Arxiv:2101.06032V3 [Quant-Ph] 26 May 2021 Quantum Ground States

The phases of the disordered Bose–Hubbard model with attractive interactions

Olli Mansikkamäki,Sami Laine, and Matti Silveri Nano and Molecular Systems Research Unit, University of Oulu, P.O. Box 3000, FI-90014 University of Oulu, Finland (Dated: May 27, 2021) We study the quantum ground state phases of the one-dimensional disordered Bose–Hubbard model with attractive interactions, realized by a chain of superconducting transmon qubits or cold atoms. We map the phase diagram using perturbation theory and exact diagonalization. Compared to the repulsive Bose–Hubbard model, the quantum ground state behavior is dramatically different. At strong disorder of the on-site energies, all the bosons localize into the vicinity of a single site, contrary to the Bose glass behavior of the repulsive model. At weak disorder, depending on hopping, the ground state is either superfluid or a W state, which is a multi-site and multi-particle entangled superposition of states where all the bosons occupy a single site. We show that the robustness of the W phase against disorder diminishes as the total number of bosons increases.

Introduction The Bose–Hubbard model is a paradig- Bose–Hubbard model is immediately applicable also for matic model of quantum matter and quantum phase tran- cold atoms in optical lattices, where the interaction can sitions, with applications ranging from magnetism to be tuned from repulsive to attractive via the Feshbach disordered superfluid helium [1–4]. It is canonically char- resonance [4, 39]. acterized by a repulsive boson-boson interaction disfavor- In this letter, we use exact diagonalization and perturba- ing local multi-occupancy, together with boson hopping tion theory to construct the ground state phase diagram of which models excitation kinetics. When repulsive inter- the one-dimensional disordered attractive Bose–Hubbard action dominates hopping, the ground state is the Mott model, and provide analytical expressions for the states insulating phase where a fixed integer number of bosons belonging to the W phase, the superfluid phase, and the are located on each lattice site. Otherwise, the ground localized phase. Our main result is that the robustness of state is the delocalized superfluid [5, 6]. In the presence of the W phase against disorder diminishes exponentially as disorder, a third phase—the Bose glass—emerges between the total number of bosons is increased. Finally, we pro- the Mott insulator and the superfluid phases [6–14]. The pose a possible realization of these phases using transmon Bose glass is an insulating phase with finite compressibil- chains with experimentally feasible parameters. ity caused by disorder localization. We note that phases and phase transitions are, strictly The attractive Bose–Hubbard model has remained speaking, only defined in the thermodynamic limit, that much less studied than its repulsive counterpart. At is, when both the number of lattice sites and the number strong disorder, many-body localization emerges [15–17] of particles approach infinity. For the Bose–Hubbard for highly-excited states. The quantum ground state model with attractive interactions, this is ill-defined since behavior changes dramatically when switching from re- the two limits are non-commutative [21] and the ground pulsive to attractive interaction [18–22]. When attractive state energy is not bounded below. However, seeing that interaction dominates hopping, in the absence of disorder, the finite-size behavior of the model resembles that of a the ground state is the W state, a fascinating self-trapped system with well defined phases, the concept of phase is multiparticle entangled state comprising a cat-state-like frequently used [21, 25, 40]. superposition of states with all the bosons occupying a Model The disordered attractive Bose–Hubbard model single site [23–25]. However, the interplay of disorder and with L sites is defined in the basis of the local bosonic attractive interactions has not been studied before for the annihilation aˆ`, creation aˆ`†, and occupation number nˆ` = arXiv:2101.06032v3 [quant-ph] 26 May 2021 quantum ground states. aˆ`†aˆ` operators by the Hamiltonian [15, 16, 33, 34] The attractive Bose–Hubbard model is an important ˆ L model for arrays of superconducting transmon devices, H U = ω`nˆ` nˆ`(ˆn` 1) + J aˆ`†+1aˆ` +a ˆ`†aˆ`+1 . a leading platform for large-scale quantum science ex- ~ − 2 − `=1 periments. A transmon is an anharmonic bosonic os- X (1) cillator with negative anharmonicity [26]. Fabrication Here, ω` represent random disorder of on-site energies. disorder [27] has hindered their utilization in large- We draw them from a uniform distribution in the in- scale quantum simulators [28, 29] of other than disorder terval ω [ D,D], with D being the strength of the ` ∈ − physics [15, 17, 30–32]. The size of experimentally demon- disorder. Since the model conserves the total number ˆ L strated transmon arrays has grown rapidly from a few to of bosons N = `=1 nˆ`, we consider a fixed N and can over 50 sites [30–38]. Thus, an array of coupled transmons thus ignore the mean on-site energy. The strength of the realizes the disordered attractive Bose–Hubbard model attractive interactionsP is given by U, and J is the hopping in a natural manner [16, 33]. Furthermore, the attractive frequency. Finally, ~ is the reduced Planck’s constant. 2

Figure 1. A schematic illustration and the local occupation density pn` = n ρˆ` n as a function of the transmon site ` and the h | | i local Fock state n for (a) the localized state, (b) the W state, and (c) the superﬂuid state at the scaled hopping frequency | i 3 τ = 0.05, 0.15, and 1, respectively. A single disorder realization at the scaled disorder strength δ = 0.33 10− was used. The · local density matrixρ ˆ` is calculated by tracing over all the other sites:ρ ˆ` = Tr i=` ( ψ ψ ), where ψ is the ground state. { 6 } | i h | | i

The bounded spectrum at fixed N implies that the most where only a single correction is added. After averaging excited eigenstate of the Hamiltonian (1) is the quantum over the disorder, the second-order energy of the localized ground state of the corresponding repulsive model, and ground state phase is vice versa. For the sake of experimental relevance, we 1 L 1 2 2 L 1 focus on open chains with aˆL+1 = 0 in Eq. (1) (see Ref. 41 εloc = δ − (1 2τ ) 2τ − . (5) for the corresponding results for a periodic chain). In −2 − L + 1 − − L what follows, we measure all energies in units of the char- When reducing disorder, the ground state phase changes acteristic energy ~UN(N 1), yielding the scaled energy either into the superfluid phase or into the W phase, − ε = E/~UN(N 1), hopping frequency τ = J/U(N 1), depending on hopping. − − disorder strength δ = D/U(N 1), and on-site energies W phase Let us next consider weak hopping, τ 1, in − σ` = ω`/U(N 1). the absence of disorder. Then the states n = N − | `=1,...,L i Localized phase Let us first consider strong attractive with the energy ε0 = 1/2 are degenerate and coupled by − interactions and strong disorder δ 1/(N 1) τ. a weak high-order hopping interaction, yielding a quantum − The attractive interactions force all the bosons to occupy ground state which is a superposition of the localized 1 the site `0 with the lowest on-site energy σ`0 , leading to states ψ` of Eq. (4), see Fig.1(b). This state is called 0 | i total energy ε = 1/2 + σ` . With strong disorder and the W state [25, 42, 43] because of its resemblance with loc − 0 ignoring hopping, the ground state is a product state of the W state of entangled qubits [44]. the form To solve the W state analytically, we resort to high- order degenerate perturbation theory [45], detailed in 0 ψ = n` = N , (2) | `0 i | 0 i Ref. 41. First, the sites at the ends of the open chain have only one neighbor each. Thus, the effective hopping en- where n = N denotes the state where N bosons oc- | `0 i ergy near the ends is higher than elsewhere, reducing the cupy the site ` and other sites are empty. We refer to 0 number of degenerate states by ` = N/2 1 counting this as the localized state, not to be confused with An- s d e − from both ends of the chain. For simplicity, let us consider derson localization of the non-interacting situation or the even L and odd N, such that the number of degenerate many-body localization of the highly-excited states [15– states L = L 2` 2. Then the degenerate perturba- 17]. When we take into account the hopping up to first d − s ≥ tion theory in the Nth order shows that the remaining order in non-degenerate perturbation theory, the localized states n = N are each coupled with their neighboring- state ψ0 gets a correction of the form | ` i | `0 i site counterparts, which gives us the superposition [see Fig.1(b)] n` 1 = 1, n` = N 1 1 √ 0± 0 ψ`0 1 = τ N | − i, (3) | ± i (σ` σ` 1) 1 0 0± − − Ld e 1 2 ` π` 1 that is, a state localized onto the site `0 will be ψw = ( 1) sin ψ`+`s . (6) | i Ld + 1 − Ld + 1 | i r `=1 1 0 1 1 X ψ`0 = ψ`0 + ψ`0+1 + ψ`0 1 , (4) | i | i | i | − i The existence of the W state requires a non-zero hop- up to a normalization factore [see Fig.e1(a)]. The same ping frequency, which is why we use the first-order per- form applies to every localized state excepting the ends, turbed states of Eq. (4) in the superposition. The shape 3 of the W state depends strongly on the total density The ground state can be calculated by first transforming N/L of the bosons— the higher the density, the more the these into the reciprocal basis and considering their effects bosonsbunch toward the middle of the chain, forming a perturbatively [41]. The second-order disorder-averaged self-trapping state [22]. For even L, the limit is an equal energy is given by superposition of the two middlemost localized states of π 3 a Eq. (4). For odd L, the limit is just the middlemost ε = 2τ cos δ2 + b , (10) SF − L + 1 − 4(L + 1) − τ localized state. The mean second-order energy of the W state is where a and b are coefficients dependent on N and L 1 2 (at N = 4 and L = 8, a 0.23 and b 0.05, see εW = 2τ , (7) ≈ ≈ −2 − Ref. 41). Notice that this condensate of bosons can indeed be called a superfluid since it is able to support metastable with no disorder contribution after averaging over different persistent currents in a periodic chain [46, 47]. realizations. Phase diagram Analytical results indicate that we When introducing disorder, we can probe the weak could expect ground state phase transitions at parame- interaction between the degenerate states formed through ters where the energies of any two phases [Eqs. (5), (7), high-order bosonic hopping. The magnitude of the Nth and (10)] are close. To study this in detail, we calculate order hopping interaction between the state n` = N N | i the quantum ground state of the Hamiltonian (1) by nu- and the states n` 1 = N is [α(N)τ] , where α(N) = | ± i merical exact diagonalization. To identify all three phases N 1 1/N (N 1) − /(N 1)! is a coefficient dependent on − − in numerical calculations, we need at least two indicator the boson number. At large N, the value of α(N) ap- quantities. We can use the fact that the states are charac- proaches Euler’s number e. The magnitude of disorder terized by different types of localization: the localized and needs to be of the same order or stronger than this effec- superfluid phases are localized in spatial and reciprocal tive hopping energy to cause the W state to disintegrate bases, respectively, and the W state is a superposition into a localized state. To sum up, at weak disorder and of spatially localized states. One quantity measuring the N hopping, the W phase can exist only when δ . [α(N)τ] , degree of localization is the inverse participation ratio showing that the robustness of the W phase against disorder diminishes exponentially with increasing total boson 1 N 2 s/r = 1 , (11) number. L s/r 2 P L 1 ψ nˆm ψ − ! Superfluid phase When hopping dominates interac- − m=1 | h | | i | tions, τ 1, the bosons can move from site to site largely which we can calculateP in both the position space s unhindered by each other, that is, the ground state is s P and in the reciprocal space r separately. Here, nˆm = a superfluid. It is easier here to work in the reciprocal r P aˆm† aˆm and nˆm = cˆm† cˆm are the number operators in the space, i.e., in the eigenbasis of the hopping term of the corresponding spaces. The inverse participation ratio Hamiltonian, accessible via the transformation yields zero if the state is localized and one if the state is completely delocalized. We can use to distinguish the L Ps 2 π`k spatially localized states from the superfluid states and cˆk = sin aˆ` (8) L + 1 L + 1 the W states. Conversely, we can use to distinguish r `=1 Pr X the superfluid states from the W states and the localized in the open chain. Analogously to the spatially localized states. states of Eq. (2), we may form localized states in the Keeping in mind the theoretical limitations of increasing reciprocal space, the total boson number N mentioned in the introduction, we are nevertheless interested in the phase boundaries ψ0 = η = N , (9) | ki | k i of the three phases as the system size in N is increased, which physically means approaching the semiclassical where η = N denotes the state where N bosons occupy | k i limit. For this purpose, we define the critical scaled the kth reciprocal mode and other modes are empty. hopping frequency [48] In the limit of vanishing interactions and disorder, the 0 0 superfluid ground state ψSF = ψk=L is completely ∂ | i | i s/r s/r localized to the lowest-energy mode of the reciprocal τc = argmax P (12) τ ∂ τ k L ε0 τ π/ L space, = , with the energy SF = 2 cos[ ( + 1)]. − Although the excitations created via cˆk† are localized in to locate the transition point from localized to delocalized s the reciprocal space, they are delocalized in the position state both in the position space (τc ) and in the reciprocal r space, see Fig.1(c) and Ref. 41 for visualizations. Both space (τc ). the interaction term and the disorder term have a weakly In Fig.2(a), we show the inverse participation ratios delocalizing effect in the reciprocal space, similar to the for the ground states of the Hamiltonian (1) as a func- Ps/r effect of the hopping term on the spatially localized state. tion of the scaled hopping frequency τ at weak disorder. 4

Figure 3. The critical scaled hopping frequencies (a) τ r (col- Figure 2. (a) and (b) In white, the inverse participation ratios c ored filled triangles) and (b) τ s (colored filled circles), de- (solid) and (dashed), defined in Eq. (11), of the quantum c Ps Pr fined in Eq. (12), as a function of the total boson number ground state of the disordered attractive Bose–Hubbard model N = 4,..., 9 and the scaled disorder strength δ. The chain (1) as a function of the scaled hopping frequency τ at two values length is L = 8 and the results are averaged over 2000 disorder of the scaled disorder strength, (a) δ = 0.001 and (b) δ = 0.036. realizations. The analytically predicted phase boundary be- In black, the corresponding critical scaled hopping frequencies s/r tween the localized phase and the W phase is shown as colored τ N c defined in Eq. (12). (c) The two inverse participation lines at δ = 2[α(N)τ s] in (b). ratios , represented as a single colormap, as a function of c Ps/r the scaled hopping frequency τ and the disorder strength δ. The amount of blue (red) indicates the value of ( ): The Ps Pr bluer (redder) the pixel, the more delocalized the ground state grows, the W state gets more and more localized towards is spatially (in the reciprocal space). When both of the inverse the middle of the chain, and thus the local maximum of participation ratios are non-zero, the color of the corresponding at τ > 0 disappears. At moderate disorder, depicted in Pr pixel is purple, indicating the W phase. The overlaid lines Fig.2(b), increasing hopping simultaneously delocalizes indicate the phase boundaries estimated from the analytical N the system in the position space and localizes it in the results: localized-to-W at δ = 2(ατ) (white dashed line), reciprocal space, with a crossover at τ 0.23, indicating superfluid-to-localized at εSF = εloc (gray dash-dotted line), ≈ absence of the W state. and W-to-superfluid at εW = εSF (yellow dotted line). The arrows indicate the locations of the horizontal cut-offs shown To explore the ground state phases in the full parameter in (a) and (b). The ground states were numerically computed space, we have computed the inverse participation ratios for an open chain of length L = 8 with the total number of as a function of both the scaled hopping frequency τ Ps/r bosons N = 4, and averaged over 1000 disorder realizations. and the scaled disorder strength δ. These are represented See Ref. [41] for the results for corresponding periodic chains. in Fig.2(c) as a colormap, where the amount of blue Note that τ 0.05 in this and all the following figures. ≥ (red) color in each pixel indicates the value of s ( r). In P P the W phase, both indicators are non-zero, shown with purple color. Notice that with N = 4 and L = 8, the W s/r The critical scaled hopping frequencies τc indicate that phase exists only at δ . 0.03. At stronger disorder, there the system localizes in the position space (solid) when is a direct transition from the localized phase into the τ < τ s 0.11 and in the reciprocal space (dashed) when superfluid phase. The overlaid lines of Fig.2(c) indicate c ≈ τ > τ r 0.17. In the region between, we identify the W the phase boundaries predicted by equating the analytical c ≈ state which is delocalized spatially but not yet localized in ground state energies. the reciprocal space. Note that attains its maximum At strong disorder, the localized phase changes into the Pr at a non-zero τ, within the W phase. This stems from superfluid phase near the curve εloc = εSF (dash-dotted the fact that, due to the shape of the transformation (8), line). At weak disorder, the superfluid phase changes a superposition of spatially localized states can be more into the W phase in the proximity of the curve εSF = εW uniformly distributed in the reciprocal space than a single (dotted line), while the W phase disintegrates into the localized state. However, as the number of bosons N localized phase at δ = 2[α(N)τ]N (dashed line). Here, the 5

numerical coefficient 2 in the localized-to-W boundary was choice of N < 10. obtained by fitting the analytically predicted expression To detect and distinguish the state, the population of (see Ref. 41) δ = (3A/2)[α(N)τ]N , with A a constant of every transmon or a subset of them can be measured with the order of unity, to the numerical data. high accuracy by coupling them individually to dispersive r Figures3(a–b) show the critical hopping frequencies τc readout resonators [50], as demonstrated recently in many- s and τc , respectively, as a function of the disorder strength body settings [15, 34]. This way, one can measure the δ for varying total boson number N. In Fig.3(a), the local occupation density that uniquely identifies each transition between the W phase and the superfluid phase phase as shown in Fig.1. By utilizing the sophisticated shifts towards larger hopping frequency as N is increased. driving protocols, one can realize cooling and stabilization We can explain this qualitatively through the fact that the schemes that can be used to achieve the quantum ground coupling between the reciprocal modes by the interaction states at fixed boson numbers, as was experimentally term of the Hamiltonian (1) increases with N, contained demonstrated in the seminal works [33, 34]. in the coefficients a and b of Eq. (10). In Fig.3(b), the Conclusions In this work, we have concentrated on data points indicate the transition between the localized the static ground state properties of the attractive Bose– and the W phase. Here, we observe clearly that the W Hubbard model with weak to moderate energy disorder. N phase exists only at δ . 2[α(N)τ] (solid colored lines), We mapped the ground state phase diagram and focused confirming our analytical prediction. on the W phase, which can be experimentally realized Realization with transmon arrays With transmon ar- with transmon arrays only within stringent bounds for the rays, the hopping term is realized by the capacitive cou- disorder and coupling strengths. Seeing that decoherence pling between the transmons [26, 29], resulting in the value and dissipation are common issues in superconducting of J/2π ranging between 10 MHz and 100 MHz. The inter- quantum devices, and that the loss of bosons can change action term originates from an approximation of the cosi- the nature of the ground state, even a system starting nusoidal potential of the Josephson junctions of the trans- from the ground state for a given number of bosons N mons [26], with typical values of U/2π between 200 MHz may experience quite interesting dynamics. Specifically, and 300 MHz. The disorder ω` models the small uninten- even if the bare system parameters J and U are not tional variations in manufactured devices, but it can also changed, boson losses due to dissipation effectively act as be artificially magnified or reduced by applying magnetic quenches in the scaled hopping frequency τ = J/U(N 1) − flux through the loop formed by the two parallel Josephson and disorder strength δ = D/U(N 1). This allows for − junctions of the transmons [15, 16, 26, 33, 34, 36], yielding the exploration of the parameter space (τ, δ), although D/2π in the range from 100 kHz to 2 GHz. The exper- only in discrete steps. A more detailed discussion of imentally realistic range of parameters belongs roughly the dynamics due to dissipation and quenches in the 4 2 to the interval J/U [0.03, 0.5] and D/U [10− , 10 ], ∈ ∈ parameters of the model, including a potential realization allowing the realization of all the ground state phases of dynamical quantum phase transition [51, 52], is an of the disordered attractive Bose–Hubbard model. For interesting subject of future work. Another intriguing modern transmons, the rate Γ1 of losing bosons (pho- question is the structure of the ground state phase diagram tons) is low [32, 34, 37], with Γ1/2π in the range of a in two and higher dimensions, relevant even for current few kHz. This should be contrasted with the effective transmon arrays [36]. hopping frequencies in the system. For example, in the W Acknowledgments The authors would like to thank phase, the ratio between the effective hopping frequency Steven Girvin and Tuure Orell for useful discussions J = UN(N 1)[α(N)τ]N and the many-body dissipation − on this and other closely related topics. This research rate is J/(NΓ1) & 50 with the parameters of Ref. 37, work was financially supported by the Alfred Kordelin indicatinge an ample window of coherent dynamics to form foundation, the Emil Aaltonen foundation, the Kvantum and detecte the W phase before a disruptive photon loss Institute of the University of Oulu, and the Academy event. of Finland under Grants No. 316619 and No. 320086. The Hamiltonian (1) becomes increasingly worse a The numerics are implemented in Julia [53] using the model of a transmon array as the number of bosons KrylovKit and Optim [54] packages. on a single site increases, since the interaction term Unˆ(nˆ 1)/2 is just the lowest anharmonicity term of − − the cosine potential of the Josephson junction [26]. How- ever, a more fundamental limitation is that the cosine potential also implies that a transmon has only a finite [1] S. M. Girvin and K. Yang, Modern Condensed Matter number of discrete bound states [49]. This number for Physics (Cambridge University Press, Cambridge, UK, 2019). modern transmon parameters is ω /(√8U) 10, with 0 ≈ [2] M. Greiner, O. Mandel, T. Esslinger, T. W. Hänsch, and ω0 the mean on-site energy. Adding more bosons will I. Bloch, Quantum phase transition from a superfluid to change the transmon spectrum and couplings, rendering a Mott insulator in a gas of ultracold atoms, Nature 415, the Bose–Hubbard model insufficient. This warrants our 39 (2002). 6

[3] T. Stöferle, H. Moritz, C. Schori, M. Köhl, and A. Zalcman, V. Smelyanskiy, H. Neven, S. Gopalakrish- T. Esslinger, Transition from a strongly interacting 1D su- nan, D. Abanin, M. Knap, J. Martinis, and P. Roushan, perfluid to a Mott insulator, Phys. Rev. Lett. 92, 130403 Direct measurement of non-local interactions in the many- (2004). body localized phase, arXiv:1910.06024 (2019). [4] I. Bloch, J. Dalibard, and W. Zwerger, Many-body physics [18] J. Dorignac, J. C. Eilbeck, M. Salerno, and A. C. Scott, with ultracold gases, Rev. Mod. Phys. 80, 885 (2008). Quantum signatures of breather-breather interactions, [5] T. D. Kühner, S. R. White, and H. Monien, One- Phys. Rev. Lett. 93, 025504 (2004). dimensional Bose-Hubbard model with nearest-neighbor [19] M. W. Jack and M. Yamashita, Bose-Hubbard model with interaction, Phys. Rev. B 61, 12474 (2000). attractive interactions, Phys. Rev. A 71, 023610 (2005). [6] M. A. Cazalilla, R. Citro, T. Giamarchi, E. Orignac, [20] P. Buonsante, V. Penna, and A. Vezzani, Attractive ul- and M. Rigol, One dimensional bosons: From condensed tracold bosons in a necklace optical lattice, Phys. Rev. A matter systems to ultracold gases, Rev. Mod. Phys. 83, 72, 043620 (2005). 1405 (2011). [21] N. Oelkers and J. Links, Ground-state properties of the [7] T. Giamarchi and H. J. Schulz, Anderson localization and attractive one-dimensional Bose-Hubbard model, Phys. interactions in one-dimensional metals, Phys. Rev. B 37, Rev. B 75, 115119 (2007). 325 (1988). [22] P. Buonsante, V. Penna, and A. Vezzani, Quantum signa- [8] M. P. A. Fisher, P. B. Weichman, G. Grinstein, and D. S. tures of the self-trapping transition in attractive lattice Fisher, Boson localization and the superfluid-insulator bosons, Phys. Rev. A 82, 043615 (2010). transition, Phys. Rev. B 40, 546 (1989). [23] L. Bernstein, J. C. Eilbeck, and A. C. Scott, The quantum [9] D. B. Haviland, Y. Liu, and A. M. Goldman, Onset of theory of local modes in a coupled system of nonlinear superconductivity in the two-dimensional limit, Phys. Rev. oscillators, Nonlinearity 3, 293 (1990). Lett. 62, 2180 (1989). [24] O. S. Sørensen, S. Gammelmark, and K. Mølmer, Relative [10] A. F. Hebard and M. A. Paalanen, Magnetic-field-tuned and center-of-mass motion in the attractive Bose-Hubbard superconductor-insulator transition in two-dimensional model, Phys. Rev A 85, 043617 (2012). films, Phys. Rev. Lett. 65, 927 (1990). [25] A. A. Gangat, I. P. McCulloch, and G. J. Milburn, De- [11] S. Rapsch, U. Schollwoeck, and W. Zwerger, Density terministic many-resonator W entanglement of nearly matrix renormalization group for disordered bosons in arbitrary microwave states via attractive Bose-Hubbard one dimension, Europhys. Lett. (EPL) 46, 559 (1999). simulation, Phys. Rev. X 3, 031009 (2013). [12] L. Fallani, J. E. Lye, V. Guarrera, C. Fort, and M. In- [26] J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. guscio, Ultracold atoms in a disordered crystal of light: Schuster, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, Towards a Bose glass, Phys. Rev. Lett. 98, 130404 (2007). and R. J. Schoelkopf, Charge-insensitive qubit design [13] R. Yu, L. Yin, N. S. Sullivan, J. S. Xia, C. Huan, derived from the Cooper pair box, Phys. Rev. A 76, A. Paduan-Filho, N. F. Oliveira Jr, S. Haas, A. Steppke, 042319 (2007). C. F. Miclea, F. Weickert, R. Movshovich, E.-D. Mun, [27] D. L. Underwood, W. E. Shanks, J. Koch, and A. A. B. L. Scott, V. S. Zapf, and T. Roscilde, Bose glass and Houck, Low-disorder microwave cavity lattices for quan- Mott glass of quasiparticles in a doped quantum magnet, tum simulation with photons, Phys. Rev. A 86, 023837 Nature 489, 379 (2012). (2012). [14] C. Meldgin, U. Ray, P. Russ, D. Chen, D. M. Ceperley, and [28] A. A. Houck, H. E. Türeci,and J. Koch, On-chip quantum B. DeMarco, Probing the Bose glass–superfluid transition simulation with superconducting circuits, Nat. Phys. 8, using quantum quenches of disorder, Nat. Phys. 12, 646 292 (2012). (2016). [29] M. Dalmonte, S. I. Mirzaei, P. R. Muppalla, D. Marcos, [15] P. Roushan, C. Neill, J. Tangpanitanon, V. M. Bastidas, P. Zoller, and G. Kirchmair, Realizing dipolar spin models A. Megrant, R. Barends, Y. Chen, Z. Chen, B. Chiaro, with arrays of superconducting qubits, Phys. Rev. B 92, A. Dunsworth, A. Fowler, B. Foxen, M. Giustina, E. Jef- 174507 (2015). frey, J. Kelly, E. Lucero, J. Mutus, M. Neeley, C. Quin- [30] K. Xu, J.-J. Chen, Y. Zeng, Y.-R. Zhang, C. Song, W. Liu, tana, D. Sank, A. Vainsencher, J. Wenner, T. White, Q. Guo, P. Zhang, D. Xu, H. Deng, K. Huang, H. Wang, H. Neven, D. G. Angelakis, and J. Martinis, Spectro- X. Zhu, D. Zheng, and H. Fan, Emulating many-body scopic signatures of localization with interacting photons localization with a superconducting quantum processor, in superconducting qubits, Science 358, 1175 (2017). Phys. Rev. Lett. 120, 050507 (2018). [16] T. Orell, A. A. Michailidis, M. Serbyn, and M. Silveri, [31] C. Zha, V. Bastidas, M. Gong, Y. Wu, H. Rong, R. Yang, Probing the many-body localization phase transition Y. Ye, S. Li, Q. Zhu, S. Wang, Y. Zhao, F. Liang, J. Lin, with superconducting circuits, Phys. Rev. B 100, 134504 Y. Xu, C.-Z. Peng, J. Schmiedmayer, K. Nemoto, H. Deng, (2019). W. Munro, X. Zhu, and J.-W. Pan, Ergodic-localized [17] B. Chiaro, C. Neill, A. Bohrdt, M. Filippone, F. Arute, junctions in a periodically driven spin chain, Phys. Rev. K. Arya, R. Babbush, D. Bacon, J. Bardin, R. Barends, Lett. 125, 170503 (2020). S. Boixo, D. Buell, B. Burkett, Y. Chen, Z. Chen, [32] M. Gong, G. D. d. M. Neto, C. Zha, Y. Wu, H. Rong, R. Collins, A. Dunsworth, E. Farhi, A. Fowler, B. Foxen, Y. Ye, S. Li, Q. Zhu, S. Wang, Y. Zhao, F. Liang, J. Lin, C. Gidney, M. Giustina, M. Harrigan, T. Huang, Y. Xu, C.-Z. Peng, H. Deng, A. Bayat, X. Zhu, and J.-W. S. Isakov, E. Jeffrey, Z. Jiang, D. Kafri, K. Kechedzhi, Pan, Experimental characterization of quantum many- J. Kelly, P. Klimov, A. Korotkov, F. Kostritsa, D. Land- body localization transition, arXiv:2012.11521 (2020). huis, E. Lucero, J. McClean, X. Mi, A. Megrant, [33] S. Hacohen-Gourgy, V. V. Ramasesh, C. De Grandi, I. Sid- M. Mohseni, J. Mutus, M. McEwen, O. Naaman, M. Nee- diqi, and S. M. Girvin, Cooling and autonomous feedback ley, M. Niu, A. Petukhov, C. Quintana, N. Rubin, D. Sank, in a Bose-Hubbard chain with attractive interactions, K. Satzinger, A. Vainsencher, T. White, Z. Yao, P. Yeh, Phys. Rev. Lett. 115, 240501 (2015). 7

[34] R. Ma, B. Saxberg, C. Owens, N. Leung, Y. Lu, J. Si- H. Wang, Stark many-body localization on a supercon- mon, and D. I. Schuster, A dissipatively stabilized Mott ducting quantum processor, arXiv:2011.13895 (2020). insulator of photons, Nature 566, 51 (2019). [39] M. Theis, G. Thalhammer, K. Winkler, M. Hellwig, [35] M. Reagor, C. B. Osborn, N. Tezak, A. Staley, G. Ruff, R. Grimm, and J. H. Denschlag, Tuning the G. Prawiroatmodjo, M. Scheer, N. Alidoust, E. A. Sete, scattering length with an optically induced feshbach reso- N. Didier, M. P. d. Silva, E. Acala, J. Angeles, A. Best- nance, Phys. Rev. Lett. 93, 123001 (2004). wick, M. Block, B. Bloom, A. Bradley, C. Bui, S. Cald- [40] X.-L. Deng, D. Porras, and J. I. Cirac, Quantum phases well, L. Capelluto, R. Chilcott, J. Cordova, G. Crossman, of interacting phonons in ion traps, Phys. Rev. A 77, M. Curtis, S. Deshpande, T. E. Bouayadi, D. Girshovich, 033403 (2008). S. Hong, A. Hudson, P. Karalekas, K. Kuang, M. Leni- [41] See supplemental material. han, R. Manenti, T. Manning, J. Marshall, Y. Mohan, [42] L. Heaney, A. Cabello, M. F. Santos, and V. Vedral, W. O’Brien, J. Otterbach, A. Papageorge, J.-P. Paque- Extreme nonlocality with one photon, New J. Phys. 13, tte, M. Pelstring, A. Polloreno, V. Rawat, C. A. Ryan, 053054 (2011). R. Renzas, N. Rubin, D. Russel, M. Rust, D. Scarabelli, [43] Z. Wang and D. Markham, Nonlocality and entanglement M. Selvanayagam, R. Sinclair, R. Smith, M. Suska, T.-W. for symmetric states, Phys. Rev. A 87, 012104 (2013). To, M. Vahidpour, N. Vodrahalli, T. Whyland, K. Yadav, [44] W. Dür,G. Vidal, and J. I. Cirac, Three qubits can be W. Zeng, and C. T. Rigetti, Demonstration of universal entangled in two inequivalent ways, Phys. Rev. A 62, parametric entangling gates on a multi-qubit lattice, Sci. 062314 (2000). Adv. 4, eaao3603 (2018). [45] J. J. Sakurai and J. Napolitano, Modern Quantum Me- [36] F. Arute, K. Arya, R. Babbush, D. Bacon, J. C. chanics 2nd ed. (Cambridge University Press, Cambridge, Bardin, R. Barends, R. Biswas, S. Boixo, F. G. S. L. 2017). Brandao, D. A. Buell, B. Burkett, Y. Chen, Z. Chen, [46] G. Arwas, A. Vardi, and D. Cohen, Superfluidity and B. Chiaro, R. Collins, W. Courtney, A. Dunsworth, chaos in low dimensional circuits, Sci. Rep. 5, 1 (2015). E. Farhi, B. Foxen, A. Fowler, C. Gidney, M. Giustina, [47] G. Arwas and D. Cohen, Superfluidity in Bose-Hubbard R. Graff, K. Guerin, S. Habegger, M. P. Harrigan, circuits, Phys. Rev. B 95, 054505 (2017). M. J. Hartmann, A. Ho, M. Hoffmann, T. Huang, [48] P. Buonsante and A. Vezzani, Ground-state fidelity and T. S. Humble, S. V. Isakov, E. Jeffrey, Z. Jiang, bipartite entanglement in the Bose-Hubbard model, Phys. D. Kafri, K. Kechedzhi, J. Kelly, P. V. Klimov, S. Knysh, Rev. Lett. 98, 110601 (2007). A. Korotkov, F. Kostritsa, D. Landhuis, M. Lind- [49] I. Pietikäinen, J. Tuorila, D. S. Golubev, and G. S. mark, E. Lucero, D. Lyakh, S. Mandrà, J. R. Mc- Paraoanu, Photon blockade and the quantum-to-classical Clean, M. McEwen, A. Megrant, X. Mi, K. Michielsen, transition in the driven-dissipative Josephson pendulum M. Mohseni, J. Mutus, O. Naaman, M. Neeley, C. Neill, coupled to a resonator, Phys. Rev. A 99, 063828 (2019). M. Y. Niu, E. Ostby, A. Petukhov, J. C. Platt, C. Quin- [50] P. Krantz, M. Kjaergaard, F. Yan, T. P. Orlando, S. Gus- tana, E. G. Rieffel, P. Roushan, N. C. Rubin, D. Sank, tavsson, and W. D. Oliver, A quantum engineer’s guide K. J. Satzinger, V. Smelyanskiy, K. J. Sung, M. D. Tre- to superconducting qubits, App. Phys. Rev. 6, 021318 vithick, A. Vainsencher, B. Villalonga, T. White, Z. J. Yao, (2019). P. Yeh, A. Zalcman, H. Neven, and J. M. Martinis, Quan- [51] M. Heyl, Dynamical quantum phase transitions: a review, tum supremacy using a programmable superconducting Rep. Prog. Phys. 81, 054001 (2018). processor, Nature 574, 505 (2019). [52] M. Lacki and M. Heyl, Dynamical quantum phase tran- [37] D. L. Campbell, Y.-P. Shim, B. Kannan, R. Winik, D. K. sitions in collapse and revival oscillations of a quenched Kim, A. Melville, B. M. Niedzielski, J. L. Yoder, C. Tahan, superfluid, Phys. Rev. B 99, 121107(R) (2019). S. Gustavsson, and W. D. Oliver, Universal nonadiabatic [53] J. Bezanson, A. Edelman, S. Karpinski, and V. B. Shah, control of small-gap superconducting qubits, Phys. Rev. Julia: A fresh approach to numerical computing, SIAM X 10, 041051 (2020). Rev. Soc. Ind. Appl. Math. 59, 65 (2017). [38] Q. Guo, C. Cheng, H. Li, S. Xu, P. Zhang, Z. Wang, [54] P. K. Mogensen and A. N. Riseth, Optim: A mathematical C. Song, W. Liu, W. Ren, H. Dong, R. Mondaini, and optimization package for Julia, J. Open Source Softw. 3, 615 (2018). Supplementary material: The phases of the disordered Bose–Hubbard model with attractive interactions

Olli Mansikkam¨aki,Sami Laine, and Matti Silveri Nano and Molecular Systems Research Unit, University of Oulu, P.O. Box 3000, FI-90014 University of Oulu, Finland (Dated: May 27, 2021)

I. MODEL it is simply a constant when N is ﬁxed, and thus we can neglect it in our analysis. In this supplementary material, we take a closer look at the asymptotic phases of the attractive Bose–Hubbard model with L sites. We consider both an open chain and II. OPEN CHAIN a periodic chain, the former of which was discussed in the main text. In both cases, the Hamiltonian of the system Let us ﬁrst study the open chain, which was analysed can be written as a sum of four parts, in the main text.

Hˆ = HˆM + HˆD + HÛ + HˆJ , (S1) A. Localised phase where L Let us set the hopping frequency J to zero and consider ˆ HM /~ = ω0 nˆ`, (S2) the Hamiltonian X`=1 L Hˆ0 = HˆD + HÛ . (S6) HˆD/~ = ω`nˆ`, (S3) ˆ X`=1 The eigenstates of H0 at fixed N are given by the tensor U L product states n1, . . . , nL n1 nL with n` ˆ | Li ≡ | i ⊗ · · · ⊗ | i HU /~ = nˆ`(ˆn` 1), (S4) ` n N − 2 − excitations at site and `=1 ` = . The corresponding X`=1 eigenenergies are L P ˆ L L HJ /~ = J aˆ`†+1aˆ` +a ˆ`+1aˆ`† . (S5) U `=1 En1,...,nL /~ = ω`n` n`(n` 1). X − 2 − X`=1 X`=1 Here, aˆ`† and aˆ` are the bosonic creation and annihilation operators at site ` satisfying the commutation relations Using the facts that ω` minm ωm (by definition) and 2 ≥ L 2 L [aˆ`†, aˆm† ] = [aˆ`, aˆm] = 0, [aˆ`, aˆm† ] = δ`,m, with δ`,m the n n (expanding the r.h.s.), we see `=1 ` ≤ `=1 ` Kronecker delta, and nˆ` = aˆ`†aˆ` is the corresponding that the ground state of the system is a state completely occupation number operator. For an open chain we define P P localized to the site `0 with the lowest on-site energy ω`, aˆL† +1 = aˆL+1 = 0, while for a closed chain we set aˆL† +1 = (0) a† a a E = n`0 = N , (S7) ˆ1 and ˆL+1 = ˆ1. | i | i The first two terms in the Hamiltonian, HˆM and HˆD, represent the on-site energies of the bosons neglecting where `0 = argmin` ω`. The energy of the ground state is

arXiv:2101.06032v3 [quant-ph] 26 May 2021 interactions. The on-site energy of a single boson at (0) U site ` is denoted by ~(ω0 + ω`), where ~ω0 is the mean E /~ = ω`0 N N(N 1). (S8) − 2 − value calculated over the L sites and ~ω` measures the local deviation from the mean. We assume that ω` are We assume here that the minimum is unique, that is, the sampled from the uniform random distribution ( D,D), ground state is non-degenerate. D U − where is the strength of the disorder. The third term Let us then treat the hopping interaction HˆJ as a Hˆ in the Hamiltonian, U , describes the attractive on-site small perturbation to Hˆ0 using the standard Rayleigh- interaction between the bosons, with U > 0 the strength Schrödingerperturbation theory [1], where we expand the ˆ of the interaction. Finally, the fourth term, HJ , models ground state and the ground-state energy in powers of J the nearest-neighbour interaction between the sites, with as E = E(0) + E(1) + ... and E = E(0) + E(1) + .... J > 0 the hopping frequency. The| first-orderi | i correction| i to the ground-state energy is Since the Hamiltonian commutes with the total number given by ˆ L operator N = `=1 nˆ`, we can study separately the (1) (0) (0) manifolds of fixed eigenvalues N of Nˆ. Since HˆM Nˆ, E /~ = E HˆJ /~ E = 0. (S9) P ∝ h | | i 2

Figure S1. The occupation density of the open chain model in the position space pn` = n ρˆ` n , (a)–(c), as a function of h | | i the transmon site ` and the Fock state n , and the occupation density in the reciprocal space pnk = n ρˆk n , (d)–(f), as a | i h | | i function of the reciprocal mode k and the corresponding Fock state n for (a),(d) the localized state, (b),(e) the W state, and | i (c),(f) the superﬂuid state at the scaled hopping frequency τ = 0.05, 0.15, and 1, respectively, and at a single disorder realization 3 at the scaled disorder strength δ = 0.33 10− . The local density matrix ρˆ` is calculated by tracing over all the other sites: · ρˆ` = Tr i=` ( ψ ψ ), where ψ is the full quantum ground state, and similarly for the reciprocal space. { 6 } | i h | | i

The first-order correction to the ground state is given by Again, this needs to be interpreted so that any term with (1) ω1 1 or ωL+1 vanishes. E = (S10) − | i Since each different realisation of ω`:s produces a differ- N ˆ (0) (1) n1, . . . , nL HJ /~ nl0 = N ent result for the ground state energy E E + E + n , . . . , n h | | i. ≈ 1 L (0) E(2), we are more interested in its disorder average. In gen- | i (E En1,...,nL )/~ n1,...,nL=1 − f f ω , . . . , ω n1+ X+nL=N eral, the disorder average of a function = ( 1 L) n··· =N `0 6 can be calculated as Since the unperturbed ground state is completely localised, D D the only two states accessible via the perturbation HˆJ f(ω1, . . . , ωL) f D = dω1 dωL L . (S14) are n`0 = N 1, n`0 1 = 1 . We thus obtain h i D ··· D (2D) | − ± i Z− Z− n = 1, n = N 1 (1) √ `0 1 `0 E = J N | − − i Since E involves the minimum of ω`, we split the integra- | i ω`0 ω`0 1 U(N 1) L − − − − tion region [ D,D] into L distinct partitions ω` ωm, n = N 1, n = 1 − ≤ +| `0 − `0+1 i . (S11) m = `. A straightforward integration then yields ω ω U(N 1) 6 `0 − `0+1 − − If it so happens that the state is localised at either one 1 L 1 L 1 ∞ ( 1)nL ε = δ − 2τ 2 − − (2δ)n, of the boundaries (`0 = 1 or `0 = L), we simply interpret −2 − L + 1 − L (n + 1)(n + L) n , n N n N , n n=1 0 = 1 1 = 1 = L = 1 L+1 = 1 = 0. X (S15) | Finally, the second-order− i | correction− to the ground-statei where ε = E D/~UN(N 1), τ = J/U(N 1), and energy is given by δ = D/U(Nh i1). The infinite− series has a closed-form− (2) (0) (1) − E /~ = E HˆJ /~ E . (S12) expression in terms of the natural logarithm, but we h | | i are mainly interested in the region where 2δ is small, Using the result above, we obtain and thus the series expansion is more convenient. The exact ground state of the localized phase is visualized (2) 2 1 E /~ = J N in Fig. S1(a),(d) through the occupation density in the ω` ω` 1 U(N 1) 0 − 0− − − position and the reciprocal space, and the comparison 1 + . (S13) with the analytic approximation of Eqs. (S7) and (S11) ω ω U(N 1) `0 − `0+1 − − is shown in Fig. S2(a). 3

B. W phase where

i 2 1 − Let us set both the hopping frequency J and the disor- Kˆ = Pˆ Hˆ Pˆ E Hˆ Pˆ − Pˆ Hˆ Pˆ i 0 J 1 − 0 1 1 J 1 der strength D (and thus all ω`) to zero and consider the h i 1 Hamiltonian Pˆ E Hˆ Pˆ − Hˆ Pˆ . (S21) × 1 − 0 1 J 0 Hˆ Hˆ . h i 0 = U (S16) i The operators Kî behave essentially like (HˆJ ) . They are explicitly of order i in J, although higher-order terms We see from the results above that the ground state of Hˆ 0 emerge when expanding E in powers of J. Working in is L-fold degenerate and belongs to a manifold spanned by the basis n = N ,..., n = N , we see that Kˆ are the localised states n = N , ` = 1,...,L. The ground- 1 L i ` diagonal if| i < Ni, since| it takesi at least N moves to state energy is given| by i transfer N excitations from one site to another. For odd U i, all the diagonal terms vanish, since it is impossible E(0)/~ = N(N 1), (S17) to move excitations localised to one particular site by − − 2 an odd number of times and still return to the original ˆ but the exact form of the ground state is left undeter- state. If i is even, i = 2m, the diagonal of Ki is of the form mined. diag k1, . . . , km, km+1, km+1, . . . , km+1, km+1, km, . . . , k1 , { k > k } If we now let J > 0, we can determine the shape of the where j j+1. This is explained by the fact that ground state superposition in the limit J 0 using the for sites farther away from the boundaries, there are standard degenerate perturbation theory→ [1]. However, more ways to move excitations around before hitting Kˆ unlike in the usual textbook examples, it is not enough to the boundary. Since each such route contributes to i Kˆ consider just first-order perturbations. Instead, one must with a negative weight, the result follows. Finally, N is compute the corrections up to the Lth or the Nth order, tridiagonal, with equal super- and subdiagonal elements. whichever is the lowest, as we shall see below. This follows from the fact that there is exactly one way to transfer N excitations from one site to another site Defining the projection operator Pˆ = 0 with N moves. In particular, the target site has to be L n = N n = N to the manifold spanned `=1 | ` i h ` | one of the nearest neighbours. by the degenerate ground states of Hˆ (we will refer P 0 With these observations in mind, we can proceed with to this simply as the degenerate manifold), and its solving Eq. (S20) for E(0) . We work our way up order | i complement Pˆ1 = Iˆ Pˆ0 with Iˆ being the identity by order, up to order N. All odd-order corrections to the − ˆ ˆ operator, the full eigenvalue problem of H0 + HJ can be ground state energy vanish, since Kî = 0 for odd i. In written as [1] particular,

E(1)/~ = 0. (S22) E E(0) Pˆ Hˆ Pˆ (S18) − − 0 J 0 Each even-order term Kˆ up to order N 1 reduces the 1 2m − Pˆ Hˆ Pˆ E Hˆ Hˆ Pˆ − Hˆ Pˆ Pˆ E = 0, dimension of the ground state manifold by dropping out − 0 J 1 − 0 − J 1 J 0 0 | i the m sites closest to each of the boundaries, since these h i 1 − are higher in energy. Thus, if N is high enough compared Pˆ E = Pˆ E Hˆ Hˆ Pˆ Hˆ Pˆ E . (S19) 1 | i 1 − 0 − J 1 J 0 | i to L, only one or two sites participate in the ground state. h i To be more precise, this happens if L 2 N/2 , where Here, E is the exact ground state and E is the exact x x ≤ d e | i denotes the ﬂoor of , i.e., the smallest integer greater ground-state energy, which we again expand in powers of thand e or equal to x. For odd L, this completely determines (0) (1) (0) (1) J as E = E + E +... and E = E +E + .... the unperturbed ground state, Equation| i (S18| ) isi an| eigenvaluei problem within the degen- (0) erate manifold, and determines the shape of the solution E = n(L+1)/2 = N . (S23) therein, in particular the correct form of the ground state | i | i (0) E of Hˆ0. After solving Eq. (S18), Eq. (S19) can For even L, the two middlemost sites have equal energy be| usedi to calculate the solution outside the degenerate up to N 1st order, but at Nth order the degeneracy is − manifold. lifted, and we ﬁnd that the unperturbed ground state is Noting that Pˆ0HˆJ Pˆ0 vanishes identically, and expand- ˆ ˆ ˆ ˆ (0) 1 N ing the inverse of P1(E H0 HJ )P1 using Neumann E = nL/2 = N + ( 1) nL/2+1 = N series, Eq. (S18) can be− written− as (still no approxima- | i √2 | i − | i tions) (S24) by solving a simple 2 2 eigenvalue problem. If, on the other hand,× L > 2 N/2 , then the un- ∞ d e E E(0) Kˆ Pˆ E = 0, (S20) perturbed ground state is a linear combination of the − − i 0 | i i ! Ld = L + 2 2 N/2 middlemost states n`+`s = N , X=2 − d e | i 4

2 Figure S2. The overlap ψ ψ0 of the exact ground state ψ and the analytically solved ψ0 of the open chain model as | h | i | | i | i a function of the scaled hopping frequency τ and disorder strength δ for (a) the localized phase ψ0 = ψ of Eqs. (S7) | i | loci and (S11), (b) the W phase ψ of Eq. (S30), and (c) the superﬂuid phase ψ of Eqs. (S41), (S48) and (S49). The overlap | Wi | SFi was computed for an open chain of length L = 8 with the total number of bosons N = 4, and averaged over 1000 disorder realisations

where ls = N/2 1 and ` = 1,...,Ld. The exact form If N > 2, the ratio c/ b is always quite small compared of this lineard combinatione − is determined at Nth order. to unity, attaining the| value| 2/9 for N = 4 and decreasing If N is odd, the diagonal of KˆN vanishes, and the prob- exponentially with increasing N. Thus, a good approxi- lem reduces to finding the eigenvalues and -vectors of a mation for the ground state is obtained by simply setting symmetric tridiagonal Toeplitz matrix with non-negative c = 0. In this case, we have coefficients. This is a well-known problem, with the solu- Ld tion [2] (0) 2 π` E = sin n`+`s = N . | i Ld + 1 Ld + 1 | i r `=1 X L (S30) 2 d π` E(0) ` n N . For completeness, if 0 < c < b , the ground state is of = ( 1) sin `+`s = | | | i Ld + 1 − Ld + 1 | i r `=1 the form X (S25) Ld (0) 2 If N is even, the diagonal of KˆN does not vanish, and we E = c sin[(` 1)θ] b sin(`θ) | i ALd + B { − − } need to solve the eigenproblem of a matrix of the form r `=1 X n` ` = N , (S31) a + c b 0 0 0 ... 0 × | + s i b a b 0 0 0 where θ is the smallest positive solution to the equation  0 b a b 0 ··· 0  ··· Ld + 1 c Ld 1  0 0 b a b 0  , (S26) cos θ + cos − θ = 0 (S32)  . . . ···.  2 b 2  ......     0 0 0 b a b  and A = b2 + c2 2bc cos θ, B = b2 c2.  ···   0 0 0 0 b a + c Going to N +1st− order, we find that− for odd N, the first  ···  (1)   order correction Pˆ0 E is a linear combination of the where a < 0, same states as E(0) |, buti orthogonal to this (choosing the usual normalisation| i E(0) E = 1). In the case of even N 1 h | i (N 1) − ˆ (1) b = UN(N 1) − τ N , (S27) N, which is the case we are more interested in, P0 E − − (N 1)! ˆ | i − vanishes. This is due to the fact that KN+1 vanishes, c = UN(N 1)τ N , (S28) since N + 1 is odd. − In the space orthogonal to the degenerate manifold, the and τ = J/U(N 1). Note that the value of a has no first-order correction to the ground state can be calcu- − effect on the eigenstates, it simply shifts the eigenvalues. lated using Eq. (S19). Each state n` = N in the linear | i Also note that 0 c b . A fully closed-form solution is combination attains a correction identical to the one we available if c = 0≤ or c≤= | b| [2]. If N = 2, then c = b and calculated for the localised phase, see Eq. (S11). the ground state is given| | by | | Finally the second-order correction to the ground-state energy is given by Ld (0) 2 π(2` 1) 2 E n N . (2) (0) (1) 2J N = sin − `+`s = (S29) E / = E Hˆ / E = . (S33) | i Ld 2Ld | i ~ J ~ r `=1 h | | i −U(N 1) X − 5

(0) N 1 N The expected value of the ground state energy E E + sites with strength ~UN(N 1)(N 1) − τ /(N E(1) + E(2) is thus given by ≈ 1)!. When this coupling increases− enough− compared to− the energy difference between neighbouring sites, the 1 ε = 2τ 2, (S34) average value of which is 2~DN/3, the localised state −2 − starts to spread considerably. Increasing the coupling even further, the ground state starts to approach the W where ε = E / UN(N 1) and τ = J/U(N 1). D ~ state. Comparing the coupling strength with the energy The exact groundh i state of− the W phase is visualized− difference, we find that the phase boundary obeys the in Fig. S1(b),(e) through the occupation density in the formula position and the reciprocal space, and the comparison with the analytic approximation of Eq. (S30) is shown in N δ = (Cτ) , (S35) Fig. S2(b). where C. Effect of disorder on the W phase 3 (N 1)N 1 C = N A − − (S36) s2 (N 1)! Above, we discussed how the hopping term HˆJ lifts the − L-fold degeneracy of the ground state of the Hamiltonian and A is a constant close to unity. Note that, by Stirling’s Hˆ , fixing the correct form of the superposition of the U approximation for factorials, C e as N . states n` = N to the W state. In the analysis, we ne- → → ∞ glected| the disorderi altogether. Here, we briefly discuss the consequences of disorder on the W phase. D. Superfluid phase Treating both HˆD and HˆJ as small perturbations to (0) HÛ , the unperturbed ground state E of HÛ can be solved from Eq. (S18) after making| thei replacement Let us set the on-site interaction strength U and the D HˆJ HˆD + HˆJ . Note that unlike Pˆ0HˆJ Pˆ0, the term disorder strength to zero, and consider the Hamiltonian → L Pˆ Hˆ Pˆ N ω n N n N 0 D 0 = `=1 ` ` = ` = does not van- ˆ ˆ ish. This immediately shows| usi that h if D| J, the disor- H0 = HJ . (S37) der is the dominantP term, present already∼ at first order ˆ in the perturbation theory, making the localised phase One way to find the eigenstates and eigenenergies of H0 energetically favourable compared to the W phase. is to first consider the Heisenberg equations of motion for Let us again denote τ = J/U(N 1) and δ = D/U(N aˆ`, 1), and assume that L > 2 N/2 −. If δ τ n for n −2, d e ∼ ≥ daˆ` ˆ the disorder first appears at order n in the perturbation = i H0/~, aˆ` = iJaˆ` 1 iJaˆ`+1 τ J dt − − − theory, using (or ) as the expansion parameter. On h i the other hand, we saw above that the exact form of the This is a simple tridiagonal Toeplitz system, which can W state is determined at order N. This leads us to the be diagonalised with a transformation [2] following conclusion. If n > N, the disorder only slightly modifies the W phase. If n < N, the disorder is the L 2 π`k dominant perturbation, and thus the localised phase is aˆ` = sin cˆk. (S38) L + 1 L + 1 favoured (although the hopping term may still reduce r k=1 the possible sites of localisation to ones sufficiently far X away from the boundaries, depending on the value of In terms of the reciprocal space creation and annihilation ˆ ˆ n). If n = N, both HD and HJ contribute more or less operatorsc ˆk† andc ˆk, the Hamiltonian can be written as equally, and we are at the transition region between the two phases. Therefore, roughly speaking, the W phase L πk N ˆ is stable when δ τ . Otherwise, the localised phase is H0/~ = 2J cos cˆ† cˆk. (S39) . L + 1 k favoured. kX=1 Let us consider the transition region in a bit more detail. Assuming δ τ N , the ground state of the system is some The total number operator is still of the same form as ∼ ˆ L L superposition of the states n = N , ` = 1,...,L , before, N = k=1 cˆk† cˆk = k=1 ηˆk. | `+`s i d where ` = N/2 1 and L = L 2` (see subsection The eigenstates of Hˆ0 at fixed N are given by the tensor s d e − d − s P P IIB). The form of this linear combination is determined product states η1, . . . , ηL η1 ηL with ηk | i ≡ | i ⊗ · · · ⊗ | i ˆ ˆ ˆ ˆ (0) k L η N by the matrix P0HDP0 + KN , the ground state of which excitations at the th reciprocal mode and k=1 k = . we need to solve within this subspace. Here, we have The corresponding eigenenergies are P ˆ (0) ˆ (0) denoted KN = KN (E = E ). Now, the ground state of L πk Pˆ0HˆDPˆ0 is a state localised to the site `0 with the lowest (0) Eη1,...,ηL /~ = 2J cos ηk. (S40) value of ω`. On the other hand, Kˆ couples neighbouring L + 1 N kX=1 6

Since cosine is a decreasing function in the interval [0, π], The first-order correction to the ground-state energy is we see that the ground state of the system is a state given by completely localized to mode L, (1) (0) (0) E /~ = E HˆD/~ + HÛ /~ E . (S45) (0) h | | i E = ηL = N , (S41) | i | i A straightforward calculation yields The energy of the ground state is L (1) 2N 2 π` 3UN(N 1) π E /~ = ω` sin − . E(0)/~ = 2JN cos . (S42) L + 1 L + 1 − 4(L + 1) − L + 1 `=1 X (S46) The first-order correction to the ground state is given by Let us then treat the disorder HˆD and the on-site ˆ ˆ interaction HU as small perturbations to H0. Expressed E(1) = (S47) in the reciprocal space, these are given by | i N η , . . . , η Hˆ + Hˆ η = N η , . . . , η 1 L D U L . L 1 L h (0)| | i 2ω` π`j π`k | i (E Eη1,...,ηL ) Hˆ / = sin sin cˆ†cˆ , η1,...,ηL=1 − D ~ j k η + X+η =N L + 1 L + 1 L + 1 1 ··· L `,j,k=1 ηL=N X 6 (S43) The only states giving a non-vanishing contribu- 2 L U 2 π`j tion are ηk = 1, ηL = N 1 for k = 1,...,L HÛ /~ = sin (S44) | − i − − 2 L + 1 L + 1 1, ηk = 2, ηL = N 2 for k = 1,...,L 1, and `,j,k,m,n=1 | − i − X ηk = 1, ηk+2 = 1, ηL = N 2 for k = 1,...,L 3. A π`k π`m π`n straightforward,| albeit a little− i tedious, calculation− shows sin sin sin cˆj†cˆk† cˆmcˆn. × L + 1 L + 1 L + 1 that E(1) = E(1) + E(1) , where | i | D i | U i

L 1 L (1) √N − ηk = 1, ηL = N 1 π`k π`L ED = | − i ω` sin sin (S48) | i −J(L + 1) cos π + cos πk L + 1 L + 1 kX=1 L+1 L+1 X`=1 L 1 (1) U N(N 1) − √2(2 + δk,1) ηk = 2, ηL = N 2 EU = − | − i | i 8J(L + 1) π πk p ( k=1 2 cos + cos X L+1 L+1 h L 3 i 2√N 1 ηL 2 = 1, ηL = N 1 − 2 ηk = 1, ηk+2 = 1, ηL = N 2 − | − − i | − i . (S49) π π(L 2) π πk π(k+2) − cos + cos − − 2 cos + cos + cos ) L+1 L+1 kX=1 L+1 L+1 L+1 Finally, the second-order correction to the ground-state energy is given by

(2) (0) (1) E /~ = E HˆD/~ + HˆU /~ E . (S50) h | | i Using the result above, we obtain

L 1 L L π`1k π`2k π`1L π`2L 2N − sin L+1 sin L+1 sin L+1 sin L+1 E(2)/ = ω ω ~ 2 `1 `2 − J(L + 1) cos π + cos πk kX=1 `X1=1 `X2=1 L+1 L+1 2 π U 2N(N 1) 15 + 4L(L + 2) 6 cos L+1 5 4(N 1) − + − + − (S51) 2 − 32J(L + 1)  6 cos π sin2 π cos π cos π cos 3π  L+1 L+1 L+1 L+1 − L+1 L π`L π`(L 2)  ω − N(N 1) `=1 ` sin L+1 sin L+1 − . 2 π π(L 2) − J(L + 1) P − cos L+1 + cos L+1 We are again interested in the expected value of the ground state energy E E(0) + E(1) + E(2). A straightforward ≈ 7

Figure S3. The occupation density of the periodic chain model in the position space pn` = n ρˆ` n , (a)–(c), as a function of h | | i the transmon site ` and the Fock state n , and the occupation density in the reciprocal space pnk = n ρˆk n , (d)–(f), as a | i h | | i function of the reciprocal mode k and the corresponding Fock state n for (a),(d) the localized state, (b),(e) the W state, and | i (c),(f) the superﬂuid state at the scaled hopping frequency τ = 0.05, 0.15, and 1, respectively, and at a single disorder realization 3 at the scaled disorder strength δ = 0.33 10− . The local density matrix ρˆ` is calculated by tracing over all the other sites: · ρˆ` = Tr i=` ( ψ ψ ), where ψ is the full quantum ground state, and similarly for the reciprocal space. { 6 } | i h | | i calculation shows that

2 π π 3 δ2 5 cos L+1 + 1 ε = 2τ cos − L + 1 − 4(L + 1) − 24τ(L + 1) π 2 π cos L+1 sin L+1 (S52) 2 π 1 15 + 4L(L + 2) 6 cos L+1 5 4(N 1) + − + − , 2 − 32τ(L + 1) (N 1)" 6 cos π sin2 π cos π cos π cos 3π # − L+1 L+1 L+1 L+1 − L+1

where ε = E D/~UN(N 1), τ = J/U(N 1), and A. Localised phase δ = D/U(Nh i1). The exact− ground state of− the super- ﬂuid phase is− visualized in Fig. S1(c),(f) through the occupation density in the position and the reciprocal space, Let us set the hopping frequency J to zero and consider and the comparison with the analytic approximation of the Hamiltonian Eqs. (S41), (S48) and (S49) is shown in Fig. S2(c).

Hˆ0 = HˆD + HˆU . (S53)

III. PERIODIC CHAIN The results for the ground state and ground-state energy are exactly the same as the ones obtained for an open chain in subsectionIIA, with the slight simpliﬁcation For a periodic chain, the analysis is very much similar that, since there are no boundaries in the chain, the sites to the case of an open chain, albeit simpler in general. ` = 1 and ` = L are now equivalent to other sites. 8

2 Figure S4. The overlap ψ ψ0 of the exact ground state ψ and the analytically solved ground states ψ0 of the periodic | h | i | | i | i model as a function of the scaled hopping frequency τ and disorder strength δ for (a) the localized phase ψ0 = ψ of | i | loci Eqs. (S7) and (S11), (b) the W phase ψ of Eq. (S57), and (c) the superﬂuid phase ψ of Eqs. (S64), (S70) and (S71). The | Wi | SFi overlap was computed for an periodic chain of length L = 8 with the total number of bosons N = 4, and averaged over 1000 disorder realisations

B. W phase leading to the ground state

L Let us set both the hopping frequency J and the disor- 1 E(0) = n = N . (S57) der strength D to zero and consider the Hamiltonian | i L | ` i r ` X=1 Hˆ = Hˆ . (S54) 0 U For odd N, b > 0 and the ground state depends on the parity of L. If L is even, the ground state is Like in the case of an open chain, the ground state of Hˆ0 is L-fold degenerate, with the states n = N all having ` L the same energy of Eq.(S17). | i 1 E(0) = ( 1)` n = N . (S58) The degeneracy is again lifted if J > 0. Following the | i L − | ` i r ` analysis of subsectionIIB, we need to solve Eq. (S20), X=1 ˆ where Ki are defined in Eq. (S21). Due to the absence of If L is odd, the ground state is doubly degenerate, with boundaries, the following analysis is now much simpler. the states Again, the operators Kî are of order i in J. They are diagonal for i < N and the diagonal elements vanish for L (0) 1 ` iπ`/L odd i. Unlike in an open chain, however, the diagonal E = ( 1) e± n` = N (S59) | ± i L − | i r ` elements for even i are all equal. This means that nothing X=1 is resolved until order N in perturbation theory, and no having equal energies. special cases have to be considered depending on the relative size of N and L. In Nth order, the ground state The rest of the results are identical to the ones calculated in subsectionIIB. is determined by KˆN . Neglecting the constant diagonal term which does not affect the shape of ground state, KˆN is of the form C. Effect of disorder on the W phase 0 1 0 0 0 ... 1 1 0 1 0 0 0 Disorder affects the W phase in a periodic chain in the 0 1 0 1 0 ··· 0 ··· same way as in an open chain, see subsectionIIC. b 0 0 1 0 1 0 , (S55) . ···  ......  . . . .    0 0 0 1 0 1 D. Superfluid phase 1 ··· 0 0 0 1 0  ···    Let us set the on-site interaction strength U and the where disorder strength D to zero, and consider the Hamiltonian N 1 N 1 (N 1) − N b = UN(N 1)( 1) − − τ (S56) ˆ ˆ − − (N 1)! H0 = HJ . (S60) − and τ = J/U(N 1). This is diagonal in the Fourier Since the chain is L-periodic and HˆJ couples nearest space. For even −N, the coefficient b is less than zero, neighbours with equal strength, it is natural to switch to 9 reciprocal space representation using the Fourier trans- The first-order correction to the ground state form η = N is given by Eq. (S47). | L/2 i L The only states giving a non-vanishing 1 2πi`k aˆ = exp ˆb . (S61) contribution are ηk = 1, ηL/2 = N 1 for ` √ L k | − i L k = L/2, ηL/ = N 2, ηL = 2 , and kX=1 6 | 2 − i ηk = 1, ηL/2 = N 2, ηL k = 1 for k = 1, . . . , L/2 1. | − − i − In terms of the reciprocal space creation and annihilation A straightforward calculation shows that E(1) = operators ˆb† and ˆb , the Hamiltonian can be written as (1) (1) | i k k E + E , where | D i | U i L ˆ 2πk ˆ ˆ H0/~ = 2J cos bk† bk. (S62) L L √N ηk = 1, ηL/2 = N 1 k=1 E(1) = | − i X | D i −2JL 2πk k=1 1 + cos L The total number operator is still of the same form as k=XL/2 ˆ L ˆ ˆ 6 before, N = k=1 bk† bk. L ` 2πi`k The eigenstates of Hˆ0 at fixed N are given by the tensor ω ( 1) exp (S70) P × ` − − L product states η1, . . . , ηL η1 ηL with ηk `=1 | i ≡ | i ⊗ · · · ⊗ | L i X excitations at the kth reciprocal mode and ηk = N. k=1 (1) U N(N 1) ηL/2 = N 2, ηL = 2 The corresponding eigenenergies are EU = − | − i (S71) P | i p 4JL ( 2√2 L 2πk L/2 1 E / J η . − ηk = 1, ηL/2 = N 2, ηL k = 1 η1,...,ηL ~ = 2 cos k (S63) | − − i L + 2πk . k=1 1 + cos L ) X kX=1 For simplicity, let us consider only the case of even L. Since cosine obtains its minimum at π, we see that the ground state of the system is a state completely localized Finally, the second-order correction to the ground-state to mode L/2, energy is given by

(0) E = ηL/ = N , (S64) (2) (0) (1) 2 E /~ = E HˆD/~ + HÛ /~ E . (S72) | i | i h | | i and the energy of the ground state is Using the result above, we obtain E(0)/~ = 2JN. (S65) − Note that if we were to consider the case of odd L, U 2N(N 1)(L2 1) E(2)/~ = − − (S73) the ground state would be degenerate, with the states − 48JL2 η(L 1)/2 = N and η(L+1)/2 = N having equal energies. L L L 2πik(` ` )/L | − i | i N e 1− 2 ˆ `1+`2 Let us then treat the disorder HD and the on-site ω`1 ω`2 ( 1) . ˆ ˆ − 2JL2 − 1 + cos 2πk interaction HU as small perturbations to H0. Expressed k=1 `1=1 `2=1 L k=XL/2 X X in the reciprocal space, these are given by 6 L ˆ 1 2πi`(k j)/Lˆ ˆ Calculating the expected value of the ground state energy HD/~ = ωè − bj†bk, (S66) (0) (1) (2) L E E + E + E over different realisations of ω`, `,j,k=1 X we≈ obtain L U 2πi`(m+n j k)/L Hˆ / e ˆb†ˆb†ˆb ˆb . U ~ = 2 − − j k m n −2L 1 δ2(L2 4) L2 1 `,j,k,m,nX =1 ε = 2τ − − , (S74) (S67) − − 2L − 36τL − 48τ(N 1)L2 − The first-order correction to the ground-state energy is given by where ε = E D/~UN(N 1), τ = J/U(N 1), and δ = D/U(Nh i1). − − (1) (0) (0) − E /~ = E HˆD/~ + HÛ /~ E . (S68) h | | i Figure S3 shows the visualizations of the localized phase, the W phase and the superfluid phase of the periodic A straightforward calculation yields model through the occupation density in the position L and the reciprocal space. Figure S4 shows the overlap (1) N UN(N 1) E /~ = ω` − . (S69) between the analytic approximations and the numerically L − 2L computed exact ground state for the periodic chain model. X`=1 10

Figure S5. The spatial (a),(c) and reciprocal (b),(d) inverse participation ratios of the quantum ground state of the Ps Pr disordered attractive Bose–Hubbard model as a function of the scaled hopping frequency τ and disorder strength δ. The ground states were numerically computed for an open chain (a)–(b) and a periodic chain (c)–(d) of length L = 8 with the total number of bosons N = 4, and averaged over 1000 disorder realisations. 11

IV. INVERSE PARTICIPATION RATIOS The Fig. 2(a) of the main text shows both inverse participation ratios as a function of the scaled hopping frequency τ = J/U(N 1) at two values of the scaled disorder The spatial s and reciprocal r inverse participation − ratios are deﬁnedP as P strength δ = D/U(N 1) for an open chain geometry. In Fig. S5 (a)–(b), we show− these open chain inverse partici- 1 N 2 pation ratios in the whole parameter range. The Fig. 2(c) s = L 1 , (S75) of the main text is made by combining these two as a P L 1 ψ nˆs ψ 2 − ! − m=1 | h | m| i | single ﬁgure. Figures S5 (c)–(d) show the corresponding 1 P N 2 inverse participation ratios in a periodic chain model. r = L 1 . (S76) P L 1 ψ nˆr ψ 2 − ! − m=1 | h | m| i | P

[1] J. J. Sakurai and J. Napolitano, Modern Quantum Me- 2017). chanics 2nd ed. (Cambridge University Press, Cambridge, [2] L. Losonczi, Acta Math. Hung. 60, 309 (1992).