. Kn¨uppel, H. Abbaspour, M. Lupatini, [127] F. G. Eich, Markus Holzmann, and G. Vignale, “Ef- S. F¨alt,W. Wegscheider, M. Kroner, and A. Imamoˇglu, fective mass of quasiparticles from thermodynamics,” “Accelerating polaritons with external electric and mag- Phys. Rev. B 96, 035132 (2017). netic fields,” Phys. Rev. X 10, 011040 (2020). [128] C. Hainzl and R. Seiringer, “Mass renormalization and [140] A. Damascelli, “Probing the electronic structure of com- energy level shift in non-relativistic qed,” Adv. Theor. plex systems by ARPES,” Physica Scripta T109, 61 Math. Phys. , 847–871 (2002). (2004). [129] F. Wilczek, “Fundamental constants,” arXiv:0708.4361 [141] J. Callaway, Energy Band Theory (Academic Press Inc., [hep-ph] (2007). 1964). [130] A. Bettini, Introduction to Elementary Particle Physics [142] M. Dresselhaus, G. Dresselhaus, Cronin S.B., and (Cambridge university press, 2008). A. Gomes Souza Filho, Solid State Properties (Springer, Berlin, Heidelberg, 2018).