Collisional loss of one-dimensional fermions near a p -wave Feshbach resonance
Ya-Ting Chang, Ruwan Senaratne, Danyel Cavazos-Cavazos, Randall G. Hulet
CCollisional loss of one-dimensional fermions near a p -wave Feshbach resonance Ya-Ting Chang, Ruwan Senaratne, Danyel Cavazos-Cavazos, and Randall G. Hulet ∗ Department of Physics and Astronomy, Rice University, Houston, Texas 77005, USA (Dated: September 23, 2020)We study collisional loss of a quasi-one-dimensional (1D) spin-polarized Fermi gas near a p -waveFeshbach resonance in ultracold Li atoms. We measure the location of the p -wave resonance inquasi-1D and observe a confinement-induced shift and broadening. We find that the three-bodyloss coefficient L as a function of the quasi-1D confinement has little dependence on confinementstrength. We also analyze the atom loss with a two-step cascade three-body loss model in whichweakly bound dimers are formed prior to their loss arising from atom-dimer collisions. Our dataare consistent with this model. We also find a possible suppression in the rate of dimer relaxationwith strong quasi-1D confinement. We discuss the implications of these measurements for observing p -wave pairing in quasi-1D. The realization of ultracold atomic Fermi gases has pro-vided experimental access to a wide array of phenomena,largely because of the presence of Feshbach resonances(FRs) that provide for externally tunable interactions [1–4]. In addition to the usual s -wave interactions betweendistinguishable fermions, higher partial-wave interactionsmay be tuned via FRs [5]. p -Wave interactions are ofparticular interest as they are the dominant low-energyscattering process between identical fermions and are pre-dicted to exhibit phenomena distinct from those observedin s -wave interacting Fermi gases [6]. In particular, pair-ing between identical fermions is an essential ingredientof the Kitaev chain Hamiltonian [7], which supports Ma-jorana zero-modes at the ends of the chain. These zero-modes have been observed in semiconducting nanowires[8], and are a promising candidate platform for fault-tolerant quantum computing [9, 10]. p -Wave FRs have been observed in K [11–13] and Li[14–19]. The severe atom losses associated with these res-onances, however, have limited their usefulness. Three-body losses, which are suppressed by symmetry in thecase of a fermionic two-spin system with s -wave inter-actions [20], are not suppressed for p -wave interactions.Much work has been done in characterizing the atom lossassociated with p -wave FRs [21–24], and there is renewedinterest in studying these resonances in reduced dimen-sions. Recent theoretical work has suggested that three-body losses may be suppressed in quasi-1D [25]. The ab-sence of a centrifugal barrier in 1D results in Feshbachdimers that have extended wavefunctions which overlapless with deeply-bound molecules. If three-body loss issuppressed by this mechanism, it might open a path to-wards realizing p -wave pairing in quasi-1D and emulatingthe Kitaev chain Hamiltonian.We present an experimental study of three-body losses near a p -wave FR of identical Li fermions in quasi-1D.We measure the three-body loss coefficient ( L ) as a func-tion of 1D confinement for a direct three-body process.We also analyze the observed atom loss within the frame-work of a cascade model with explicit dimer formationand relaxation steps [26, 27], using in situ imaging to re-duce the effect of the inhomogeneous density. Finally, wecharacterize the confinement-induced shifts in the reso-nance position that appear in quasi-1D [28–32]. Theseshifts allow us to extract a value for the effective range.The apparatus and the experimental methods we useto prepare degenerate Fermi gases have been describedpreviously [33–35]. A Li degenerate Fermi gas is firstprepared in the two lowest hyperfine states of the S / manifold (states | i and | i , respectively) at 595 G, andthen loaded into a crossed-beam dipole trap formed bythree linearly-polarized mutually-orthogonal laser beamsof wavelength λ = 1 . µ m. Each beam is retro-reflected, with the polarizations of the incoming andretro-reflected beams initially set to be perpendicular toeach other to avoid lattice formation. We eliminate state | i from the trap with a resonant burst of light. At thisstage, we obtain × atoms in state | i in a nearlyisotropic harmonic trap with a geometric-mean trappingfrequency of π × Hz, and at a temperature
T /T F ≈ . where, T F is the Fermi temperature. The optical trapdepths are increased and the polarizations of the retro-reflected beams are rotated to achieve a 7 E r deep 3Doptical lattice, where E r = h / (2 mλ ) = k B × . µ Kis the recoil energy, and m is the atomic mass. Dur-ing the lattice ramp-up, a co-propagating beam of 532nm light is introduced along each trapping-beam dimen-sion to flatten the trapping potential [33, 34]. By tuningthese compensation beam powers, we create a 3D bandinsulator with a central density of approximately 1 atom a r X i v : . [ phy s i c s . a t o m - ph ] S e p per site. In order to produce a 2D lattice, which is anarray of quasi-1D tubes, we slowly turn off the compen-sation beams and the vertical lattice beam, while increas-ing the intensity of the two remaining beams to achievea desired 2D lattice depth, V L . This depth determinesthe confinement in the quasi-1D traps, which is parame-terized by a ⊥ = p ~ /mω r transversely and R F axially,where ω r = √ E r V L / ~ is the trapping frequency of a lat-tice site when approximated as a harmonic potential, and R F ( N t,j , ω z ) = p (2 N t,j + 1) ~ /mω z is the Fermi radius oftube j with number of atoms N t,j and an axial frequency ω z . The aspect ratio of the quasi-1D tubes, ω r /ω z ≈ .We load a maximum of around 30 atoms per quasi-1Dtube with T < T F to avoid exciting any radial modes.We use a two-step servo scheme to stabilize the cur-rent in the coils producing the Feshbach magnetic field,because the Li | i − | i p -wave FR near 159 G is verynarrow. The first servo, S , provides the large dynamicrange required to run our experimental sequence, whilethe second servo, S , controls the current in a bypass cir-cuit added in parallel to the magnetic coils. This improvesthe stability of the magnetic field to ±
10 mG and providesfiner magnetic-field resolution. After reaching the holdfield B , the atoms are transferred into | i with a π -pulseof duration 75 µ s using RF radiation resonant with the | i − | i transition. After a hold time τ , we ramp the fieldback to 595 G, where the distribution of the remainingatoms is imaged using in situ phase-contrast imaging witha probe beam propagating perpendicular to the tube axis[35]. By using the inverse Abel transform, which exploitsthe approximate cylindrical symmetry of the 2D lattice,we measure the distribution with a spatial resolution ofapproximately three lattice constants. We sector the 2Dlattice into concentric shells in which the tubes have sim-ilar chemical potentials, µ . This procedure is useful asscattering processes are in general energy-dependent, soobservables depend on rate coefficients that are averagedover the Fermi-Dirac distribution for atoms in each tube.We characterize the | i−| i p -wave FR in 3D and quasi-1D by measuring atom loss as functions of B and τ . In 3D,we find the onset of loss at 159.05(1) G, which agrees withprevious measurements of the location of this resonance in3D [15, 17] but differs with other measurements [19, 36] bya few 10’s of mG. We are not able to resolve the expecteddoublet feature arising from the dipole-dipole interaction[12, 19, 37] because of limitations of the field stability. Allthe 1D data in this paper were measured with the mag-netic field aligned with the z-axis, and thus only involvecollisions with the m l = 0 projection of the angular mo-mentum. As V L is increased, we observe a confinement- FIG. 1. (a) p -Wave resonances in 3D and quasi-1D measuredwith magnetic-field-dependent loss. Dashed lines show theresonance position for each V L . We define the resonance fieldfor zero-momentum collisions, which corresponds to the on-set (15% loss, to overcome atom number fluctuation) of theobserved atomic loss. Data are averaged over 6 experimentalruns and error bars are the standard error of the mean. (b)Diamonds show B D vs V L . The solid curve shows the resultof fitting the data to Eq. 3, where the effective range α p = a − and B D = τ is chosen such that peak loss is 30-50% of total atom numberfor each value of V L : 2.5 ms for 3D, 0.5 ms for E r and 0.2ms for 15-75 E r . induced shift in the resonance field and broadening of theatom-loss feature, as shown in Fig. 1(a).We review p -wave scattering in 3D and quasi-1D toshow how the measured confinement-induced shift can beused to extract α p , the 3D effective range. For low-energycollisions in 3D, the cotangent of the phase-shift δ p associ-ated with p -wave scattering can be expanded as a functionof scattering volume, V p , and effective range, α p [38]: k cot( δ p ( k )) = − V p − α p k + O ( k ) , (1)where α p > and has units of inverse length. Thesescattering properties are modified in quasi-1D, k cot( δ p ( k )) = − l p − ξ p k + O ( k ) , (2)where l p is the 1D scattering length and ξ p is the1D effective range, which has units of length. Thesequasi-1D scattering parameters are given by l p =3 a ⊥ (cid:2) a ⊥ / V p + α p a ⊥ + 6 | ζ ( − / | (cid:3) − and ξ p = α p a ⊥ / [30–32], where ζ is the Riemann zeta function ( ζ ( − / ≈− . ). The second and third terms in /l p lead to aconfinement-induced shift in the resonance location. Inthis formalism, only dynamics along the axial dimensionare relevant, and scattering quantities, such as the elasticscattering cross-section, are expressed in units appropri-ate for 1D.By performing a coupled-channel calculation, which re-quires detailed knowledge of the inter-atomic potentials[39], we obtain an expansion /V p ( B ) up to second or-der in B . The effective range α p can be approximatedas a constant independent of B for the relevant range ofmagnetic field. The FR in 3D occurs at the magneticfield B at which V p diverges. Similarly, in quasi-1D,the resonance occurs when l p diverges at a magnetic field B , which is a function of V L and α p . The confinement-induced shift, δ B ( V L , α p ) = B D − B D , can be approx-imated to leading order in confinement strength V L by[40] δ B = − mE r ~ ∂ (1 /V p ) ∂B | B = B D α p p V L . (3)We cannot accurately measure B D for m l = 0 alone dueto the unresolved | m l | = 1 collisions in 3D, so we fit themeasured δ B as a function of V L to Eq. 3 by taking α p and B D as fitting parameters. The result of the fit to thequasi-1D data is shown by the solid curve in Fig. 1(b).We obtain α p = a − which is consistent with ourcoupled-channel result of 0.1412 a − , where a is the Bohrradius, and B D = 159 . which is consistent with ourloss-onset measurement and a dipolar splitting of 10 mGin 3D [19]. We also find a consistent value by analyzingprevious measurements performed on a 2D gas of Li instate | i [21, 40].The observed atom loss is presumably due to the for-mation of deeply-bound molecules. To characterize theloss, we measured N , the number of atoms remainingin the trap after a hold time τ for various B and V L .Background-gas collisions lead to a /e atom lifetime of38 s in this apparatus, and are negligible for this analysis. Atom loss due to three-body collisions is described by ˙ NN = − L n , (4)where n = ( N t,c / R F,c ) is the squared atomic line den-sity for a central tube, determined using a length-scale oftwice the local Fermi radius R F,c . We measure the timeevolution with V L between 15 and 75 E r and extract L by fitting loss vs τ to Eq. 4. Fig. 2(a) shows such a fitto typical loss data. Since L also depends on ∆ B , thefield detuning from resonance, we extract L from thetime evolution at several ∆ B to find the peak value foreach V L . The peak L for all V L are found to be approxi- Shells (a)(b)
FIG. 2. Typical time evolution of (a) total number in the en-tire sample and (b) averaged tube population h N t i in 4 shells.For these data, ∆ B =
30 mG and V L = E r . The differentcolors and symbols in (b) indicate different shells with approx-imately uniform initial atom number per tube. The shells arelabeled from i = 1 , the inner-most, to i = 4 , the outer-most.Solid curves show fits to Eq. 4 to extract L with the squaredatomic density (a) n = ( N t,c / R F,c ) of a central tube and(b) n = ( h N t i i / R F,i ) of a typical tube in each shell. Thecorresponding L values are plotted in Fig. 3. Data points areaveraged over 5 shots, and the standard error of the mean is(a) approximately equal to the symbol size and (b) indicatedby the error bars. mately × − cm /s. We observe no dependence on1D confinement in this range [40]. Due to the inhomo-geneity of the initial distribution of atoms across the 2Dlattice, however, we find a rather poor agreement of thedata to Eq. 4.The results of a more comprehensive analysis of thesame data that provides an improved fit to Eq. 4 is shownin Fig. 2(b). Here, we group the tubes into separatecylindrical shells (labeled by i = 1 − ) with an averagedatom number per tube h N t i i [40] and a correspondingFermi temperature T F,i . Figure 3(a) shows L for eachshell extracted from data with V L = 75 E r vs ∆ B . Thepeak L for each shell is in the range of × − cm /s to × − cm /s, and is similar to the peak L extractedfrom the whole atomic cloud.In [25], Zhou and Cui suggest that the rate of three-body loss near a p -wave FR can be suppressed by reduc-ing the overlap between the wavefunctions of a deeply-bound molecule and a Feshbach dimer with increasingconfinement. To investigate this hypothesis, we analyzeour observed loss data using a cascade model of two con-secutive two-body processes instead of a direct three-bodyevent: two atoms resonantly form a dimer, followed by acollision between the dimer and an atom, resulting in adeeply-bound molecule and an atom [26]. This approachhas previously been applied to the particular p -wave FRwe study, but in 3D and quasi-2D [27]. It is the naturalformalism in which to evaluate the predicted suppression,as it models the formation and relaxation of dimers. Theequations governing this loss process are dN a dt = 2 Γ ~ N d − K aa N a ( N a − R F − K ad N a N d R F , (5a) dN d dt = − Γ ~ N d + K aa N a ( N a − R F − K ad N a N d R F , (5b)where N a is the number of atoms, N d is the number ofdimers, K aa is the two-body event rate for atom-atomcollisions converting atoms into dimers, and K ad is thetwo-body atom-dimer inelastic collision event rate. Γ , theone-body decay rate of dimers is the width of the FR. Therate of dimer formation is proportional to the number ofpossible pairs of atoms, given by N a ( N a − / . K ad is of particular interest, as it depends on the over-lap between dimers and deeply-bound molecules. Both Γ and K aa are related to the elastic scattering cross-section, σ D ( E ) , which can be calculated, thus constraining the fitto the cascade process to a single parameter, K ad . σ D ( E ) may be approximated by a Lorentzian in collision energy, E = ~ k /m , centered at the above-threshold binding en-ergy of the Feshbach dimer E res = − ~ /l p ξ p m > andwith width Γ = ( ~ /ξ p ) p E res /m [6, 40]. K aa may be calculated by averaging σ ( k r ) over theensemble of pairs of atoms with relative momentum k r and velocity v r K aa = h σ ( k r ) v r i = ~ Z ∞−∞ dk r σ ( k r ) v r P ( k r ) , (6)where P ( k r ) is the probability density function of k r ob-tained from the density distribution of a trapped Fermigas [40]. We assume a global temperature T across theentire sample. However, µ varies significantly from tube-to-tube due to the density inhomogeneity across the 2Dlattice. This effect is mitigated by sectoring the cloudinto shells of similar µ , as discussed earlier, thus givinga distinct value of K aa for each shell. For each quasi-1Dtube, µ is determined by N t,j and T .Although we cannot directly measure T , we exploit thefact that at a sufficiently large ∆ B , the rate equations canbe approximated as a direct three-body loss process with aloss coefficient ˜ L = (3 / ~ K ad K aa / Γ under the assump-tions of a steady-state dimer population ( dN d /dt = 0 )and Γ / ~ (cid:29) K ad N a / R F [27]. Assuming that these as-sumptions hold for large ∆ B , we fit the measured valuesof L for each shell with T and K ad as fitting parametersto ˜ L . We find that T = 0 . T F, , and that K ad = 0 . cm/s is independent of field for ∆ B >
100 mG. The as-sumptions given above are confirmed in this range. Thesolid lines in Fig. 3(a) show ˜ L for each shell.The extracted K ad values from fitting loss data for V L = E r to Eqs. 5 using the calculated values of Γ and K aa are shown in Fig. 3(b) for the full range of ∆ B [40]. Wefind that under these conditions, Eqs. 5 model the timebehavior of the observed loss as well as Eq. 4. The valuesof K ad extracted for ∆ B > mG are field indepen-dent. The observed field independence strongly supportsthe cascade model as the atom-dimer collision process isinherently non-resonant. In the dimer formation step,the atoms must collide with a momentum dictated bythe binding energy of the dimer, which is field-dependent.The dimer relaxation step, however, may proceed for anycollision momentum, as the atom receives the binding en-ergy of the deeply bound molecule.The behavior of K ad for ∆ B < mG is consistentwith a suppression of the rate of dimer relaxation. Thespatial overlap of the dimer and deeply-bound wavefunc-tions increases with κa ⊥ , where κ = √ mE res / ~ , so thepredicted suppression is strongest for small ∆ B , where (a) Shells (b)
FIG. 3. (a) L vs ∆ B for V L = E r . L is obtained byfitting N t,i vs τ to Eq. 4 for each shell. An example of thisdata is given in Fig. 2(b) for ∆ B = 30 mG. Solid curves show (3 / ~ K ad K aa / Γ with a constant K ad = T = 0 . T F, , where T F, = 4 . µ K. (b) K ad vs ∆ B . K ad is extracted by fitting h N t i i vs τ to Eq. 5, using thecalculated values of Γ and K aa . Black dashed line indicates ∆ B =
27 mG, which corresponds to κa ⊥ = V L = E r [25]. Error bars are one-sigma confidence intervals for thefitting parameters L and K ad . The large uncertainty in thefitted values for the outermost shell is indicative of small N t . E res is smallest. The suppression is expected to be sig-nificant for κa ⊥ < / [25], which for V L = E r cor-responds to ∆ B <
27 mG. Another interpretation of thesmall-detuning behavior of K ad is that the cascade modelbreaks down due to, for example, the existence of a shal-low three-body bound state [41].This work is the first detailed experimental study of p -wave collisions in quasi-1D. We confirm the confinement-induced shift and broadening as a function of V L . Theconfinement-induced shift agrees well with quasi-1D the-ory [32] and the extracted value of α p agrees with previouswork [21]. We measure L as a function of V L and findno dependence up to 75 E r . The magnetic field indepen-dence of K ad for ∆ B > mG confirms the cascade model[26, 27] for three-body loss in quasi-1D in the regime oflarge ∆ B ( > mG), as well as for intermediate ∆ B (50-100 mG) where the cascade model is not well approx-imated by the three-body loss rate equation.The suppression in K ad at ∆ B < mG is possiblyexplained by p -wave dimer stretching [25]. Achievinggreater suppression in Li by increasing V L is challeng-ing since at a fixed ∆ B , κa ⊥ ∝ /V / L [40], but futurework at even higher V L or with improved magnetic fieldresolution and stability would enable further study of thisnarrow feature. Our result also provides insight into a po-tential pathway towards observing pairing between iden-tical fermions in cold atom systems. Suppressing loss inheavier fermions with FRs, such as K [11–13],
Dy[42], and
Er [43], is promising, as small values of κa ⊥ may be more readily achieved in these atoms. Note added. − During the peer-review process, anothergroup reported on a similar experiment [44]. Althoughboth groups observe similar overall atom loss, they reporta suppression of L ∝ V − L , while we find L independentof V L over a wide range (Fig. S2). The difference liesin the choice between defining L using the 3D or the1D densities. In their analysis, L is defined in termsof the 3D density of a tube, which increases with V / L ,while we use the 1D line density. While the two resultsare consistent, we argue that 1D densities are most ap-propriate based on physical and practical considerations.Physically, the dimensionless quantity κa ⊥ parameterizesthe effective dimensionality of the system near a FR, andthe peak values of L we report were measured in regionswhere κa ⊥ < . Practically, 1D units make it clear thatthe peak loss rate is independent of V L .We would like to thank T. L. Yang for his contribu-tions to the apparatus and W. I. McAlexander for hiscoupled-channel code. This work was supported in partby the Army Research Office Multidisciplinary Univer-sity Research Initiative (Grant Nos. W911NF-14-1-0003and W911NF-17-1- 0323), the NSF (Grant No. PHY-1707992), and the Welch Foundation (Grant No. C-1133).D. C. acknowledges financial support from CONACyT(Mexico, Scholarship No. 472271). ∗ [email protected][1] B. DeMarco and D. S. Jin, Science , 1703 (1999).[2] A. G. Truscott, K. E. Strecker, W. I. McAlexander, G. B.Partridge, and R. G. Hulet, Science , 2570 (2001).[3] F. Schreck, L. Khaykovich, K. L. Corwin, G. Ferrari,T. Bourdel, J. Cubizolles, and C. Salomon, Phys. Rev.Lett. , 080403 (2001).[4] S. R. Granade, M. E. Gehm, K. M. O’Hara, and J. E. Thomas, Phys. Rev. Lett. , 120405 (2002).[5] C. Chin, R. Grimm, P. Julienne, and E. Tiesinga, Rev.Mod. Phys. , 1225 (2010).[6] V. Gurarie and L. Radzihovsky, Ann. Phys. (NY) , 2(2007).[7] A. Y. Kitaev, Phys. Usp. , 131 (2000).[8] S. D. Sarma, M. Freedman, and C. Nayak, Quantum Inf. , 15001 (2015).[9] S. B. Bravyi and A. Y. Kitaev, Ann. Phys. (NY) , 210(2002).[10] A. Stern and N. H. Lindner, Science , 1179 (2013).[11] C. A. Regal, C. Ticknor, J. L. Bohn, and D. S. Jin, Phys.Rev. Lett. , 053201 (2003).[12] K. Günter, T. Stöferle, H. Moritz, M. Köhl, andT. Esslinger, Phys. Rev. Lett. , 230401 (2005).[13] C. Luciuk, S. Trotzky, S. Smale, Z. Yu, S. Zhang, andJ. H. Thywissen, Nat. Phys. , 599 (2016).[14] J. Zhang, E. G. M. van Kempen, T. Bourdel,L. Khaykovich, J. Cubizolles, F. Chevy, M. Teichmann,L. Tarruell, S. J. J. M. F. Kokkelmans, and C. Salomon,Phys. Rev. A , 030702 (2004).[15] C. H. Schunck, M. W. Zwierlein, C. A. Stan, S. M. F. Rau-pach, W. Ketterle, A. Simoni, E. Tiesinga, C. J. Williams,and P. S. Julienne, Phys. Rev. A , 045601 (2005).[16] Y. Inada, M. Horikoshi, S. Nakajima, M. Kuwata-Gonokami, M. Ueda, and T. Mukaiyama, Phys. Rev.Lett. , 100401 (2008).[17] J. Fuchs, C. Ticknor, P. Dyke, G. Veeravalli, E. Kuhnle,W. Rowlands, P. Hannaford, and C. J. Vale, Phys. Rev.A , 053616 (2008).[18] R. A. W. Maier, C. Marzok, C. Zimmermann, and P. W.Courteille, Phys. Rev. A , 064701 (2010).[19] M. Gerken, B. Tran, S. Häfner, E. Tiemann, B. Zhu, andM. Weidemüller, Phys. Rev. A , 050701 (2019).[20] D. S. Petrov, C. Salomon, and G. V. Shlyapnikov, Phys.Rev. Lett. , 090404 (2004).[21] M. Waseem, Z. Zhang, J. Yoshida, K. Hattori, T. Saito,and T. Mukaiyama, J. Phys. B , 204001 (2016).[22] M. Waseem, T. Saito, J. Yoshida, and T. Mukaiyama,Phys. Rev. A , 062704 (2017).[23] J. Yoshida, T. Saito, M. Waseem, K. Hattori, andT. Mukaiyama, Phys. Rev. Lett. , 133401 (2018).[24] M. Waseem, J. Yoshida, T. Saito, and T. Mukaiyama,Phys. Rev. A , 020702 (2018). [25] L. Zhou and X. Cui, Phys. Rev. A , 30701 (2017).[26] J. Li, J. Liu, L. Luo, and B. Gao, Phys. Rev. Lett. ,193402 (2018).[27] M. Waseem, J. Yoshida, T. Saito, and T. Mukaiyama,Phys. Rev. A , 052704 (2019).[28] M. Olshanii, Phys. Rev. Lett. , 938 (1998).[29] T. Bergeman, M. G. Moore, and M. Olshanii, Phys. Rev.Lett. , 163201 (2003).[30] B. E. Granger and D. Blume, Phys. Rev. Lett. , 133202(2004).[31] L. Pricoupenko, Phys. Rev. Lett. , 170404 (2008).[32] D. V. Kurlov and G. V. Shlyapnikov, Phys. Rev. A ,032710 (2017).[33] R. A. Hart, P. M. Duarte, T.-L. Yang, X. Liu, T. Paiva,E. Khatami, R. T. Scalettar, N. Trivedi, D. A. Huse, andR. G. Hulet, Nature , 211 (2015).[34] T. L. Yang, P. Grišins, Y. T. Chang, Z. H. Zhao, C. Y.Shih, T. Giamarchi, and R. G. Hulet, Phys. Rev. Lett. , 103001 (2018).[35] R. G. Hulet, J. H. V. Nguyen, and R. Senaratne, Rev.Sci. Instrum. , 011101 (2020).[36] T. Nakasuji, J. Yoshida, and T. Mukaiyama, Phys. Rev.A , 12710 (2013).[37] C. Ticknor, C. A. Regal, D. S. Jin, and J. L. Bohn, Phys.Rev. A , 042712 (2004).[38] C. J. Joachain, Quantum Collision Theory (North-Holland, 1975).[39] M. Houbiers, H. T. C. Stoof, W. I. McAlexander, andR. G. Hulet, Phys. Rev. A , R1497 (1998).[40] See Supplemental Material at [URL] for information re-garding the derivation of Eq. 3, the confinement-inducedshift in quasi-2D, the dependence of L vs V L , in situ imaging data, the fitting of time evolution data to the cas-cade model, the quasi-1D p -wave scattering cross-section,the probability density function of k r , and the dependenceof κa ⊥ on V L .[41] M. Schmidt, H.-W. Hammer, and L. Platter, Phys. Rev.A , 062702 (2020).[42] K. Baumann, N. Q. Burdick, M. Lu, and B. L. Lev, Phys.Rev. A , 020701 (2014).[43] S. Baier, D. Petter, J. H. Becher, A. Patscheider, G. Na-tale, L. Chomaz, M. J. Mark, and F. Ferlaino, Phys. Rev.Lett. , 093602 (2018).[44] A. S. Marcum, F. R. Fonta, A. M. Ismail, and K. M.O’Hara, arXiv:2007.15783 [physics.atom-ph] (2020). upplementary material for “Collisional loss of one-dimensional fermions near a p − wave Feshbach resonance” Ya-Ting Chang, Ruwan Senaratne, Danyel Cavazos-Cavazos, and Randall G. Hulet ∗ Department of Physics and Astronomy, Rice University, Houston, Texas 77005, USA (Dated: September 23, 2020)
CONFINEMENT-INDUCED SHIFT δ B IN QUASI-1D
The Feshbach resonance occurs at a magnetic field B D where the 1D p -wave scattering length l p diverges [1–3] l p = a ⊥ / V p + α p a ⊥ + 6 | ζ ( − / | a ⊥ = 0 . (S1)Since α p a ⊥ (cid:29) | ζ ( − / | for the lattice depths V L we can achieve in this experiment, Eq. S1 can be approximatedusing /V p = − α p /a ⊥ . By Taylor expanding this around the 3D resonance field B D to first order V p | B = B D = 1 V p | B = B D + ∂ (1 /V p ) ∂B | B = B D ( B D − B D ) = − α p a ⊥ , (S2)we obtain a simple analytical form for the confinement-induecd shift δ B ( V L , α p ) = B D − B D δ B = − α p∂ (1 /V p ) ∂B | B = B D a ⊥ = − mE r ~ ∂ (1 /V p ) ∂B | B = B D α p p V L , (S3)where ∂ (1 /V p ) ∂B | B = B D < . CONFINEMENT-INDUCED SHIFT δ B, D IN QUASI-2D
Similarly to the confinement-induced shift in quasi-1D, an equivalent expression to Eq. S3 can be derived forthis geometry by considering the quasi-2D scattering parameters [3]. The confinement-induced shift in quasi-2D δ B, D = B D − B D can be approximated by δ B, D = 12 δ B . (S4)The open circles in Fig. S1 show the data of δ B in quasi-2D from [4], and the solid curve shows the result of a fit toEq. S4 with the effective range α p = a − as a fitting parameter. This value of α p is within 15% of the valuethe authors of [4] obtained by fitting measurements of the dissociation energy, as well as the value extracted from thefit to our quasi-1D data shown in the main text. ∗ [email protected] a r X i v : . [ phy s i c s . a t o m - ph ] S e p Figure S1. The open circles show the confinement-induced shift for the quasi-2D p -wave Feshbach resonance position as afunction of the corresponding lattice depth (data from [4]). The solid curve shows the result of fit to Eq. S4, with the effectiverange α p = a − . THREE-BODY LOSS COEFFICIENT L VERSUS V L We measure the time evolution of atom number N with V L between 15 and 75 E r and extract L by fitting to Eq.4 in the main text. The peak L as a function of V L is shown in Fig. S2. Figure S2. Peak three-body loss coefficient L as a function of lattice depth V L . L is extracted by fitting the time evolution of N to Eq. 4 as described in the Fig. 2(a) caption. Error bars indicate the one-sigma confidence interval for the fitting parameters L . ANALYSIS USING
IN SITU
IMAGING WITH INVERSE ABEL TRANSFORM
We probe atoms using in situ imaging and perform the inverse Abel transform on the column density, assumingcylindrical symmetry, to obtain the distribution of the number of atoms per tube N t ( r ) . We group the tubes intoseparate cylindrical shells as shown in Fig. S3. (a) (b) (c)(d) (e) (f) Figure S3. Typical time evolution of N t in a tube at a distance r from the center of the 2D lattice. Regions with coloredbackgrounds correspond to the shells i = 1 − in the main text Fig. 2(b). Data points are averaged over 5 shots, and errorbars indicate the standard error of the mean. TYPICAL TIME EVOLUTION OF h N t i FITS TO THE CASCADE MODEL
We extract K ad from the fits to Eq. 5 in the main text, using theoretical values of K aa and Γ for T = 0 . T F, . Atypical time evolution with the fitted curve is shown in Fig. S4. QUASI-1D p -WAVE SCATTERING CROSS-SECTION The quasi-1D p -wave scattering amplitude is [3] f ( k ) = − ik /l p + ξ p k + ik , (S5)In 1D, the equivalent of the scattering cross-section is simply the modulus of the scattering amplitude squared σ ( k ) = | f D ( k ) | = k k + (1 /l p + ξ p k ) , (S6)which is bounded from above by 1. Near a Feshbach resonance for l p < , this expression may be approximated by aLorentzian in terms of the collision energy, E = ~ k /m as follows: σ ( E ) ≈ (cid:0) Γ2 (cid:1) ( E − E res ) + (cid:0) Γ2 (cid:1) . (S7)Here, E res = − ~ /l p ξ p m > for l p < is the above-threshold binding energy of the Feshbach molecule, and Γ =( ~ /ξ p ) p E res /m is the width of the resonance [5]. Figure S4. Typical time evolution of averaged tube population h N t i in 4 shells at ∆ B =
30 mG with V L = E r . Data are thesame as Fig. 2(b) in the main text. Solid curves show fits to Eq. 5 in the main text with K aa and Γ calculated for T = 0 . T F, . PROBABILITY DENSITY FUNCTION OF k r FOR A TRAPPED FERMI GAS
The probability density function of k r for a trapped Fermi gas is P ( k r ) = ~ N a Z ∞−∞ dk n ( k ) n ( k − k r ) , (S8)and the k -space number density is given by n ( k ) = 12 π ~ Z ∞−∞ dx β ( mω x + ~ k m − µ )] + 1 , (S9)where β = 1 /k B T and µ is the chemical potential. DEPENDENCE OF κa ⊥ ON V L κ is the magnitude of the wavevector related to the binding energy of the Feshbach molecule, which can be calculatedby κ = √ mE res ~ = s − l p ξ p . (S10)where l p and ξ p are the 1D scattering parameters modified from 3D scattering quantities V p and α p with confinementstrength a ⊥ as mentioned in the main text. By Taylor expanding κ around the Feshbach resonance field B for aparticular V L , we find a constant κ at a fixed magnetic field detuning ∆ B which is independent of V L : κ (∆ B ) = κ | B = B D + ∂κ ∂B | B = B D ∆ B + ∂ κ ∂B | B = B D ∆ B + O (∆ B )= ∂ (1 /V p ) ∂B | B = B D ∆ B + ∂ (1 /V p ) ∂B | B = B D ∆ B + O (∆ B ) (S11)Therefore, κa ⊥ is proportional to V / L for a particular ∆ B , as shown in Fig. S5. Figure S5. κa ⊥ as a function of lattice depth V L . Each curve is V / L with a scaling factor set by ∆ B .[1] B. E. Granger and D. Blume, Phys. Rev. Lett. , 133202 (2004).[2] L. Pricoupenko, Phys. Rev. Lett. , 170404 (2008).[3] D. V. Kurlov and G. V. Shlyapnikov, Phys. Rev. A , 032710 (2017).[4] M. Waseem, Z. Zhang, J. Yoshida, K. Hattori, T. Saito, and T. Mukaiyama, J. Phys. B , 204001 (2016).[5] V. Gurarie and L. Radzihovsky, Ann. Phys. (NY)322