Theoretical study of the spin and charge dynamics of two-leg ladders as probed by resonant inelastic x-ray scattering
TTheoretical study of the spin and charge dynamics of two-leg laddersas probed by resonant inelastic x-ray scattering
Umesh Kumar,
1, 2
Alberto Nocera,
1, 3, 4
Elbio Dagotto,
1, 3 and Steven Johnston
1, 2 Department of Physics and Astronomy, The University of Tennessee, Knoxville, TN 37996, USA Joint Institute for Advanced Materials, The University of Tennessee, Knoxville, TN 37996, USA Materials Science and Technology Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, USA Stewart Blusson Quantum Matter Institute, University of British Columbia, Vancouver, British Columbia V6T 1Z4, Canada (Dated: April 2, 2019)Resonant inelastic x-ray scattering (RIXS) has become an important tool for studying elementaryexcitations in correlated materials. Here, we present a systematic theoretical investigation of theCu L-edge RIXS spectra of undoped and doped cuprate two-leg spin-ladders in both the non-spin-conserving (NSC) and spin-conserving (SC) channels. The spectra are rich and host many exoticexcitations. In the NSC-channel of the undoped case, we identify one-triplon and bound triplet two-triplon excitations in the strong-rung coupling limit, as well as confined spinons in the weak-rung coupling limit. In the doped case, we observe a quasiparticle excitation formed from a boundcharge and spin- in the strong-rung coupling limit. In the SC-channel, we also identify severalnew features, including bound singlet two-triplon excitations and confined spinons in the undopedladders in the strong- and weak-rung coupling limits, respectively. Conversely, in the doped case,the SC channel primarily probes both gapless and gapped charge excitations. Finally, we revisit theavailable data for the ladder compound Sr Cu O in the context of our results. I. INTRODUCTION
Strongly correlated spin ladders are excellent platformsfor studying quantum many-body phenomena, such ashigh critical temperature (high-T c ) superconductivity [1]and spinon confinement [2]. Quantum ladders are in-termediate between one- and two-dimensional materials,and their study allows for detailed comparisons betweentheoretical models and experimental probes [2–20]. Thediscovery of superconductivity in the ladder “telephonenumber” compound Sr . Ca . Cu O . [21], whichhad been theoretically predicted [22], created new op-portunities to study the relationships between lattice,orbital, charge, and magnetic degrees of freedom and un-conventional superconductivity in copper-oxide materi-als. Accordingly, a significant effort has been launchedto understand the magnetic excitation spectrum of ma-terials hosting quantum ladders and its connection to su-perconductivity. For example, inelastic neutron scatter-ing (INS) studies have reported the observation of a spingap in Sr Cu O [13], triplon and two-triplon excita-tions in La Sr Cu O [23], and spinon confinement inCaCu O [2].With continued improvements in instrumentation, res-onant inelastic x-ray scattering (RIXS) is being in-creasingly employed to study collective magnetic exci-tations [3–5, 24–30]. RIXS is complementary to INS inthat the scattering processes allows for both ∆ S = 0and ∆ S = 1 excitations, depending on the elementaledge [25, 31], the strength of the spin-orbit couplingin the core level [31], and the local crystal stuctureof the material [32]. As such, the technique accessesmany magnetic excitations including magnons [26–30],bimagnons in two-dimensional (2D) cuprates [25, 33],and multi-spinon excitations in one-dimensional (1D)cuprates [4, 34–36]. RIXS also provided surprising re- sults for 2D cuprates, where the paramagnon excitationsare found to persist deep into the overdoped region ofthe phase diagram [26–29, 37]. Recently, an electron-holeasymmetry in the doping dependence of the spin excita-tions of 2D cuprates was reported, as well as an addi-tional collective charge excitation in the electron-dopedcase that is absent in the hole-doped case [30].The rich variety of excitations observed in 1D and 2Dcuprates described above, and their possible connectionto unconventional superconductivity, provides a strongmotivation for exploring the RIXS spectra of two-leg spinladders, both as a function of the rung coupling and dop-ing. Such studies provide information not only aboutmagnetic excitations but also about potentially cooper-ative/competing charge, orbital, and lattice excitations.Early RIXS Cu K -edge experiments on the telephonenumber compounds focused primarily on the high-energycharge excitations across the Mott gap [38–40]. Later,as the instrumental resolution improved, studies startedaddressing low-energy magnetic excitations. For exam-ple, the magnetic response of Sr Cu O at the Cu L -edge was measured [5] and interpreted in terms ofthe lower boundaries of a two-triplon continuum. An-other Cu L -edge study on CaCu O – a weakly coupledspin-ladder system – showed that the spectra could bedecomposed into contributions from the spin-conserving(SC) and non-spin-conserving (NSC) channels [41]. Sub-sequent work at the same edge on the same material fo-cused on spin-orbital fractionalization, but did not carryout an analysis of spinon confinement [42].From a theoretical perspective, studies of the RIXS re-sponse of spin ladders have mainly focused on undopedsystems using a projector method [8] or exact diagonal-ization (ED) of small (4 ×
2) Hubbard clusters [5], andwere restricted to a limited set of rung couplings. To ourknowledge, no systematic RIXS study of the low-energy a r X i v : . [ c ond - m a t . s t r- e l ] A p r excitations of doped and undoped ladders has been car-ried out. Here, we present such a study. Specifically, weuse the Kramers-Heisenberg formalism to compute theRIXS response of undoped and doped two-leg t - J lad-ders while varying the superexchange coupling along therungs over a wide range of values. The RIXS intensity isevaluated numerically, exactly or with a very small error,using ED and the density matrix renormalization group(DMRG) [43–45] methods. Using these tools, we investi-gate the charge and magnetic excitations in both the SCand NSC channels and catalog an assortment of quasipar-ticle and collective excitations. The present systematicstudy can guide future RIXS experiments and help toclassify compounds as being in the weak- or strong-rungcoupling regime, depending on the observed excitations.Iridates provide another group of spin- ( J eff = ) ma-terials that have been studied with RIXS [46–49]. More-over, progress was recently made in engineering quasi-1D iridates in heterostructures [50], establishing anotherplatform for examining and controlling the properties ofquantum spin ladders. The results presented here canserve as a valuable roadmap in these contexts as well.Our organization is as follows: Section II introducesthe spin ladder model and the relevant scattering cross-sections for RIXS within the Kramers-Heisenberg for-malism. Sections III A and III B present results for theRIXS spectra in the NSC and SC channels, respectively.Section III C revisits and discusses the RIXS data re-ported [5] on the spin-ladder compound Sr Cu O inthe context of our results. Section IV summarizes ourfindings. II. METHODSA. Model Hamiltonian
We study the t - J model in a two-leg ladder geometry.The Hamiltonian is H = J rung (cid:88) i (cid:0) S i, · S i, − n i, n i, (cid:1) + J leg (cid:88) i,τ (cid:0) S i,τ · S i +1 ,τ − n i,τ n i +1 ,τ (cid:1) + t rung (cid:88) i,σ (cid:16) c † i, ,σ c i, ,σ + h.c. (cid:17) + t leg (cid:88) i,τ,σ (cid:16) c † i,τ,σ c i +1 ,τ,σ + h.c. (cid:17) . (1)Here, τ = 0 , i = 1 , . . . , L indexes the unit cell along each leg; S i,τ is a spin operator; c i,τ,σ ( c † i,τ,σ ) annihilates (creates) ahole with spin σ (= ↑ , ↓ ) at site ( i, τ ) subject to theconstraint of no double occupancy; J leg ( J rung ) and t leg ( t rung ) are the superexchange and hopping integralsalong the leg (rung) direction of the ladder, respectively;and n i,τ = (cid:80) σ c † i,τ,σ c i,τ,σ is the hole number operator. Material J leg J rung J ring Ref.Sr Cu O
130 72 INS [13].110 140 RIXS [5].110 ±
20 4J leg /5 Raman [51].145 123 This work.La Sr Cu O
186 124 31 INS [23].La Ca Cu O
110 110 16.5 INS [14].CaCu O
134 11 RIXS [42].TABLE I. Different values of the exchange parameters (inunits of meV) reported in the literature for various spin- ladder systems. Entries where the value of J ring is missingcorrespond to studies where the ring exchange terms werenot included in the analysis. Note that we neglected the ring-exchange terms in ourmodel for simplicity.The two-leg spin-ladder model can be used to describeseveral compounds, and a range of model parametershave been reported, as summarized in Table I. The pa-rameters can be different even for the same compounddepending on the nature of the experiment or model usedto analyze the data. Due to the variability in the reportedcouplings, we opted to carry out a systematic studyover a range of rung parameters spanning from weak to strong rung couplings. Unless otherwise stated, weadopt the specific couplings (in units of meV) J leg = 140, J rung = 140 r , t leg = − t rung = − √ r , where r = J rung J leg is a parameter used to adjust the ratio of therung-leg couplings. The choice t rung = − √ r preservesthe relationship J rung ∝ t U , assuming a fixed valuefor the Coulomb interaction ( U ) strength in an effectiveHubbard model. B. RIXS Intensity
We evaluated the RIXS response at the Cu L -edgeof cuprate materials. In a RIXS experiment, photonswith energy ω in and momentum k in ( (cid:126) = 1) scatter in-elastically from a sample, transferring momentum q = k out − k in and energy ω = ω out − ω in to its elementaryexcitations. The RIXS spectrum is evaluated using theKramers-Heisenberg formula [52] and is given by I = (cid:88) f (cid:12)(cid:12)(cid:12)(cid:12) (cid:104) f |D † k out | n (cid:105)(cid:104) n |D k in | g (cid:105) E g + ω in − E n + iΓ (cid:12)(cid:12)(cid:12)(cid:12) δ ( E f − E g + ω ) , (2)where | g (cid:105) , | n (cid:105) , and | f (cid:105) are the ground, intermediate,and final states with energies E g , E n , and E f , re-spectively, and Γ is the core-hole lifetime broadening.The eigenstates are obtained by diagonalizing H + H ch ,where H ch = V c (cid:80) i,τ n i,τ n pi,τ accounts for the interac-tion between the valence and the core holes in the in-termediate state. Here, V c is the inter-orbital repul-sion between the holes in the Cu 2 p and 3 d orbitals, FIG. 1. Schematic diagrams of the elementary magnetic exci-tations that are possible in spin-ladders. Panel (a) shows theground state configuration of the spins with antiferromagneticcorrelations. Panel (b) shows the single spin-flip excitationsthat appear in the non-spin-conserving (NSC) channel. Pan-els (c) and (d) show the double spin-flip processes relevant tothe spin-conserving channel (SC). In each panel, the wigglygrey lines indicate “broken” magnetic bonds. n pi,τ = (cid:80) α p † i,τ,α p i,τ,α , and p † i,τ,α ( p i,τ,α ) creates (annhi-lates) a hole in the J = ( ) core level of site ( i, τ ) for L ( L )-edge. In the two-leg t - J ladder, the dipole oper-ator is given by D k = (cid:80) i,τ,σ,α e i k · R i,τ [ c i,τ,σ p † i,τ,α + h . c . ],where we have neglected the prefactor that depends onthe polarization of the photon and the scattering angle.Due to the large spin-orbit coupling in the core 2 p or-bital, both NSC (∆ S = 1) and SC (∆ S = 0) excitationscan occur in this edge, [35, 41, 43] and the RIXS spectrahas contributions from both of these channels. However,it has been recently shown how the Cu L -edge spectracan be resolved into its individual SC and NSC compo-nents [41]. For this reason, we will consider these twochannels separately in what follows.The momentum transfer has two components in a two-leg ladder geometry: q = ( q x , q y ), where q x = 2 πn/La ,with n ∈ [0 , L ) but q y = 0 or π/a , only. For our ED cal-culations, we evaluate Eq. (2) directly, while the detailsof our DMRG approach are given in Ref. 43.Throughout this study, we use V c = 6 . t , Γ = | t | for all n , and a Lorentzian broadening with η = J leg / L -edge measurements on the cuprates. Most of thespectra were computed using ED on N = L × × N = L × × (cid:104) n (cid:105) = 0 . ω in was tuned to matchthe maximum of the x-ray absorption spectrum, as dis-cussed in Appendix B. III. RESULTS AND DISCUSSION
We calculated the RIXS spectra in various rung cou-pling regimes, ranging from strong ( r = J rung J leg = 4, 2), to isotropic ( r = 1), to weak ( r = 0 . , . , . ladders inthe strong rung coupling limit are well understood start-ing from a dimerized rung basis [22, 53]. For r → ∞ ,the individual rungs of the ladder are decoupled, eachforming a spin dimer. For the antiferromagnetic case, theground state of the L -rung ladder is then a direct productof rung singlets with total spin S = 0. The elementaryexcitations of this state are “triplons” [54–56], where oneor more of the rungs are excited into the triplet man-ifold [57]. For example, the first excited state is L -folddegenerate, where one of the rungs is in a triplet spin con-figuration, while the higher lying excited states involveinteger numbers of rung triplets distributed throughoutthe system. The degeneracy of the excited states is liftedwhen J leg (cid:54) = 0, leading to a dispersive quasiparticle ex-citation. To order O (cid:16) J /J rung (cid:17) , the triplon dispersionis [56] ω t ( q ) = J rung (cid:104) J leg J rung cos( qa ) + 34 (cid:16) J leg J rung (cid:17) (cid:105) , (3)where a is the lattice constant along the leg direction.The two-triplon excitation manifold is even richer.Here, the two-triplon excitations appear in three angu-lar momentum channels corresponding to S = 0 , , singlet , triplet and quintet channels, respec-tively [53, 58, 59]. Previous work [58] showed that a finitevalue of J leg can lead to two-triplon bound states whosedispersions in the singlet ( S = 0) and triplet ( S = 1)channels are to order O (cid:16) J /J (cid:17) given by [53] ω S t ( q ) = J rung (cid:104) − r + r − r − (cid:0) r − r + r (cid:1) × cos( qa ) − (cid:0) r + r (cid:1) cos(2 qa ) − r cos(3 qa ) (cid:105) , (4)and ω T t ( q ) = J rung (cid:104) − r + r + r − (cid:0) r + r − r (cid:1) × cos( qa ) − (cid:0) r + r (cid:1) cos(2 qa ) − r cos(3 qa ) (cid:105) , (5)respectively.The two-triplon excitations in the triplet channel wereanalyzed in a recent INS study [23], reporting a dispersiveexcitation along the q = ( q x ,
0) direction in the Brillouinzone, in very good agreement with the lower boundary line of the two-triplon continuum. It was also argued thatthe four-spin ring-exchange term frustrates the formationof a S = 1 bound state below the continuum. In thiseffort, we neglect the four-spin cyclic exchange, thus wefind that our data overlays well with the dispersion ofthe bound S = 1 two-triplon state. Our results are alsoconsistent with the available RIXS experimental data [5].We know of no experimental study probing two-triplonexcitations in the singlet and quintet channels; however,our results below show that the RIXS SC channel canaccess the two-triplon bound state in the singlet channel.In the weak rung coupling limit for undoped two-legspin- ladders, the system can be viewed as a set ofweakly coupled Heisenberg chains. In this regime, theexcitation spectrum is understood in terms of a confinedspinon continuum with a finite spin gap [60]. Intuitively,the excitations of a Heisenberg chain are spin- spinons,which always appear in pairs and are basically free tomove along a single chain. When the two chains are cou-pled antiferromagnetically, however, the spinons within asingle chain feel an effective confining potential [2]. Thispotential is created by the region of ferromagneticallycoupled spins that forms on the rungs between the twospinons as they separate.When r (cid:28)
1, the two-leg ladder problem can bemapped onto one of weakly interacting singlet and triplet
Majorana fermions with effective masses m s = 3 m and m t = m , respectively, where m ≈ . J rung [61]. Theexcitation spectrum, as encoded in the dynamical spinstructure factor S ( q, ω ) [11], is characterized by combi-nation of sharp modes and a broader continuum arisingfrom several multi-particle Majorana excitations. Thelower boundaries of these excitations are defined by ω l ( q ) ≈ (cid:113) m + v ( q x − q min x ) , (6)where v = πJ leg is the spin velocity of the chain and m thres and q min depend on the particles involved in theexcitation. A summary of the relevant values can befound in Table I of Ref. 11, which we have reproducedin Table II for convenience. From this table, one cansee that a Majorana triplet (1T) excitation appears near q min = ( π/a, π/a ) and m thres = m , while the excita-tions near q min = ( π/a,
0) correspond to a three-particlebound state consisting of two Majorana triplets and aMajorana singlet (2T + 1S) with a threshold set by m thres = 5 m . While the mapping to the Majoranafermion picture holds for r (cid:28)
1, recent DMRG resultsfor S ( q, ω ) [11] have shown that this picture providesa qualitative description of the excitation spectrum fora wide range of r <
1. These same calculations alsoshowed that the spectral weight of the multi-particle con-tinuum increases as r → ω ls ( q ) = π J | sin( qa ) | and ω us ( q ) = πJ | sin( qa/ | , respectively.Understanding the behavior of a small number of holesdoped into an antiferromagnetic background is one of the Excitation q min m thres
1T ( π/a, π/a ) 1 m
2T (0 ,
0) 2 m
3T ( π/a, π/a ) 3 m
1T + 1S (0 , π/a ) 4 m
2T + 1S ( π/a,
0) 5 m TABLE II. The momentum q min and m thres values that definethe lower boundaries of the single- and multiparticle excita-tions that occur in the Majorana fermion description of thespin- ladders. Reproduced from Ref. 11. central problems in the quest to comprehend unconven-tional superconductivity. In this context less is knownabout the excitations in doped spin-ladders as comparedto the undoped case, where the former are usually studiedusing numerical methods [22, 62–66]. A single hole dopedinto a two-leg ladder introduces a spin and charge + e to the system. DMRG results [63] indicate that in thestrong-rung coupling limit, the doped hole behaves as aquasiparticle, where the spin and charge remain tightlybound within a typical distance of about one lattice con-stant. In the isotropic limit, the quasiparticle developsmore internal structure with a length scale of ∼ a . Inthe decoupled case ( J rung = 0), the doped hole fraction-alizes completely into a spinon and holon [36, 67].In the analysis below, we explore the RIXS spectra inboth the NSC and SC channels and identify the relevantelementary excitations in these spectra. A. Results for the non-spin-conserving channel
We begin our study with the NSC or “spin-flip” chan-nel, which typically dominates the Cu L-edge RIXS spec-tra in cuprates [41, 68]. The NSC channel produces localsingle spin-flips due to a large spin-orbit coupling in the2 p core level, as shown in Fig 1(b). The elementary exci-tations generated in this scattering channel correspond tomagnetic excitations with ∆ S = 1 relative to the groundstate. In terms of the spectra, the NSC channel is com-parable to the spin-flip channel of INS, and hence RIXSspectra at the Cu L-edge of cuprates compare well with S ( q , ω ) (see Appendix A) [37, 68].
1. Undoped t - J ladders The RIXS spectra in the NSC channel for undopedladders are plotted in Fig. 2. Panels (a)-(f) and (g)-(l)show results for momentum transfers q = ( q x ,
0) and q =( q x , π/a ), respectively. Several excitations are identified.In the limit of strong rung coupling, the spectra alongthe q = ( q x ,
0) direction [Figs. 2(a) and 2(b)] exhibita dispersive quasiparticle-like excitation. To determineits nature, we overlayed the dispersion ω T t ( q ) given by ω / J r ung
0 31.25 (a) r = 4
0 125 (b) r = 2
0 312.5 (c) r = 1024 0 0.5 1 1.5 2 ω / J l e g (q x , 0) [ π /a]
0 625 (d) r = 0.5 0 0.5 1 1.5 2(q x , 0) [ π /a]
0 625 (e) r = 0.25 0 0.5 1 1.5 2(q x , 0) [ π /a]
0 625 (f) r = 0.1 ω / J r ung
0 1250 (g) r = 4
0 1250 (h) r = 2
0 1250 (i) r = 1024 0 0.5 1 1.5 2 ω / J l e g (q x , 1) [ π /a]
0 1250 (j) r = 0.5 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 1250 (k) r = 0.25 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 1250 (l) r = 0.1
FIG. 2. The RIXS spectra in the non-spin-conserving channel for a half-filled t - J ladder, using ED and a 2 ×
10 cluster. I ∆ S =1 ( q x , , ω ) and I ∆ S =1 ( q x , π/a, ω ) are shown in panels (a)-(f) and (g)-(l), respectively. Panels (a)-(c) and (g)-(i) haveoverlays of the dispersion relationships for the bound triplet two-triplon [Eq. (5)] and the one-triplon excitations [Eq. (3)],respectively, calculated using perturbation theory. Panels (c) and (i) have overlays (solid white) of the dispersion relationshipsfor the same excitations extracted from Ref. 55, which were evaluated using a continuous unitary transformation (CUT)method. The thin dotted and solid white lines in panels (f) and (l) plot the upper and lower boundaries of the two spinoncontinuum expected for completely decoupled chains. The thick white lines in panels (d)-(f) and (j)-(l) plot the lower boundariesof the multiparticle continua near their respective minima. Note that the y -axis of the top and botton rows are scaled withrespect to J rung and J leg , respectively. Eq. (5). We find that the observed excitation closelyfollows the dispersion relationship for r = 4 but for r = 2 there are some deviations, most notably at thezone boundary. (The disagreement becomes even moreapparent for r = 1, as discussed below.) The agreementbetween the dispersion of the excitations and ω T t ( q ), andthe fact that we are in the NSC channel, allows us toconclude that these excitations are the two-triplon boundstate in the triplet channel. Similarily, the q = ( q x , π/a )excitation in the strong-rung coupling case [Figs. 2(g)and 2(h)] corresponds to a single triplon excitation. Toconfirm this, we overlayed the dispersion ω t ( q ) given byEq. (3), showing it captures well the observed excitationsfor r ≥ q x ,
0) [Figs. 2(d)-(f)] and ( q x , π/a )[Figs. 2(j)-(l)] directions, the spectra can be described us-ing the picture of confined spinons with a continuum ofexcitations appearing above a sharper dispersing mode.The lower boundaries of the continua near their respec-tive minima are overlaid as thick white lines. Accordingto Table II, we assign the excitations near q min = (0 ,
0) to2T excitations with m thres = 2 m , while the excitationsnear q min = ( π/a,
0) correspond the 2T + 1S excita-tions with m thres = 5 m . Similarily, the excitations near q min = ( π/a, π/a ) are 1T excitations with m thres = m and the excitations near q min = (0 , π/a ) correspond to1T + 1S excitations with m thres ≈ m . The isotropic coupling case behaves qualitatively likethe strong-rung coupling cases, but the calculated spectradeviate significantly from the dispersion predicted by per-turbation theory [Eq. (5)]. Nevertheless, we are still ableto assign the intense dispersing features to the S = 1 two-triplon bound state and the one-triplon excitations, as inthe strong-rung coupling limit. In Fig. 2(c) and Fig. 2(i)we have overlayed the dispersions for the bound triplet two-triplon and one-triplon excitations, respectively, thistime extracted from Fig. 4 of Ref. 55. In this case, thedispersions were computed using a continuous unitarytransformation (CUT) method, and agree well with ourevaluated spectra. In Fig. 2(c) we also observe additionalspectral weight at higher energies, which corresponds tounbound excitations inside the two-triplon continuum.Our results in this regime should be of considerable in-terest for future RIXS experiments on cuprates spin- ladder materials, as most of the estimated values of the J rung /J leg ratios fall in this intermediate category.
2. Doped t - J ladders The excitations of doped ladder compounds are rel-evant to explain pressure-induced superconductivity.Moreover, while low-energy spin-fluctuations are widelyconsidered pivotal for superconductivity, the relationshipbetween the doping evolution of charge and high-energyspin excitations and the superconducting mechanism hasrecently become the subject of considerable debate, espe- ω / J r ung
0 40 (a) r = 4
0 40 (b) r = 2
0 100 (c) r = 1024 0 0.5 1 1.5 2 ω / J l e g (q x , 0) [ π /a]
0 300 (d) r = 0.5 0 0.5 1 1.5 2(q x , 0) [ π /a]
0 300 (e) r = 0.25 0 0.5 1 1.5 2(q x , 0) [ π /a]
0 300 (f) r = 0.1 ω / J r ung
0 400 (g) r = 4
0 400 (h) r = 2
0 400 (i) r = 1024 0 0.5 1 1.5 2 ω / J l e g (q x , 1) [ π /a]
0 400 (j) r = 0.5 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 400 (k) r = 0.25 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 400 (l) r = 0.1
FIG. 3. RIXS spectra in the non-spin-conserving channel for a doped t - J ladder, using ED on a 2 ×
10 cluster and a filling of (cid:104) n (cid:105) = 0 . I ∆ S =1 ( q x , , ω ) and I ∆ S =1 ( q x , π/a, ω ) are in panels (a)-(f) and (g)-(l), respectively. Panels (a), (b) and (g), (h)have overlays (solid white) of the dispersion relationships for the bound triplet two-triplon [Eq. (5)] and one-triplon excitations[Eq. (3)], respectively, derived using perturbation theory. Panels (a) and (b) have an additional overly (dashed white) of thedispersion for a quasiparticle ω ( k ) = 2˜ t [1 − cos( ka )]. The y -axis of the top and botton rows are scaled with respect to J rung and J leg , respectively. cially in 2D cuprates. Our results for the RIXS spectra ofthe doped spin-ladder in the NSC-channel are in Fig. 3.As in the undoped case, panels (a)-(f) and (g)-(l) showspectra along the ( q x ,
0) and ( q x , π/a ) directions, respec-tively.For strong-rung couplings [Figs. 3(a) and 3(b)], thespectra along the q = ( q x ,
0) directions have two distinctsets of excitations. The first corresponds to the same triplet two-triplon excitations identified in the undopedcase, as confirmed by overlaying the dispersion given byEq. (5) as solid white lines. The second is the Blochquasiparticle excitation formed from the tightly boundspin and charge of the doped hole [63]. Its dispersion iswell described by ω ( k ) = 2˜ t [1 − cos( ka )] (the dashed lineoverlay), where ˜ t = t rung / q = ( q x , π/a ) direction, shown in Figs. 3(g) and 3(h),have only a single set of excitations, whose dispersionsagree well with the one-triplon excitation Eq. (3), whichis again overlaided as a solid white line.Results for the weak-rung coupling regime along the q = ( q x ,
0) and ( q x , π/a ) directions are shown inFigs. 3(d)-(f) and Figs. 3(j)-(l), respectively. We findthat the spectra soften as compared to undoped spin lad-ders. Moreover, the spin gap no longer appears to scalewith J rung but instead appears to vanish at ( π/a,
0) forall r < q = ( π/a, π/a ).In the isotropic case at q = (0 , π/a ), shown in Fig. 3(c),the brightest dispersing peak does not have the samedownturn in the two-triplon dispersion that was observed in the undoped case. Instead, there is an increased weightappearing in at higher energy losses, corresponding tothe two-triplon continuum. In contrast, the excitationsat q = ( π/a, π/a ), shown in Fig. 3(i), have the ubiqui-tous incommensurate peaks that are also commonly ob-served in doped ladders and 2D cuprates [66, 70–72]. Itis interesting to contrast the results for the doped two-leg spin ladder found above with available results in thedoped 2D cuprates at the Cu L-edge. In 2D cuprates,a weakly dispersive high-energy paramagnon band alongthe q = ( q x ,
0) line was reported to be persistent uponhole doping [26–29, 37]. This type of excitation comparesrelatively well with our results in the two-leg ladder casein Fig. 3(c) for the isotropic case.
B. Results for the spin-conserving channel
We now analyze the RIXS spectra in the SC channel,both for the undoped and doped cases. As shown pic-torially in Figs. 1(c) and 1(d), the magnetic excitationsthat are accessible in this channel occur via double spin-flip processes, which correspond to ∆ S = 0 excitations inthe antiferromagnetic ladders. For the undoped cupratesmeasured at the Cu L -edge, the SC channel probes exci-tations encoded in the dynamical exchange structure fac-tor S exch ( q , ω ) (see Appendix A), which is a second orderterm in the ultrashort core-hole lifetime (UCL) expan-sion [34, 68]. Because these are higher order processes,this channel is expected to be weaker as compared to theNSC channel, [41, 68] and our results are consistent withthis expectation. In the doped case, magnetic and chargeexcitations coexist in the RIXS spectra and the SC chan- ω / J r ung
0 0.8 (a) r = 4
0 2 (b) r = 2
0 5 (c) r = 1024 0 0.5 1 1.5 2 ω / J l e g (q x , 0) [ π /a]
0 10 (d) r = 0.5 0 0.5 1 1.5 2(q x , 0) [ π /a]
0 12 (e) r = 0.25 0 0.5 1 1.5 2(q x , 0) [ π /a]
0 15 (f) r = 0.1 ω / J r ung
0 0.1 (g) r = 4
0 2 (h) r = 2
0 10 (i) r = 1024 0 0.5 1 1.5 2 ω / J l e g (q x , 1) [ π /a]
0 20 (j) r = 0.5 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 20 (k) r = 0.25 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 20 (l) r = 0.1
FIG. 4. RIXS spectra in the spin-conserving channel for the half-filled t - J ladder, using ED and a 2 ×
10 cluster. I ∆ S =0 ( q x , , ω )and I ∆ S =0 ( q x , π/a, ω ) are in panels (a)-(f) and (g)-(l), respectively. Panels (a) and (b) have overlays (solid white) of thedispersion relations of the singlet bound two-triplon excitations [Eq. (4)] derived using perturbation theory, while panel (c)has an overlay of the dispersion relation for the same excitation from Ref. 55, using the CUT method. The thin dotted andsolid white lines in panels (f) and (l) plot the upper and lower boundaries of the two spinon continuum expected for completelydecoupled chains. The thick white lines in panels (d)-(f) and (j)-(l) plot the lower boundaries of the multiparticle continuanear their respective minima. The y -axis of the top and botton rows are plotted in units of J rung and J leg , respectively. nel also has a significant contribution at second ordergiven by a modified charge structure factor ˜ N ( q , ω ) [seeEq. (9) of Ref. 68 and Appendix A]. The SC channel isalso particularly relevant at the O and Cu K-edges, wheredirect spin-flip excitations are often forbidden [4, 25, 36].Our numerical study motivates RIXS experiments thatcould be able to disentangle SC and NSC components ofthe spectra by the use of photon polarization, which hasbeen successfully demonstrated in Ref. 41 for the weaklycoupled ladder cuprate CaCu O .
1. Undoped t - J ladders The RIXS spectra in the SC channel for the undopedladders are shown in Fig. 4. Panels (a)-(f) and (g)-(l)show the RIXS spectra for momentum transfers alongthe q = ( q x ,
0) and ( q x , π/a ) directions, respectively. Asexpected, the intensity of the excitations in this channelis weaker as compared to the NSC channel by approxi-mately one order of magnitude.As with the previous sections, we first consider thestrong-rung coupling limit. Along the ( q x ,
0) direction[Figs. 4(a) and 4(b)], we observe a weakly dispersing fea-ture that agrees well with the two-triplon bound state inthe singlet channel given by Eq. (4). This is one of theimportant results of our current investigation: becausethe SC channel probes ∆ S = 0 excitations, we are ableto clearly identify and distinguish the two-triplon boundstates in both the singlet and triplet channels. We alsosee additional spectral weight at higher binding energiesnear ( π/a, q = ( q x , π/a ) direction in the strong-rungcoupling regime shown in panels (g) and (h), the spectracan again be understood in terms of singlet two-triplonexcitations, which we observe at energy losses around3 J rung . We observe zero spectral weight at ( π/a, π/a ) forall the rung couplings we investigated, in contrast to theweak spectral weight observed at ( π/a, S exch ( q , ω ), whichhas been computed and shown in Fig. 9 of Appendix A.Indeed, S exch ( q , ω ) shows the spectral weight cancella-tion at ( π/a, π/a ) for all the rung couplings investigated.In the weak rung regime ( r < q = ( q x ,
0) and ( q x , π/a ), shown in Figs. 4(d)-(f) and 4(j)-(l), respectively, resemble the continuum expected forconfined spinons also observed in the NSC channel. Tohighlight this, we overlaid the boundaries of the two-spinon continuum as well as the lower boundaries of themulti-particle continua that were introduced when de-scribing the NSC channel. In this case, all of the excita-tions appear above the lower boundary lines, indicatingthat these excitations are multiparticle in nature. Wealso note that spectra along both momentum directionshave a suppressed intensity at q x = π/a , which is simi-lar to what occurs in one-dimensional antiferromagneticchains when probed in the SC channel [35, 73, 74].The spectra along both momentum directions for theisotropic case ( r = 1) behaves qualitatively similar to the ω /t l e g
0 32 (a) r = 4
0 32 (b) r = 2
0 32 (c) r = 1 ω /t l e g (q x , 0) [ π /a]
0 32 (d) r = 0.5
0 0.5 1 1.5 2(q x , 0) [ π /a]
0 32 (e) r = 0.25
0 0.5 1 1.5 2(q x , 0) [ π /a]
0 32 (f) r = 0.1 ω /t l e g
0 9 (g) r = 4
0 9 (h) r = 2
0 9 (i) r = 102468 0 0.5 1 1.5 2 ω /t l e g (q x , 1) [ π /a]
0 30 (j) r = 0.5 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 30 (k) r = 0.25 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 30 (l) r = 0.1
FIG. 5. RIXS spectra in the spin-conserving channel for the doped t - J ladder, using ED, a 2 ×
10 cluster, and n = 0 . I ∆ S =0 ( q x , , ω ) and I ∆ S =0 ( q x , π/a, ω ) are in panels (a)-(f) and (g)-(l), respectively. In panels (a)-(c) the dispersions of thequasiparticle state with a bandwidth W = 2 t rung are shown (white dashed). In panels (d)-(f) and (j)-(l) the holon dispersionwith a bandwidth W = 4 t leg is also shown (solid white). The white dashed overlay in panels (g)-(i) plot the boundary of thespinon-holon continuum gapped by 2 t rung , while the dotted overlay in panels (g) and (h) corresponds to the dispersion relationof the one-triplon excitations. The y -axis of the top row and botton rows are plotted in units of t leg . strong-rung coupling case, where we observe a continuumof excitations. In Fig. 4c, we plot an overlay extractedfrom Fig. 4(b) of Ref. 55 for bound two-triplon exci-tation in the singlet channel, again evaluated using theCUT method. This dispersion agrees well with the lowerboundary of the evaluated spectra suggesting that thecontinuum of excitations is related to singlet two-triplonexcitations and that the bound singlet state is not farremoved from the continuum.
2. Doped t - J ladders Finally, we examine the RIXS spectra of the doped lad-ders in the SC channel. As before, Figs. 5(a)-(f) and 5(g)-(l) show results for momentum transfers q = ( q x ,
0) and( q x , π/a ), respectively. The spectra are quite rich and weobserve several new excitations that were not present inthe undoped ladders. Indeed, we expect that magneticand charge excitations coexist in the SC channel responseand that most of the spectral features observed in the fullRIXS response can be described using the modified dy-namical charge correlation function ˜ N ( q , ω ) (see Fig. 10in Appendix A and Ref. 68).As before, we begin our discussion from the strong-rung coupling limit. In this case, the ladder can be con-sidered as composed of weakly decoupled dimers wherethe orbitals on each leg form bonding and antibondingstates. If t leg is finite, these bonding (-) and antibonding(+) states form the basis for Bloch states with dispersionrelations given by ω ( q ) = ∓ t rung + 2˜ t [1 − cos( qa )] [62],where ˜ t is the effective hopping parameter obtained fromthe change of basis to the bonding and antibonding states. When a small number of holes are doped into thesystem, they first occupy the bonding band as quasiparti-cles, as shown in Fig. 6. The charge excitations observedin this channel can then be understood by invoking quasi-particle scattering within and between the bonding andantibonding bands, respectively.Along the ( q x ,
0) direction in the strong-rung couplingregime, Figs. 5(a) and 5(b), we observe dispersive chargeexcitations consistent with particle-hole scattering withinthe bonding band, as shown in Fig. 6. To confirm this,we overlaid the dispersion ω ( q ) = 2˜ t [1 − cos( qa )], where˜ t ≈ t leg /
2, which agrees with the numerical data [62].Along the ( q x , π/a ) direction [Figs. 5(g) and 5(h)] we findthe corresponding particle-hole excitation where scatter-ing occurs into the antibonding band. In this case, weoverlaid the dispersion ω ( q ) = 2 t rung + 2˜ t [1 − cos( qa )](white dashed line). We also notice that the bondingand anti-bonding bands are separated by 2 t rung , whichaccounts for the shift in cosine-like dispersion observedwhen q y = π/a .It is important to note that, even in this case, thesecharge excitations are weaker in intensity when comparedto the magnetic excitations in NSC channel by approxi-mately one order of magnitude, but stronger than the SCchannel of the undoped case. Our results show that RIXScan explicitly probe charge excitations at low energies assuggested in the literature for the Cu L -edge [75, 76].In addition to the charge excitations, we also observe acontinuum of magnetic excitations for momentum trans-fers along ( q x , π/a ). The lower boundary of this contin-uum is defined by the one triplon dispersion given byEq. (3), which has been overlaid as a dotted white line.In the weak-rung coupling regime, shown in Figs. 6(d)- FIG. 6. A schematic diagram of the particle-hole excitationspossible in the doped t - J ladder in the weak-rung couplingregime. In this limit, the orbitals along the legs form bonding √ ( c † i,σ, − c † i,σ, ) | (cid:105) and anti-bonding √ ( c † i,σ, + c † i,σ, ) | (cid:105) states, which form the basis for the Bloch states propagatingalong the leg direction. In this case, the dispersions of thetwo Bloch states are split by an amount proportional to therung hopping ± t rung . The doped holes form quasiparticlescarrying both spin- and charge e , and the broad blue colourhighlights the filled states in the ground state. RIXS probesscattering within the bonding and to the antibonding band,as shown by the arrows. (f) and 6(j)-(l), the quasiparticle excitations display abandwidth 4 t leg with dispersion ω ( q ) = 2 t leg [1 − cos( qa )]along both the ( q x ,
0) and ( q x , π/a ) directions. This issimilar to the results for the 1D AFM chain reported inRef. 36 where the excitations are holons, and consistentwith the notion that the individual legs of the ladder areweakly coupled. The fact that these observed excitationsare completely governed by t leg indicates that the holesoccupy the chains of the ladder rather than the bondingand antibonding orbitals on each rung.Finally, we consider the isotropic rung-coupling limit,which is of much interest for future RIXS experimentsand future theoretical investigations. In this case, ourspectra show a gapless and a gapped continuum alongthe ( q x ,
0) and ( q x , π/a ) directions, respectively. Theseresults hence resemble qualitatively the spectral featuresalso observed in the strong-rung coupling limit. It is in-teresting to compare our full RIXS spectra with the dy-namical charge structure factor results reported in Fig. 7of Ref. 20: our results compare well with the lower Hub-bard band excitation observed in which N ( q x , π/a, ω ) isgapped, in contrast to the gapless N ( q x , , ω ). C. Revisiting Sr Cu O RIXS data Cu L -edge RIXS data has been reported [5] for theprototypical spin-ladder Sr Cu O . At that time, theobserved spectra were interpreted in terms of two-triplon∆ S = 0 excitations in the strong rung coupling regime( r ≈ .
37) as it was believed that ∆ S = 1 excita- ω [ e V ] (q x , 0) [ π /a] FIG. 7. Calculated RIXS spectra of an undoped t - J ladderin the non-spin-conserving channel I ∆ S =1 ( q x , , ω ) evaluatedusing DMRG on an N = 16 × -edge data for Sr Cu O , reproducedfrom Ref. 5. tions were forbidden at the Cu L -edge. It was latershown that not only is this channel allowed but thatit dominates the magnetic RIXS response in the un-doped cuprates [31]. The RIXS spectra were later the-oretically evaluated [8] employing a projection methodfor the two-leg ladder using the parameter set derivedfrom La Sr Cu O in Ref. 23 (the model involvedadditional ring spin-exchange term as compared to ourmodel). These calculations showed that the RIXS spec-tra was associated with the two-triplon excitations with∆ S = 1 when momentum transfers of q y = 0 were used,as shown in Fig. 6 of Ref. 8. But there was also a sig-nificant difference in the dispersion of the experimentaland theoretical data. Armed now with these theoreticalinsights and our new model calculations, we revisit theexisting Sr Cu O data.To access large system sizes with improved momen-tum resolution, we computed the RIXS spectra in the∆ S = 1 channel for an undoped spin-ladder using ourrecently formulated DMRG approach [43]. To obtain aunified description of RIXS and INS experiments, we firstadopted a model given by Eq. (1) without ring exchangeterms and parameters from Ref. [13], however, we foundthat this model gave poor agreement with the experi-mentally observed RIXS spectra. Instead, we are ableto find good agreement when we set J leg = 145 meVand J rung = 0 . J leg . This result places Sr Cu O inthe weak rung-coupling regime but close to the isotropiclimit. The resulting RIXS spectra shown in Fig. 7 agreewell with the experimental data, but the theoreticalmodel predicts a vanishing spectral weight at q x = 0.This observation is consistent with the model of Ref. 8but inconsistent with the finite intensity observed in theexperiment. At this time, the source of this discrepancyis unclear.Because r is close to the isotropic case, the data can0be qualitatively understood using either the dimer ex-citation picture or the confined spinon picture. In theformer case, the excitations are understood as bound S = 1 two-triplet excitations. In the latter case, theyare viewed as a three particle bound state composed ofMajorana fermions near the zone boundary. However, aquantitative description of the data can only be achievedwith nonperturbative numerical methods. In this sense,our results place Sr Cu O in a regime similar tothe organometallic compound (C H N) CuBr [11], butwith larger exchange couplings. IV. CONCLUSIONS
We have systematically studied the RIXS spectra ofboth undoped and doped spin- ladders, covering theweak- to strong-rung coupling regimes. Our study showsthat RIXS experiments performed on these compoundscan access a wealth of magnetic and charge excitations.This study was motivated by RIXS experiments at theCu L-edge in low-dimensional cuprates, where the RIXSdata can be decomposed into the non-spin-conserving(NSC) and spin conserving (SC) channels [25, 35, 41, 68].Therefore, we evaluated the RIXS spectra in both ofthese channels and provided an energy-momentum re-solved roadmap that can guide future RIXS experimentson spin-ladder compounds.In the first part of our effort, we reported the RIXSexcitations in the NSC or “spin-flip” channel, which typ-ically dominates the Cu L-edge RIXS spectra in thecuprates [41, 68]. In the undoped two-leg ladder, wehave shown that RIXS can access dispersive one triplonexcitations and a two-triplon bound state in the triplet ( S = 1) channel in the intermediate to strong rung cou-pling regime. In the weak-rung coupling regime, the NSCchannel probes single- and multiparticle excitations con-sistent with the Majorana fermion description of confinedspinons.The study of the RIXS spectra for doped spin- lad-der compounds is of much importance in the contextof pressure-induced superconductivity in low-dimensionalhigh-T c cuprates. In the doped ladder, we accessed one-and (triplet) two-triplon excitations in the strong-rungcoupling limit and softened confined spinons in the weak-rung coupling limit. We also identified signatures of abound spin-charge quasiparticle excitation in the strong-rung coupling limit.In the second part of our work, we studied the RIXSspectra of the spin-ladder in the SC channel, whichprobes ∆ S = 0 excitations of the system [41, 68]. Thiscomponent of the RIXS spectra has received less atten-tion in the literature, and our work provides a startingpoint for future theoretical and experimental investiga-tions of this channel on spin-ladders. In the undopedladders, magnetic excitations are created in this channelvia double spin-flip processes. Because these are higherorder processes, their contribution to the RIXS spectra is expected to be weaker compared to the NSC chan-nel [41, 68]. Our results are consistent with this expecta-tion, and we found that the spectral intensity is at leastone order of magnitude smaller than the correspondingspectra in the SC channel. Nevertheless, in the interme-diate to strong rung coupling regime, we are able to iden-tify bound two-triplon excitations in the singlet ( S = 0)channel.In the SC RIXS channel for doped spin-ladders, weidentified a set of dispersive low-energy charge excitationsthat were interpreted by invoking quasiparticle scatter-ing within and between the bonding and antibondingbands, respectively, in the strong rung coupling case.Conversely, the spectra are dominated by holon excita-tions in the weak-rung coupling limit. The direct accessto charge excitations offered by this channel provides anew opportunity to study superconductivity in cuprateladders, where the role of spin and charge excitationsis still debated. We believe our numerical study mo-tivates new RIXS experiments on spin-ladders such asSr Cu O allowing disentanglement of data into theNSC and SC channels, due to the richness predicted inthe RIXS spectra.Finally, we revisited the available RIXS data forSr Cu O and found that it was best described usinga model in the weak rung coupling regime with r = 0 . ACKNOWLEDGMENTS
We thank T. Schmitt and J. Schlappa for useful discus-sions. A. N. and E. D. are supported by the U.S. Depart-ment of Energy, Office of Science, Basic Energy Sciences,Materials Sciences and Engineering Division. S. J. is sup-ported by the National Science Foundation under GrantNo. DMR-1842056. This work used computational re-sources supported by the University of Tennessee andOak Ridge National Laboratory Joint Institute for Com-putational Sciences.
Appendix A: Results for the dynamical correlationfunctions
The interpretation of RIXS spectra computed with theKramers-Heisenberg formalism can be difficult. To sim-plify matters, the full RIXS intensity can be expandedin powers of J/ Γ using the ultra-short core-hole life-time (UCL) approximation. This procedure expressesthe RIXS intensity as a series of increasingly complicatedmulti-particle correlation functions, which can then befurther subdivided into correlation functions of the NSCand SC channels. The detailed procedure to be followedhas been reported in several prior studies [68, 73, 77, 78].Here, we evaluate some effective correlation functions1
024 0 0.5 1 1.5 2 ω / J r ung (q x , 0) [ π /a]
0 31.25 (a) r = 4 0 0.5 1 1.5 2(q x , 0) [ π /a]
0 312.5 (b) r = 1 024 0 0.5 1 1.5 2 ω / J l e g (q x , 0) [ π /a]
0 625 (c) r = 0.1024 0 0.5 1 1.5 2 ω / J r ung (q x , 1) [ π /a]
0 1250 (d) r = 4 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 1250 (e) r = 1 024 0 0.5 1 1.5 2 ω / J l e g (q x , 1) [ π /a]
0 1250 (f) r = 0.1
FIG. 8. The dynamical spin correlation function S ( q , ω ) eval-uated for the undoped ladder with r = J rung J leg = { , , . } rung couplings, using ED and a 2 ×
10 cluster. Panels (a)-(c) and (d)-(f) show the spectra along the q = ( q x ,
0) and q = ( q x , π/a ) directions, respectively. The overall intensityhas been rescaled by a factor 1 / Γ , which corresponds to theprefactor relating S ( q , ω ) to the RIXS intensity. These plotscapture all of the features of the spectra presented in Fig. 2. motivated by Eqs. (B1) and (B2) of Ref. 68. In manycases, these simplified correlation functions give an accu-rate description of the RIXS intensity.The spectral weight of the NSC channel is dominatedby the first-order term in the UCL expansion, which isequivalent to the dynamical spin structure factor S ( q , ω ) = 1 L (cid:88) f (cid:12)(cid:12)(cid:12) (cid:104) f | (cid:88) i,τ e i q · R i,τ S αi,τ | g (cid:105) (cid:12)(cid:12)(cid:12) δ ( E f − E g + ω ) . (A1)Here, S αi ( α = {± , z } ) is a component of the spin op-erator at site ( i, τ ). The S ( q , ω ) responses for an un-doped ladder along the ( q x ,
0) and ( q x , π/a ) directionsare plotted in Figs. 8(a)-(c) and 8(c)-(f), respectively.These results compare well with the RIXS intensity com-puted within the Kramers-Heisenberg formalism shownin Figs. 2 for all values of the rung coupling.To account for the magnetic excitations of the SC chan-nel, the second-order term in the UCL expansion of theKramers-Heisenberg formula is needed [see Eq. (B2) ofRef. 68]. In the undoped case, the first-order term onlycontributes to the elastic line in this channel and a dou-ble spin-flip process appearing at second-order generatesmagnetic excitations. The RIXS spectra in the SC chan-nel of the undoped ladders is hence dominated by thedynamical spin-exchange structure factor [68, 73, 74] S exch ( q , ω ) = 1 L (cid:88) f (cid:12)(cid:12)(cid:12) (cid:104) f | (cid:88) i,τ e i q · R i,τ O exch i,τ | g (cid:105) (cid:12)(cid:12)(cid:12) × δ ( E f − E g + ω ) . (A2)
024 0 0.5 1 1.5 2 ω / J r ung (q x , 0) [ π /a]
0 0.8 (a) r = 4 0 0.5 1 1.5 2(q x , 0) [ π /a]
0 5 (b) r = 1 024 0 0.5 1 1.5 2 ω / J l e g (q x , 0) [ π /a]
0 15 (c) r = 0.1024 0 0.5 1 1.5 2 ω / J r ung (q x , 1) [ π /a]
0 0.1 (d) r = 4 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 10 (e) r = 1 024 0 0.5 1 1.5 2 ω / J l e g (q x , 1) [ π /a]
0 20 (f) r = 0.1
FIG. 9. The dynamical spin exchange correlation function S exch ( q , ω ) evaluated for the undoped ladder with r = J rung J leg = { , , . } rung couplings, using ED and a 2 ×
10 cluster. Pan-els (a)-(c) and (d)-(f) show the spectra along the q = ( q x , q = ( q x , π/a ) directions, respectively. The overall inten-sity was rescaled by a factor of 1 / Γ , which corresponds tothe prefactor relating S exch ( q , ω ) to the RIXS intensity. Theseplots capture many of the features of the spectra Fig. 4. Here, O exch i,τ = S i,τ · [ J leg ( S i +1 ,τ + S i − ,τ ) + J rung S i, ¯ τ ] / S exch ( q , ω ) in the undoped ladderalong the ( q x ,
0) and ( q x , π/a ) directions are shown inFigs. 9(a)-(c) and 9(d)-(f), respectively. These resultscompare well with the RIXS intensities shown in Fig. 4evaluted using the full Kramers-Heisenberg formalism. S exch ( q , ω ) captures the correct excitations for weak-rung couplings, but the intensities are overpredicted forstrong-rung couplings, the regime where the UCL ap-proximation is expected to fail (large J/ Γ).The SC channel for the doped ladders is dominatedby charge excitations. In this case, one must also go tosecond order and the RIXS intensity is well approximatedby a modified dynamical charge structure factor [68]˜ N ( q, ω ) = 1 L (cid:16) (cid:88) f (cid:12)(cid:12)(cid:12) (cid:104) f | (cid:88) i,τ e i q · R i,τ O i,τ | g (cid:105) (cid:12)(cid:12)(cid:12) + 1Γ (cid:12)(cid:12)(cid:12) (cid:104) f | (cid:88) i,τ e i q · R i,τ O i,τ | g (cid:105) (cid:12)(cid:12)(cid:12) (cid:17) × δ ( E f − E g − ω ) . (A3)Here, O i,τ = (cid:80) σ c † i,τ,σ c i,τ,σ , O i,τ = (cid:80) σ [ t leg ( c † i +1 ,τ,σ + c † i − ,τ,σ ) + t rung c † i, ¯ τ,σ ] c i,τ,σ /
2, and c i,τ,σ annihilates a spin σ hole at site ( i, τ ), subject to the constraint of no dou-ble occupancy. Results for ˜ N ( q , ω ) along the ( q x ,
0) and( q x , π/a ) directions are shown in panels Fig. 10(a)-(c) andFig. 10(d)-(f), respectively. In the doped case, the first-order term is non-zero but the majority of the intensity is2 ω /t l e g (q x , 0) [ π /a]
0 32 (a) r = 4 0 0.5 1 1.5 2(q x , 0) [ π /a]
0 32 (b) r = 1 02468 0 0.5 1 1.5 2 ω /t l e g (q x , 0) [ π /a]
0 32 (c) r = 102468 0 0.5 1 1.5 2 ω /t l e g (q x , 1) [ π /a]
0 9 (d) r = 4 0 0.5 1 1.5 2(q x , 1) [ π /a]
0 9 (e) r = 1 02468 0 0.5 1 1.5 2 ω /t l e g (q x , 1) [ π /a]
0 30 (f) r = 0.1
FIG. 10. The modified dynamical charge structure factor˜ N ( q , ω ) evaluated for the doped ladder with r = J rung J leg = { , , . } rung couplings, using ED, a 2 ×
10 cluster, and n = 0 .
9. Panels (a)-(c) and (d)-(f) show the spectra alongthe q = ( q x ,
0) and q = ( q x , π/a ) directions, respectively.The overall intensity has been rescaled by a factor 1 / Γ , corre-sponding to the prefactor relating ˜ N ( q , ω ) to the RIXS inten-sity. These plots capture many of the features of the spectrapresented in Fig. 5. set by the second-order term. The ˜ N ( q , ω ) results com-pare reasonably well to the RIXS spectra shown in Fig. 5evaluated using the full Kramers-Heisenberg formalism. Appendix B: X-ray absorption -2 0 2 4 I ( a r b . un it s ) ω in /J leg r = 0.10r = 0.25r = 0.50r = 1.00r = 2.00r = 4.00 -2 0 2 4 I ( a r b . un it s ) ω in /J leg r = 0.10r = 0.25r = 0.50r = 1.00r = 2.00r = 4.00 FIG. 11. XAS spectra for the undoped (left panel; n = 1 . n = 0 .
9) ladders for various valuesof r = J rung J leg , using ED and a 2 ×
10 cluster. Increasing therung coupling factor r , the peak position of the XAS shifts tohigher incident photon energies. For all the RIXS figures discussed in the main text, wetuned the incident photon energy to the peak positionobserved in the x-ray absorption (XAS) spectra given by I ( ω in ) = (cid:88) n (cid:12)(cid:12) (cid:104) n | D k =0 | g (cid:105) (cid:12)(cid:12) δ ( E n − E g − ω in ) . (B1)Figure 11 shows the XAS spectra for the undopedand doped cases as a function of the rung coupling r = J rung /J leg . For the undoped two-leg ladder, the res-onance peak in the XAS spectra shifts to larger valuesof ω in with increasing rung coupling. In the strong-rungcase ( r > r while new peaks appear on the high-energyside of the resonance. [1] E. Dagotto and T. M. Rice, Science , 618 (1996).[2] B. Lake, A. M. Tsvelik, S. Notbohm, D. Alan Tennant,T. G. Perring, M. Reehuis, C. Sekar, G. Krabbes, andB. B¨uchner, Nature Physics , 50 (2009).[3] J. Schlappa, K. Wohlfeld, K. J. Zhou, M. Mourigal,M. W. Haverkort, V. N. Strocov, L. Hozoi, C. Mon-ney, S. Nishimoto, S. Singh, A. Revcolevschi, J.-S. Caux,L. Patthey, H. M. Rønnow, J. van den Brink, andT. Schmitt, Nature , 82 (2012).[4] J. Schlappa, U. Kumar, K. J. Zhou, S. Singh, M. Mouri-gal, V. N. Strocov, A. Revcolevschi, L. Patthey, H. M.Rønnow, S. Johnston, and T. Schmitt, Nature Commu-nications , 5394 (2018).[5] J. Schlappa, T. Schmitt, F. Vernay, V. N. Strocov,V. Ilakovac, B. Thielemann, H. M. Rønnow, S. Vanishri,A. Piazzalunga, X. Wang, L. Braicovich, G. Ghiringhelli,C. Marin, J. Mesot, B. Delley, and L. Patthey, Phys.Rev. Lett. , 047401 (2009).[6] M. Mourigal, M. Enderle, A. Kl¨opperpieper, J.-S. Caux,A. Stunault, and H. M. Rønnow, Nature Physics , 435 (2013).[7] S. R. White and D. J. Scalapino, Phys. Rev. Lett. ,136403 (2003).[8] T. Nagao and J.-i. Igarashi, Phys. Rev. B , 224436(2012).[9] D. Schmidiger, P. Bouillot, S. M¨uhlbauer, S. Gvasaliya,C. Kollath, T. Giamarchi, and A. Zheludev, Phys. Rev.Lett. , 167201 (2012).[10] M. Jeong, H. Mayaffre, C. Berthier, D. Schmidiger,A. Zheludev, and M. Horvati´c, Phys. Rev. Lett. ,167206 (2017).[11] D. Schmidiger, S. M¨uhlbauer, A. Zheludev, P. Bouillot,T. Giamarchi, C. Kollath, G. Ehlers, and A. M. Tsvelik,Phys. Rev. B , 094411 (2013).[12] B. Thielemann, C. R¨uegg, H. M. Rønnow, A. M. L¨auchli,J.-S. Caux, B. Normand, D. Biner, K. W. Kr¨amer, H.-U.G¨udel, J. Stahn, K. Habicht, K. Kiefer, M. Boehm, D. F.McMorrow, and J. Mesot, Phys. Rev. Lett. , 107204(2009).[13] R. S. Eccleston, M. Uehara, J. Akimitsu, H. Eisaki, N. Motoyama, and S.-i. Uchida, Phys. Rev. Lett. ,1702 (1998).[14] M. Matsuda, K. Katsumata, R. S. Eccleston, S. Brehmer,and H.-J. Mikeska, Phys. Rev. B , 8903 (2000).[15] J. E. Lorenzo, L. P. Regnault, C. Boullier, N. Mar-tin, A. H. Moudden, S. Vanishri, C. Marin, andA. Revcolevschi, Phys. Rev. Lett. , 097202 (2010).[16] G. Deng, N. Tsyrulin, P. Bourges, D. Lamago, H. Ron-now, M. Kenzelmann, S. Danilkin, E. Pomjakushina,and K. Conder, Phys. Rev. B , 014504 (2013).[17] A. T. Savici, G. E. Granroth, C. L. Broholm, D. M. Pa-jerowski, C. M. Brown, D. R. Talham, M. W. Meisel,K. P. Schmidt, G. S. Uhrig, and S. E. Nagler, Phys.Rev. B , 094411 (2009).[18] J. E. Lorenzo, L. P. Regnault, C. Boullier, N. Martin,S. Vanishri, and C. Marin, Phys. Rev. B , 140413(2011).[19] T. Hong, K. P. Schmidt, K. Coester, F. F. Awwadi, M. M.Turnbull, Y. Qiu, J. A. Rodriguez-Rivera, M. Zhu, X. Ke,C. P. Aoyama, Y. Takano, H. Cao, W. Tian, J. Ma,R. Custelcean, H. D. Zhou, and M. Matsuda, Phys. Rev.B , 174432 (2014).[20] A. Nocera, Y. Wang, N. D. Patel, G. Alvarez, T. A.Maier, E. Dagotto, and S. Johnston, Phys. Rev. B ,195156 (2018).[21] M. Uehara, T. Nagata, J. Akimitsu, H. Takahashi,N. Mri, and K. Kinoshita, Journal of the Physical Soci-ety of Japan , 2764 (1996).[22] E. Dagotto, J. Riera, and D. Scalapino, Phys. Rev. B , 5744 (1992).[23] S. Notbohm, P. Ribeiro, B. Lake, D. A. Tennant, K. P.Schmidt, G. S. Uhrig, C. Hess, R. Klingeler, G. Behr,B. B¨uchner, M. Reehuis, R. I. Bewley, C. D. Frost,P. Manuel, and R. S. Eccleston, Phys. Rev. Lett. ,027403 (2007).[24] L. Braicovich, J. van den Brink, V. Bisogni, M. M. Sala,L. J. P. Ament, N. B. Brookes, G. M. De Luca, M. Sal-luzzo, T. Schmitt, V. N. Strocov, and G. Ghiringhelli,Phys. Rev. Lett. , 077002 (2010).[25] V. Bisogni, L. Simonelli, L. J. P. Ament, F. Forte,M. Moretti Sala, M. Minola, S. Huotari, J. van den Brink,G. Ghiringhelli, N. B. Brookes, and L. Braicovich, Phys.Rev. B , 214527 (2012).[26] M. Le Tacon, M. Minola, D. C. Peets, M. Moretti Sala,S. Blanco-Canosa, V. Hinkov, R. Liang, D. A. Bonn,W. N. Hardy, C. T. Lin, T. Schmitt, L. Braicovich,G. Ghiringhelli, and B. Keimer, Phys. Rev. B , 020501(2013).[27] M. Minola, G. Dellea, H. Gretarsson, Y. Y. Peng,Y. Lu, J. Porras, T. Loew, F. Yakhou, N. B. Brookes,Y. B. Huang, J. Pelliciari, T. Schmitt, G. Ghiringhelli,B. Keimer, L. Braicovich, and M. Le Tacon, Phys. Rev.Lett. , 217003 (2015).[28] H. Y. Huang, C. J. Jia, Z. Y. Chen, K. Wohlfeld,B. Moritz, T. P. Devereaux, W. B. Wu, J. Okamoto,W. S. Lee, M. Hashimoto, Y. He, Z. X. Shen, Y. Yoshida,H. Eisaki, C. Y. Mou, C. T. Chen, and D. J. Huang, Sci-entific Reports , 19657 (2016).[29] M. Le Tacon, G. Ghiringhelli, J. Chaloupka, M. M. Sala,V. Hinkov, M. W. Haverkort, M. Minola, M. Bakr, K. J.Zhou, S. Blanco-Canosa, C. Monney, Y. T. Song, G. L.Sun, C. T. Lin, G. M. De Luca, M. Salluzzo, G. Khali-ullin, T. Schmitt, L. Braicovich, and B. Keimer, NaturePhysics , 725 (2011). [30] W. S. Lee, J. J. Lee, E. A. Nowadnick, S. Gerber,W. Tabis, S. W. Huang, V. N. Strocov, E. M. Motoyama,G. Yu, B. Moritz, H. Y. Huang, R. P. Wang, Y. B.Huang, W. B. Wu, C. T. Chen, D. J. Huang, M. Greven,T. Schmitt, Z. X. Shen, and T. P. Devereaux, NaturePhysics , 883 (2014).[31] L. J. P. Ament, G. Ghiringhelli, M. M. Sala,L. Braicovich, and J. van den Brink, Phys. Rev. Lett. , 117003 (2009).[32] X. Lu, P. Olalde-Velasco, Y. Huang, V. Bisogni, J. Pel-liciari, S. Fatale, M. Dantz, J. G. Vale, E. C. Hunter,J. Chang, V. N. Strocov, R. S. Perry, M. Grioni, D. F.McMorrow, H. M. Rønnow, and T. Schmitt, Phys. Rev.B , 041102 (2018).[33] V. Bisogni, M. Moretti Sala, A. Bendounan, N. B.Brookes, G. Ghiringhelli, and L. Braicovich, Phys. Rev.B , 214528 (2012).[34] F. Forte, M. Cuoco, C. Noce, and J. van den Brink,Phys. Rev. B , 245133 (2011).[35] S. Kourtis, J. van den Brink, and M. Daghofer, Phys.Rev. B , 064423 (2012).[36] U. Kumar, A. Nocera, E. Dagotto, and S. Johnston, NewJournal of Physics , 073019 (2018).[37] C. J. Jia, E. A. Nowadnick, K. Wohlfeld, Y. F. Kung, C.-C. Chen, S. Johnston, T. Tohyama, B. Moritz, and T. P.Devereaux, Nature Communications , 3314 (2014).[38] A. Higashiya, S. Imada, T. Murakawa, H. Fujiwara,S. Kasai, A. Sekiyama, S. Suga, K. Okada, M. Yabashi,K. Tamasaku, T. Ishikawa, and H. Eisaki, New Journalof Physics , 053033 (2008).[39] L. Wray, D. Qian, D. Hsieh, Y. Xia, H. Eisaki, and M. Z.Hasan, Phys. Rev. B , 100507 (2007).[40] K. Ishii, K. Tsutsui, T. Tohyama, T. Inami, J. Mizuki,Y. Murakami, Y. Endoh, S. Maekawa, K. Kudo,Y. Koike, and K. Kumagai, Phys. Rev. B , 045124(2007).[41] V. Bisogni, S. Kourtis, C. Monney, K. Zhou, R. Kraus,C. Sekar, V. Strocov, B. B¨uchner, J. van den Brink,L. Braicovich, T. Schmitt, M. Daghofer, and J. Geck,Phys. Rev. Lett. , 147401 (2014).[42] V. Bisogni, K. Wohlfeld, S. Nishimoto, C. Monney,J. Trinckauf, K. Zhou, R. Kraus, K. Koepernik, C. Sekar,V. Strocov, B. B¨uchner, T. Schmitt, J. van den Brink,and J. Geck, Phys. Rev. Lett. , 096402 (2015).[43] A. Nocera, U. Kumar, N. Kaushal, G. Alvarez,E. Dagotto, and S. Johnston, Scientific Reports , 11080(2018).[44] S. R. White, Phys. Rev. Lett. , 2863 (1992).[45] S. R. White, Phys. Rev. B , 10345 (1993).[46] B. J. Kim, H. Jin, S. J. Moon, J.-Y. Kim, B.-G. Park,C. S. Leem, J. Yu, T. W. Noh, C. Kim, S.-J. Oh, J.-H. Park, V. Durairaj, G. Cao, and E. Rotenberg, Phys.Rev. Lett. , 076402 (2008).[47] L. J. P. Ament, G. Khaliullin, and J. van den Brink,Phys. Rev. B , 020403 (2011).[48] J. Kim, D. Casa, M. H. Upton, T. Gog, Y.-J. Kim, J. F.Mitchell, M. van Veenendaal, M. Daghofer, J. van denBrink, G. Khaliullin, and B. J. Kim, Phys. Rev. Lett. , 177003 (2012).[49] H. Gretarsson, N. H. Sung, J. Porras, J. Bertinshaw,C. Dietl, J. A. N. Bruin, A. F. Bangura, Y. K. Kim,R. Dinnebier, J. Kim, A. Al-Zein, M. Moretti Sala,M. Krisch, M. Le Tacon, B. Keimer, and B. J. Kim,Phys. Rev. Lett. , 107001 (2016). [50] J. H. Gruenewald, J. Kim, H. S. Kim, J. M. Johnson,J. Hwang, M. Souri, J. Terzic, S. H. Chang, A. Said, J. W.Brill, G. Cao, H.-Y. Kee, and S. S. A. Seo, AdvancedMaterials , 1603798 (2017).[51] A. Gozar, G. Blumberg, B. S. Dennis, B. S. Shastry,N. Motoyama, H. Eisaki, and S. Uchida, Phys. Rev.Lett. , 197202 (2001).[52] L. J. P. Ament, M. van Veenendaal, T. P. Devereaux,J. P. Hill, and J. van den Brink, Rev. Mod. Phys. ,705 (2011).[53] W. Zheng, C. J. Hamer, R. R. P. Singh, S. Trebst, andH. Monien, Phys. Rev. B , 144410 (2001).[54] K. P. Schmidt and G. S. Uhrig, Phys. Rev. Lett. ,227204 (2003).[55] K. P. Schmidt and G. S. Uhrig, Modern Physics LettersB , 1179 (2005).[56] T. Barnes, E. Dagotto, J. Riera, and E. S. Swanson,Phys. Rev. B , 3196 (1993).[57] The triplon and bound two-triplon excitations are oftenreferred to as magnon and bound two-magnon excita-tions, respectively, in the literature. We have adopted thetriplon nomenclature to be consistent with the previousRIXS studies [5, 8].[58] Knetter, C., Schmidt, K. P., and Uhrig, G. S., Eur. Phys.J. B , 525 (2003).[59] O. P. Sushkov and V. N. Kotov, Phys. Rev. Lett. ,1941 (1998).[60] M. Greven, R. J. Birgeneau, and U. J. Wiese, Phys. Rev.Lett. , 1865 (1996).[61] D. G. Shelton, A. A. Nersesyan, and A. M. Tsvelik, Phys.Rev. B , 8521 (1996).[62] M. Troyer, H. Tsunetsugu, and T. M. Rice, Phys. Rev.B , 251 (1996). [63] S. Liu, H.-C. Jiang, and T. P. Devereaux, Phys. Rev. B , 155149 (2016).[64] S. R. White, D. J. Scalapino, and S. A. Kivelson, Phys.Rev. Lett. , 056401 (2015).[65] N. J. Robinson, F. H. L. Essler, E. Jeckelmann, andA. M. Tsvelik, Phys. Rev. B , 195103 (2012).[66] A. Nocera, N. D. Patel, E. Dagotto, and G. Alvarez,Phys. Rev. B , 205120 (2017).[67] E. H. Lieb and F. Y. Wu, Phys. Rev. Lett. , 1445(1968).[68] C. Jia, K. Wohlfeld, Y. Wang, B. Moritz, and T. P.Devereaux, Phys. Rev. X , 021020 (2016).[69] T. F. A. M¨uller and T. M. Rice, Phys. Rev. B , 3425(1998).[70] T. Tohyama, M. Mori, and S. Sota, Phys. Rev. B ,235137 (2018).[71] J. M. Tranquada, H. Woo, T. G. Perring, H. Goka, G. D.Gu, G. Xu, M. Fujita, and K. Yamada, Nature , 534(2004).[72] G. Xu, G. D. Gu, M. H¨ucker, B. Fauqu´e, T. G. Perring,L. P. Regnault, and J. M. Tranquada, Nature Physics ,642 (2009).[73] F. Forte, L. J. P. Ament, and J. van den Brink, Phys.Rev. B , 134428 (2008).[74] A. Klauser, J. Mossel, J.-S. Caux, and J. van den Brink,Phys. Rev. Lett. , 157205 (2011).[75] D. Benjamin, I. Klich, and E. Demler, Phys. Rev. Lett. , 247002 (2014).[76] M. Kan´asz-Nagy, Y. Shi, I. Klich, and E. A. Demler,Phys. Rev. B , 165127 (2016).[77] J. van den Brink and M. van Veenendaal, EPL (Euro-physics Letters) , 121 (2006).[78] L. J. P. Ament, F. Forte, and J. van den Brink, Phys.Rev. B75