Arxiv:1801.01995V1 [Cond-Mat.Str-El] 6 Jan 2018

Quantum Spin Superﬂuid from Bose-Einstein Condensation of Spinons in Pyrochlore Spin Ice

Imam Makhfudz Univ Lyon, ENS de Lyon, Univ Claude Bernard, Lyon, France

(Dated: January 9, 2018) Quantum spin ice in pyrochlore lattice exemplifies three dimensional frustrated spin systems.In existing studies, Bose-Einstein condensation of bosonic spinons gives rise to magnetically ordered ground state.A truly liquid quantum spin superfluid state that manifests a deconfined Higgs condensate of spinons is demonstrated.This state is shown to occur in the fully antiferromagnetic case of the non-Kramers quantum spin ice with a fluctuations-induced first-order quantum phase transition from the U(1) spin liquid to the spin superfluid.The spin superfluid density jump is obtained analytically.

Introduction.— Superfluidity discovered in the 30’s [1] nonzero spin superfluid density, due to the breaking of gives an example of manifestation of quantum mechan- U(1) boson number gauge invariance.Using variational ics at macroscopic level and for decades has fascinated energy calculation, we deduce that the quantum spin physicists not only in low-temperature physics commu- superfluid occurs next to the U(1) QSL regime of the nity but also nowadays has inspired new ideas based on fully antiferromagnetic non-Kramers doublet quantum it in such diverse fields as astrophysics and cosmology.Its spin ice.The U(1) QSL-quantum spin superfluid transi- theoretical explanation in liquid Helium 4 [2][3][4] and tion is shown clearly to be first order, marked by a finite liquid Helium 3 [5] constitutes a crowning achievement of spin superfluid density jump which we derive analytically. theoretical condensed matter physics in the 20th century, Generic Model: Bosonic Spinon-Gauge Field Hamil- accompanying that of BCS theory of superconductivity tonian.— We adopt a bosonic formalism applied to py- [6].In the context of magnetism, the interplay between rochlore quantum spin ice and apply the so-called gauge geometric frustration and quantum fluctuations can give mean-field theory (gMFT), which is a mean-field the- rise to a disordered spin state even at T = 0, which would ory that incorporates the gauge degree of freedom and otherwise be forbidden by classical physics.This quan- thus takes into account the gauge (quantum) fluctuations tum spin liquid (QSL) state was first suggested by Phil in the ground state of quantum spin ice [11].When im- Anderson in terms of resonating valence bond state in plemented to generic quantum spin ice systems of non- 1973 [7] and also made appearance in the theories of high Kramers doublet with integer spins, one obtains an ef- Tc superconductivity in copper oxides.Essentially, quan- fective Hamiltonian in terms of bosonic spinon and the tum spin liquid is a liquid state of spins, which has the gauge field operator, which takes the form of lattice quan- maximum possible symmetry.In spin systems with strong tum electrodynamics (QED) [12].Using simpler but oth- quantum fluctuations, enhanced by lattice geometry such erwise equivalent notation to that in [12], the bosonic as in frustrated spin systems [8], such quantum spin dis- QED model of quantum spin ice is given by ordered state is prevalent.Fractionalization of the spin J into emergent quantum particles and their deconfinement zz X 2 X X † −ηr +ηr HQED = Q − J± b br s s occur in quantum spin liquid [9].In 3d frustrated spin 2 r rµ ν rrµ rrν r r µ6=ν systems such as those forming a family called pyrochlore quantum spin ice where a U(1) spin liquid is argued to exist, the emergent particle manifests as bosonic spinon J ±± X X −2ηr † † ηr ηr that act like magnetic monopoles [10].Bose-Einstein con- + γµν brbrbrµ brν srr srr + h.c. 2 µ ν densation of such bosons which manifests Higgs mecha- r µ6=ν nism in confined state has so far been found to give rise arXiv:1801.01995v1 [cond-mat.str-el] 6 Jan 2018 to some form of magnetic order [11][12][13], as is the case X X −J Sz γ−ηr b†b sηr + h.c.+constant (1) also for the condensation of magnons [14][15][16]. z± rrµ µν r rν rrν In this work, we propose a different scenario where r µ6=ν Bose-Einstein condensation of bosonic spinons, which † also manifests a Higgs mechanism but in deconfined where br(br) is bosonic spinon creation (annihilation) op- state, gives rise to a spin analog of the superfluid in liq- erator defined at the position vector r of the center of uid Helium, which we refer to as quantum spin superfluid a tetrahedron of the pyrochlore lattice, rµ = r + ηreµ state.This article will show how such exotic state emerges where eµ with ηr = ±1 for the ‘up’ and ‘down’ tetrahe- in the ground state of pyrochlore spin ice.This state pre- dra respectively is the local basis vector defined at each serves the translational and spin rotational symmetries corner of the tetrahedron.The spin operators are defined by S± = b†s± b ,Sz = sz where ˜r = r + η e /2 and thus represents a liquid state of spins but yet it has ˜rµ r rrµ rµ ˜rµ rrµ µ r µ 2 marks the pyrochlore lattice sites, and γµν are con- where Re[··· ] refers to the real part of ··· [19].Intuitively 3 stants [12].The boson charge Q = η P Sz speaking, g±µ is a measure of quantum fluctuations of r r µ=0 r,r+ηreµ/2 is subject to the constraint that the total electric charge the spin at a corner of a tetrahedron of the pyrochlore Q = P Q commutes with the Hamiltonian H .In ex- lattice.Therefore, in the Ising limit g±ν = 0 and in any r r QED 2 other cases one expects g±ν 6= 0.We will interpret |Ψ| as isting quantum spin ice materials, Jzz,J±± > 0, Jz± < 0 the spin superfluid density and show soon that it gives while J± > 0 (ferromagnetic) or J± < 0 (antiferromagnetic) [17]. a mean-field ansatz which results in proper superfluid ‘Unilinear’ Gauge Mean-Field Theory:The Spin Super- properties.Clearly, only if the right hand side gives a real fluid Density and Mean-Field Ansatz.— From Eq.(1), we positive number do we have a superfluid state while oth- perform a mean-field decoupling that manifests the cre- erwise we have a U(1) QSL state.From Eq.(5) in conjunc- ation or annihilation of bosonic particles [18], to be called tion with Eq.(1), we can conclude that the spin super- ‘unilinear mean-field decoupling’ here, where we have fluid state is driven by the hopping terms J±,Jz± while single boson creation or annihilation operators rather being inhibited by the repulsive boson interaction term than bilinear products of them in the resulting mean-field J±±.This is totally consistent with the theoretical under- Hamiltonian [19].Upon the mean-field decomposition, the standing of superfluidity in liquid Helium and interacting boson and the gauge field are decoupled and the resulting boson system in general where the mobility of the bosons mean-field QED Hamiltonian can be written as [19] favors superfluidity while a repulsive interaction between them inhibits this exotic state. MF MF MF ± z HQED = HQED(b) + HQED(s , s ) (2) We will assume that the superfluid state does not break Ising symmetry, gzµ = 0, which will be shown later to be where consistent.Eq.(5) then simplifies to be MF X † ∗ 3J±± P −iθ H (b) = −J± g−µg+ν (Ψb + Ψ brν ) + × 1 J g−µg+ν e QED rµ 2 2 ± r,µ6=ν r,µ6=ν |Ψ| = h i (6) 3 J±± P −2ηr −iθ r,µ6=ν Re γµν g+µg+ν e X 2 −2ηr † ∗ which works for both ferromagnetic (J > 0) and anti- |Ψ| γµν g+µg+ν 2Ψbr + Ψ (brµ + brν ) + h.c. ± r,µ6=ν ferromagnetic (J± < 0) case [19].Eq.(6) assumes translational symmetry and thus applies only to distinguishing X ‘normal’ spin liquid (|Ψ| = 0) and superfluid spin liquid − J γ−ηr g g (Ψb† + Ψ∗b ) + h.c. (3) z± µν zµ +ν r rν (|Ψ| 6= 0).The mean-field ansatz for a quantum spin su- r,µ6=ν perfluid state is obtained by requiring each of the terms P under the sum µ6=ν on the numerator of the right hand MF ± z Jzz X z z side of Eq.(6) to be real in order to give a real value H (s , s ) = s gzν + s gzµ QED 2 rrµ rrν r,µ,ν for |Ψ|.We write g±µ = ∆ exp(±iAr,rµ ) which, using the MF ± z fact that HQED(s , s ) is a spin S = 1/2 operator and z X 2 +η −η our assumption hs i = 0, imposes ∆ = 1/2.We find that −J |Ψ| s r g + g s r 2 ± rrν −µ +ν rrµ a nontrivial ansatz which gives |Ψ | > 0 is given by r,µ6=ν

Ar,r0 = −π/2, Ar,r1 = π/2, Ar,r2 = −π/2, Ar,r3 = π/2 3 h i (7) X −2ηr 4 ηr ηr + J±± γ |Ψ| (s g+ν + s g+µ) + h.c. with J± < 0 (antiferromagnetic case) [19].The boson 2 µν rrµ rrν r,µ6=ν phase field is given by [19] θ(r) = G · r (8) X h i − J γ−ηr |Ψ|2(sz g + g sηr ) + h.c. (4) z± µν rrµ +ν zµ rrν where G is the reciprocal lattice wave vector of the r,µ6=ν dual diamond lattice, giving θ(rµ) = 2mπ, and m = plus constant, with g = hs±ηr i, g = hsz i and Ψ = 0, ±1, ±2, ±3, ··· .Using the above ansatz, we obtain ±µ rrµ zµ rrµ J hbri. |Ψ|2 = − ± (9) In order to characterize the bosonic spinon condensate, 3J±± we evaluate Ψ = |Ψ|eiθ.On the other hand, a liquid state for J± < 0.The corresponding gauge flux Φ through of spins by definition preserves translational symmetry, the hexagon of the dual diamond lattice Φ = ∇ × which amounts to spatially uniform mean field solution P A = A 0 is Φ = π using Eq.(7).The re- for |Ψ|.Taking δhHMF (b)i/δhb i = 0 gives the following rr0∈hex r,r QED rν sulting gauge flux ensures that the mean-field ansatz result preserves the symmetries of the effective Hamiltonian P J g g e−iθ + 2J Re[γ−ηr g g e−iθ] Eq.(1).With the net π flux, the spin superfluid state spec- 2 r,µ6=ν ± −µ +ν z± µν zµ +ν |Ψ| = h i ified by the ansatz Eq.(7) preserves time reversal sym- P −2ηr −iθ 3J±± r,µ6=ν Re γµν g+µg+ν e metry.Alternative ansatz that works for the ferromag-

(5) netic case (J± > 0) is exempliﬁed by Ar,r0 = 0, Ar,r1 = 3

we have

+ † + hS i = hb br+e ihs i = 0 (10) r+eµ/2 r µ r,r+eµ

† given that hbrbr+eµ i = 0 as the immediate consequence of the translational invariance of liquid state of spins, be it spin superfluid or ‘normal’ spin liquid.This, combined with the conditions |Ψ| = |hbi| = |hb†i|= 6 0, hszi = 0 self- consistently concluded earlier, implies that we find a state with Bose-Einstein condensation of bosonic spinons but FIG. 1. The 2d projection on (kx, ky) plane of the spinon which does not give a magnetic order.Instead, since the spectrum for the a) U(1) QSL and b) QSSF states at fixed state has a vanishing spin expectation value but a finite kz = 0. spin superfluid density, it gives a quantum spin superfluid (QSSF) state, a new state of matter that we find in this † work.Having hbrbr+ηreµ i = 0 representing a translational π, A = 0, A = π, corresponding to Φ = 0 flux state r,r2 r,r3 invariant state while hbri 6= 0 might be surprising but but this turns out to be energetically unfavorable. can be proven by considering boson correlation function MF ± z ± Next, we solve δhHQED(s , s )i/δhs i = 0 for [19].Furthermore, just like the QSL, the QSSF found here hs±i.Our ansatz hs±ηr i = (1/2) exp(±iA ) with A manifests a deconfined state of spinons that corresponds rrµ r,rµ r,rµ given by Eq.(7) does satisfy this extremum condition to its propagating, fluid nature [20]. MF ± z ± ‘Bilinear’ Gauge Mean-Field Theory:Spinon Disper- δhHQED(s , s )i/δhs i = 0 [19].Noting the complexity of Eq.(4), the solution of the extremum condition for sion, Variational Energy Calculation, Spin Superfluid hs±i is complex-valued in general.We suggest that a com- Properties, and Ground State Phase Diagram.— To fur- plex value for hs±i is an indicator for a liquid state of ther verify the ‘unilinear’ gauge mean-field analysis, we spins, in agreement also with the complex ansatz for perform a variational energy calculation by computing ±ηr the expectation value of the Hamiltonian Eq.(1) hHQEDi U(1) QSL where hsrr i = ∆ exp(±iAr,rµ ) with ∆ = µ following bilinear operator mean-field decomposition [12], 1/2, Ar,rµ = µQ · r with µ = (0110), Q = 2π(100),r = 2 but focusing only on the liquid states in which case (n1, n2, n3)/4 [12], which however gives |Ψ| = 0 when hb†bi = hbbi = 0 [19].From the energy dispersion (k) of substituted into Eq.(6).This is to be compared with a ± the bosonic spinon, illustrated in Fig.1, we find that the real-valued ansatz for hs i which turns out to give rise U(1) QSL and the spin superfluid state are distinguished to some type of (quadrupolar) magnetically-ordered state ± by the nature of their bosonic spinon energy spectrum [12].This complex-valued hs i is in agreement with our (k); as Fig. 1 shows, while there is in general an energy explicit results in Eq.(7) above for a liquid state of spins. MF ± z z gap between the ‘singlet’ ground state and the three-fold Finally, we solve δhHQED(s , s )i/δhs i = 0 for degenerate ‘triplet’ state (where the sublattice plays the z z hs i.We find that hs i = 0 is indeed the exact solution role of the spin) in the U(1) QSL (except on a surface of the mean-field equation when Ar,rµ is given by Eq.(7) in the 3d k space), the spectrum is gapless everywhere [19].Thus, our quantum spin superfluid state breaks no for the spin superfluid state.This extensive gaplessness z time reversal or Ising symmetry, hs i = 0, consistent with drives the spinon condensation even in the absence of our earlier assumption used in obtaining Eq.(6).In ad- translational symmetry-breaking and sublattice-mixing dition, the superfluid state corresponds to Φ = π flux particle-hole order.Furthermore, the intuitive prediction state, according to the conclusion in the previous para- from the unilinear gauge mean-field analysis regarding graph.This matches with the π-flux state realized in the the sign of J± and the associated flux that can give rise antiferromagnetic case (J± < 0) [12], which suggests that to spin superfluidity remarkably agrees fully with a rigor- the spin superfluid state is realized only in antiferromag- ous variational energy calculation [19].Our variational en- netic case, where indeed the frustration takes effect and ergy calculation decisively shows that the quantum spin broadens the regime of the stability of the liquid state of superfluid state with its π-flux ansatz for antiferromag- spins [12].As noted earlier, the mean-field ansatz in this netic case (J± < 0) has lower energy than that of the case has a gauge flux Φ = π through the hexagon of the U(1) QSL state in an appropriate regime of the coupling dual diamond lattice, which suggests a complex ground J’s space, while this does not hold for the ferromagnetic state flux pattern expected of a frustrated spin system (J± > 0) case [19]. with liquid character for its spin state. Using the energy dispersion (k), we can compute the Now, we show how a Bose-Einstein condensate of first critical velocity vc1 of the spin superfluid given by bosonic spinons can give rise to a quantum spin superfluid effective ‘phonon’ velocity vc1 = ∂(k)/∂k|k0 , where state, a new state of matter that we propose in this work, k0 = 2π(101) (and symmetry-related points) is the wave rather than some form of magnetic order [11][12][13].We vector at the global minimum of (k) [19].This nonzero noted earlier that |g | = |hs±ηr i| 6= 0.The expecta- ±µ r,r+ηreµ k0 is the origin of the deconfinement and the propagation tion value of the spin operator however vanishes because of the spinons even when they condense as concluded ear- + x y † + with S ≡ S +iS = b s br+e [12], r+eµ/2 r+eµ/2 r+eµ/2 r r,r+eµ µ lier.The flat part at the ‘crossroad’ in the spinon disper- 4

the U(1) QSL and the AFQ order while bordering the FQ order directly.We conjecture a schematic ground state phase diagram of quantum spin ice with quantum spin superfluid state in it that is illustrated in Fig.2, motivated by the result in [12].Based on Eqs.(5),(9),(11),(12) and our variational energy calculation [19], the QSSF should occur below a critical J±± line and above a critical J± 2 value with spin superfluid density |Ψ| = |J±|/(3J±±) c for |J±| > |J±| [21].There is thus a jump in the spin superfluid density |J c | δ|Ψ|2 = ± (14) 3J±± that implies a first-order quantum phase transition out of U(1) QSL induced by quantum fluctuations [23].The quantum phase transition between the QSSF and AFQ (noncoplanar FQ) occurs because as one in- crit crit creases J±(J±±) past a critical J± (J±± ), a transla- FIG. 2. The schematic ground state phase diagram of quan- tional symmetry-breaking particle-hole order parameter tum spin ice with its spin superfluid state for J± < 0 at that mixes the two sublattices of the dual diamond lat- Jz± = 0. tice (corresponding to the ‘up’ and ‘down’ tetrahedra) † develops a nonzero expectation value hbrbr+ηreµ i 6= 0, which destroys the translationally invariant spin super- sion Fig.1b) serves as the ‘roton’ that provides a second fluid state; making it a spin analog of solid, while the critical velocity vc2 = ∆(k)/∆k with ∆k = kroton − k0 Bose-Einstein condensation of bosonic spinons survives and thence vc = min[vc1, vc2].From the boson phase field hbi, hb†i= 6 0 in both phases.The key distinctions between Eq.(8), the superfluid velocity [22] is given by the U(1) QSL, QSSF, and the AFQ/FQ phases are shown in Table 1. ~ ∗ ∗ ~ ~ vsf = −i 2 (Ψ ∇Ψ − Ψ∇Ψ ) = ∇θ = G 2mb|Ψ| mb mb z ± † z ± State hs ± i hs ± i hbri hbrbri hbrbr± i hS˜r i hS˜r i (11) r,rµ r,rµ µ µ µ where mb is the effective mass of the bosonic spinon given U(1) QSL 0 6= 0 0 0 0 0 0 by QSSF 0 6= 0 6= 0 0 0 0 0 AFQ/FQ 0 6= 0 6= 0 6= 0 6= 0 0 6= 0 2 −1 ± ∂ (k) 2 where rµ = r±eµ gives the position vector of the centers mb = 2 |k0 = (12) ∂k 3J± of the tetrahedra (the sites of the dual diamond lattice) while ˜rµ = r + ηreµ/2 gives the pyrochlore lattice site The spin superflow is stable only when vsf < vc [3], and (the corner of a tetrahedron) position vector.It is to be this constrains the reciprocal lattice wave vector G in noted that while hbri is nonzero both in the QSSF and Eq.(11); only the few smallest G’s satisfying 0 < |G| < AFQ/FQ phases, this quantity is lattice-translational |Gc| where invariant in QSSF as the consequence of Eq.(8) but this |∇ (k)| is not the case in AFQ/FQ. |G | = k k0 (13) c |∇2 (k)| k k0 Discussion.— From Bose-Einstein Condensation of which is finite for finite geometry, contribute to a stable Bosonic Spinons to Spin Superfluidity.— The possibil- spin superflow current [19]. ity of Bose-Einstein condensation in spin systems is an It has been shown that as one increases the trans- actively pursued idea [16].In this case, the bosons are the verse spin coupling J± at small J±± starting from a U(1) magnons; the quanta of spin waves around a magnetically QSL state, bosonic spinons will condense and give rise ordered ground state excited by thermal fluctuations at to a translational invariance-breaking sublattice-mixing low temperatures.Their condensation normally gives rise particle-hole order, which manifests as a antiferromag- to some type of magnetic order [14][15] while its rela- netic quadrupolar (AFQ) order while increasing J±± to tion to (finite temperature) spin superfluidity is an active large enough value gives a noncoplanar ferroquadrupolar topic [24], but spin superfluidity from magnon condensa- (FQ) order [12].Our result indicates that a translational tion emerging in appropriate spintronics setup has been invariant bosonic spinon condensate state can preempt proposed (e.g. [25]).In our work, we consider conden- the QSL-AFQ transition and manifests as a quantum sation of spinons; fractionalized excitations of quantum spin superfluid state.Based on variational energy calcula- spin system [9] and find that such condensation may give tion [19] and the spin superfluid density Eq.(6), we con- rise to a new state of matter; a quantum spin superfluid, clude that the quantum spin superfluid occurs between rather than some type of magnetic order. 5

The relation between the Bose-Einstein condensa- Monte-Carlo numerical simulation of pyrochlore spin ice tion and superfluidity is itself an interesting problem at finite temperature and magnetic field along [111] di- [26][27].In superfluid Helium, it is suggested that the su- rection at J±± = Jz± = 0 [32].In their case, they found a perfluidity is caused by the Bose-Einstein condensation ‘monopole superfluid’ state, but which manifests a con- of the Helium atoms [28][29][30] but the effort in un- fined Higgs condensed state of spinons, corresponding derstanding this remains in progress [31].In our work, to an XY ferromagnet order.Interestingly though, their we note that since the Bose-Einstein condensate hbi is monopole superfluid state in the zero field limit occurs on translationally invariant and gives rise to a spin rotation- the same part of J± axis as ours and they also obtain a ally invariant state, it must correspond to a liquid state superfluid density proportional to |J⊥| ≡ |J±|.Our work of the spins, which however cannot be a ‘normal’ spin investigates analytically the case at finite J±±,Jz± at liquid, because it has some condensation of deconfined zero field and finds a truly ‘fluidic’ spin superfluid state (propagating) spinons with finite spin superfluid veloc- with no magnetic order, manifesting a deconfined state ity.We thus conclude that this state is a spin superfluid of Higgs condensed spinons.The singularity in |Ψ|2 at c crit state.This state is also found to be energetically favor- J±± = 0 noted in [21] along the |J±| < |J±| < |J± | seg- able over the ‘normal’ U(1) quantum spin liquid state in ment in Fig.2 marks a deconfinement-confinement transi- the fully antiferromagnetic non-Kramers quantum spin tion of the spinons from our QSSF state to the XY ferro- ice on an appropriate regime of the spin coupling param- magnet of [32].Our work is thus distinct but complements eter space.Other than of fundamental interest, the quan- that of [32].We also just learned that a quantum spin tum spin superfluid state if realized experimentally could superfluid state from condensation of hard-core bosons open up a prospect for dissipationless spin transport ap- rather than spinons was found in pyrochlore using quan- plications, even though advanced low temperature cool- tum Monte-Carlo simulation a while ago [33] which also ing technology might have to be employed.We conclude indicates a first order superfluid transition.These works c the story by reflecting that the system of charged bosons estimate that |J±| ' 0.104Jzz.A spin superfluid state in coupled to (emergent) gauge field as realized in quantum 2d spin system (kagoméspin 1/2 XY model) was also spin ice is a fascinating model and may perhaps be re- deduced from quantum Monte-Carlo simulation, but realized in other platforms, such as ultracold atom system quires ring-exchange interaction and has a second-order where the physics discussed in this work could probably superfluid phase transition within XY universality class be investigated. [34]. Note added: While we were finishing this manuscript, Acknowledgements.—I.M. would like to thank Prof. I. we became aware of a recent work based on quantum Lukyanchuk for the endless encouragement all along.

[1] Kapitza, P., Nature, 141, 74 (1938); Allen, J. F., and [17] K. A. Ross, L. Savary, B. D. Gaulin, and L. Balents, Misener, A. D., Nature, 141, 75 (1938). Phys. Rev. X 1, 021002 (2011). [2] L. Tisza, Phys. Rev. 72, 838 (1947). [18] S. Sachdev, Quantum Phase Transitions (Cambridge [3] L. Landau, Phys. Rev. 60, 356 (1941); J. Phys. USSR, University Press, Cambridge, England (2013)). Vol.5 (1941) p.71. [19] Supplementary Material. [4] Ginzburg, V. L. and L. P. Pitaevskii, Zh. Eksp. Teor. Fiz. [20] The possibility of having a deconfined Higgs condensate 34, 1240 (1958); [Sov. Phys. JETP 7, 858 (1958)]. of bosons coupled to gauge field is explained in [19]. c [5] A. J. Leggett, Rev. Mod. Phys. 47, 331 (1975). [21] The value of J± can only be determined from a numerical [6] J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. self-consistent calculation on a system with finite volume 2 Rev. I08, 1175 (1957). [12].The apparent singularity of |Ψ| at J±± = 0 indicates [7] P. W. Anderson, Mater. Res. Bull. 8, 153 (1973). a transition into a new state at J±± = 0 (Please see also [8] L. Balents, Nature (London) 464, 199 (2010). Note added). [9] L. Balents, M. P. A. Fisher, and S. M. Girvin, Phys. Rev. [22] L. D. Landau and E. M. Lifshitz, Statistical Physics Part B 65, 224412 (2002). 2 (Pergamon Press, Oxford, England (1976)). [10] M. Hermele, M. P. A. Fisher, and L. Balents, Phys. Rev. [23] I. Makhfudz, Phys. Rev. B 89, 024401 (2014). B 69, 064404 (2004). [24] E.B. Sonin, Advances in Physics, Vol. 59, No. 3, May- [11] L. Savary and L. Balents, Phys. Rev. Lett. 108, 037202 June 2010, 181-255. (2012). [25] B. Flebus, S.A. Bender, Y. Tserkovnyak, and R.A. Duine, [12] S. B. Lee, S. Onoda, and L. Balents, Phys. Rev. B 86, Phys. Rev. Lett. 116, 117201 (2016). 104412 (2012). [26] A. J. Leggett, Rev. Mod. Phys. 71, S318 (1999). [13] G. Chen, Phys. Rev. B 94, 205107 (2016). [27] C. N. Yang, Rev. Mod. Phys. 34, 694 (1962). [14] T. Matsubara and H. Matsuda, Prog. Theor. Phys. 16, [28] F. London, Phys. Rev. 54, 947 (1938). 569 (1956). [29] N. N. Bogoliubov, J. Phys. (USSR) 11, 23 (1947). [15] T. Nikuni, M. Oshikawa, A. Oosawa, and H. Tanaka, [30] A. J. Leggett, Phys. Rev. Lett. 25, 1543 (1970); J. Stat. Phys. Rev. Lett. 84, 5868 (2000). Phys. 93, 927 (1998). [16] T. Giamarchi, C. Ruegg and O. Tchernyshyov, Nature [31] Bose-Einstein Condensation, edited by A. Griffin, D.W. Physics 4, 198-204 (2008). Snoke, and S. Stringari (Cambridge University Press, 6

Cambridge, England (1996)). S. V. Isakov, K. Damle, and Y. B. Kim, Phys. Rev. Lett. [32] Y. Kato and S. Onoda, Phys. Rev. Lett. 115, 077202 100, 047208 (2008). (2015); T. A. Bojesen and S. Onoda, Phys. Rev. Lett. [34] L. Dang, S. Inglis, and R. G. Melko, Phys. Rev. B 84, 119, 227204 (2017). 132409 (2011). [33] S. V. Isakov, S. Wessel, R. G. Melko, K. Sengupta, and Y. B. Kim, Phys. Rev. Lett. 97, 147202 (2006); A. Banerjee,

Supplementary Material: Quantum Spin Superﬂuid from Bose-Einstein Condensation of Spinons in Pyrochlore Spin Ice Imam Makhfudz Univ Lyon, ENS de Lyon, Univ Claude Bernard, Lyon, France

In this Supplementary Material, we first discuss some subtle points regarding the local gauge invariance in our system and the possibility of having a deconfined Higgs condensed state of bosons coupled to gauge field, then show how translational invariance can be preserved in a condensed state of spinons, and finally give some details on mean-field and variational energy calculations.

I. LOCAL GAUGE INVARIANCE IN SYSTEMS spinons), we perform a mean-field decomposition that ex- WITH NON-LOCAL INTERACTIONS AND THE plicitly breaks the U(1) symmetry [2].This gives rise to DECONFINED HIGGS CONDENSATE OF a non-conservation of local (i.e. at each site) as well as BOSONIC SPINONS total numbers of bosons.The total charge of bosons is however conserved because the bosons are created or an- A subtle point to note regarding the effective Hamilto- nihilated in pairs of opposite charges.This is indeed the nian for bosonic spinon coupled to gauge field in Eq.(1) case in quantum spin ice where flipping a spin at the cor- in the main text is that the J±± term at zeroth order ner of a tetrahedron of the pyrochlore lattice creates a level (hs+ i = 1 + O(A )) essentially corresponds to pair of bosonic spinons (or magnetic monopoles) of op- rrµ rrµ a non-local boson-boson interaction because the boson posite signs.The spin superfluid instability thus breaks operators are defined at three generally different sites; U(1) boson number conservation at both local (gauge) the centers of three adjacent tetrahedra in a pyrochlore and global levels but it conserves the total charge of the lattice.Physically, this J±± term flips two spins at two bosons; i.e., it preserves the U(1) boson charge gauge in- corners of a tetrahedron both up (++) or both down variance.The presence of nonlocal boson interaction and (−−) with respect to local spin vector basis defined the preservation of the U(1) boson charge gauge invari- at each corner.If we start from a state with no mag- ance turn out to play role in the deconfinement of the netic monopoles, this process translates into a creation bosonic spinons, as explained below. of two magnetic monopoles of positive charge at the cen- In the existing studies of gauge theories where a bo- ter of tetrahedron located at r and two negative charged son is coupled to gauge field, it is argued that there oc- monopoles at neighboring tetrahedra centers located at cur Higgs condensed state and confined state of bosons r + ηreµ and r + ηreν .This boson ‘scattering’ process is [3].Both of these phases are the mechanisms to remove equivalent to a nonlocal boson interaction.In the origi- the gauge field from low energy theory by gapping it nal lattice spin model, this interaction originates from out.It is suggested that these two phases are continu- the short-range nearest-neighbor spin exchange interac- ously connected [3] when the Higgs fields (bosons) are tion, but it translates into nonlocal interaction between in the fundamental representation and all the gauge in- the bosonic spinons, despite the constant interaction cou- variance has been broken.This is despite the fact that pling strength J±±. they describe two different phenomena; Higgs condensa- As a result, the Elitzur theorem [1], which permits only tion gaps out the gauge field by condensing the bosons, locally gauge invariant quantity to acquire nonzero ex- e.g. by Bose-Einstein condensation, whereas confined pectation value does not apply in this system, because phase refers to the confinement of the bosons.Although the theorem assumes local interaction.This is true even it sounds very sensible to think that condensed bosons at gauge theory level and one does not need to resort are also confined, the two phenomena still have distinct to mean-field theory for which the local gauge invari- physical meaning; it is not impossible to have condensed ance is broken explicitly and the Elitzur theorem ceases bosons that are not confined, that is, the bosons are de- to apply.As a consequence, this permits a ground state confined so that they can still move around.BCS theory with hbi 6= 0 corresponding to condensation of bosonic superconductivity [4] gives a good example of boson con- spinons.Whether such state prevails or not can be de- densation where the bosons (the Cooper pairs) are not termined from variational energy calculation.To find an confined at all, otherwise we will not get a superconduct- instability of U(1) quantum spin liquid towards a spin ing state.There are at least two justifications for having a superfluid state (that is, the superfluid state of bosonic deconfined Higgs condensed boson state.First, if there is 7 still unbroken gauge invariance.In our system, the U(1) is clear that the contributions from all other symmetry- boson number gauge invariance is broken corresponding related points cancel out each other but we are still left to the creation or annihilation of bosonic spinons but with the ‘zero mode’ contribution from k0 that results † the U(1) boson charge gauge invariance is still preserved, in hbrbr0 i= 6 0.However, if k0 6= 0 as is the case with our since the bosonic spinons are created or annihilated in QSSF state (k0 = 2π(101)), there is no ‘zero mode’ con- pairs of opposite charges, so that their total charge is tribution and all the terms in the sum cancel out each † still conserved.Second, the presence of nonlocal, long- other to zero; hbrbr0 i = 0.This can also be verified explic- 0 range interaction, as alluded to above, which was not itly by substituting k0 = 2π(101) and r − r = eµ which 0 considered in [3].This nonlocal interaction is produced gives exp(ik0 ·(r −r)) = ±1 and yields exact cancellation by the J±± term which when it is nonzero, as we con- to zero upon summation.This proves our assertion that cluded in the main text, gives rise to the deconfined state we can have translational invariant state characterized † of spinons corresponding to the quantum spin superfluid by hbrbr+ηreµ i = 0 even with hbri= 6 0.This conclusion state that we find, but once it is zero, the resulting state remains valid even if we do not have total condensation, is a confined ‘monopole superfluid’ which corresponds to as long as the condensed bosons are still distributed uni- 0 ordinary magnetic order (XY ferromagnet) as found in formly among the global minima k0s (which is a valid as- [5]. sumption in the absence of time reversal or parity break- † ing), we still have hbrbr0 i = 0 for k0 6= 0.The sum of the contributions from the uncondensed bosons vanishes due II. TRANSLATIONAL INVARIANCE IN A to the random phase assumption. CONDENSED STATE OF SPINONS

† Consider boson correlation hbrbr0 i.Taking its Fourier III. GAUGE MEAN-FIELD THEORY transform, we have EQUATIONS

† 1 X † i(k0·r0−k·r) A. Mean-Field Decoupling hb b 0 i = hb b 0 ie r r N k k k,k0 From Eq.(1) in the main text, we perform the following ‘unilinear’ mean-field decomposition that stresses the 1 X 0 = hb† b ieiq·rei(k+q)·(r −r) (15) U(1)-breaking character of a superfluid state [2]: N k k+q k,q sz sz → sz hsz i + hsz isz − hsz ihsz i (18) rrµ rrν rrµ rrν rrµ rrν rrµ rrν Now, consider the condensed boson state with hbri = |Ψ| exp(iG · r) that we have found.This hbri is (lattice) s+ s− b† b → hs+ ihs− i(b† hb i+hb† ib −hb† ihb i) translational invariant (in the dual diamond lattice).A rrν rrµ rµ rν rrν rrµ rµ rν rµ rν rµ rν translational invariant state, by definition, has no density wave order in terms of the above boson correlation.That † † + (s+ hs− i + hs+ is− − hs+ ihs− i)hb† ihb i (19) is hbkbk+qi = hbkbkiδ(q) so that rrν rrµ rrν rrµ rrν rrµ rµ rν

† 1 X † ik·(r0−r) hbrbr0 i = hbkbkie (16) + + † † N s s b b br br → k rrµ rrν r r µ ν Suppose that at the ground state (T = 0) we can have hs+ ihs+ i 6b†hb†ihb ihb i + 3hb†i2(b hb i + b hb i) a total condensation (this is indeed the case in ideal Bose rrµ rrν r r rµ rν r rµ rν rν rµ gas) and that the boson dispersion has periodic lattice of global minima located at k0 and the symmetry-related +3hb†i2hb ihb i(s+ hs+ i + s+ hs+ i) points (as is the case in our work, c.f. Fig. 1 in the r rµ rν rrµ rrν rrν rrµ main text), such that all the bosons are condensed at those wave vectors.Then, the boson correlation function − 9hb†i2hb ihb ihs+ ihs+ i (20) becomes r rµ rν rrµ rrν

1 0 † X † ik0·(r −r) 0 hbrbr i = hb bk0 ie (17) N k0 z ± † z ± † † 0 S s b b → hS ihs i(b hb i + hb ib ) k0s rrµ rrν r rν rrµ rrν r rν r rν

0 where k0s denotes k0 and all the wave vectors symmetry related to k .It is also intuitively clear to expect that +hb†ihb i(Sz hs± i + hSz is± ) 0 r rν rrµ rrν rrµ rrν † hbkbki is the same for all those wave vectors, since their dispersions are precisely identical to each other.Now, if − 2hSz ihs± ihb†ihb i (21) k0 = 0, using the standard random phase assumption, it rrµ rrν r rν 8

−2ηr 2 −ηr where we have single boson creation or annihilation oper- 3J±±Re γµν |Ψ| g+µ − 2Jz±Re γµν gzµ (26) ators rather than bilinear products of them.It is also to be † MF ± z ± noted that hb i, hbi= 6 0 in general because the nonlocal Setting δhHQED(s , s )i/δhs i = 0, the solution for boson J±± interaction annuls the applicability of Elitzur g = hs± i is complex-valued in general, as stated ±µ rrµ theorem.If we further assume that the boson expecta- in the main text.Indeed, this solution is complex-valued tion value does not break translational symmetry, we can in any liquid state of spins; in both the U(1) quantum simplify the mean-field decomposition further.Denoting spin liquid and quantum spin superfluid discussed in this † † hbi = Ψ, hb i = Ψ , we get work. + − † + − † ∗ 2 From the gauge mean-field decoupled Hamilto- s s b br → hs ihs i(b Ψ + Ψ br − |Ψ| ) rrν rrµ rµ ν rrν rrµ rµ ν nian in Eqs.(2-4) in the main text, we take MF ± z z δhHQED(s , s )i/δhs i. We obtain + (s+ hs− i + hs+ is− − hs+ ihs− i)|Ψ|2 (22) rrν rrµ rrν rrµ rrν rrµ δhHMF (s±, sz)i QED = s+ s+ b†b†b b → δhsz i rrµ rrν r r rµ rν rrµ

+ + † 2 2 ∗ hs ihs i 6b |Ψ| Ψ + 3|Ψ| Ψ (br + br ) rrµ rrν r µ ν X 2 X −ηr Jzz gzν − 2Jz±|Ψ| Re γµν g+ν (27) r,ν r,ν6=µ +3|Ψ|4(s+ hs+ i+s+ hs+ i)−9|Ψ|4hs+ ihs+ i (23) rrµ rrν rrν rrµ rrµ rrν We then substitute our gauge mean-field ansatz given in Eq.(7) in the main text.We find that, remarkably, for Sz s± b†b → hSz ihs± i(b†Ψ + Ψ∗b ) rrµ rrν r rν rrµ rrν r rν P the ansatz Eq.(7) in the main text, the sum µ6=ν in the second term in Eq.(27) above (that multiplies Jz±) +|Ψ|2(Sz hs± i+hSz is± )−2hSz ihs± i|Ψ|2 (24) rrµ rrν rrµ rrν rrµ rrν vanishes identically.As a result, the extremum condition MF ± z z z δhH (s , s )i/δhs i = 0 implies that gzµ = hs i = 0 Applying the above decoupling to Eq.(1) in the main QED rrµ (if we impose uniform gzν to minimize the Ising energy text, we obtain Eqs.(2-4). 2 term ∼ gzν ), regardless of Jz±.As far as perturbative regime is concerned, no Ising order (where hszi= 6 0) B. Gauge Mean-Field Equation in the Matter has been found in the existing works that employ the Sector: Derivation of the Spin Superfluid Density generic model [7][6], thus corroborating the correctness of our mean-field analysis result above.Such Ising order is argued to occur in simpler XXZ model where only J From the gauge mean-field decoupled Hamiltonian in zz and J are taken into account [8].The logical explanation Eqs.(2-4) in the main text, we take δhHMF (b)i/δhbi. We ± QED for this is that any finite amount of coupling J read- obtain z± ily drives strong longitudinal spin quantum fluctuations δhHMF (b)i that enforce hsz i = 0 while J reduces the transverse QED X ∗ 2 rrµ ±± = {−J±g−µg+ν Ψ + 3J±±|Ψ| × ± δhbr i spin fluctuations and makes hsrr i= 6 0.In [6] it was found ν r,µ6=ν µ that the hszi = 0 state indeed minimizes the variational energy hHQEDi for Jz± = 0 that justifies them to neglect Re γ−2ηr g g Ψ∗ − 2J Re γ−ηr g g Ψ∗} (25) µν +µ +ν z± µν zµ +ν Jz± altogether.We work under the assumption that this conclusion holds even to finite but small J /J 1 ±ηr z± zz where Ψ = hbri, g±µ = hs i = ∆ exp(±iAr,r ) and rrµ µ where we are still in the translationally invariant liquid g = hsz i.We then use the translationally invari- zµ rrµ state of spins, be it QSL or QSSF, which then protects the ant bosonic spinon condensation density ansatz Ψ = conclusions derived from Eqs.(5-6) in the main text.This MF |Ψ| exp(iθ(r)) and solve the equation δhHQED(b)i/δhbi = assumption is further supported by the observation that 2 0 for |Ψ| , which gives Eq.(5) in the main text. at finite but small Jz±/Jzz, one still has U(1) QSL at sufficiently large |J±|/Jzz [7].

C. Gauge Mean-Field Equation in the Gauge Sector IV. DEDUCTION OF THE GAUGE MEAN-FIELD LINK VARIABLE ANSATZ AND From the gauge mean-field decoupled Hamilto- SUPERFLUID PROPERTIES nian in Eqs.(2-4) in the main text, we take MF ± z ± δhHQED(s , s )i/δhs i. We obtain The phase field (average gauge field) ansatz given in Eq.(7) in the main text is derived simply by requiring δhHMF (s±, sz)i QED X 2 X 2 that the terms under the sums P on right hand ηr = −J± |Ψ| g−µ + |Ψ| × r,µ6=ν δhsrr i ν r,µ6=ν r,µ6=ν side of Eqs.(5-6) is a real quantity (whose precise sign 9 depends on whether J± > 0 (ferromagnetic case) or J± < spinon energy dispersion (k).The linear derivative term 0 (antiferromagnetic case) but overall must be positive). gives the first critical velocity of ‘phonon’ type vc1 =

From Eq.(6) in the main text, we write g±µ = ∂(k)/∂k|k0 .The second critical velocity is determined by the ‘roton’ type of excitations; local but not global ∆ exp(±iAr,rµ ) and γµν = |γµν | exp(iφµν ).Eq.(6) in the main text then becomes minima, represented by the ‘flat’ part in the spinon energy dispersion of the QSSF state Fig. 1b) in the main 2 |Ψ| = text; vc2 = ((kroton) − (k0)) /(kroton − k0).The spin superfluid superflow is stable only when vsf < vc.The actual critical velocity is the minimum of the vc1 and i(A − +A + −θ(rµ)) P r,rµ r,rν J± r,µ6=ν e vc2.Let us for illustration assume that vc1 < vc2 so that v = v , as is the usual case in superfluid Helium.The P −2ηr c c1 3J±± r,µ6=ν |γµν | cos(Ar,r+ + Ar,r+ − θ(rµ) − 2φµν ηr) µ ν condition vsf < vc turns out to constrain the reciprocal (28) lattice wave vector G; only the few smallest |G|’s con- ± where we have used ∆ = 1/2 and rµ = r ± ηreµ.We have tribute to a stable spin superflow.To derive the bound, in total five unknown variables; four average gauge field we note that while the explicit expressions for vsf and vc Arr with Arr−η e = −Arr+η e and µ = 0, 1, 2, 3 plus µ r µ r µ are delicate, they can be written as vsf = α|J±||G| (we one boson phase field θ(rµ).The first condition to solve set ~ = 1 for compactness) and vc = β|J±|, where α = for these five unknown variables is obtained by requir- 2 |∇ (k)|k /|J±| while β = |∇k(k)|k /|J±|.Substituting ing each term under the sum P in the numerator of k 0 0 µ6=ν these into vsf < vc, we obtain |G| < |Gc| where Eq.(28) to be real, we have 2 |Gc| = β/α = |∇k(k)|/|∇k(k)|k0 .The minima of (k) occur at k0 = 2π(101) modulo the reciprocal lattice A + − A + − θ(rµ) =nπ ˜ (29) r,rν r,rµ vector G, which gives mb = 2/(3J±) corresponding to α = 3/2.The critical velocity vc1 however is found to be wheren ˜ = 0, ±1, ±2, ±3, ··· and we have used A − = r,rµ zero, which originates from a cosine profile of the energy −A + .It turns out that the above equation is also the dispersion around the minima, so that the linear com- r,rµ MF ± z ± ponent is zero.This latter point is valid only for infinite condition for the δhHQED(s , s )i/δhs i = 0 to be satis- fied and is therefore mandatory.The second condition is system; for finite system, e.g. in a thick slab or thin film fixing the boson phase field to be geometry normal to the h111i direction for example, the momentum space will have an infrared cutoff δk and the

θ(rµ) = G · rµ critical velocity is not defined at k0, but at k0 ± δk and this will give a nonzero critical velocity vc1.This is really where G is the reciprocal lattice wave vector of the reminiscent of the Berezinskii-Kosterlitz-Thouless theory dual diamond lattice.This gives G · rµ = 2mπ with that explains the occurrence of superfluidity in thin film m = 0, ±1, ±2, ±3, ··· and of liquid helium despite the absence of long-range order [9]. A + − A + = nπ (30) r,rν r,rµ where n =n ˜ + m = 0, ±1, ±2, ±3, ··· .A simple but yet V. VARIATIONAL ENERGY CALCULATION nontrivial ansatz that satisfies Eq.(30) is given in Eq.(7) in the main text.Substituting the ansatz into Eq.(28), we The most radical feature of the quantum spin super- find that the sum in the numerator is positive whereas fluid, if one wishes, is that it breaks neither spin rota- that in the denominator is negative, which means only tional symmetry nor translational symmetry, but yet it the J < 0 (antiferromagnetic case) can give |Ψ|2 > 0, ± breaks gauge invariance associated with boson number that is, a spin superfluid state.The superfluid velocity is conservation.In the context of gMFT [7][6], this implies given by that hbi= 6 0.We will verify whether such state is energetically favorable, using the complex average gauge ~ ∗ ∗ ~ ~ vsf = −i 2 (Ψ ∇Ψ − Ψ∇Ψ ) = ∇θ = G field ansatz found in Eq.(7) in the main text.Since as 2mb|Ψ| mb mb (31) discussed in the main text the supposed spin superfluid should, if ever exists, appear next to the spin liquid, the where mb is the effective mass of the bosonic spinon.This is obtained by expanding the energy dispersion of the translational symmetry-breaking particle-hole order parameter hb†b i is supposed to vanish.In this case, the bosonic spinon, which is computed in the following sec- r r±eµ tion, to quadratic order only order parameter of interest is the link gauge field expectation value hs±i, which is given by Eq.(7) in the † 2 main text.The other order parameters hb b i and hb b ± i 1 ∂ (k) r r r rµ (k) = (k0)+∆k·∇(k)|k0 + ∆ki∆kj |k0 (32) 2 ∂ki∂kj appearing in the quadratic operator mean-field Hamil- tonian [6] are zero as well for the U(1) QSL and our 2 2 −1 which gives mb = (∂ (k)/∂k |k0 ) where k0 corre- proposed quantum spin superfluid (QSSF) state. sponds to the wave vector at the global minima of the We start from the Hamiltonian Eq.(1) in the main text 10 and perform mean-field decoupling that produces bilinear The J±± part drops out because the corresponding operator Hamiltonian [6], which we referred to as ‘bilin- particle-hole order parameter of the spinons that mix ear gauge mean-field theory’ in the main text.We focus the two sublattices and pairing order parameters of the on the bosonic spinon mean-field Hamiltonian because it spinons on the same sublattice vanish for the U(1) QSL is the bosonic spinon (variational) energy that we want and our QSSF states of our interest.The Jz± part also to minimize.The Hamiltonian is given by drops out because we have hszi = 0.We then write the action given by i(A + −A + ) Jzz X 2 J± X † rr r,r Hb = Q − b br e ν µ 2 r 4 rµ ν r r,µ6=ν

  Z ! Z Jzz J± i(A + −A + ) X X X 2 X † rrν r,rµ X 2 S[Q, b] = dτ Qr∂τ br − Hb = dτ  Qr∂τ br − Qr + br brν e + λr(|br| − 1) 2 4 µ r r r r,µ6=ν r

where the Lagrange multiplier λ term is intro- which is nicely diagonal for our QSL and QSSF 2 duced explicitly to impose the constraint |br| = states.The diagonal elements automatically are the four 1.We then integrate out the Qr in the parti- eigenvalues m, which consist of triplet degenerate eigen- R R R tion function Z = DQ Db Dλ exp(S[Q, b, λ]) = values (2,3,4(k) = M11(k)) and one nondegenerate sin- R R Db Dλ exp(S[b, λ]).The result is the effective action glet state eigenvalue (1(k) = M11(−k)).We have checked for the spinons that M11(k),M11(−k) are real-valued, as should be the case for an energy eigenvalue.It is to be noted that the   Z eigenvalue is independent of J±±.This is valid so long 1 X J± X S[b, λ] = dτ |∂ b |2 − b† b ei∆Aµν crit  τ r rµ rν  as J±± < J±± so that we are in either QSL or QSSF 2Jzz 4 r r,µ6=ν states.We find that, interestingly, while there is a gap between the singlet state and the three degenerate triplet Z states for the U(1) QSL, the four states become degen- X 2 erate in the QSSF state; that is, the singlet-triplet en- + dτ λr(|br| − 1) (33) r ergy gap vanishes in the spin superfluid state.This is an important observation.The corresponding eigenstates are where we have defined ∆Aµν = Arr+ −Ar,r+ for compact- 4-component vectors with elements that can be written ν µ m th th ness.In the saddle point approximation, the functional as φα which represents the α -element of the m eigen- integral in Z will be dominated by the saddle point value state.The diagonal-ness of the matrix M greatly simpli- P fies the calculations since φm = δm.In order to compute of λ, which we then take out of the sum r.We then α α apply the Fourier transform on the bosonic operator and the variational energy later, a compute the single-particle obtain the bosonic spinon action Green’s function of the bosonic spinon m∗ m X φα φβ G (k, ω) = (37) Z 2 αβ ω2 dωn ω X ˆ† n ˆ m 2J + λ + m(k) Sb = bk M(k) + I + λ bk (34) zz 2π 2Jzz k>0 We then sum over Matsubara frequencies, which for T = 0 becomes an integral over frequency, which can be done ˆ A A∗ B B∗ T where bk = (bk , b−k, bk , b−k) describing the bosonic using contour integral, giving spinons living at the sites of the dual diamond lattice, Z m∗ m corresponding to the centers of ‘up’ and ‘down’ tetrahe- dω X φα φβ G (k) = dra, defining the A and B sublattices respectively of the αβ ω2 2π + λ + m(k) diamond lattice.The 4 × 4 matrix M given by m 2Jzz

M11(k) 0 0 0  s X Jzz m∗ m  0 M11(k) 0 0  = φα φβ (38) M(k) = (35) 2(λ + m(k))  0 0 M11(k) 0  m 0 0 0 M (−k) 11 The variational energy of interest is given by where Jzz X X X hH i = hQ2i − J hb† b ihs−ηr ihs+ηr i QED r ± rµ rν rrµ rrν X − + −ik·(e −e ) 2 M (k) = −J hs ihs ie µ ν (36) r r µ6=ν 11 ± rrµ rrν µ6=ν (39) 11

2 The task now is to compute the expectation values hQri and hb† b i in Eq.(39) using the spinon single parti- rµ rν cle Green’s function Eq.(38).Note that the hb† b i is rµ rν a particle-hole order parameter between spinons on the same sublattice, and is nonzero even for the U(1) QSL 2 and the QSSF under consideration.To compute hQri, we note that br and Qr satisfy a commutation relation † [ϕr,Qr] = i where br = exp(−iϕr) satisfying brbr = 1.Due to this commutation relation, we can view br (via ϕr) and Qr as a pair of canonically conjugate position- momentum variables x and p respectively.To be pre- x y x y cise, we write Qr = (Qr ,Qr ) and br = br + ibr .Then √ x(y) √ px(y) = JzzQr and x(y) = br/ Jzz.We then use a standard result from virial theorem or equipartition the- 2 2 orem; hp i = hx˙ i.As a consequence, FIG. 3. The variational energy hHQEDi of the U(1) QSL crit and QSSF states vs. J± with Jzz = 1,J±± < J±± ,Jz± = Jzz X 1 X 1 X 0, λQSL = 3.0, λQSSF = 3.5, and J± < 0 (antiferromagnetic). hQ2i = hp2 i + hp2i = hx˙ 2i + hy˙2i 2 r 2 x y 2 r r r

Z 2 X 1 ˙† ˙ dω X ω † = hbrbri = hbkbki 2Jzz 2π 2Jzz r k∈A,B

Z dω X ω2 = (G11(k, ω) + G33(k, ω)) 2π 2Jzz k∈A,B

1 X X = p2J (λ + (k)) (φm∗φm + φm∗φm) 2 zz m 1 1 3 3 k m (40) On the other hand, using Eqs.(36) and (38), we have X X X FIG. 4. The variational energy hHQEDi of the U(1) QSL −J hb† b ihs−ηr ihs+ηr i = M (k)hb† b i and QSSF states vs. J with J = 1,J < J crit,J = ± rµ rν rrµ rrν 11 k k ± zz ±± ±± z± r µ6=ν k 0, λQSL = 3.0, λQSSF = 3.5, and J± > 0 (ferromagnetic).

s Jzz X X m∗ m m∗ m Ar,r = 0, Ar,r = π, Ar,r = 0, Ar,r = π corresponding = M11(k)(φ1 φ1 + φ3 φ3 ) (41) 0 1 2 3 λ + m(k) k m to zero flux, one will find that the QSSF state has consistently higher energy than the QSL state, as shown in Fig. m m Using φα = δα and substituting it into Eqs.(40) and (4).This means the QSSF state does not occur for fer- (41) in conjunction with Eq.(39), we obtain romagnetic case (J± > 0).The quantum spin superfluid state thus exists only in the antiferromagnetic quantum X 1 JzzM11(k) hH i = p2J (λ + (k))+ spin ice, in precise agreement with the mean-field anstaz QED 2 zz m p k,m=1,3 2Jzz(λ + m(k)) and flux analysis presented in the main text.This result (42) is the variational energy calculation basis for our pro- The result for the variational energy hHQEDi as func- posed ground state phase diagram given in Fig. (2) in the tion of J± at fixed Jzz,J±± and λ for the U(1) QSL main text with the quantum spin superfluid state in it.In and the QSSF states using the ansatz given in Eq.(7) this ground state phase diagram, the straight line of the in the main text, applicable to the antiferromagnetic phase boundary in the QSL-QSSF transition arises from (J± < 0) case, is shown in Fig. (3).We clearly see that the J±±-independence of the variational energy hHQEDi the quantum spin superfluid has lower energy than the of both liquid states due to the absence of both pairing U(1) quantum spin liquid state for large enough |J±| in and sublattice-mixing particle-hole orders.However, the this antiferromagnetic J± < 0 case.The crossing point spin superfluid jump decreases as one increases J±±. marks a quantum phase transition from U(1) QSL state As supplementary note, in obtaining Figs.(3) and (4), at small |J±| to QSSF state at larger |J±| for this an- we have chosen λQSSF > λQSL because the spinon con- tiferromagnetic (J± < 0) case.Crucially, when one re- densation in QSSF state raises its Lagrange multiplier, 0 2 peats the calculations using the ferromagnetic ansatz according to λ = λ0 + λ /Nu.c. where λ0 is for the 12 case without spinon condensation (as in QSL), λ0 is a from a self-consistent variational calculation on a finite 0 coupling-dependent parameter, and Nu.c is the number system and λ then depends on the system size Nu.c.The c of unit cells [6].The λ0 is determined by the minimum resulting critical |J±| ' 0.4Jzz marking the QSL-QSSF of the eigenvalue (k) which we find to give λ0 = 3 in transition as illustrated in Fig.(3) turns out to be of the 0 units where Jzz = 1.But, the λ has to be determined same order of magnitude as that from numerical quantum c Monte-Carlo result which gives |J±| ' 0.104Jzz [5][10].

[1] S. Elitzur, Phys. Rev. D, 12, 3978 (1975). [7] L. Savary and L. Balents, Phys. Rev. Lett. 108, 037202 [2] S. Sachdev, Quantum Phase Transitions (Cambridge (2012). University Press, Cambridge, England (2013)). [8] G. Chen, Phys. Rev. B 94, 205107 (2016). [3] E. H. Fradkin and S. H. Shenker, Phys. Rev. D 19, 3682 [9] Berezinskii V. L., Sov. Phys. JETP 32, 493 (1971); Sov. (1979). Phys. JETP 34, 610 (1972); Kosterlitz J. M., Thouless [4] J. Bardeen, L. N. Cooper, and J. R. Schrieﬀer, Phys. D. J., J. Phys. C 6, 1181 (1973). Rev. I08, 1175 (1957). [10] S. V. Isakov, S. Wessel, R. G. Melko, K. Sengupta, and Y. [5] Y. Kato and S. Onoda, Phys. Rev. Lett. 115, 077202 B. Kim, Phys. Rev. Lett. 97, 147202 (2006); A. Banerjee, (2015); T. A. Bojesen and S. Onoda, Phys. Rev. Lett. S. V. Isakov, K. Damle, and Y. B. Kim, Phys. Rev. Lett. 119, 227204 (2017). 100, 047208 (2008). [6] S. B. Lee, S. Onoda, and L. Balents, Phys. Rev. B 86, 104412 (2012).