Quantum Many-Body Physics with Ultracold Polar Molecules: Nanostructured Potential Barriers and Interactions

Andreas Kruckenhauser,1, 2 Lukas M. Sieberer,1, 2 Luigi De Marco,3 Jun-Ru Li,3 Kyle Matsuda,3 William G. Tobias,3 Giacomo Valtolina,3 Jun Ye,3 Ana Maria Rey,3, 4 Mikhail A. Baranov,1, 2 and Peter Zoller1, 2 1Center for Quantum Physics, University of Innsbruck, Austria 2Institute for Quantum Optics and Quantum Information of the Austrian Academy of Sciences, Innsbruck, Austria 3JILA, University of Colorado and National Institute of Standards and Technology, and Department of Physics, University of Colorado, Boulder, CO 80309, USA 4Center for Theory of Quantum Matter, University of Colorado, Boulder, CO 80309, USA (Dated: June 9, 2020) We design dipolar quantum many-body Hamiltonians that will facilitate the realization of exotic quantum phases under current experimental conditions achieved for polar molecules. The main idea is to modulate both single-body potential barriers and two-body dipolar interactions on a spatial scale of tens of nanometers to strongly enhance energy scales and, therefore, relax temperature requirements for observing new quantum phases of engineered many-body systems. We consider and compare two approaches. In the first, nanoscale barriers are generated with standing wave optical light fields exploiting optical nonlinearities. In the second, static electric field gradients in combination with microwave dressing are used to write nanostructured spatial patterns on the induced electric dipole moments, and thus dipolar interactions. We study the formation of inter- layer and interface bound states of molecules in these configurations, and provide detailed estimates for binding energies and expected losses for present experimental setups.

I. INTRODUCTION dipole moment in Debye induced by a DC electric field 12 kV/cm) implies binding energies less than kHz. Thus, the design of strongly interacting many-body Recent experimental progress with ultracold polar systems with dipole moments less than a Debye will re- molecules [1–14] opens up unique opportunities to de- quire, or strongly benefit from going to much smaller sign novel quantum many-body systems [15–18]. Elec- distance scales than those provided by the optical wave- tric dipole moments, induced by external electric fields in length scale [41–43]. In the present paper we explore the manifold of rotational molecular ground states, give various possibilities of designing quantum many-body rise to long-range and anisotropic dipolar interactions, systems with polar molecules, involving both nanostruc- which are potentially larger than those realized with mag- tured potential barriers and dipolar interactions modu- netic interactions in atomic systems [19–32]. Thus po- lated on the scale of tens of nanometers, where the goal lar molecules promise the realization of strongly inter- is to enhance relevant energy scales. We do this by ex- acting quantum many-body systems, e.g., as Hubbard ploiting the unique properties offered by polar molecules: or spin models in optical lattices with strong nearest- This includes the long life-time of the molecular rota- neighbor or long-range interactions [33–36], or in bi- tional states in the ground state many-fold, and the pos- layer systems with strong controllable inter-layer cou- sibilities of controlling induced electric dipole moments pling which can be tuned attractive or repulsive [37– of rotational states. 40]. In an optical lattice, or a bilayer system created We will first discuss an all optical scheme (see Sec.II), with standing wave laser fields, the interaction energy were following Refs. [44–48] for atoms, a nanoscale barrier 2 3 between dipolar particles scales as Eint d /w with d can be realized by exploiting the nonlinear response of a the (induced) dipole moment and w the lattice-∼ or bilayer molecule exposed to spatially nonuniform optical light spacing provided by w = λ/2 as half the optical wave- fields in a Λ-configuration. These nanostructured barri- length [37–40]. However, only polar molecules with the ers allow to split a single well in a 1D optical lattice, thus largest electric dipole moments (LiRb 3.99 D and LiCs forming a bilayer system with separation on a scale of 5.39 D) fulfill the promise of large off-site interactions ap- tens of nanometers. This leads to significant interactions,

arXiv:2001.11792v2 [cond-mat.quant-gas] 8 Jun 2020 proaching the scale of tens of kHz, comparable or larger and inter-layer binding energies even for molecules with than the other relevant energy scales. These interactions comparatively small electric dipole moments. Assuming can be quantified by the binding energy EB of a pair w = 60 nm, the binding energy becomes EB > 10 kHz h of molecules in a head-to-tail configuration in a bilayer for KRb for an induced dipole moment given above. Ta- system formed by a 1D optical lattice with w = λ/2 bleI provides a list of interaction and binding energies for (see Fig.1(a)). In addition, EB must be larger than various molecules. Besides binding energies, we analyze −2 available temperatures, EB,Eint & kBT 1 kHz h [8]. in detail loss rates expected for this setup, Γ 10 EB. For polar molecules with small electric dipole≈ moments, For polar molecules there is in addition the opportunity≈ meeting these requirements can be challenging: For KRb to choose a pair of rotational ground states in the Λ- and w λ/2 = 250 nm, and assuming d = 0.33 D (as system with induced dipole moments having opposite ≡ 2

(a) (b) (a) (b) 3

e 2 | i ⌦c(z) (c) s 1 ⌦p 0 1, 0 | i Figure 1. (a) Polar molecules in a bilayer system in a head- 0, 0 -1 to-tail dipolar configuration attract each other, and can form | i e an inter-layer bound state. Here a nanoscale optical barrier e -4 -2 0 2 4 separating the two layers w  λ/2 provides provides strongly Figure 2. (a) Raman coupled Λ-system. The two rotational states |0, 0i and |1, 0i are coupled by the control and probe enhanced energy scales for the binding energy EB (see Sec.II). e e (b) Strong electric field gradients allow the induced electric laser via the electronically excited state |ei. (b) The barrier + Vna(z) arises as a nonadiabatic correction to the slow motion dipole moment of a molecule dz (x), to change sign on a short spatial distance scale s (see text). The resulting dipolar in- of a molecule in the dark-state Born-Oppenheimer channel teraction allows the formation of interface bound states (see of a Λ-system with spatially inhomogeneous Rabi frequen- Sec.III). cies. The control Rabi beam vanishes at z = 0 which deter- mines the position of the potential barrier Vna(z). In the lower panel, the dark-state decomposition as a function of position is schematically illustrated. (c) A double well potential for sign. This results in repulsive interactions where one dipolar molecules is created by inserting an optical nanoscale molecule is inside the barrier, i.e.,effectively increasing barrier Vna(z) into a single potential well generated by VL(z). the barrier height, and a corresponding suppression of In order to split the potential well into two sites, the width the tunneling and thus decay rate. This can be relevant l of VL(z) has to be smaller than the ground state size aL of for chemically reactive molecules [16]. VL(z). The effective separation of the molecules is w = 2aL. Second, we discuss an all electric scheme (see Sec.III), where instead of the field gradients provided by an opti- cal standing wave we exploit electric field gradients [49], plicable in this case. In TableII we summarize binding resulting in position-dependent energy shifts of molecu- energies EB for various molecules for typical parameters. lar states. In contrast to the nanoscale optical (single- This all-electric scheme may also be of interest when the particle) potential barrier discussed above, our aim is optical manipulation of molecules is not available. now to modulate the (two-particle) dipolar interaction between molecules on a short distance scale. The under- lying mechanism is to couple a pair of rotational states (1) (2) with opposite induced dipole moments dz = dz with II. INTER-LAYER BOUND STATES IN resonant microwave fields. The resulting dressed− states OPTICAL Λ-SYSTEMS of this two-level atom acquire position-dependent dipolar + (1) 2 moments dz (x) = dz (x/s)/ 1 + (x/s) illustrated We discuss below bilayer systems for polar molecules with layer separation of tens of nanometers (see Fig.2). in Fig.1(b), with the spatialh scalep of the variationi set by the electric field gradient. Thus we design a two-body Our work builds on earlier proposals [44, 47] and exper- + + 3 iments [48] to realize an optical nanoscale barrier in Λ- interaction V (ρ1, ρ2) = dz (x1)dz (x2)/ ρ1 ρ2 , where | th− | ρj = (xj, yj) labels the position of the j molecule in systems. In Sec.IIA we first summarize how molecular the xy-plane, see Fig.1(b). Molecules on different sides Λ-systems which consist of two rotational levels coupled of the plane x = 0 will attract each other allowing for the by a Raman transition can be trapped in a double-layer formation an interface bound state. This state is stabi- geometry with sub-optical-wavelength spacing as illus- lized by the position dependent dipole-dipole interaction trated in Fig.2. In Sec.IIB we give a detailed account which vanishes the molecules approach x = 0, and be- of loss channels in such systems, and we identify a hier- come repulsive for molecules on the same side. We find archy of scales which ensures the suppression of losses. that a bound state occurs above a critical value of dipo- Finally, in Sec.IIC we study the formation of inter-layer lar interactions. For molecules with d > 3 D (as for LiRb bound states of molecules. and LiCs) this results in binding energies of tens of kHz The model underlying our discussion is a hetero- for electric field gradients 5 kV/(cm mm). We note nuclear diatomic molecule in its electronic (X1Σ+ for that while also in this scheme∼ achievable binding energies KRb [49]) and vibrational ground state placed in a strong are in principle enhanced by the nanoscale separation of external electric field. Under such conditions, the Stark molecules, this enhancement is partially compensated by effect dominates over the hyperfine interactions (this the reduction of the dipolar moment in the vicinity of the happens already for electric field strengths of tens of interface and the aforementioned scaling of Eint is not ap- V/cm [50, 51]) and, omitting the latter, the Hamiltonian 3

(a) (b) A. Summary of nanoscale potentials in optical Λ-systems

To prepare the ground for the discussion in Secs.IIB andIIC and to fix the notation we find it worthwhile for the reader to summarize how to engineer a poten- tial barrier on the nanoscale of size l as proposed in the Refs. [44] and [47]. The potential barrier is used to cut Figure 3. Energy (a) and induced dipole moment (b) as a a single well of an optical potential VL(z) into two sites function of applied electric field for the lowest energy eigen- as illustrated in Fig.2(c). The optical potential can be Λ states of the molecular Hamiltonian Eq. (1) with m = 0. 0 approximated in the vicinity of its potential minimum z0 and 2LS denote the offset fields chosen in Secs.II andIII, 2 2 0 by VL(z) mωL(z z0) /2, where ωL is the harmonic respectively. oscillator≈ frequency.− The size of the ground state wave function in this potential is aL = ~/(mωL) and in the limit l a the potential well is split into two sites.  L p for a molecule is given by To this end, we consider the motion of a single molecule along the z-axis, which is described by ˆ p2 HM = HR d , (1) H = z + V (z) + H0 (z), (2) − · Λ 2m L Λ where H = BNˆ 2 is the Hamiltonian of a rigid rotor, B where pz = i~∂z is the z-component of the momentum R ~ p − the rotational constant and Nˆ is the orbital angular mo- operator = i~ . For the moment, we consider only the motion along− ∇ the z axis, and we restore the full 3D mentum operator. Further,  is the external electric field, 0 dˆ = ˆr/ ˆr is the dipole moment operator expressed in form of the Hamiltonian in the next section. HΛ(z) is a termsD of| the| position operator ˆr, and is the perma- position dependent Λ-system Hamiltonian which is given nent molecule frame dipole moment.D For  = 0, the in a proper rotating frame by eigenstates and energies of H = H are N, m and M R | i ∆ Ωc(z)/2 Ωp/2 E = BN (N + 1), respectively, where N = 0, 1, 2,... 0 − N ~ H = ~ Ωc(z)/2 0 0 . (3) denotes the angular momentum quantum number, and Λ   Ωp/2 0 0 m = N, N + 1,...,N its projection onto the quanti- − −   zation axis. Here and in the following, we choose both As illustrated in Fig.2(a), the first leg of the Λ-system the quantization axis and the z-coordinate axis along is a weak control laser Ω which couples the states 0, 0 p | i the direction of the electric field  = ez, such that and an electronically excited state e ; The second leg is the angular momentum projection quantum number m | i a strong standing wave control laser Ωc(z) = Ωc sin(ekcz) is conserved. For  = 0, we denote the eigenstates which couples the states 1, 0 and e . ∆ denotes the de- 6 | i | i of HM by N, m  and the corresponding energy lev- tuning of the Raman transition, see Fig.2(a) and (b).  | i els EN,m( ) are shown in panel (a) of Fig.3. Panel In the case of KRb, a possiblee candidate for e is a (b) depicts thee induced dipole moment which is given vibrational excitation of the 11Π electronically| excitedi by thee change of the energy with the electric field , state [53], for which the Franck-Condon factor is maxi- N,m 1 d () = ∇E () =  N, m dˆ N, m . The in- mized [54]. Rabi coupling between the state 1 Π and the − N,m h | | i ducede dipole moment is aligned along the electric field, absolute ground state of KRb has been demonstrated in i.e., along the z axis.e From thise point on,e to simplify the Ref. [53] in the absence of an electric offset field. We dis- notation, we drop the explicit dependency on  for all cuss the decay of molecules which results from the finite quantities. lifetime of the electronically excited state in Sec.IIB. A Raman coupled Λ-system always hosts one dark As mentioned above, for strong electric fields the Stark state at zero energy and two bright states, which we de- effect dominates over the hyperfine interactions and, note by 0 and , respectively. The corresponding therefore, the nuclear spins become decoupled from the z z eigenenergies| i are given|±i by orbital angular momentum. In this case, both m and m1 + m2, where m1,2 is the the nuclear spin projection ~ EΛ = 0 and EΛ = ∆ Ω2 + Ω2(z) + ∆2 . (4) of the atom 1 and 2, respectively, become good quantum 0 ± 2 − ± p c numbers [50, 51]. With an additional magnetic field of h q i and in particular, the dark state reads a few hundred Gauss, the remaining degeneracy among hyperfine states is lifted, making the individual compo- 1 0 z = z/l 0, 0 1, 0 , (5) nents m1 and m2 good quantum numbers. We therefore | i 1 + (z/l)2 | i − | i consider the molecules to be in a single hyperfine state,   e e e.g., the hyperfine groundstate mK = 4 and mRb = 3/2 where we used thatp kc z z0 . kcaL 1. The character- for 40K87Rb [52]. − istic length scale on which| − | the structure of 0 changes | iz 4 is given by and resonant transitions between the states ψL,R(z) and states in the channel are possible. The decay rate − l = Ωp/(Ωc kc). (6) from the zero-energy BO channel to the channel can be obtained by using Fermi’s golden rule.− The derivation For z l, the dark state is essentially equal to the | |  presented in AppendixA yields for ∆ = 0 internal state 0, 0 ; The contribution to 0 z from 1, 0 is | i | i | i Γ l relevant only in a region of size l around the zero crossing Λ = 2.5β2 √κ exp ( 1.75 κ) , (9) of Ωc(z) as illustratede in Fig.2(b). e ωL aL − We consider the limit of slow motion of the molecules, where κ is the square root of the ratio of the gap between in which the eigenvalues and the eigenstates of H0 (z) Λ the zero-energy and channels and the height of the form decoupled Born-Oppenheimer (BO) channels [55– barrier, κ = ( Ω /2)−/( 2/2ml2), and the number β 59]. Nonadiabatic corrections yield two types of contri- ~ p ~ 0.34 is determined by the exact position of the barrier≤ butions: On the one hand, they describes nonadiabatic and the wave function of the trapped state. This result is channel couplings; on the other hand, they gives rise valid in the limit κ 1 or equivalently l/a ω /Ω , to repulsive potential barriers. Below, we discuss con- L L c when Γ is exponentially suppressed. The gap between ditions under which the nonadiabatic channel couplings Λ the zero-energy and channels and thus the parameterp are negligible. Then, the Hamiltonian for the motion of a κ is enhanced in the− far blue-detuned regime in which molecule prepared in the dark state, i.e., the zero-energy ∆ Ω , and where according to Eq. (4) the size of channel, can be written as  | p,c| the gap is equal to ~∆. In this case, Eq. (9) provides only 2 an estimate of the decay rate, because it does not take ad pz HΛ,00 = + VL(z) + Vna(z), (7) into account the change of the non-adiabtic couplings 2m with ∆. where the nonadiabatic potential barrier is independent In contrast to rotational excitations, electronically ex- on the value of ∆ and reads cited states have a non-negligible decay rate γe. If the motion of molecules in the zero-energy or dark-state ~2 1 channel is perfectly adiabatic, they are not affected by V (z) = , (8) na 2ml2 [1 + (z/l)2]2 this decay. However, nonadiabatic corrections couple the dark-state channel to the or bright-state chan- as introduced in Refs. [44] and [47], see also Appendix nels, and thus lead to a small admixture± of the electron- A 1. The characteristic scale l from Eq. (6), which defines ically excited state e to the dark state, which in turn both the width and the height of the potential barrier can lead to inelastic| i scattering of photons. This effect Vna(z), is determined by the Rabi frequencies Ωp and Ωc is strongly suppressed in the far blue-detuned regime: and the wave vector kc, see Fig.2. By choosing these pa- On the one hand, for ∆ Ωp,c , the BO-channel is rameters such that a l, a single potential well gener- shifted down energetically, and | the| mixing− between the L  ated by VL(z) is split into two sites. This is illustrated in dark-state and channels is negligible. On the other − Fig.4(a), where we show the ground ( ψR(z)) and first ex- hand, the gap between the + and zero-energy channels cited (ψL(z)) state in the zero-energy BO channel, which is reduced. To estimate the mixing between the dark- are located, respectively, to the right and left of the bar- state and + channels, we first analyze wave functions in rier. We obtain the states ψL,R(z) by numerically di- the + channel. For ∆ & Ωp,c the + bright-state energy agonalizing the Hamiltonian in Eq. (7). An analytical in Eq. (4) can be approximated| | around z = 0 as a har- 2 2 2 discussion of the suppression of wave functions inside the monic potential, E+(z) Ωc /(4∆) + mω+z /2, where barrier is provided in AppendixA. 2 2 ≈ ω+ = Ωc ~ kc /(2m∆) takes the role of the harmonic oscillator frequency. The + channel hosts a barrier which is for far bluep detuning equivalent to the one in the zero- B. Loss channels and hierarchy of scales energy channel. Therefore, as discussed in AppendixA2, eigenfunctions in the + channel are suppressed within the The Hamiltonian Eq. (7) ignores nonadiabatic channel barrier by a factor of l/a+, where a+ = ~/(mω+) is the couplings. These couplings induce decay of states in the oscillator length. The admixture of + channel wave func- zero-energy BO channel to the other channels. In the tions can be approximated to first orderp in the diabatic vicinity of z = 0, the energies of the bright states give rise corrections by c (l/a )(l/a )V (0)/∆E, where the + ≈ L + na to trapping and antitrapping of molecules in the + and factors l/aL and l/a+ describe the reduction of the dark- channels, respectively. Therefore, at low energies, the state and bright-state wave functions, respectively. The +− channel hosts a discrete set of trapped states. None of coupling matrix element is approximately given by the these states are resonant with the states ψL,R(z) in the height of the barrier Vna(0) [47], and ∆E is the energy zero-energy channel if the minimal gap ~Ωp/2 between gap. For ∆ > Ωc the gap o the lowest energy state in the the zero and the + channels is larger than the spacing of + BO-channel is ∆E ~ω+/2, which yields a relatively 2≈ levels ~ωL in the zero-energy channel. In contrast, the small admixture, c+ < 0.1 for l/aL = 0.1. In the far antitrapped≈ states in the channel form a continuum, blue-detuned limit,| the| amplitude of the excited state e − | i 5 in the + channel is small and therefore a reduced decay (a) (b) + 2 2 rate γe γeΩp/(8∆ ) (see Sec. B2 in the supplemen- tary material≈ of Ref. [44]) has to be used to estimate the inelastic scattering rate in the zero-energy BO channel, 0 + 2 −4 γe γe c+ 10 γe, where γe is typically on the or- der≈ of tens| | of≈ MHz [53, 60]. Inelastic light scattering is 0 negligible when γe ωL, which is indeed the case for ω 2π 88 kHz as considered in Sec.IID. L ≈ × In addition to the decay of the states ψL,R(z) in the zero-energy BO channel due to non-adiabatic channel couplings and inelastic light scattering, the stability of a Figure 4. (a) Spatial probability densities for the ground many-body system that is loaded into the states ψL,R(z) (ψR(z)) and first excited state (ψL(z)) in the zero-energy is reduced by inelastic collisions, like chemical reactions BO channel. The dashed baselines indicate the corresponding [61], which can occur at short distances. This effects are eigenenergies. The peaks of the wave functions are separated minimized when ψL(z) and ψR(z) are maximally local- by w = 2aL. Parameters are l/aL = 0.3 and z0/aL = 0.3. ized on either side of the barrier. As a measure of the Energies and the probability density of ψR(z) correspond to degree of localization we calculate the wave function leak- the left-hand and right-hand axis, respectively, and L denotes ∞ age = ,0 dz ψ (z) 2. The leakage is affected the numerical box size. The scale for the probability density OL,R 0,−∞ | L,R | by the length scales a , l and z , which determine the of ψL(z) is the same as for ψR(z), shifted to the correspond- R L 0 ing baseline. (b) Leakage for ψ (z) and ψ (z) through the shape of the wave functions. For z /a = 0.3, in Fig.4(c) L R 0 L barrier as a function of l/a with fixed ratio z /a = 0.3. we observe minimal leakage for a range of small values L 0 L of l/aL. In particular, leakage is negligible in the limit l/a 1. L  fermionic molecules. In particular, we study the forma- Even when leakage of the eigenstates of the tion of bound states between molecules on the left and single-particle Hamiltonian (7) is strongly suppressed, right of the optical nanoscale barrier. We assume a 3D ge- molecules which are loaded into the potential well on one ometry, in which the motion of molecules is not restricted side of the barrier can tunnel through the barrier and col- in the xy plane. For molecules in the zero-energy BO lide inelastically with molecules on the other side. Tun- channel, the position-dependent internal state is given neling through the barrier can be estimated by the trans- by 0 in Eq. (5). The induced dipole moment of this mission coefficient t 4E/(π2V (0)) for an incoming | iz ≈ na state is oriented along the static electric field, i.e., along plane wave with energy E Vna(0) [46]. The tunneling p 2 2 the z axis. Therefore, the interaction of molecules within rate is then given by J = t E/h, where t is the prob- one of the layers L or R is always repulsive [62], whereas ability for a molecule to tunnel through the barrier and molecules in different layers experience attractive inter- E/h is the attempt frequency. For a particle prepared actions when their separation ρ = xex + yey in the xy in ψL,R(z), we set E = EL,R, which yields a tunneling 2 2 3 2 2 2 plane has sufficiently small magnitude ρ = x + y , so constant of JL,R/ωL = 8/π (l/aL) E /(~ωL) . For a L,R that their relative position corresponds to a head-to-tail stable bilayer system we require J /ω 1 which is L,R L configuration. As illustrated in Fig.1, thisp can lead to achievable for l/a 1. In particular, for l/a = 0.1 we L L the formation of a bound state. get J /ω 0.01. L,R L To describe the bound state quantitatively, we start We obtain≤ a hierarchy of conditions which have to be with the Hamiltonian for the motion of two molecules met by combining the requirements of small leakage, tun- j = 1, 2 with position coordinates r = (x , y , z ) in neling and decay rates (9): j j j j the 3LS configuration described in the previous section, l ω which reads 1 L . (10)  a  Ω L r c p2 H = j + V (z ) + H0 (z ) (11) Engineering barriers to cut a single well into two sites 2,Λ 2m L j Λ j j=1,2 ! is possible if the above states hierarchy of scales is ful- X filled. We emphasize that our scheme has sufficiently (1) (2) (2),† (1),† + U (t) U (t) Vdd(r1, r2) U (t) U (t), many “tuning knobs” to adjust each parameter indepen- Λ ⊗ Λ Λ ⊗ Λ dently. In the next section, we discuss many-body effects where the dipolar interaction is given by arising from dipolar interactions across the barrier. 1 dˆ r dˆ r V (r) = dˆ dˆ 3 1 · 2 · (12) dd r 3 1 · 2 − r 2 C. Dipolar interaction and inter-layer bound state | | " | | #

and U (j)(t) = exp iω t 0, 0 0, 0 iω t 1, 0 1, 0 is We now turn to many-body physics of molecules in Λ − 1 | ih | − 2 | ih | the Λ configuration discussed above. As an illustra- the rotating-frame transformationh acting on the jth par-i tive example, we consider the motion of two identical ticle. The frequencies ω1,2e aree chosen suche thate ∆ = 6

ω1 (Ee E0,0)/~ = ω2 (Ee E1,0)/~ where Ee de- momentum adds an additional repulsive centrifugal bar- notes− the− energy of the electronically− − excited state. rier, which increases the energy. Then, antisymmetry of e e When we project the Hamiltonian in Eq. (11) to the the wave function Ψ(r1, r2) under the exchange of parti- 0 zero-energy channel, the contribution HΛ(zj) is replaced cles requires that the motional state of the two molecules by the effective BO potential in Eq. (8). The dipolar in- along the z axis is an antisymmetric superposition of the teraction for two molecules in the zero-energy BO channel single-particle states ψL(z) and ψR(z), that is, the coef- reads ficients cαβ in Eq. (16) are given by cLL = cRR = 0 and cLR = cRL = 1/√2, so that V 0 (r , r ) = 0 0 U (1)(t) U (2)(t) − dd 1 2 z1 h | ⊗ z2 h | Λ ⊗ Λ (2),† h (1),† 1 V (r , r ) U (t) U (t) 0 0 Ψ(r1, r2) = χ0(ρ)ΨCOM(R) × dd 1 2 Λ ⊗ Λ | iz1 ⊗ | iz2 (13) √2 i 2 d (z )d (z ) (z z ) [ψL(z1)ψR(z2) ψR(z1)ψL(z2)] . (17) z 1 z 2 1 3 1 − 2 , × − ≈ r3 − r2 1,2 " 1,2 # With this ansatz, the Hamiltonian (11) yields the follow- ing Schr¨odingerequation (SE) for the radial component where r2 = (z z )2 + ρ2 denotes the relative distance 1,2 1 2 of the two-body wave function: of the two molecules− in 3D, and 2 2 1 ~ ∂ 1 ∂ d (z) = 0 d 0 = (z/l)2d0,0 + d1,0 . (14) + + V2D(ρ) χ0(ρ) = EBχ0(ρ). z z z z 2 z z −2m ∂ρ2 ρ ∂ρ h | | i 1 + (z/l)     h e e i (18) The above expression for the dipolar interaction is valid if The effective 2D dipolar interaction is given by fast oscillating terms are neglected and the penetration of the single-particle wave functions into the barrier is 1 0 V2D(ρ) = dz1dz2 Vdd(r1, r2) negligible which is the case for l/aL 1. In summary, 2 the projection of the two-molecule Hamiltonian (11) to Z [ψ (z )ψ (z ) ψ (z )ψ (z )]2 . (19) the zero-energy channel reads × L 1 R 2 − R 1 L 2 It can be decomposed into “direct” and “exchange” con- p2 0 j 0 D E H = + V (z ) + V (z ) +V (r , r ). (15) tributions, V2D(ρ) = V2D(ρ) V2D(ρ), which read 2,Λ 2m L j na j dd 1 2 − j=1,2 " # X D,E We assume in the following that the single-particle en- V2D (ρ) = dz1dz2 ψL,L(z1)ψR,R(z2) ergy scales associated with the nanoscale potential are Z V (ρ, z , z ) ψ (z )ψ (z ), (20) dominant as compared to the dipolar interaction, such × dd 1 2 R,L 2 L,R 1 that motion of the two molecules in the regime of low In the limit of small spatial wave function overlap which energies is restricted to the states ψL(z) and ψR(z). is realized for l/aL 1, the exchange part of the inter- Under these conditions, the nanoscale potential effec- action gives a strongly suppressed repulsive contribution tively implements a two-layer geometry, with layers L which we neglect in the following. and R which are parallel to the xy plane. As illustrated A solution to Eq. (18) with EB < 0 corresponds to a in Fig.4(a), the separation of the peaks of the wave bound state. In Fig.5(a) and (b), respectively, we show functions defines an effective layer separation given by the bound state energy E and the corresponding wave w 2a . The two-body wave function Ψ(r , r ) in the B ≈ L 1 2 0,0 zero-energy channel can then be written as function for different values of the dipolar length ad = 0,0 2 e m (dz /~) . We note that the dipolar length scales as 0,0 2 Ψ(r1, r2) = Ψk(ρ1, ρ2) cαβ ψα(z1) ψβ(z2). (16) a /ae m with a prefactor which depends on l/a d L ∝ D L α,βX=L,R ande 0. The dependence on 0 gives an experimental handle to tune the value of the dipolar length. The component Ψk(ρ1, ρ2), which describes motion in the xy plane, can be decomposed further: First, the system is translationally invariant, and therefore D. Experimental parameters Ψk(ρ1, ρ2) = Ψrel(ρ)ΨCOM(R) factors into contribu- tions corresponding to the relative and COM motion with respective coordinates ρ = ρ ρ and R = In this section we present experimental parameters 1 − 2 for the creation of a bilayer system separated on the (ρ1 + ρ2)/2. Second, due to the symmetry of the Hamil- tonian under rotations around the z axis, the wave func- nanoscale. As a generic example, we consider KRb [8] tion corresponding to the relative motion can be decom- with mass m = 127 u and permanent molecule-frame posed into radial and angular components, Ψ (ρ) = dipole moment = 0.57 Debye. We choose the off- rel D 0,0 imz φ set field 0 = 3 ~B/ , which implies dz = 0.58 = m ∈Z χmz (ρ)e . To find a bound state it is suffi- z 0,0 D D cient to consider mz = 0, since a nonvanishing angular 0.33 Debye and a = 207 nm. For l/aL e= 0.1, z0/aL = P d e 7

(a) (b) Eq. (9). To fulfill the hierarchy of scales we choose Ω = 100 ω which leads to Γ /ω < 4 10−3. For the p L Λ L × sake of self-consistency we require ΓΛ < EB = 0.20 ωL. In absolute numbers, the Rabi frequencies| are| given by Ω = 2π 8.8 MHz and Ω = 2π 330 MHz, where we p × c × used λc = 660 nm [53]. To ensure single-level address- ability of rigid rotor eigenlevels it is necessary to have Ω < B = 2π 1.1 GHz for KRb. c × We note that with the chosen opposite dipole moments (d0,0 d1,0/4) for the dark state constitutes, the inter- z ≈ − z particlee interactione effectively increases the height of the Figure 5. (a) Energy and effective energy of the inter-layer barrier because of the head-to-head or tail-to-tail orien- 0,0 bound state as a function of the dipolar length ad for tation of the dipole moments when one of the molecule 0 = 3~B/D and l/aL = 0.1. The dotted lines indicatee that is under the barrier and the other not. The resulting the results for |EB | > ~ωL should be interpreted as order-of- repulsive dipole-dipole interaction V adds to the barrier magnitude estimates. For these values of EB , excited single- potential and, hence, suppresses the wave function am- particle levels, which we neglect in our analysis, begin to con- plitude under the barrier and, therefore, the tunneling tribute to the bound state. (b) Comparision of the interaction eff JL,R and ΓΛ. The interaction V can be estimated as potentials V2D(ρ) and V (ρ). As shown in the inset, both 2D√ √ 0,0 1,0 3 potentials change sign at 2 w ≈ 2 2a . The orange solid V 2d d /a 2π 300 kHz ~ for the case of KRb, L ≈ z z L ≈ × line is the radial probability density for the bound state χ (ρ), which is 0.1 V (0). This reduces approximately J 0 e ≈ e na L,R and the dashed baseline indicates the corresponding binding by 10% and ΓΛ by 5%. 0,0 energy. Parameters are the same as in (a) at ad /aL = 6.9. We finally compare achievable binding energies for bi- L denotes the numerical box size. e layer systems separated by nanoscale potential barriers to those that can be realized with standard optical lat- λ/2 a0,0 |E | tices. To do so, we define an effective layer separation m D ~B 0 d |EB | B u Debye h GHz kV/cm aeL h kHz h kHz as follows: The interaction potential for dipolar particles KRb[8] 127 0.57 1.11 11.6 6.9 17.6 < 0.1 which are held fixed at a relative separation w along the Toy 100 1.00 1.00 6.0 16.8 78∗ 0.2 z axis is given by [38] ∗ NaK[13, 63] 63 2.72 2.83 6.2 78 > 103 19 ∗ 2 2 2 LiRb 91 3.99 7.61 11.4 242 > 103 62 eff 0,0 ρ 2w ∗ V2D (ρ) = dz − , (21) LiCs 139 5.39 6.54 7.23 676 > 104 139∗ (ρ2 + w2)5/2 e Table I. Parameters for a variety of experimentally relevant and it changes sign at ρ = √2w. As illustrated in fermionic polar molecules [64, 65]. For all molecular species, Fig.5(b), also the effective dipolar interaction V2D(ρ) the offset field is chosen as 0 = 3.0 ~B/D, which implies √ 0,0 in Eq. (19) exhibits a sign change at ρ 22aL, and dz = 0.58D, aL = 30 nm and l/aL = 0.1. The last column ≈ λ/2 we are thus led to identify the effective layer separation presentse the binding energy E achievable for a conventional B as w = 2aL. Deviations of V2D(ρ) from the form given bilayer-stem with inter layer separation w = λ/2 = 250 nm in Eq. (21) are most prominent at small separations ρ, and an electric offset field of  = 3.0 B/D. The asterisk ∗ 0 ~ where V (ρ) exhibits a logarithmic divergence due to the marks binding energies which exceed the single-particle level 2D nonvanishing overlap of ψL(z) and ψR(z). Nevertheless, spacing in the potential well, |EB | > ~ωL, and thus go beyond the range of validity of our analysis which assumes that the the direct comparison in Fig.5(a) shows good quantita- eff molecules populate only the two lowest-energy single-particle tive agreement of the binding energies EB and EB which eff states. Therefore, these values of EB should be considered as we obtain for V2D(ρ) and V2D (ρ), respectively. order-of-magnitude estimates. In conventional bilayer systems which are generated with optical lattices, the layers are separated by λ/2, where λ is the optical wavelength [38]. The inter-layer interaction is then described by the effective potential 0.3 and an harmonic oscillator length of aL = 30 nm the V eff (ρ) in Eq. (21), where w is given by λ/2. This should binding energy is EB = 2π 18 kHz ~, well beyond typ- 2D | | × be compared to the value w = 2a which we obtained ical temperatures of 50 nK kB = 2π 1 kHz ~ [8]. In Ta- L bleI binding energies for different polar× molecules are above for the nanoscale potential barrier. For a0,0/w d  summarized. 1, the corresponding binding energies can be estimatede 0,0 From a single-particle point of view, we have to ful- as Ew 2a ~2/(mw3)[38]. Therefore, reducing the B ≈ d fill the hierarchy of scales 1 l/a ω /Ω from effective layere spacing w from w = λ/2 250 nm to  L  L c & Eq. (10). The first inequality ensures that the nona- achievable harmonic oscillator lengths of w = 2aL = diabatic barrier splits a single potentialp well into two 60 120 nm increases the binding energy by 1-2 orders sides; The second inequality guarantees stability against of− magnitude. Results for a variety of experimentally nonadiabatic channel couplings, i.e., Γ /ω 1, see relevant molecular species are summarized in TableI. Λ L  8

III. INTERFACE BOUND STATE OF with linearly position-dependent detunings, ∆N,0(x) = MOLECULES IN ELECTRIC GRADIENT FIELDS ∆0 x, where ∆0 = dN,0 0/~. Due to its depen- N,0 N,0 − · e e 0 dence on the position x, the Hamiltonian H2LS(x) couples We now consider the setup illustrated in Fig.1, in e e which the electric field gradient created by a standing internal and external degrees of freedom. Typically, the optical wave is replaced by a static electric gradient field. energy scales associated with the internal degrees of free- Further, in contrast to the Λ-systems with optical transi- dom are much larger than those which characterize the tions which we considered in the previous section, we fo- motion of molecules. This allows us to treat the internal cus now on effective two-level systems, which are formed dynamics in a BO approximation. To wit, we first diag- onalize the 2LS Hamiltonian (23); Thereby, we treat the by two rotational levels that are coupled by MW fields. 2LS position x as a parameter. The eigenvalues E± (x) of As we show, in the adiabatic limit, when the single- 0 particle dynamics occurs in a single BO channel, bound H2LS(x) are given by states of two molecules can form at the interface at which ~Ω 1 δ x 2 the induced dipole moment changes sign. E2LS(x) = − 1 + (x/s) , (24) ± 2 1 + δ s ±  q  as illustrated in Fig.6(c), where δ = ∆0 /∆0 = A. Single-particle physics − 0,0 2,0 d0,0/d2,0 is the ratio of induced dipole moments in the z z e e The key elements to engineer interface bound states statese e0, 0 and 2, 0 , and 0 | i | i are a position-dependent electric field, (x) = 0 +  x,  Ω which comprises both a homogeneous offset field 0 as e e s = (25) discussed at the beginning of Sec.II and a gradient field ∆0 (1 + δ) 0 0 2,0  x =  ezx, and a MW-induced Rabi coupling Ω of the molecular levels 0, 0 and 2, 0 , see Fig.6(a) and (b). corresponds to the lengthe scale which delimits the spa- | i | i Note, that these states can be coupled by MW field be- tial region in which the Rabi coupling is resonant. The cause of the admixturee of thee state 1, 0 in both of them corresponding eigenvectors read for  = 0. The reason for the choice| i of these internal 0 6 states will be clarified below. As illustrated in Fig.3(b), + x = sin(θx/2) 2, 0 + cos(θx/2) 0, 0 , | i | i | i (26) the strong offset field 0 induces state-dependent dipole x = cos(θx/2) 2, 0 sin(θx/2) 0, 0 , moments dN,m, which couple to the gradient field 0x |−i |e i − |e i and thus givee rise to position-dependent Stark shifts. To where θ = atan(s/x) and 0 θ π. The states x e ≤ x ≤ e calculate these Stark shifts, we note that in the present are superpositions of the rotational states 0, 0 and setup the motion of the molecules along the x-direction |±ix | i 2, 0 with spatially varying amplitudes. In particular, is restricted to a region of spatial extent l0 which we de- | i + 0, 0 and + 2, 0 for x +e and termine below to be on the order of tens of nanometers, | ix → | i | ix → | i → ∞ 0 xe , respectively, with the transition occurring in and we assume 0  l0, so that the gradient field can be → −∞ taken into account perturbatively. Then, to first order in a spatial regione of extent s. Correspondingly,e the lim- iting behavior of is given by 2, 0 and the gradient field, the Stark shift of the rotational energy |−ix |−ix → | i levels is E [ + 0x] E ( ) [dN,m( ) 0] x. x 0, 0 for x + and x , respectively. N,m 0 ≈ N,m 0 − 0 · |−iAs→ already | i discussed→ in∞ Sec.IIA →the −∞ BO approximatione To simplify the notation, we omit the dependency on  e 0 [55–59] assumes that the internal state of the molecules in the following.e Higher orderse in the expansion in 0x e adiabatically follows the external motion, where the para- are negligible for the large electric offset fields consid- metric dependence of the internal state on the position is ered in the following sections. As a result, the molecules given by Eq. (26). In particular, the Hamiltonian for the move in a state-dependent linear potential, and the Rabi motion of a molecule prepared in the + channel is given coupling between the internal states is resonant only at a by particular point, which we chose as the coordinate origin. The Hamiltonian which describes the motion of a single 2 ad px 2LS molecule in this configuration of electric and MW fields H = + E (x) + Vna(x), (27) 2LS,++ 2m + is given by 2 where the nonadiabatic potential barrier is given by px 0 H2LS = + H2LS(x), (22) 2m ~2 1 1 Vna(x) = . (28) where px = i~∂x is the x-component of the momentum 2ms2 4 [1 + (x/s)2]2 operator. In− a proper rotating frame, the MW coupling Hamiltonian reads A detailed derivation of the adiabatic Hamiltonian from Eq. (27) is provided in AppendixA1. We focus on a 0 ∆2,0(x)Ω/2 parameter regime that is specified below, in which both H2LS(x) = ~ − , (23) Ω/2 ∆0,0(x)! the channel coupling and the nonadiabatic contribution e − e 9

(a) (b) is sufficient to obtain a strongly suppressed decay rate of 2,0(x) −6 Γ2LS/ω0 < 10 . In the next section, we focus on the e + channel and discuss bound states of pairs of molecules 2, 0 due to spatially inhomogeneous dipole moments. | i (c) ⌦ e 0, 0 B. Dipolar interaction and interface bound state | i (x) e0,0 Since the state + x defined in Eq. (26) changes from e | i + x 0, 0 for x + to + x 2, 0 for x , Figure 6. (a) The two rotational states |e0, 0i and |e2, 0i are | i → | i → ∞ | iD → | i → −∞ coupled by a microwave with Rabi frequency Ω. (b) The de- also the induced dipole moment dz (x) defined in Eq. (35) below varies in space and takes limiting values with op- tunings ∆N,0(x), where N = 0, 2 are position dependent due e e 0 D 0,0 D 2,0 to a spatially non-uniform static electric field (x) = 0 +  x. posite signs given by dz (x) dz and dz (x) dz e → → (c) Therefore, the micro wave coupling is resonant with the for x + and x , respectively.e Therefore,e position-dependent Stark shifted energies (dashed lines) of two molecules→ ∞ in the +→ channel −∞ experience attractive in- two rotational states with opposite induced dipole moments duced dipolar interactions if they are located on oppo- (blue and orange arrows) at a particular value of the x coor- site sides of the point x0 where the dipole moment dz(x) dinate which we choose as the origin of the coordinate sys- changes sign. If they are on the same side, they repel tem. This gives rise to the two dressed channels with energies each other [62]. As we show in the following, this gives E2LS(x). ± rise to the formation of bound states of two molecules close to x0. 2 We consider now the full 3D geometry with a har- to the effective BO potential from A (x) are negligible. 2 2 The validity of this approximation is discussed in detail monic confinement V⊥(z) = mω⊥z /2 in the z direction, in AppendixA. whereas motion along the y axis is unrestricted. The The effective BO potential corresponding to the eigen- Hamiltonian for the motion of two molecules j = 1, 2 with position coordinates rj = (xj, yj, zj) and momenta value E+(x) forms a trap. This is because the induced 2,0 0,0 pj is then given by dipole moments dz and dz have opposite sign as can be seen in Fig.3(b), and thus the position-dependent 2 e e pj 1 Stark shifts ∆ (x) for x s and ∆ (x) for x s H = + H0 (x ) + mω2 z2 (31) 0,0 & 2,0 . 2,2LS 2m 2LS j 2 ⊥ j − j=1,2 " # have opposite signs. For x s we approximate the BO X e 2LS | |  e (1) (2) (2),† (1),† trapping potential E+ (x) by a harmonic potential, + U (t) U (t) V (r r ) U (t) U (t), 2LS ⊗ 2LS dd 1 − 2 2LS ⊗ 2LS 2LS 2 2 2 2 E+ (x) ~Ω/2 1 x0/s + (x x0) /s where the dipolar interaction is given in Eq. (12) ≈ − − (29) (j) = ∆E + mω2(x x )2/2. and the rotating-frame transformation U (t) =  0 − 0  2LS exp i(E E )t/~ 0, 0 0, 0 for the jth particle. The harmonic approximation is characterized by the ef- − 2,0 − 0,0 | ih | 2 2 The wave function for two molecules in the + BO chan- fective trap frequency ~ω0 = ~Ω ~ /(2ms ), the posi- h i nel can bee writtene as e e tion of the minimum x0/s = (1 δ)/2√δ, and the energy 2 p− shift ∆E = ~Ω/2[1 (x0/s) ]. In this harmonic potential, − Ψ = dr1dr2 Ψ(r1, r2) r1, r2 + x1 + x2 . (32) the size of the ground state is given by l0 = ~/(mω0), | i | i ⊗ | i ⊗ | i Z and for self-consistency we require l0/s 1. The effec-  p In the following, we neglect nonadiabatic corrections to tive potential E−(x) in the BO channel corresponding the BO approximation. Upon projecting the Hamil- to the eigenstate x forms an inverted trap, and con- |−i tonian in Eq. (31) to the + channel, the contribution sequently this channel hosts a continuum of scattering 0 states. The coupling between the BO channels leads to H2LS(xj) is replaced by the effective potential E+(xj). decay of the bound states in the + channel to the contin- Further, the dipolar interaction for two molecules in the uum in the channel. The decay rate can be estimated + channel is given by by Fermi’s golden− rule as + (1) (2) Vdd(r1, r2) = x1 + x2 + U2LS(t) U2LS(t) 2 h | ⊗ h | ⊗ Γ l s 2LS 0 † h † 2√π exp 8 . (30) V (r r ) U (2), (t) U (1), (t) + + . ω ≈ s − l dd 1 2 2LS 2LS x1 x2 0 "  0  # × − ⊗ | i ⊗ | i i (33) A detailed derivation is provided in AppendixA. The (j) adiabatic limit requires Γ2LS/ω0 1 which is achieved Due to the rotating-frame transformation U (t), cer-  2LS for l0/s 1, in agreement with the harmonic approx- tain contributions to the dipolar interaction acquire  imation of E+(x). Because of the exponential factor in rapidly oscillating phase factors. These contributions av- the rate Eq. (30) a moderately small ratio of l0/s = 1/√2 erage to zero, and we only keep the time-independent 10

0 0 components. Then, the dipolar interaction can be writ- Ψk(ρ1, ρ2) φ⊥(z1) φ⊥(z2) where ρj = (xj, yj). The har- ten as monic oscillator ground state wave function reads 2 D D E E 0 1 z + dz (x1)dz (x2) + 2dz (x1)dz (x2) φ (z) = exp , (39) V (r , r ) = ⊥ 2 dd 1 2 3 l⊥√π −2l⊥ r1 r2   | − | 2 2 (z1 z2) where l = /(mω p) is the corresponding oscillator 1 3 − , (34) ⊥ ~ ⊥ × − r r 2 length. Under these conditions, which also imply l⊥  | 1 − 2|   l0, the motion of the molecules is confined to the xy plane, where the “direct” dipole moments are and the system becomes effectively 2D. Up to the zero- point energy in the harmonic confinement in the z di- D ˆ dz (x) = x + 2, 0 2, 0 dz 2, 0 2, 0 rection, the Hamiltonian for the 2D motion of the two h | | ih | | ih | molecules is given by  + 0, 0 0, 0 dˆz 0, 0 0, 0 + x | ihe | e | ihe | e| i (35) p2 j 2LS  H2D = + E (xj) + V2D(ρ , ρ ). (40) 2,0 1 δ 1 + δ x 2m + 1 2 = dz e e− e e , j=1,2 " # 2 − 2 √s2 + x2 X e   We obtain the effective 2D dipolar interaction by inte- and the “exchange” moments, which correspond to inter- grating out the tightly confined z direction, which yields action processes that exchange the internal states of the molecules, read V (ρ , ρ ) = dz dz φ0 (z ) 2 φ0 (z ) 2 V + (r , r ) 2D 1 2 1 2 | ⊥ 1 | | ⊥ 2 | dd 1 2 Z dE(x) = + 2, 0 2, 0 dˆ 0, 0 0, 0 + D D z xh | ih | z| ih | ix = dz (x1)dz (x2)v2D(ρ). 1 s (36) (41) = 2, 0 dˆz 0, 0 . h |e | e i2 √ e2 e2 s + x That is, the effective 2D dipolar interaction can be writ- In the following,e wee neglect contributions to ten as the product of position-dependent dipole moments D + E d (xj) and an interaction potential v2D(ρ) which de- V (r1, r2) which involve d (x). This is justified z dd z pends only on the relative distance in the xy plane, since 2, 0 dˆ 0, 0 d2,0 , d0,0 . We note that a z z z ρ2 = (x x )2 + r2, where r = y y is the relative corresponding|h | | relationi|  does | not| | apply,| for example, for 1 2 y y 1 2 e e coordinate− in the y direction. The interaction− potential the pair of states 0, 0 and 1, 0 . We further remark e e is given by that the diagonal matrix| i elements| i of the dipole moment operator which entere Eq. (35e) are not affected by the 1 ρ2 ρ2 ρ2 v (ρ) = exp 2 + K rotating-frame transformation which lead to the 2LS 2D 3 2 2 0 2 √8πl⊥ 4l⊥ l⊥ 4l⊥ Hamiltonian (23). The position x0 of the zero-crossing       D 2 2 of dz (x) coincides with the minimum of E+(x) from ρ ρ 2 K1 2 . (42) Eq. (24) given above. For l0 s we expand −l 4l  ⊥  ⊥  3/2 Here, Kn(z) is the modified Bessel function of the second D 2,0 4δ x dz (x0 + x) dz 2 (37) kind. At large distances ρ l⊥, the interaction potential ≈ − (1 + δ) s  3 e assumes the characteristic dipolar form v (ρ) 1/ρ ; 2D ∼ At short distances ρ l⊥ it diverges logarithmically, to linear order in x/s around x0. Here and in the fol- 3  lowing, the x-coordinates x of the molecules are mea- v2D(ρ) 2/π/l⊥ ln(ρ/l⊥). 1,2 The setup∼ we consider is translationally invariant along sured from the zero-crossing x0 of the induced dipole p moment (35). Then, the projection of the two-molecule the y axis and, therefore, the motion of two molecules in Hamiltonian (31) to the + channel reads this direction factorizes into center-of-mass (COM) and relative components, Ψk(ρ1, ρ2) = ψ(x1, x2, ry) ξ(Ry), where R = (y + y )/2 is the COM coordinate. Pos- p2 1 y 1 2 H+ = j + E2LS(x ) + mω2 z2 +V + (r , r ). sible bound states are negative-energy solutions of the 2,2LS 2m + j 2 ⊥ j dd 1 2 j=1,2 " # two-body SE X (38) Since the induced dipole moment (35) is oriented along 2 2 2 2 ~ ∂ 1 2 2 ~ ∂ the z axis, inelastic head-to-tail collisions of the two + mω0xj  −2m ∂x2 2 − m ∂r2 j=1,2 j ! y molecules can be suppressed through tight confinement X in this direction. We assume that the trapping frequency  ω⊥ is sufficiently large such that excited states in the + V2D(x1, x2, ry) ψ(x1, x2, ry) potential V⊥(z) are energetically inaccessible, and the  molecules reside in the ground state. Then, the wave = (E + ω ) ψ(x , x , r ), (43) function Ψ(r1, r2) in Eq. (32) factorizes as Ψ(r1, r2) = B ~ 0 1 2 y 11 where we replaced the effective potential E+(x) in the + be suppressed by increasing the strength of the harmonic BO channel by its harmonic approximation (29), and we confinement. omitted the energy offset ∆E. The binding energy EB is The above simplified model does not take symme- measured from the ground state energy of the noninter- try requirements on the wavefunction for two identical acting two particle problem, which is 2 ~ω0/2. fermions into account. To discuss this point, we re- × We solve this Eq. (43) numerically. Details are dis- turn to the full SE (43). We note that a Hamilto- cussed in AppendixB, and we present our results in nian which describes the motion of identical particles Fig.7. As shown in Fig.7(a), a bound state occurs for has to be symmetric under the exchange ρ1 ρ2 of sufficiently strong induced dipolar interactions, where the the coordinates of the particles. The Hamiltonian↔ in strength of dipolar interactions is characterized by the Eq. (43) obeys an even stronger symmetry: It is symmet- 2,0 ric under the exchange of both only the x coordinates, ratio a /l0 with d x x , and only the y coordinates, y y . There- e 1 ↔ 2 1 ↔ 2 2 fore, also its eigenfunctions have definite parity under d2,0 4δ3/2 2,0 z these operations, i.e., ψ(x1, x2, ry) = ψ(x2, x1, ry) and ad = m 2 . (44) ± e (1 + δ) ψ(x , x , r ) = ψ(x , x , r ). Overall, the two-body e " ~ # 1 2 y 1 2 y wave function for± identical− fermions has to be antisym- We note that while achievable binding energies are metric, ψ(x1, x2, ry) = ψ(x2, x1, ry). Numerically, − − boosted if the characteristic length scale l0 of this setup we find that the bound state wave function is antisym- is on the order of tens of nanometers, the scheme suffers metric with respect to x1 x2, and symmetric under ↔ from a reduction of the dipolar moment in the vicinity y1 y2 (or, equivalently, ry ry). Finally, we note ↔ → − of the interface. Figure7(b) shows the wave function of that fermionic statistics imply that the probability of a 2,0 close encounter of the molecules is strongly suppressed, the bound state for l0 = 1.5l⊥ and ad = 69l0. Numeri- i.e., ψ(x1, x2, ry) 0 for x1 x2 and ry 0. In com- cally, we find that the threshold valuee for the formation parison to bosonic→ molecules,→ this enhances→ the stability of a bound state depends only slightly on l⊥. This is of fermionic molecules against chemical reactions [61]. because V2D(x1, x2, ry) is only modified in regions where (x x )2 +r2 l2 , and in these regions, the probability Apart from reduced stability, bound states also occur for 1 − 2 y . ⊥ pairs of bosonic molecules in the 2LS configuration. How- amplitude ψ(x , x , r ) 2 is suppressed. The existence of 1 2 y ever, for bosons we expect an increased threshold value | | 2,0 a threshold value of the ratio ad /l0 implies that strong of the dipolar length. This expectation is based on the harmonic confinement suppressese the formation of the simplified model described above and on symmetry ar- bound state. guments: The simplified model suggests that in order to These results can be understood qualitatively within form a bound state, the relative motion of two molecules a simplified model which we obtain from Eq. (43) by in the x direction has to populate excited harmonic os- setting both the COM coordinate in the x direction, cillator states. For fermions, antisymmetry of the total Rx = (x1 + x2)/2, and the relative coordinate in the wavefunction implies that excited states with odd har- y direction to zero, Rx = ry = 0. This yields an effec- monic oscillator quantum numbers are admissable, and tive SE for the component of the wave function which the lowest lying state that is compatible with this re- describes the relative motion of the two molecules in the quirement is the first excited state. However, this state x direction. The corresponding effective potential, which is excluded for bosons by symmetry. The necessity to depends only on the relative coordinate rx = x1 x2, con- invest at least two harmonic oscillator excitation quanta − tains contributions from the harmonic confinement and results in a higher threshold to form a bound state. the dipolar interaction and is given by

1 V (r ) = mω2r2 + V (r /2, r /2, 0). (45) C. Experimental parameters eff x 4 0 x 2D x − x

2,0 In this section, we discuss the optimal choice of ex- Figure7 shows the effective potential for ad /l0 = 0 and perimental parameters for the realization of an interface 2,0 e ad /l0 = 69. In the former case, Veff (rx) reduces to a har- bound state. From a single-particle point of view we re- monice potential. Then, the state with lowest energy is quire the decay rate of molecules in the + channel given just the corresponding harmonic oscillator ground state. in Eq. (30) to be small, Γ2LS/ω0 1, which is guaran- 2,0 teed for l /s 1. However, even a moderately small For a /l0 = 69, the effective potential exhibits two min- 0 d  √ −6 ima ate rx = rx,0 = 0. In this situation, it is energeti- ratio of l0/s = 1/ 2 leads to Γ2LS/ω0 < 10 , such cally advantageous± 6 for the molecules to “pay the price” that the lifetime of molecules in the + channel is well of climbing up to the first excited state in the harmonic above any experimentally relevant time scale. The rela- potential and thus effectively increase their relative dis- tion l0/s = 1/√2 can be rearranged as tance, since this allows them to reduce their total energy 2/3 due to the contribution from the dipolar interaction in 0 2,0 ~ ~Ω = 2  d (1 + δ) , (46) Eq. (45). Evidently, the formation of a bound state can z √m  e  12

(a) (b) d2,0(1 + δ) = d2,0 d0,0 has to be maximized. This | z | | z − z | boilse down to findinge thee optimal value of the electric offset field, which is 0 = 7.5 ~B/ . In turn, this implies D d0,0 = 0.74 , d2,0 = 0.26 and δ = 2.81. As an ex- z D z − D ample,e we considere the LiRb molecule with parameters m = 91 u, = 3.99 Debye. Further, we take the electric field gradientD to be 0 = 3.5 (kV/cm)/mm. Then, with the optimal value 0 = 7.5 ~B/ of the offset field, the D harmonic oscillator length evaluates to l0 = 36 nm, which is considerably below optical length scales. For the same 2,0 (c) parameters, the dipolar length is given by ad /l0 = 69, which is above the threshold value for thee appearance of a bound state. As can be seen in Fig.7(a), the cor- responding binding energy is EB = 2π 66 kHz ~ for the given parameters. In comparison,| | recent× experiments with KRb [8] reached temperatures which correspond to a much lower energy of 50 nK kB = 2π 1 kHz ~. We list additional parameters for different species× of polar Figure 7. (a) Energy of the bound state in units of E0 = molecules in TableII. 2/2ml2 as a function of the dipolar length a2,0 defined in ~ 0 √ d We finally point out that the dipolar length scales as Eq. (44) for several values of l0/l⊥ and s = 2le0. (b) Proba- 2,0 2 0 1/3 4/3 R 2 ad /l0 (  ) m with a prefactor that depends bility density dRx|ψ(rx,Rx, ry)| of the bound state, where ∝ D D one the value of the offset field 0. In experiments, this rx = x1 − x2 and Rx = (x1 + x2)/2 are the relative and COM dependence on 0 can be used to tune the dipolar length coordinates in the x direction, for l0 = 1.5l⊥ and ad/l0 = 69. L denotes the box size used in the numerical diagonalization by adjusting the value of 0, and thus cross the threshold of Eq. (43). (c) Effective potential (45) for the simplified for the formation of a bound state. model discussed in the main text.

a2,0 m D ~B 0 l0 d |EB | Ω D. Adiabatic loading u Debye hGHz (kV/cm) nm le0 hkHz kHz KRb[8] 127 0.57 1.11 28.9 61 1.2 0 86 To conclude this section, we present an experimen- Toy 100 1.00 1 14.9 54 3.1 0 135 tal protocol to adiabatically prepare molecules in the + [13, 63] NaK 63 2.72 2.83 15.5 46 17 0 307 channel. The electric offset field is kept at a constant LiRb 91 3.99 7.61 28.4 36 69 66 351 value 0 throughout the protocol, while the gradient field LiCs 139 5.39 6.54 18.1 28 241 564∗ 374 0 as well as the Rabi coupling Ω in the 2LS Hamilto- nian (23) are set to zero initially. Further, we assume Table II. A list of relevant parameters for different fermionic that the molecules are prepared in the rotational state polar molecules [64, 65]. For all molecular species, the offset 0, 0 , and that they are trapped in an auxiliary optical field is chosen as 0 = 7.5 ~B/D and the gradient field is fixed | i 0 ∗ potential Vaux(x), which is switched off at the end of the to  = 3.5 (kV/cm)/mm and l0 = 1.5 l⊥ holds. The asterisk marks binding energies which exceed the single-particle level protocol.e spacing in the potential well along the confined z-axis, |EB | > The first step is to turn on the Rabi coupling in the 2LS 2 2 ~ /(2ml⊥), and thus go beyond the range of validity of our Hamiltonian (23), where the detunings are chosen such analysis which assumes that the molecules populate only the that ∆ = ∆ ∆ < 0 and ∆ /Ω 1, implying 2,0 2,0 − 0,0 | 2,0|  lowest-energy single-particle state. + 0, 0 . Next, the magnitue of the detunings is adi- e e e e e e |abaticallyi ≈ | i reduced ( ∂ ∆ /Ω2 1) to zero. Under this | t 2,0|  0 condition,e the molecules remain in the instantaneous ex- such that Ω is fixed for a given value of  . This can be cited eigenstate + and ate e the end of the adiabatic sweep used to determine the harmonic oscillator length in the | i this state is given by + = ( 0, 0 + 2, 0 )/√2. The adiabatic regime as final state to prepare molecules| i | ini + | fromi Eq. (26) | ix 1/3 is to adiabatically turn off the auxiliary optical poten- 2 e e ~ 1 tial while ramping up the electric gradient field 0, such l0 = √2 . (47) 2m 0 2,0 1+δ  dz ! that the effective harmonic confinement in the + channel 2 2LS e (E+ (x) + Vaux(x)) stays approximately constant. At As l0 sets the length and energy scale of the prob- the same time, states in the channel evolve from be- lem, the denominator in the above expression should be ing trapped to being antitrapped.− Due to nonadiabatic as large as possible. In current experiments, the elec- channel couplings, there are narrow avoided crossings be- tric field gradient is limited by the apparatus. There- tween excited motional states in the channel and states fore, the difference between induced dipole moments in the + channel. This avoided crossings− are exponen- 13 tially small, see AppendixA3c, and have to be passed each other. The observation of these bound states under nonadiabatically. current experimental conditions is possible for molecules with dipole moments of a few Debye. In a system of fermionic molecules, the formation of interface bound IV. CONCLUSION AND OUTLOOK state can cause a transition from fermionic to bosonic behavior, and it is an interesting question for further To summarize, we presented two approaches for studies which quantum phases can be realized in such engineering quantum many-body systems with polar systems. We finally note that this all electric scheme can molecules by employing the coupling of rotational states be applied with molecules for which the complexity of via external laser or MW fields and electric field gra- the level structure makes the optical manipulation with dients. The first approach builds on a Raman coupled laser light challenging. Moreover, this technique can also Λ-systems, where the strong electric field gradient of a be adapted to atoms or molecules with magnetic dipole standing laser field leads to the generation of a single moments by coupling to an external magnetic field gra- particle potential barrier on the nanoscale. The second dient [66, 67], which requires gradients on the order of scheme generates spatially modulated electric dipoles and thousands of G/mm [68]. thus dipolar interactions by MW mixing in the presence In summary, the promise of enhanced energy scales of electric DC field gradients. and novel interaction terms modulated on spatial scales Nanostructured optical barriers in combination with of tens of nanometers provides an interesting new avenue optical lattice potentials enable the generation of bilayer for many-body physics, including BCS–BEC or fractional systems, where the layers are separated by only few tens quantum Hall phases. of nanometers. The resulting many-body systems are Acknowledgments –The authors thank M. Lacki, D. described by a Hamiltonian of the form Petter and M. Mark for discussions. P.Z. thanks JILA for hospitality as Visiting Fellow in September 2018, where H = H + H + V (ρ ρ ), (48) part of this work was done. Work at Innsbruck is sup- L R LR iL − jR iL,jR ported by the European Union program Horizon 2020 X under Grants Agreement No. 817482 (PASQuanS) and 2 2 ρ ρ where Hα = i( ~ /2m) iα + i

It is convenient to introduce the new independent vari- To derive the above conditions, let us consider the able z = z/l and rewrite this equation as asymptotics of ϕ (z) for z 1, λ  d2 l4 1 λ2 2ml2 + z2 + − ψ (z) = Eψ (z). πλ πλ −dz2 a4 (1 + z2)2 λ 2 λ ϕλ(z) Aλ sin B cos  L  ~ ≈ 2 − 2 (A8)      πλ B πλ For energies smaller than the height of the barrier, E + A cos + sin z, (A14) 2 2  2 λ 2 ~ /(2ml ), and inside the barrier, z . 1, one can neglect      both the harmonic potential and| the| right-hand side of and for z 1 this equation, such that the behavior of low-energy wave  − functions becomes energy independent. To establish this πλ πλ universal behavior, we consider the zero-energy solution ϕ (z) Aλ sin + B cos λ ≈ 2 2 ϕλ(z) of the SE (A8),      πλ B πλ d2 1 λ2 + A cos + sin z. (A15) − ϕ (z) = 0. (A9) − 2 λ 2 dz2 − (1 + z2)2 λ        These asymptotics have to be matched with the wave ± We can find a general solution of this equation by intro- function and its derivatives for z 1, ψ(z) ψλ(0 ) + ducing the new independent variable y = arctan(z), such 0 ± ∼ ± ≈ ψλ(0 )z. This gives that the equation takes the form

± πλ πλ d2 d ψλ(0 ) = Aλ sin B cos , 2 tan(y) 1 λ2 ϕ (z) = 0. (A10) 2 ∓ 2 dy2 − dy − − λ     (A16)   πλ B πλ
ψ0 (0±) = A cos + sin . λ ± 2 λ 2 After introducing the new unknown function gλ(y) =     ϕ (y)/ cos(y), we finally obtain λ After excluding the unknown constants A and B, and d2 restoring the original units, we arrive at the conditions λ2 g (y) = 0. (A11) presented in Eq. (A13). dy2 − λ   The boundary conditions in Eq. (A13) allow us to esti- This equation can easily be solved and the general solu- mate the suppression of the wave function in the barrier tion of Eq. (A9) then reads region. For this purpose we restore the harmonic poten- tial in Eq. (A6) characterized by the frequency ωL and the harmonic length aL, which we assume to be much ϕ (z) = 1 + z2 A cos[λ arctan(z)] larger than the width of the barrier l, aL l. The λ   eigenenergies and eigenfunctions of this problem can be p B obtained by matching the solution + sin[λ arctan(z)] , (A12) λ  2 1 ν 1 2 y+(z) = C+ exp( z /2) Φ ; ; z where A and B are unknown constants. The general solu- − Γ( 1−ν ) − 2 2 tion ϕ (z) can be used to formulate the effective “bound-  2   λ 2z 1 ν 3 ary conditions” which connect the wave function and its e e Φ − ; ; z2 e (A17) −Γ( ν ) 2 2 derivative on different sites of the barrier. These condi- − 2   tions are applicable for wave functions which change on e e a scale that is much larger than the width of the barrier. which decays exponentially for z + where z = z/aL, → ∞ In other words, the corresponding eigenenergies are much with the solution smaller than the height of the barrier. The “boundary e e conditions” are given by 2 1 ν 1 2 y−(z) = C− exp( z /2) 1−ν Φ ; ; z − Γ( 2 ) − 2 2 πλ    ψ0 (0+) + ψ0 (0−) = sλ tan ψ0 (0+) ψ0 (0−) , 2z 1 ν 3 λ λ 2 λ − λ e e+ Φ − ; ; z2 e (A18)   Γ( ν ) 2 2   2   + − πλ 0 + 0 − − ψλ(0 ) ψλ(0 ) = sλ cot ψ (0 ) + ψ (0 ) , e − − 2 λ λ which decays exponentially for z e, in the re-    (A13) gion of the barrier. Here, Γ(z) is→ the −∞ Gamma func- tion, Φ(a; b; z) the degenerate hypergeometric function, ± 0 ± e where ψλ(0 ) and ψλ(0 ) are the values of the wave and ν = 2E/~ωL the energye measured in units of function and its derivative, respectively, on the right and the harmonice oscillator spacing. For low-energy eigen- left side of the barrier. states, E ~2/(2ml2), we can use the boundary condi-  16 tions (A13), which give the following equations: in the zero-energy BO channel into the continuum in the channel. Below, we estimate the corresponding 1 l 2 πλ decay− rate using Fermi’s golden rule. In the 2LS, we + λ tan (C+ + C−) = 0, Γ( 1−ν ) a Γ( ν ) 2 are interested in the + channel, for which the effective  2 L − 2   (A19) potential is given by the BO potential in Eq. (27). Decay occurs again to the open channel. 1 1 πλ l 2 − tan (C+ C−) = 0. Γ( 1−ν ) λ 2 − a Γ( ν ) −  2   L − 2  (A20) a. DOS in the open channel These equations determine the eigenergies ν. The eigenenergies of states which are symmetric with respect According to Fermi’s golden rule, the rate of decay of to z z fulfill a given initial state to an open channel hosting a con- → − tinuum of final states is determined by the product of 1 l 2 πλ the transition matrix element between the initial and fi- e e1−ν + ν λ tan = 0, (A21) Γ( ) aL Γ( ) 2 nal states, and the DOS in the open channel. In this 2 − 2   section, we derive the DOS in the open channel, which where C+ = C− follows from Eq. (A19). The correspond- corresponds to the BO channel in both the Λ-system ing eigenfunctions for z l have the form and the 2LS. We neglect− nonadibatic corrections to the | | & effective potential in the channel. Moreover, for the − 2 2 l πλ Λ-sytem we are only interested in a spatial region of ysym(z) exp( z /2aL) λ tan ∼ − aL 2 extension z z0 . aL 2π/kc, thus we expand    sin(k z) | k−z, and| for simplicity we consider for the ν 1 z2 z 1 ν 3 z2 c ≈ c Φ ; ; + | | Φ − ; ; . (A22) 2LS the symmetric case δ = 1. Then, for both systems, × − 2 2 a2 a 2 2 a2  L  L  L  the effective BO potential in the channel can be writ- ten as U(z) = (Ω/2)2 + (∆−0z)2, where we use the The eigenenergies of antisymmetric solutions, for which ~ following− identification− table: C+ = C− results from Eq. (A20), satisfy the equation p − 0 1 1 πλ l 2 Ω ∆ tan = 0, (A23) 0 Γ( 1−ν ) λ 2 − a Γ( ν ) 2LS Ω ∆ 2   L − 2 2,0 Λ-system Ωp Ωckc/2 and the wave functions for z l are e | | & For the 2LS the z coordinate has to be replaced by x, and for the Λ-system we consider the resonant case with 2 2 l λ yasym(z) exp( z /2aL) sign(z) πλ ∆ = 0. Since we are interested in the resonant decay ∼ − aL tan " 2 from energetically higher channels into the channel, 2 2 − ν 1 z z 1 ν 3 z
only states with energy E > Ω/2 are relevant. Such Φ ; ; 2 + Φ − ; ; 2 . (A24) − × − 2 2 aL aL 2 2 aL states can be described in the WKB approximation. To     write down the corresponding wave functions, we assume From Eqs. (A22) and (A24), we see that as stated above that the system is contained in a large box of size 2L, the wave function in the region of the barrier, z . l, is x L. (The size L will disappear from the final result | | | | ≤ reduced by a factor l/aL 1, as compared to its typical for the decay rate.) Then, the normalized wave function values outside the barrier for z l. can be written as | | &

0 1/4 − 1 1 ∆ 3. Nonadiabatic channel couplings ψ( )(z) = E L1/4 √2 E + U(z)   1 z We now turn to nonadiabatic channel cou- sin / cos dz0 2m[E + U(z)] , (A25) × plings, which correspond to the terms Cna = ~ Z0  2 p pz,A(z) + A (z) /(2m) and px,A(x) /(2m) 0 in−{ Eq. (A2}) for the Λ-system and−{ 2LS, respectively.} where we neglect terms of order Ω/∆ L 1. The choice   of sin ... or cos ... corresponds to either symmetric or In the Λ-system, we are interested in the zero-energy { } { } BO channel. The external potential confines the zero antisymmetric parity of the wave function, respectively. energy BO channel, whereas the effective potential in The corresponding eigenenergies E = En can be obtained the channel does not confine states with energy bigger from the WKB quantization condition than− 0. Therefore, the channel hosts a continuum of − L scattering states, and the nonadiabatic channel coupling 1 0 1 2 dz 2m[En + U(z)] = π n + . (A26) pz,A(z) + A (z) /(2m) induces decay of states 2 −{ } ~ −L   Z p   17

It follows from this equation that the DOS is given by reads

L 2 (−) ~ 1 d dn 2√2 mL M = dz ψ (z) E=0 2 2 , (A27) − −L ml √2 1 + z dz dE ≈ π ~3∆0 Z  r z ϕ (z) − (1 + z2)2 0 where we neglect terms vanishing for L .  → ∞ 2 L/l ~ (−) B = dz ψ (z) . E=0 2 3/2 −ml√2 − (1 + z ) Z L/l (A30) b. Decay rate in the Lambda-system We note that the term which is proportional to A in We now derive the decay rate of the zero-energy BO ϕ0(z) gives a vanishing contribution. A nonvanishing channel in the Λ-system for the resonant case ∆ = 0. As result can be obtained only if one includes the correction 2 2 discussed in Sec.A2, for aL l the nonadiabatic poten- to ϕ0(z) of first order in 2ml E/~ 1 in Eq. (A8). tial barrier has a strong influence on the wave function in  (−) After substituting the expression (A28) for ψE=0(z), we the zero-energy channel. However, to estimate the decay obtain rate, we neglect the effects of the barrier on the open- ∞ (−) ~2 B 1 channel wave function ψE (x). This is legitimate be- M = dz 2 2 4 2 13/8 cause the height of the barrier /(2ml ) is much smaller 2ml √Ll −∞ (1 + z ) ~ Z than the gap Ωp/2 between the channel and the zero- z (A31) − 0 0 energy channel. Therefore, we can use Eq. (A25) with cos κ dz (1 + z 2)1/4 , (−) × ( 0 ) E = 0 for the final wave function ψE (x) in the open Z channel, where, using L/l 1 and convergence of the integral over z, we extend the limits of the integration to infinity. 1 1 ∆0 1/4 ψ(−) (z) To calculate the integral in Eq. (A31), we write it as E=0 ≈ L1/4 √2 EΛ(z)  −  − z ∞ z 1 0 Λ 2 −13/8 0 02 1/4 sin / cos dz 2mE−(z) dz (1 + z ) exp iκ dz (1 + z ) , × ~ 0 − −∞ ( 0 )  Z  (A28) Z Z 1 1 1 q (A32) = (Ll)1/4 √2 (1 + z2)1/8 and consider as a contour integral in the complex z plane. z sin / cos κ dz0(1 + z02)1/4 , To uniquely define the multi-valued integrand, we make × ( 0 ) two branch cuts on the imaginary axis: from i to i Z and from i to i . We then move the contour of inte-∞ − − ∞ Λ gration from the real axis to the upper half-plane where where z = z/l, E−(z) is defined in Eq (4), and κ is the the integrand decays exponentially. The new contour is square root of the ratio of the gap between the channels 2 1/2 around the upper branch cut and consists of three parts: and the height of the barrier, κ = [ml Ωp/~] . The The first one is from i to i over the left-side of the cut nonadiabatic channel coupling p ,A(z) /(2m), where ∞ −{ z } where z = i + y exp( i3π/2) with real y [, ] and A(z) is given by Eq. (A3) and is nonzero only inside  infinitesimal and positive.− The second one∈ is∞ around the barrier for z l, where the behavior of the wave | | . the circle of radius  around i in the counterclockwise di- function in the zero-energy channel is governed by the rection, z = i +  exp(iφ) with φ [ 3π/2, π/2], and universal solution in Eq. (A12) with λ = 0, the third one is over the right-hand∈ − side of the cut, z = i + y exp(iπ/2). We note that every integral over 2 ϕ0(z) = 1 + z [A + B arctan(z)] . (A29) individual parts diverges when  0, and only their sum is finite. → p On the new integration contour, the integral in the The coefficients A and B in the above expression depend exponent in Eq. (A32) can be written as on details of the behavior of the wave function outside the barrier and on the position of the barrier. We can z i estimate these coefficients as follows: The typical value dz0 (1 + z02)1/4 = dz0 (1 + z02)1/4 of the wave function localized in a spatial area of size aL Z0 Z0 is ϕ 1/ a ; Further, according to our discussion in z √ L 0 02 1/4 (A33) Sec.A2∼ , the wave function within the barrier is reduced + dz (1 + z ) Zi by a factor of l/aL. This yields the estimate A B 4 3/2 ∼ ∼ iI + 2−1/4eiπ/8(z i)5/4, l/aL . The nonadiabatic coupling matrix element then ≈ 5 − 18 where for which ΓΛ/ωL is exponentially suppressed. 1 √2π3/2 I = dx(1 x2)1/4 = 0.874, (A34) − 6Γ(3/4) ≈ Z0 c. Decay rate in the 2LS and the second integral is in calculated by keeping only the leading term in the expansion of (1+z02)1/4 in powers We proceed to calculate the rate of decay from the + of z0 i because for κ 1 only the region z0 i κ4/5 BO channel to the channel in the 2LS. For simplicity, −  | − | .  1 is important. After substituting (A33) into (A32) and we focus on the symmetric− case δ = 1, and we consider expanding (1 + z2)13/8 around z = i to leading order, we the ground state in the harmonic approximation (29) to find for the integral (A32) the result the effective potential in the + channel as the initial state, ∞ 1 x2 −κI −5/8 4 −1/4 ds −s π π e 2 κ 2 e [cos(s + ) + sin(s + )] φ0(x) = exp 2 , (A40) 5 √s 8 8 l0√π −2l0 r Z0   4 2π where l = /(mω )p and we set x = 0. The validity of = e−κI 2−5/8 κ 2−1/4 π√2 = κe−κI . 0 ~ 0 0 5 5 the harmonic approximation is controlled by the inequal- r q r p (A35) ity l0 s, where s, which is given in Eq. (25), denotes the width of the resonant region. This inequality will be To obtain this result, we first integrate by parts in the y- of crucial importance for the following discussion. integrals over the sides of the branch cut (this, in combi- The state φ0(x) in the + channel is coupled to the state nation with the integral over the circle, eliminates the di- (−) vergences for  0), then we take the limit  0, and in- described by the wave function ψE (x) in Eq. (A25) in → →5/4 −1/4 the channel by the operator C = p ,A(x) /(2m), troduce the new integration variable s = 4κy 2 /5. − na − { x } With this result, the final expression for the coupling which occurs as a nonadiabatic channel coupling in matrix element reads Eq. (A2). Using the aforementioned inequality, this op- erator can be approximated as 2 ~ B 2π − M = κe κ I . (A36) 2ml √4 Ll 5 ~2 s d sx 0 1 r Cna = − 2m s2 + x2 dx − (s2 + x2)2 1 0 An analogous calculation shows that the channel cou-   ! pling A2 (z) from Eq. (A2) yields a matrix element which (A41) Λ ~2 1 d x 0 1 is smaller than M in Eq. (A36) by a factor of order − . ≈ 2m s dx − s3 1 0 ((l/aL) ωL/Ω) and thus negligible. With the matrix   ! elementO Eq. (A36) and the DOS from Eq. (A27), using As a result, the coupling matrix element M which enters Fermi’s goldenp rule we obtain for the decay rate Fermi’s golden rule is 2 4π ~B −2κ I ΓΛ = √κe . (A37) 2 L ~ 1 d x (−) 5 ml M = dx φ (x) ψ (x) = M M . 2m 0 s dx − s3 E 1 − 2 We note that the exponential factor in Eq. (A37) coin- Z−L   cides with the results obtained in the absence of har- (A42) monic confinement as considered in Ref. [47]. For the Since φ0(x) is a symmetric function of x, the matrix el- (−) harmonic confinement in the dark-state channel with ement M is different from zero only if ψE (x) is an an- 3/2 tisymmetric function. Therefore, one has to choose the frequency ωL one has B = βl/aL with β . 1 and | | sin ... solution in Eq. (A25). Furthermore, one can aL = ~/mωL (for the barrier in the center of the har- { } 5 1/4 easily see that the contribution from M2 is smaller than monic trap, β = β0 (4/π ) = 0.34, such that actu- 2 p ≈ that of M1 by a factor which is ((l0/s) ), and thus ally β β0). The ratio of the decay rate to the oscillator O | | ≤ frequency ωL then reads 2 L ~ 1 d (−) M M1 = dx φ0(x) ψ (x). (A43) Γ 4π l 3/2 Ω 1/4 Ω l ≈ 2m s dx E Λ β2 p exp 2 p I Z−L ω ≈ 5 a ω − ω a L  L   L  r L L ! The calculation of the above integral can be simplified 3/2 1/4 significantly by using the inequality l s, which yields l Ωp Ωp l 0 = 2.5 β2 exp 1.75 , (−) a ω − ω a the following simplified expression of ψE (x) for x l0:  L   L  r L L ! | | ∼ (A38) 0 1/4 1 1 ∆ as written in Eq. (9). We note that the above derivation ψ(−)(x) 2,0 E ≈ L1/4 √ E + Ω/2 requires the following hierarchy of scales 2 " e~ # 1 l ωL sin x 2m[E + ~Ω/2] . (A44) 1 , (A39) × ~  aL  Ω   r p 19

2 2 2 Further, we can expand the oscillatory factor, because where we define ρ = rx + ry. Due to the spatial depen- D the neglected terms are much smaller than unity if dence of the dipole moment dz (x) it is not possible to ω0/Ω 1. Within this approximation, we obtain for separate the relative and COM motion along the x axis.  e e e E ~Ω/2 the matrix element It is convenient to expand the COM motion in eigenfunc- ≈ 2 1/4 tions of the harmonic potential described by the SE 1 ~ 1 π l0 1 2 M 1/4 p exp (˜pl0) , ≈ L 2m √2l0 2s s −2     (A45) 2 2 2 e ~ 1 ∂ 2 ~ 1 where p = √4mΩ/~. This result for the matrix element +2R Φn(Rx) = 2 n + Φn(Rx). 2ml2 −2 2 x 2ml2 2 and the DOS from Eq. (A27) yield the ratio of the decay 0  ∂Rx  0   (B4) rate Γ2LSe to the oscillator frequency ω0 given in Eq. (30), e e e That is, we usee the ansatz Γ l s 2 2LS 2√π 0 exp 8 . (A46) ω0 ≈ s − l0 "   # N

In particular, we find that the decay rate of the + chan- Ψ(Rx, rx, ry) = Φn(Rx)φn(rx, ry) , (B5) nel for the 2LS is determined by the ratio l0/s and is n=0 suppressed exponentially for l s. X 0  e e e e e e where we only keep the lowest N eigenfunctions in the Appendix B: Interface bound state: Details of above harmonic potential. We note that the required numerical analysis cutoff N to reach convergent results for the bound state and bound state energy is usually much smaller than the The wave function and eigenenergies for the bound number of grid points one needs to obtain comparable state shown in Fig.7 of the main text Sec.IIIB are the results from a brute-fore discretization in real space. The solutions of a numerical diagnoalization of Eq. (43). For above ansatz yields the following SE for the amplitudes all calculations we use the linear approximation of the φn(rx, ry): dipole moment, 3/2 e e D 2,0 4δ x N dz (x) dz 2 , (B1) 2 2 2 ≈ − (1 + δ) s ~ ∂ ∂ 1 2 1 e 2 2 + r + 2 n + δ 2ml2 − ∂r2 − ∂r2 2 x 2 m,n as introduced in the main text. It is convenient to express 0 m=0  x y    the SE (43) from the main text also along the x direction X m,n e in relative and COM coordinates, r = x x and R = + V (rx, ry) φm(rx, ry) = (EB ~ω0) φn(rx, ry). x 1 2 x 2D e e − (x + x )/2, respectively. This leads to −  1 2 (B6) e e e e e e ~2 1 ∂2 ∂2 ∂2 1 2 2 + 2R2 + r2 2 2 2 2 x x 2ml0 −2 ∂R − ∂rx − ∂ry 2 " x   Amplitudes φn(rx, ry) with different n are coupled by the rx rx e interaction matrix elements V2D(Rx + 2 , Rx 2 , ry) e + e 2 e − e Ψ(Rx, rx, ry) ~ e e e 2 e # e 2ml0e e e m,n = EΨ(Rx, rxe, rye), (B2) V 2D (rx, ry)

∞ rx rx where we measure distances in units of the confinement V2D(Rx + 2 , Rx 2 , ry) e = dRxΦm(Rx) 2 − Φn(Rx) in the x direction l0, rx = rx/l0, ry =e rey/l0, and e e ~ −∞ e 2 e Z 2ml0 Rx = Rx/l0. To obtain the binding energy one has to e e e 1 ael l3 e ρ2 l2 l2 ρ2 le2 subtract from E twice thee ground statee energy of the d 0 0 0 2 0 0 = 2 3 exp 2 2 + ρ 2 K0 2 noninteractinge system, i.e., EB = E ~ω0. The explicit √2π s l⊥ 4 l⊥ l⊥ 4 l⊥ −       form of the interaction potential is given by l2 ρ2 l2 e 1 e 2 0 0 e ρ 2 K1 2 (n + 1)(n + 2)δm,n+2 rx rx − l⊥ 4 l⊥ 4 V2D(Rx + 2 , Rx 2 , ry)    2 − e p ~ 2 e 2 e + e(2n + 1) r δm,n + n(n 1)δm,n−2 . e 2ml0e − x − 3 e 2 2  1 adl0 l 1 ρ l p = 0 R2 r2 exp 0   (B7) √2π s2 l3 x − 4 x 4 l2 e ⊥    ⊥  l2 ρ2 l2 l2 e ρ2 l2 2 0 e 0 e 2 0 0 2 + ρ 2 K0 2 ρ 2 K1 2 , For the parameters used in Fig.7, we obtain convergence × l⊥ 4 l⊥ − l⊥ 4 l⊥       for N < 9, a numerical box size of rx,y < 10 and a e e (B3) uniform grid of 600 600 for the two relative| | coordinates. e e × e 20

[1] J. W. Park, Z. Z. Yan, H. Loh, S. A. Will, and M. W. [25] S. Baier, M. J. Mark, D. Petter, K. Aikawa, L. Chomaz, Zwierlein, Science 357, 372 (2017). Z. Cai, M. Baranov, P. Zoller, and F. Ferlaino, Science [2] T. M. Rvachov, H. Son, A. T. Sommer, S. Ebadi, J. J. 352, 201 (2016). Park, M. W. Zwierlein, W. Ketterle, and A. O. Jamison, [26] L. Chomaz, R. M. W. van Bijnen, D. Petter, G. Faraoni, Phys. Rev. Lett. 119, 1 (2017). S. Baier, J. H. Becher, M. J. Mark, F. W¨achtler, L. San- [3] L. Reichs¨ollner,A. Schindewolf, T. Takekoshi, R. Grimm, tos, and F. Ferlaino, Nat. Phys. 14, 442 (2018). and H.-C. N¨agerl, Phys. Rev. Lett. 118, 073201 (2017). [27] Y. Tang, W. Kao, K. Y. Li, and B. L. Lev, Phys. Rev. [4] F. Seeßelberg, X.-Y. Luo, M. Li, R. Bause, S. Ko- Lett. 120, 230401 (2018). tochigova, I. Bloch, and C. Gohle, Phys. Rev. Lett. 121, [28] S. Lepoutre, J. Schachenmayer, L. Gabardos, B. Zhu, 253401 (2018). B. Naylor, E. Mar´echal, O. Gorceix, A. M. Rey, [5] M. Petzold, P. Kaebert, P. Gersema, M. Siercke, and L. Vernac, and B. Laburthe-Tolra, Nat. Commun. 10, S. Ospelkaus, New J. Phys. 20, 042001 (2018). 1714 (2019). [6] A. Ciamei, J. Szczepkowski, A. Bayerle, V. Barb´e,L. Re- [29] A. Patscheider, B. Zhu, L. Chomaz, D. Petter, S. Baier, ichs¨ollner, S. M. Tzanova, C.-C. Chen, B. Pasquiou, A.-M. Rey, F. Ferlaino, and M. J. Mark, Phys. Rev. Res. A. Grochola, P. Kowalczyk, W. Jastrzebski, and 2, 23050 (2020). F. Schreck, Phys. Chem. Chem. Phys. 20, 26221 (2018). [30] M. Guo, F. B¨ottcher, J. Hertkorn, J.-N. Schmidt, [7] M. Guo, X. Ye, J. He, M. L. Gonz´alez-Mart´ınez,R. Vex- M. Wenzel, H. P. B¨uchler, T. Langen, and T. Pfau, iau, G. Qu´em´ener,and D. Wang, Phys. Rev. X 8, 041044 Nature 574, 386 (2019). (2018). [31] L. Tanzi, E. Lucioni, F. Fam`a,J. Catani, A. Fioretti, [8] L. De Marco, G. Valtolina, K. Matsuda, W. G. Tobias, C. Gabbanini, R. N. Bisset, L. Santos, and G. Modugno, J. P. Covey, and J. Ye, Science 363, 853 (2019). Phys. Rev. Lett. 122, 1 (2019). [9] L. Anderegg, L. W. Cheuk, Y. Bao, S. Burchesky, [32] L. Tanzi, S. M. Roccuzzo, E. Lucioni, F. Fam`a, W. Ketterle, K.-K. Ni, and J. M. Doyle, Science 365, A. Fioretti, C. Gabbanini, G. Modugno, A. Recati, and 1156 (2019). S. Stringari, Nature 574, 382 (2019). [10] A. Yang, S. Botsi, S. Kumar, S. B. Pal, M. M. Lam, [33] A. Micheli, G. K. Brennen, and P. Zoller, Nat. Phys. 2, I. Cepait˙e,A.ˇ Laugharn, and K. Dieckmann, Phys. Rev. 341 (2006). Lett. 124, 133203 (2020). [34] B. Capogrosso-Sansone, C. Trefzger, M. Lewenstein, [11] J. A. Blackmore, L. Caldwell, P. D. Gregory, E. M. P. Zoller, and G. Pupillo, Phys. Rev. Lett. 104, 125301 Bridge, R. Sawant, J. Aldegunde, J. Mur-Petit, (2010). D. Jaksch, J. M. Hutson, B. E. Sauer, M. R. Tarbutt, and [35] A. V. Gorshkov, S. R. Manmana, G. Chen, E. Demler, S. L. Cornish, Quantum Sci. Technol. 4, 014010 (2018). M. D. Lukin, and A. M. Rey, Phys. Rev. A 84, 033619 [12] H. Yang, D. C. Zhang, L. Liu, Y. X. Liu, J. Nan, B. Zhao, (2011). and J. W. Pan, Science 363, 261 (2019). [36] A. V. Gorshkov, S. R. Manmana, G. Chen, J. Ye, [13] K. K. Voges, P. Gersema, T. Hartmann, T. A. Schulze, E. Demler, M. D. Lukin, and A. M. Rey, Phys. Rev. A. Zenesini, and S. Ospelkaus, New J. Phys. 21, 123034 Lett. 107, 115301 (2011). (2019). [37] A. Pikovski, M. Klawunn, G. V. Shlyapnikov, and [14] W. G. Tobias, K. Matsuda, G. Valtolina, L. De Marco, L. Santos, Phys. Rev. Lett. 105, 215302 (2010). J.-R. Li, and J. Ye, Phys. Rev. Lett. 124, 033401 (2020). [38] M. A. Baranov, A. Micheli, S. Ronen, and P. Zoller, [15] L. D. Carr, D. DeMille, R. V. Krems, and J. Ye, New J. Phys. Rev. A 83, 043602 (2011). Phys. 11, 055049 (2009). [39] A. C. Potter, E. Berg, D.-W. Wang, B. I. Halperin, and [16] R. Krems, B. Friedrich, and W. C. Stwalley, eds., Cold E. Demler, Phys. Rev. Lett. 105, 220406 (2010). Molecules (CRC Press, 2009). [40] N. T. Zinner, B. Wunsch, D. Pekker, and D. W. Wang, [17] M. A. Baranov, M. Dalmonte, G. Pupillo, and P. Zoller, Phys. Rev. A - At. Mol. Opt. Phys. 85, 1 (2012). Chem. Rev. 112, 5012 (2012). [41] W. Yi, A. J. Daley, G. Pupillo, and P. Zoller, New J. [18] J. L. Bohn, A. M. Rey, and J. Ye, Science 357, 1002 Phys. 10, 073015 (2008). (2017). [42] S. Nascimbene, N. Goldman, N. R. Cooper, and J. Dal- [19] A. Griesmaier, J. Werner, S. Hensler, J. Stuhler, and ibard, Phys. Rev. Lett. 115, 140401 (2015). T. Pfau, Phys. Rev. Lett. 94, 160401 (2005). [43] R. P. Anderson, D. Trypogeorgos, A. Vald´es-Curiel, [20] K. Aikawa, A. Frisch, M. Mark, S. Baier, A. Rietzler, Q.-Y. Liang, J. Tao, M. Zhao, T. Andrijauskas, R. Grimm, and F. Ferlaino, Phys. Rev. Lett. 108, 210401 G. Juzeli¯unas, and I. B. Spielman, Phys. Rev. Res. 2, (2012). 013149 (2020). [21] M. Lu, N. Q. Burdick, and B. L. Lev, Phys. Rev. Lett. [44] M.L¸acki, M. A. Baranov, H. Pichler, and P. Zoller, Phys. 108, 215301 (2012). Rev. Lett. 117, 233001 (2016). [22] A. de Paz, A. Sharma, A. Chotia, E. Mar´echal, J. H. [45] J. C. Budich, A. Elben, M.L¸acki, A. Sterdyniak, M. A. Huckans, P. Pedri, L. Santos, O. Gorceix, L. Vernac, and Baranov, and P. Zoller, Phys. Rev. A 95, 043632 (2017). B. Laburthe-Tolra, Phys. Rev. Lett. 111, 185305 (2013). [46] M. Lacki, P. Zoller, and M. A. Baranov, Phys. Rev. A [23] K. Aikawa, A. Frisch, M. Mark, S. Baier, R. Grimm, and 100, 033610 (2019). F. Ferlaino, Phys. Rev. Lett. 112, 010404 (2014). [47] F. Jendrzejewski, S. Eckel, T. G. Tiecke, G. Juzeli¯unas, [24] B. Naylor, A. Reigue, E. Mar´echal, O. Gorceix, G. K. Campbell, L. Jiang, and A. V. Gorshkov, Phys. B. Laburthe-Tolra, and L. Vernac, Phys. Rev. A 91, Rev. A 94, 063422 (2016). 011603 (2015). 21

[48] Y. Wang, S. Subhankar, P. Bienias, M.L¸acki, T.-c. Tsui, [61] A. Micheli, Z. Idziaszek, G. Pupillo, M. A. Baranov, M. A. Baranov, A. V. Gorshkov, P. Zoller, J. V. Porto, P. Zoller, and P. S. Julienne, Phys. Rev. Lett. 105, and S. L. Rolston, Phys. Rev. Lett. 120, 083601 (2018). 073202 (2010). [49] J. P. Covey, Enhanced Optical and Electric Manipula- [62] We would like to point out the difference to the case tion of a Quantum Gas of KRb Molecules, Springer The- considered in Ref. [69] where the coupling to other close ses (Springer International Publishing, Berlin Heidelberg, in energy (at short intermolecular distances) “spectator” 2018). states in the N=1 manifold reorient the induced dipole [50] J. Aldegunde, B. A. Rivington, P. S. Zuchowski,˙ and moments and results in an attractive dipole-dipole in- J. M. Hutson, Phys. Rev. A 78, 033434 (2008), 0807.1907. teraction for all directions. In our case, the presence of [51] S. A. Will, J. W. Park, Z. Z. Yan, H. Loh, and M. W. a strong electric field pushes the “spectator” states far Zwierlein, Phys. Rev. Lett. 116, 225306 (2016). away in energy thus making coupling to them irrelevant, [52] S. Ospelkaus, K.-K. Ni, G. Qu´em´ener, B. Neyenhuis, and also induces a permanent dipole moment which re- D. Wang, M. H. G. de Miranda, J. L. Bohn, J. Ye, and sults in repulsive interactions for side-by-side molecular D. S. Jin, Phys. Rev. Lett. 104, 030402 (2010). collisions. Head-to-tail collisions, where the dipole-dipole [53] D. Wang, B. Neyenhuis, M. H. G. de Miranda, K.-K. Ni, interaction is attractive, is suppressed by confining the S. Ospelkaus, D. S. Jin, and J. Ye, Phys. Rev. A 81, system in the direction of the electric field. This state- 061404 (2010). ment is supported by experimental observations currently [54] N. Okada, S. Kasahara, T. Ebi, M. Baba, and H. Katˆo, ongoing in the KRb experiment. J. Chem. Phys. 105, 3458 (1996). [63] J. W. Park, S. A. Will, and M. W. Zwierlein, Phys. Rev. [55] K. A. Burrows, H. Perrin, and B. M. Garraway, Phys. Lett. 114, 205302 (2015). Rev. A 96, 023429 (2017). [64] P. S. Zuchowski,˙ M. Kosicki, M. Kodrycka, and [56] A. P. Kazantsev, G. I. Surdutovich, and V. P. Yakovlev, P. Sold´an, Phys. Rev. A 87, 022706 (2013). Mech. Action Light Atoms (World Scientific Publishing [65] M. Aymar and O. Dulieu, J. Chem. Phys. 122, 204302 Co. Pte. Ltd., 1990). (2005). [57] R. Dum and M. Olshanii, Phys. Rev. Lett. 76, 1788 [66] S. Hofferberth, I. Lesanovsky, B. Fischer, J. Verdu, and (1996). J. Schmiedmayer, Nat. Phys. 2, 710 (2006), 0608228 [58] S. K. Dutta, B. Teo, and G. Raithel, IQEC, Int. Quan- [quant-ph]. tum Electron. Conf. Proc. (1999). [67] H. Perrin and B. M. Garraway, in Adv. At. Mol. Opt. [59] J. Ruseckas, G. Juzeli¯unas,P. Ohberg,¨ and M. Fleis- Phys., Vol. 66 (2017) pp. 181–262. chhauer, Phys. Rev. Lett. 95, 010404 (2005). [68] We use the fact that the energy of an electric dipole of [60] K.-K. Ni, S. Ospelkaus, M. H. G. de Miranda, A. Pe’er, 1 Debye in the electric field of 1 kV/cm is equal to the B. Neyenhuis, J. J. Zirbel, S. Kotochigova, P. S. Julienne, energy of a magnetic dipole moment of 1 Bohr magneton D. S. Jin, and J. Ye, Science 322, 231 (2008). in the magnetic field of 360 Gauss. [69] Z. Z. Yan, J. W. Park, Y. Ni, H. Loh, S. Will, T. Karman, and M. Zwierlein, (2020), arXiv:2003.02830.