Dipolar polaritons squeezed at unitarity

S. V. Andreev1, ∗ 1National Research Center ”Kurchatov Institute” B.P. Konstantinov Petersburg Nuclear Physics Institute, Gatchina 188300, Russia (Dated: 2 mars 2020) Interaction of dipolar polaritons can be efficiently tuned by means of a shape resonance in their excitonic component. Provided the resonance width is large, a squeezed population of strongly interacting polaritons may persist on the repulsive side of the resonance. We derive an analytical expression for the polariton coupling constant and show that it may have very large values in typical experimental conditions. Our arguments provide a new direction for the quest of interactions in quantum photonics.

PACS numbers: 71.35.Lk

The commonly adopted strategy to introduce interac- layers. The attractive side of such resonance was theore- tions into the quantum optics of semiconductors is tailo- tically explored in the context of roton-maxon excita- ring the non-linearity due to excitonic transitions [1]. In tions and supersolidity in dipolar Bose-Einstein conden- the regime of strong light-matter coupling the macrosco- sates (BEC’s) [14, 15]. On the repulsive side, the equili- pic population of the cavity mode is efficiently transferred brium ground state is a condensate of biexcitons, distin- into the exciton field, which can be regarded as a gas of guished from the exciton condensate by suppressed cohe- bosonic quasiparticles [2–5]. In particular, the blueshift of rence of the photoluminescence (PL) and a gapped exci- the polariton dispersion is governed by low-energy s-wave tation spectrum [13]. These predictions hold for a wide collisions between the pairs of excitons [2]. This naturally variety of bilayer structures, where the exciton lifetime refers to ultra-cold atomic systems, where enhancement is sufficiently long to establish a thermodynamic equili- of interactions has been demonstrated by working with brium. Thus, the numerical calculations of the exciton in- species having dipole moments [6], Rydberg excitations teraction potential in coupled QW’s [16] suggest that the [7] and using the technique of Feshbach resonance [8]. The shape resonance may be responsible for the formation of latter provides a possibility to tune the scattering length a fragmented-condensate solid of excitons [14, 15, 17, 18]. from positive to negative values through the unitary li- Several groups have recently reported an increase of mit by adjusting the position of the scattering threshold the polariton interaction due to the dipolar moment in with respect to a bound state. the excitonic component [19–21]. The results presented in On the technological side, exceptional excitonic pro- Ref. [20] are particularly compelling : a factor of 200 en- perties are found in atomically thin heterostructures of hancement of the dipolar polariton interaction strength transition metal dichalcogenides (TMD’s) [9]. The atom- as compared to unpolarized polaritons has been detected. like Lennard-Jones interaction between excitons has been Dipolar repulsion alone cannot explain such tremendous demonstrated in these materials [10]. Such interaction na- blueshift of the polariton PL. The mystery is deepened turally admits a bound state and, indeed, biexcitons have by very low values of polariton densities at which the recently been observed in several types of monolayers experiment was done. [11]. A fundamental difference from the atomic clouds is, Motivated by these experimental observations, the pa- however, a purely two-dimensional (2D) character of the per presents a phenomenological model of resonantly pai- exciton translational motion. The interactions in a 2D red dipolar polaritons. In contrast to dipolar excitons, ultra-cold gas are generically weak due to the properties microcavity polaritons are far from the thermodynamic of 2D kinematics. Thus, in contrast to three dimensions, equilibrium, their statistics being closer to lasers rather quantum scattering off a weakly-bound state has a vani- than to atomic BEC’s [22]. This makes possible existence shingly small amplitude [12]. At sufficiently low exciton of a metastable polariton population on the repulsive side densities these arguments apply also for semiconductor of the shape resonance. Coupling to a transient bipo- arXiv:1905.06418v2 [cond-mat.quant-gas] 28 Feb 2020 quantum wells (QW’s). lariton mode in this case yields divergent behaviour of As was proposed by the author [13], a 2D analog of the 2D effective interaction, akin to the unitary limit in the Feshbach resonance may be realized with dipolar three-dimensional atomic clouds. We derive an analyti- excitons formed of electrons and holes residing in spa- cal expression for the interaction enhancement factor as tially separated layers. The dipolar repulsion introduces a function of the polariton dipole moment and density, a potential barrier between the outer continuum and the and show that it can be very large in typical experimental bound state (biexciton), which enables a quasi-discrete conditions. Being exact in the dilute regime, this result level with tunable energy and lifetime. Both parameters ultimately holds for two polaritons in vacuum. Another can be controlled by changing the distance d between the interesting prediction of our theory is that the many- 2 body polariton states become squeezed by the resonance. tice, ”↑” and ”↓” typically correspond to left- and right- This could be verified in current experiments by exami- circularly polarized photons [29]. In planar dielectric mi- ning the statistics of emitted photons. crocavities the lower polariton dispersion is split into the Let us discuss the relevant timescales of the pro- linearly polarized TM and TE modes [24]. This splitting blem. First, we shall assume that the polaritons do not acts as an effective magnetic field which flips the polari- condense into the paired state, which is the equilibrium ton pseudospin σ. Provided that ground state when the discrete level ε is below the scatte- p ring threshold. Second, the width of the resonance β must p  mLTβ/~, (2) be sufficiently large for polaritons could feel the interior the decay of a bound state due to the polariton spin-flip of the barrier during their lifetime τ. We shall assume a is subdominant with respect to the tunneling under the broad resonance, which seems to be the most likely for dipolar potential barrier. Here, the parameter m ac- polaritons because of their very low effective mass [33]. LT counts both for the bare photonic and low-k excitonic Hence, we let TE-TM splitting [27, 28]. In confined geometries [25, 26] one should let p ∼ π/L, where L is the characteristic β  ε (1) length of the confinement. Under the condition (2), we may consider a single polariton branch E(p) characteri- and ~/β  τ  τp, where τp is the thermalization time zed by the effective mass m in the vicinity of its mini- [23]. mum. Analysis of a possible departure from this approxi- The system is a mixture of two polariton flavoursc ˆσ mation will be given in a separate paper. with σ = (↑, ↓) and their bipolaritonic pairs Cˆ. In prac- The many-body Hamiltonian reads

X † X † g X † † Hˆ = E(p)ˆc cˆσ,p + [2E(k/2) + ε]Cˆ Cˆk + cˆ cˆ cˆσ,p cˆσ,p σ,p k 2S σ,p1+q σ,p2−q 1 2 p,σ k p1,p2,q,σ r 2 (3) ~ β X  † † †  + cˆ cˆ Cˆ +c ˆ k cˆ k Cˆ . ↑,p+ k ↓,−p+ k k ↑,−p+ ↓,p+ k 2πmS 2 2 2 2 k,p

Here the first two terms describe the dispersions of single- quantization area S. The last term models the interac- and bipolaritons, respectively, with p = (px, py) and tion of polaritons with opposite spins by converting them E(p) = ~2p2/2m at the bottom of the band (p → 0). In into the bipolariton mode and vice versa. The square-root the limit of zero exciton dipole moment the discrete level prefactor is constructed in such a way as to reproduce the ε can be identified with the binding energy of a loosely low-energy 2D scattering amplitude for two particles in bound bipolariton molecule [30]. The next two terms is vacuum [13, 31]. the usual background interaction between the polaritons By using the standard commutation relations for bo- with alike spins (accounting both for the short-range part sons, one obtains the following set of Heisenberg equa- and the dipolar tail of the bare exciton potential) in the tions of motion

r 2 dcˆσ,p β X i = [E(p) + µ ]ˆc + ~ cˆ† Cˆ (4a) ~ dt σ σ,p 2πmS σ06=σ,k k+p k ˆ r 2 dCk ˆ ~ β X i~ = [2E(k/2) + ε]Ck + cˆ↑,−p+ k cˆ↓,p+ k , (4b) dt 2πmS 2 2 p

where we have replaced the Hartree groups of operators with nσ being the polariton densities in each component. by c-numbers and defined By introducing the slowly-varying amplitudes

g X 2 µσ = |cσ,q| = gnσ, (5) S −i[E(p)+µσ ]t/~ q cˆσ,p = ˆcσ,pe , (6) 3 we notice existence of a stationary (dˆcσ,p/dt = 0) solution of Eq. (4b) in the form

q 2 ~ β ˆ 2πmS X Ck = cˆ↑,−p+ k cˆ↓,p+ k , (7) µ + µ − ε0 2 2 ↑ ↓ k p where the motion of the bipolariton mode is reduced to that of a pair of polaritons with opposite spins. Here 0 εk = ε − εk with P Erel(p, k) < ˆc k ˆc k > p ↑,−p+ 2 ↓,p+ 2 εk = P (8) < ˆc k ˆc k > p ↑,−p+ 2 ↓,p+ 2 being the kinetic energy of the relative motion in the pair, Erel(p, k) = E(p + k/2) + E(−p + k/2) − 2E(k/2). The condition (1) provides a physical meaning to the solution (7). The objects Cˆk should be regarded as auxi- liary fields describing onset of pair correlations between Figure 1. Sketch of a TMD-based planar microcavity fea- the polaritons, rather than new (quasi)particles. Indeed, turing unitary dipolar polaritons. The active region consists of a homobilayer structure with AA0 or AB stacking, which substituting (7) into the last term of the Hamiltonian (3), allows dipolar excitons with strong oscillator strengths and one obtains an effective model convenient selection rules [38, 39]. Black arrows show the spin X Hˆ 0 = E(p)ˆc† cˆ + orientation of the conduction and valence electronic states in σ,p σ,p the K± valleys. The ”↑” and ”↓” excitons couple to the ca- p,σ vity modes with opposite helicities. Dashed line traces the 1 (9) X † † profile of the two-body interaction potential V↑↓(r) felt by cˆσ,p +qcˆ 0 gσσ0 cˆσ,p cˆσ0,p , 2S 1 σ ,p2−q 1 2 each exciton [33]. This supports a quasi-discrete level (gray p ,p ,q,σ,σ0 1 2 bar) separated from the outer continuum by a barrier due with g↑↑ = g↓↓ = g and to the dipolar repulsion. The resulting polaritons experience strong repulsion and squeezing of their collective states. ~2 β g↑↓ = 0 . (10) 2πm (µ↑ + µ↓ − εp +p ) 1 2 By substituting the slowly varying c-numbers [see the One can see, that the pair correlations between the pola- iφσ iφ definition (6)] cσ,p0 = ρσe and C2p0 = ρe into Eqs. ritons manifest themselves as resonantly strong two-body (4), and omitting the terms scaling as pε/β, one can find interactions. For a broad resonance considered in this work, large magnitude of g↑↓ is primarily due to the rela- φ(t) = φ↑(t) + φ↓(t) ± π/2 tion (1) (the µ ’s are on the order of ε). However, by re- (13) σ φ (t) = φ (0) ducing the difference µ +µ −ε0 (by, e.g., decreasing σ σ ↑ ↓ p1+p2 the polariton density) one may also observe the typical and resonant growth of the interaction strength. Though here √ 0 −1 we have in mind the case µ↑ + µ↓ > ε , the formula ρσ(t) = N cosh (t/τ0) p1+p2 √ (14) (10) can be used on the attractive side µ + µ < ε0 ↑ ↓ p1+p2 ρ(t) = N tanh(t/τ0). as well. Assume now a resonant excitation of a coherent mix- Eq. (14) shows that on the characteristic time scale ture of ”↑” and ”↓” polaritons at some point p on the 0 r dispersion curve E(p). The output signal is registered at 2πm τ0 = (15) the distance l ∼ υgrτ from the excitation spot. Here υgr is βn the corresponding group velocity. For simplicity assume the coherent mixture is entirely converted into the pai- equal densities n = n ≡ n = N/S, which in practice ↑ ↓ red state (7). According to (9) and (10), the modified may be achieved by using linearly polarized light. We polariton blueshift is given by therefore let the following initial (t = 0) configuration : √ 2 0 ~ n β cˆσ,p0 (0) |ψi = N |ψi µ = gn + . (16) (11) σ 2πm (2ng − ε) cˆσ,p6=p0 (0) |ψi = 0. Note, that spreading of the polariton distribution over The bipolariton part of the many-body wavefunction |ψi the k-space region where the dispersion can be approxi- is initially in the vacuum state : mated by a linear function will not affect the result (16) ˆ C2p0 (0) |ψi = 0. (12) since, according to (8), in this region one has εk ≡ 0. 4

The relation (1) refers to the typical case d ∼ dc, where τ0  t . τ, where τ0 is√ given by Eq. (15), one can substi- dc is the critical value at which the true bound state tute ρσ = 0 and ρ = N into the first pair of Eqs. (20) disappears [32]. In general, the width of the resonance to obtain changes from 0 to ∞ (the latter describing the ultimate 2 ±t/τ0 case where the level washes out) as the exciton dipole hδxˆσ(t)i ∼ e (21) moment is tuned from d  d to d  d . For 0 d d 2 ∓t/τ0 c c 6 . c hδyˆσ(t)i ∼ e , one may take β(d) = Bd. The corresponding dependence for the position of the level has the form [13] ε(d) = showing that the polaritons exhibit 100 % squeezing in E(d − dc). either of the two quadratures at the output. Following Ref. [20], one may then consider the ”inter- The requirement τ0  τ sets the lower value of the 0 action enhancement factor” η(d, n) = µσ/µσ − 1 as a polariton density at which the predicted effects may be function of the dipole moment d and density n. Substitu- observed. For sufficiently large β one may operate in the ting the above relations for β(d) and ε(d) into Eq. (16), ultra-dilute regime and even in the few-particle limit. In we obtain Eqs. (10),(16) and (17) the latter is formally achieved by letting n = 0. Strong repulsive interactions between just ~2 Bd η(d, n) = . (17) two polaritons may be particularly promising for imple- 2πmg [2ng − E(d − d )] c mentation of the dual-rail polariton logic [35]. Though our analytical methods do not allow us to esti- The condition (1) has allowed us to neglect the dissi- mate the typical values of the parameters B and E, by pation and make our arguments particularly transparent. virtue of (1) one may expect B  E and, consequently, Pair-breaking events due to leakage of the single photons η  1. The relation (1), in turn, is guaranteed by very from the cavity result in loss of correlations and, at a first low values of the polariton mass m as compared to bare glance, would reduce the degree of squeezing. In practice, excitons [33]. however, this reduction may be fully compensated by the Interestingly, the strong correlations in the paired state noise of the external vacuum (see Ref. [36]), which res- (7) squeeze the polariton wavefunctions. To illustrate this tores the significance of the result (21). point, consider again the situation discussed above, where Our last remark concerns the choice of the sign in Eq. one starts from a coherent state (11) for polaritons and (13). Under the condition (1), the Josephson coupling of a vacuum state (12) for their pairs. Introduce rotated the polariton states to the resonance stabilizes a definite quadratures phase relation during the signal propagation. The initial configuration is, however, chosen stochastically and may xˆ = 1 (ˆc e−iφσ + ˆc† eiφσ ) σ 2 σ,p0 σ,p0 vary from one laser pulse to another. This circumstance (18) yˆ = 1 (ˆc e−iφσ − ˆc† eiφσ ) should be taken into account when verifying the predic- σ 2i σ,p0 σ,p0 tion (21) experimentally. and Besides the already mentioned state-of-the-art in QW 1 −iφ † iφ microcavities [19–21], in Fig. (1) we sketch a possible im- Xˆ = (Cˆ2p e + Cˆ e ) 2 0 2p0 (19) plementation of unitary polaritons with TMD’s. A conve- Yˆ = 1 (Cˆ e−iφ − Cˆ† eiφ). 2i 2p0 2p0 nient choice may be a homobilayer of MoX2, showing large oscillator strengths and spin-valley selection rules ˆ ˆ Writex ˆσ = xσ + δxˆσ and the same fory ˆσ, X, Y . The analogous to the monolayer excitons [38, 39]. In addi- linearized equations of motion for the quadrature fluc- tion, natural separation between the layers here is close tuations read to the threshold dc, the latter being on the order of the r d β exciton Bohr radius [13]. The electric-field tuning of the δxˆ = ± (ρ δXˆ + ρδxˆ ) resonance position in this case of fixed d may be possible dt ↑,↓ 2πmS ↓,↑ ↓,↑ r due to suppression of the satellite carrier wavefunction in d β one of the layers (the residue of the intra-layer exciton) δyˆ↑,↓ = ± (ρ↓,↑δYˆ − ρδyˆ↓,↑) dt 2πmS [38]. r (20) d β To conclude, we predict anomalously large enhance- δXˆ = ∓ (ρ δxˆ + ρ δxˆ ) dt 2πmS ↑ ↓ ↓ ↑ ment of repulsive interactions in a system of dipolar po- r laritons. The proposed model is based on the physics of a d β δYˆ = ∓ (ρ δyˆ + ρ δyˆ ), bound state separated from the outer continuum by a po- dt 2πmS ↑ ↓ ↓ ↑ tential barrier. Our results apply to a wide variety of 2D where the sign ”+” or ”−” corresponds to the two pos- semiconductor heterostructures, such as atomically thin sible choices of the phase shift in Eq. (13), and ρ, ρσ layers of TMD’s and quantum wells. An intriguing pre- are given by (14). At t = 0 one can use Eqs. (18) and diction of our theory is that the resonantly paired pola- 2 2 (11) to find hδxˆσ(0)i = 1/4 and hδyˆσ(0)i = 1/4, the well- ritons represent an efficient source of squeezed radiation. known property of a coherent state [34]. In contrast, at This might be readily verified by examining the statis- 5 tics of emitted photons with the balanced homodyne de- tection [37]. The idea of using the shape resonance to produce strong pair correlations and squeezing at ultra- low polariton densities opens wide perspectives for future research and applications. Thus, an interesting new di- rection would be application of the physics discussed in this work to the recently established field of topological polaritons [40, 41]. The author acknowledges the support by Russian Science Foundation (Grant No. 18-72-00013). I also thank Misha Glazov for helpful reading of the manuscript.

SUPPLEMENTAL MATERIAL

Polariton Fano-Anderson model Figure 2. Schematic illustration of the two-body exciton in- Consider 2D excitons coupled to photon modes in a teraction potnetial V↑↓(r) (bold line). Due to the 2D kinema- planar microcavity of width Lc and area S with perfect tics the effective 1D potential felt by the radial wavefunction mirrors. The photon dispersion reads of the relative motion of excitons (dashed line) has an at- tractive tail which strongly suppresses the dipolar potential p 2 2 2 barrier. As a result, the tunneling amplitude is greatly enhan- ~ωph(p) = ~c p + π /Lc , (22) ced. For polaritons this yields exponential suppression of the that at low p has a massive form ratio of the resonance energy ε to its width β, Eq. (48).

~2p2 ~ωph(p) = Eph + (23) 2mph distances this potential is identical to V↑↑(r) and V↓↓(r), and is characterized by the dipolar tail 2r /mr3, where 2 ~ ∗ with mph = π ~/cLc. The exciton dispersion is given by me2d2 2p2 ~ r∗ = 2 (29) ~ωX(p) = EX + , (24) κ~ 2mX is the dipolar length. At r ∼ r , however, it features a and in typical all-dielectric microcavities one has ∗ potential well. At sufficiently small d this potential well

mX  mph. (25) admits a true bound state - biexciton. Evolution of this bound state into a resonance as one increases d was des- 4 Thus, for GaAs-based structures mX ∼ 10 mph. cribed in [13]. Here we aim at extending that phenome- The prototype Hamiltonian may be recast as a sum of nology to polaritons. ˆ ˆ ˆ two contributions H = Hsp + V , where The standard procedure is to diagonalize the single- particle term (26) by doing the Hoppfield transformation ˆ X † X ˆ† ˆ Hsp = ~ωph(p)ˆaσ,paˆσ,p + ~ωX(p)bσ,pbσ,p p,σ p,σ ˆ (26) pˆσ,p = upbσ,p + υpaˆσ,p (30) X ˆ† ˆ † + i g(p)(b aˆσ,p − bσ,paˆ ) ~ σ,p σ,p 2 2 p,σ with |up| + |υp| = 1 in order to preserve the boson commutation relation. One gets two eigenvalues E±(p) is the single-particle term describing formation of corresponding to the upper and lower polariton branches. exciton-polaritons (see below) and The Hoppfield coefficients are related by

ˆ 1 X ˆ† ˆ† ˆ ˆ ± V = b b 0 Vσσ0 (q)bσ,p bσ0,p i~gυσ,p σ,p1+q σ ,p2−q 1 2 ± 2S 0 uσ,p = . (31) p1,p2,q,σ,σ ~ωX(p) − E±(p) (27) being the two-body interaction between the excitons, ± ± In the long-wavelength limit p → 0 we may take up ≈ u0 ± ± Z and υp ≈ υ0 . Furthermore, assume Eph = EX in Eqs. −iqr Vσσ0 (q) = e Vσσ0 (r)dr. (28) (23) and (24). We get

We shall be particularly interested in V (r), schemati- ˆb ± aˆ ↑↓ pˆ± = σ,p√ σ,p (32) cally shown by the solid black line in Fig. 2. At large σ,p 2 6 and The dissociation rate of φ(r) may be estimated by consi- dering the tunnelling in the effective one-dimensional mo- E (0) = E ± g. (33) ± X ~ del

One may also estimate m+ = m− ≡ m ≈ 2mph. By 2 ~ 00 virtue of (25) χ + [E − V (r)]χ = 0, (40) m eff m  mX. (34) which√ is obtained from (39) by substituting φ(r) = We define χ(r)/ r. Here − 2 2   cˆσ,p ≡ pˆp (35a) ~ ~ 1 1 Veff (r) = V↑↓(r) − 2 ≈ 2 3 − 2 , (41) ˆ + 4mr mr∗ x 4x dσ,p ≡ pˆp (35b) 2 3 for the lower and upper polaritons, respectively. In terms where we have taken V↑↓(r) ≈ ~ r∗/mr and defined of the new operators the single-particle Hamiltonian (26) x = r/r∗. This potential is shown by the dashed line reads in Fig. 2. One can see that the tunnelling at E = 0 oc- curs in a narrow region from r ≈ r to r = 4r . For ˆ X † X ˆ† ˆ a ∗ b ∗ Hsp = E−(p)ˆcσ,pcˆσ,p + E+(p)dσ,pdσ,p. (36) a crude estimate we may use the quasiclassical approxi- p,σ p,σ mation [12] to calculate the tunnelling probability as a By substituting product of the number of the particle beatings per unit of time ∼ p/mr ∼ /mr2 and the barrier transmission ˆ ∗ ~ ∗ ˆ cˆσ,p + dσ,p bσ,p = √ (37)  r  2 b 2 Z p D(E) = exp − m(Veff (r) − E)dr . (42) into Eq. (38) and assuming occupation of the lower po- ~ lariton branch only, one gets ra

ˆ 1 X † † By evaluating the integral one gets D(0) ∼ 0.25. Thus, V = cˆ cˆ 0 Vσσ0 (q)ˆcσ,p cˆσ0,p . σ,p1+q σ ,p2−q 1 2 we may write 8S 0 p1,p2,q,σ,σ (38) 2 β ∼ ~ . (43) Thus, one can see that the low-momentum scattering of 2 mr∗ polaritons occurs in the same potential (rescaled by a numerical factor) as the scattering of excitons, but with Hence, we see that, by virtue of (34), the bipolariton re- a drastically reduced effective mass (34). By changing sonance width should greatly exceed the analogous quan- notations E−(p) ≡ E(p) and repeating the arguments tity for bare excitons βXX, of Ref. [13], one may derive the Fano-Anderson model β m (2) used in the main text. The predictions and physical ∼ X  1. (44) meaning of this model depend on the ratio ε/β. Below βXX m we shall argue that the relation (34) implies ε/β  1, On the other hand, by using the formula so that quite generally one may expect the polaritons to interact via a broad resonance. 2 ε = − ~ , (45) ma2

Dissociation of a dipolar bipolariton in two where dimensions ∆d/(dc−d) a ∼ r∗e (46)

As we have established in the previous section, the and taking into account that ∆d ∼ β [13], we may write polariton-polariton interaction potential apart of a nu- ε m  m ∆d  merical factor coincides with the bare exciton potential. ∼ X exp − X XX . (47) For the ↑↓ scattering channel of interest here this poten- εXX m m dc − d tial is schematically illustrated in Fig. 2. We first note, that the nominal huge potential barrier becomes effecti- Thus, even if one has ∆dXX ∼ (dc −d) and, consequently, vely suppressed due to the particularities of the 2D kine- εXX ∼ βXX for the bare excitons, for the polaritons one matics. To see this, consider the radial 2D Schrodinger would obtain equation for the relative motion with zero angular mo- ε ∼ e−mX/m  1. (48) mentum (s-wave) : β  2  ∂ 1 ∂ m 2 We conclude that a dipolar ”bipolariton” may exist only + − V↑↓(r) + k φ(r) = 0 (39) ∂r2 r ∂r ~2 as a broad resonance. This result generalizes the earlier 7 estimate of the bipolariton binding energy at d = 0 (non- [23] In principle, the condition (1) alone is sufficient to ex- dipolar excitons) [30]. clude the formation of an s-wave bipolariton condensate on the repulsive side of the resonance. The requirement τ  τp then ensures the absence of analogous equilibrium phases in higher partial-wave scattering channels. [24] Giovanna Panzarini, Lucio Claudio Andreani, A. Armi- ∗ Electronic adress : [email protected] tage, D. Baxter, M. S. Skolnick, V. N. Astratov, J. S. [1] I. Carusotto and C. Ciuti, Rev. Mod. Phys. 85, 299 Roberts, Alexey V. Kavokin, Maria R. Vladimirova, and (2013). M. A. Kaliteevski, Phys. Rev. B 59, 5082 (1999). [2] L. V. Keldysh and A. N. Kozlov, Sov. Phys. JETP 27 [25] Lundt, N., Dusanowski, L., Sedov, E. et al. Optical valley 521 (1968). Hall effect for highly valley-coherent exciton-polaritons in [3] E. Hanamura and H. Haug, Phys. Rep. 33, 209 (1977). an atomically thin semiconductor. Nat. Nanotechnol. 14, [4] P. B. Littlewood et al., J. Phys. : Condens. Matter 16, 770 (2019). S3597 (2004). [26] Wertz, E., Ferrier, L., Solnyshkov, D. et al. Spontaneous [5] P. Renucci, T. Amand, X. Marie, P. Senellart, J. Bloch, formation and optical manipulation of extended polari- B. Sermage, and K. V. Kavokin Phys. Rev. B 72, 075317 ton condensates. Nature Phys. 6, 860-864 (2010). (2005). [27] M. Z. Maialle, E. A. de Andrada e Silva, and L. J. Sham, [6] T. Lahaye, C. Menotti, L. Santos, M. Lewenstein, and T. Phys. Rev. B 47, 15776 (1993). Pfau, Rep. Prog. Phys. 72, 126401 (2009). [28] M. M. Glazov, T. Amand, X. Marie, D. Lagarde, L. [7] R. Low, H. Weimer, J. Nipper, J. B. Balewski, B. But- Bouet, and B. Urbaszek, Phys. Rev. B 89, 201302(R) scher, H. P. Buchler, and T. Pfau, J. Phys. B At. Mol. (2014). Opt. Phys. 45, 113001 (2012). [29] Schneider, C., Glazov, M.M., Korn, T. et al. Two- [8] C. Chin, R. Grimm, P. Julienne, and E. Tiesinga, Rev. dimensional semiconductors in the regime of strong light- Mod. Phys. 82, 1225 (2010). matter coupling. Nat. Commun. 9, 2695 (2018). [9] Gang Wang, Alexey Chernikov, Mikhail M. Glazov, Tony [30] G. C. L. Rocca, F. Bassani, and V. M. Agranovich, J. F. Heinz, Xavier Marie, Thierry Amand, and Bernhard Opt. Soc. Am. B 15, 652 (1998). Urbaszek, Rev. Mod. Phys. 90, 021001 (2018). [31] Noteworthy, anomalous behaviour of the polariton inter- [10] E. J. Sie, A. Steinhoff, C. Gies, C. H. Lui, Q. Ma, M. Ros- action had been observed also for non-dipolar species [N. ner, G. Schonhoff, F. Jahnke, T. O. Wehling, Y.- H. Lee, Takemura, et al. Nat. Phys. 10, 500 (2014)]. The authors J. Kong, P. Jarillo-Herrero, and N. Gedik, Nano Letters attempted to explain their observation by the presence 17, 4210 (2017). of a biexciton state which could mediate the polariton [11] Y. You, X.-X. Zhang, T. C. Berkelbach, M. S. Hybertsen, scattering. The term similar to the last term in the Ha- D. R. Reichman, and T. F. Heinz, Nature Physics 11, miltonian (3) was introduced in the theoretical model. 477 (2015) ; Z. Ye, L. Waldecker, E. Y. Ma, D. Rhodes, However, the physical origin of this term in that case was A. Antony, B. Kim, X.-X. Zhang, M. Deng, Y. Jiang, Z. obscure : for non-dipolar polaritons there is no potential Lu, D. Smirnov, K. Watanabe, T. Taniguchi, J. Hone, barrier linking a bound state to the outer continuum. T. F. Heinz, Nat. Commun. 9, 3718 (2018) ; Z. Li, T. According to the general arguments we have already re- Wang, Z. Lu, C. Jin, Y. Chen, Y. Meng, Z. Lian, T. ferred to in the introduction, low-momentum collisions Taniguchi, K. Watanabe, S. Zhang, D. Smirnov, S.-F. between the non-dipolar polaritons with energies close Shi, Nat. Commun. 9, 3719 (2018). to the bound state produce vanishingly small effective [12] L. D. Landau and E. M. Lifshitz, Quantum Mechanics interaction. (Pergamon Press, Oxford, 1969). [32] A. D. Meyertholen and M. M. Fogler, Phys. Rev. B 78, [13] S. V. Andreev, Phys. Rev. B 94, 140501(R) (2016). 235307 (2008) ; I. V. Bondarev and M. R. Vladimirova, [14] S. V. Andreev, Phys. Rev. B 92, 041117(R) (2015). Phys. Rev. B 97, 165419 (2018). [15] S. V. Andreev, Phys. Rev. B 95, 184519 (2017). [33] See Supplemental Information. [16] C. Schindler and R. Zimmermann, Phys. Rev. B 78, [34] M. O. Scully and M. S. Zubairy, Quantum Optics (Cam- 045313 (2008). bridge University Press, 1997). [17] S. V. Andreev, Phys. Rev. Lett. 110, 146401 (2013). [35] Menon, Vinod, Deych, Lev, Lisyansky, Alexander, Nonli- [18] S. V. Andreev, A. A. Varlamov and A. V. Kavokin, Phys. near optics : Towards polaritonic logic circuits. Nat. Pho- Rev. Lett. 112, 036401 (2014). ton. 4. 345 (2010). [19] S. I. Tsintzos, A. Tzimis, G. Stavrinidis, A. Trifonov, [36] B. Yurke, Phys. Rev. A 29, 408 (1984). Z. Hatzopoulos, J. J. Baumberg, H. Ohadi, and P. G. [37] G. Breitenbach, S. Schiller, and J. Mlynek, Nature 387, Savvidis, Phys. Rev. Lett. 121, 037401 (2018). 471 (1997). [20] I. Rosenberg, D. Liran, Y. Mazuz-Harpaz, K. West, L. [38] Iann C. Gerber, Emmanuel Courtade, Shivangi Shree, Pfeiffer, and R. Rapaport, Sci. Adv. 4, 8880 (2018). Cedric Robert, Takashi Taniguchi, Kenji Watanabe, An- [21] Emre Togan, Hyang-Tag Lim, Stefan Faelt, Werner Weg- drea Balocchi, Pierre Renucci, Delphine Lagarde, Xavier scheider, and Atac Imamoglu, Phys. Rev. Lett. 121, Marie, and Bernhard Urbaszek, Phys. Rev. B 99, 035443 227402 (2018). (2019). [22] L. V. Butov and A. V. Kavokin, Nat. Photon. 6, 2 (2012) ; [39] Jason Horng, Tineke Stroucken, Long Zhang, Eunice Y. M. Klaas, E. Schlottmann, H. Flayac, F. P. Laussy, F. Paik, Hui Deng, and Stephan W. Koch Phys. Rev. B 97, Gericke, M. Schmidt, M. v. Helversen, J. Beyer, S. Brod- 241404(R) (2018). beck, H. Suchomel, S. Hofling, S. Reitzenstein, and C. [40] A. V. Nalitov, D. D. Solnyshkov, and G. Malpuech, Phys. Schneider, Phys. Rev. Lett. 121, 047401 (2018). Rev. Lett. 114, 116401 (2015). 8

[41] Tomoki Ozawa, Hannah M. Price, Alberto Amo, Na- berberg, and Iacopo Carusotto, Rev. Mod. Phys. 91, than Goldman, Mohammad Hafezi, Ling Lu, Mikael C. 015006 (2019). Rechtsman, David Schuster, Jonathan Simon, Oded Zil-