Quantum Spin Ice: A Search for Gapless Quantum Spin Liquids in PyrochloreMagnets

M. J. P. Gingras ∗1, 2, 3 and P. A. McClarty4, 5

1Department of Physics and Astronomy, University of Waterloo,200 University Avenue West, Waterloo, ON, N2L 3G1, Canada.

2Perimeter Institute for Theoretical Physics, 31 Caroline North, Waterloo, Ontario, N2L-2Y5, Canada3Canadian Institute for Advanced Research, Toronto, Ontario, Canada, M5G 1Z8

4Max Planck Institute for the Physics of Complex Systems,Nothnitzer Strasse 38, Dresden, 01187, Germany.

5ISIS Neutron and Muon Source, STFC, Rutherford Appleton Laboratory, Harwell Oxford, Didcot OX11 0QX, UK.

The spin ice materials, including Ho2Ti2O7 and Dy2Ti2O7, are rare earth pyrochlore magnetswhich, at low temperatures, enter a constrained paramagnetic state with an emergent gauge free-dom. Remarkably, the spin ices provide one of very few experimentally realised examples of frac-tionalization because their elementary excitations can be regarded as magnetic monopoles and, oversome temperature range, the spin ice materials are best described as liquids of these emergentcharges. In the presence of quantum fluctuations, one can obtain, in principle, a quantum spinliquid descended from the classical spin ice state characterised by emergent photon-like excitations.Whereas in classical spin ices the excitations are akin to electrostatic charges, in the quantum spinliquid these charges interact through a dynamic and emergent electromagnetic field. In this review,we describe the latest developments in the study of such a quantum spin ice, focussing on the spinliquid phenomenology and the kinds of materials where such a phase might be found.

PACS numbers:

Contents

I. Introduction 1

II. Spin Ice 2A. Classical spin ice 2B. Naming Conventions 6

III. Quantum Spin Ice 6A. From a spin model to loops 7B. From loops to a gauge theory 10C. Quantum liquid: excitations and

phenomenology 12D. Stability of the U(1) liquid 13E. Naturalness of the U(1) liquid 13F. A broader perspective on quantum spin ice 15

IV. A Materials Perspective 16A. General considerations 16B. Candidate materials 19

1. Tb2Ti2O7 192. Pr2Sn2O7 & Pr2Zr2O7 203. Yb2Ti2O7 22

V. Conclusion 23

VI. Acknowledgements 24

References 24

∗gingras@uwaterloo.ca

I. INTRODUCTION

Leaving aside molecular magnets, magnetic chains andlayered magnets, there are many thousands of magneticmaterials known to us. These typically exhibit a lowtemperature phase with some long-range ordered mag-netic structure which, no matter how complicated, canbe inferred in principle from the Bragg scattering of neu-trons. Their elementary excitations, which are calledmagnons, are the normal modes of the coupled mag-netic moments. They are bosonic quasiparticles and,possess gapless modes − so-called Goldstone modes −if the moments possess a continuous global symmetry.Suppose one found, in a neutron scattering experimenton a clean cubic magnet, an absence of Bragg peaks wellbelow the Curie-Weiss temperature co-existing with lin-early dispersing excitations while heat capacity measure-ments gave no indications of a phase transition. A re-markable possibility is that such a magnet might haveno symmetry-broken order at all and that the magneticexcitations above the ground state behave like chargedparticles interacting with linearly dispersing radiation.It is the purpose of this review is to explain how thispossibility might be realized in pyrochlore magnets.

Such an unusual state of matter is one of many possibletypes of quantum spin liquid − so named because quan-tum fluctuations are responsible for keeping the spinsfrom entering a long-range ordered phase characterizedby some broken symmetry even at zero temperature.Quantum spin liquids are to be contrasted with ordinarymolecular liquids, which have only short-range order, andsuperfluid phases, which are symmetry broken phases.They are also to be contrasted with conventional param-

arX

iv:1

311.

1817

v1 [

cond

-mat

.str

-el]

7 N

ov 2

013

2

agnets and also with collective paramagnets, or so-calledclassical spin liquids, which are finite temperature statesexhibiting nontrivial short-range correlations. Quantumspin liquids are quite different: in common with frac-tional quantum Hall liquids, they appear disordered tolocal probes but they have some form of order which isinstead “encoded” nonlocally and which is often not char-acterizable in terms of symmetry. Not only are the nat-ural characterising observables in quantum spin liquidsnonlocal but they are often independent of anything butthe topology of the nonlocal observable - such spin liquidsare said to be “topologically ordered”. For pragmatists,an essential feature of quantum spin liquids is that theyare quantum phases exhibiting peculiar “fractionalized”excitations meaning that the microscopic degrees of free-dom are, for practical purposes, split into parts by thestrong correlation or, more precisely, as a consequence ofhaving long-range quantum entanglement in the groundstate.189

Until the mid-80’s, condensed matter physicists hadbeen able to understand a huge variety of different phasesof matter − indeed essentially all known states of matter− in terms of symmetry and symmetry breaking. Thediscovery of the fractional quantum Hall effect in 1986alerted condensed matter physicists to the importanceof radically new concepts underlying the organisation ofmatter at low energies. Quantum spin liquids have beenthe main medium through which theorists have been ableto generalise the physics of the fractional quantum Halleffect and the physics of low dimensional magnets andthere has been immense progress in the understanding ofthese states of matter over the last thirty years. Yet notheoretical tool exists that will allow people to determineall possible ways in which matter can organise itself andfor really ground-breaking insights we rely on guidancefrom experiment. It is for this reason that the experimen-tal realisation of quantum spin liquids has been eagerlyanticipated in the community2. One strategy to uncoverthese states of matter has been to explore magnets withmagnetic ions sitting on lattices of corner-sharing trian-gles and tetrahedra. Antiferromagnetic exchange cou-plings between the ions are highly frustrated on theselattices such that any transition temperature occurs at ascale much lower than the Curie-Weiss temperature scale.In this way, one might hope that conventional long-rangeorder is evaded entirely. A deeper reason for exploringgeometrically frustrated magnets is that the semiclassicalground states are typically subject to a local constraintthat can be interpreted as an emergent gauge invariance.Quantum fluctuations may then lead to a quantum spinliquid that is equivalent to a deconfined phase of a quan-tum mechanical gauge theory which, as we explain below,corresponds to a fractionalized phase.

At least the semiclassical part of this strategy is beau-tifully realized by a pair of materials which are becomingperhaps the archetypes of geometrical frustration amongreal magnetic materials: the spin ices Dy2Ti2O7 andHo2Ti2O7

3–5. A great deal of recent theoretical and ex-

perimental effort has been devoted to exploring their richbehaviour at low temperatures where they enter a col-lective paramagnetic phase characterized by distinctivemagnetic correlations that follow from a local constrainton the magnetic moments on each tetrahedron.

The existence of spin ices is a promising state of affairsfor the general research programme of evincing a quan-tum spin liquid in a magnetic material2. Whereas manyproposed models with quantum spin liquid ground statesare somewhat unphysical, as we explain below, quantumfluctuations acting on the set of spin ice states can bereasonably expected to lead to a quantum spin liquidwith gapless photon-like excitations. We call this quan-tum spin ice. Furthermore, among the relatives of spinice materials, there are a number of materials where spinice correlations exist at finite temperature and in whichquantum fluctuations appear significant. The low tem-perature phases of these materials remain to be under-stood.

This review is intended to bring together in one placean introduction to the quantum spin ice phase along witha survey of those materials among the pyrochlore mag-nets most likely to harbour such a quantum spin liquidstate. We begin, in Section II A, by reviewing some as-pects of the physics of the classical spin ice because (i)the magnetism in these materials is the precursor stateto quantum spin ice and (ii) a familiarity with the micro-scopic aspects of the pyrochlore magnets gained therebywill be invaluable in assessing the prospects for uncover-ing a quantum spin liquid among them. In Section III, wereview the arguments leading from an XXZ -like modelon the pyrochlore lattice to an effective low energy de-scription of the physics as an emergent electromagnetism.This proceeds in two mains steps − by mapping from thespin model to a quantum dimer model (Section III A)and by mapping from the dimer model to a lattice gaugetheory (Section III B). We then describe the propertiesof the quantum spin ice phase (Section III C) and dis-cuss both the stability (Section III D) and naturalness(Section III E) of this state of matter. We conclude bydescribing other contexts in which a U(1) liquid might befound in condensed matter systems (Section III F). Thenext main part of the review discusses quantum spin icefrom a materials perspective. A discussion of the rele-vant microscopic features of candidate materials in Sec-tion IV A is followed by a description of the specific fea-tures of various candidate quantum spin ices includingTb2Ti2O7 (IV B 1), Pr2M2O7 (M=Sn, Zr) (IV B 2) andYb2Ti2O7 (IV B 3).

II. SPIN ICE

A. Classical spin ice

Our discussion of the physics of spin ice begins with theproblem of classical Ising spins Szi = ±1/2 that reside onthe sites of a pyrochlore lattice of corner-sharing tetrahe-

3

dra (see. Fig. 1) and interact among themselves via anantiferromagnetic nearest-neighbour exchange couplingJ‖ > 0. This model, first considered by Anderson in a

1956 paper6, was aimed at describing the magnetic order-ing of spins on the octahedral sites of normal spinels andthe related problem of ionic ordering in inverse spinels,both systems having a pyrochlore lattice. The Ising an-tiferromagnet Hamiltonian, HI,AF, of Anderson’s modelis

HI,AF = J‖∑〈i,j〉

Szi Szj , (1)

with J‖ > 0 and where the sum is carried over thenearest-neighbour bonds of the pyrochlore lattice. An-derson found that HI,AF admits an exponentially largenumber of ground states given by the simple rule thateach “up” and “down’ tetrahedron (Fig. 1) must have avanishing net spin. That is, on each tetrahedron, twospins must be “up” and have Sz = +1/2 and two spinsmust be “down” and have Sz = −1/2. Anderson furtherrecognized that this problem is closely related to that ofhydrogen bonding in the common hexagonal (Ih) phase ofwater ice or, more precisely, its cubic (Ic) phase190, bothbeing characterized by hydrogen (proton) configurationsthat obey the two Bernal-Fowler ice rules7. The ‘secondice rule’ is the one relevant to Anderson’s model and tothe main topic of this review191. The second rule statesthat for each fourfold coordinated oxygen O2− ion, theremust two protons (H+) near it and covalently bonded tothat reference O2− ion, hence providing a hydrogen bondto two neighbouring O2− ions (hence H2O molecules).At the same time, there are two protons far, which arethemselves covalently bonded to the two other oxygenions and hence hydrogen-bonded ‘back’ to the originalreference O2− ion. In other words, for each O2− ion,there are two protons near and covalently bonded to itand two farther protons hydrogen-bonded to it3–5. Thisrule leads to an underconstrained system in regards tothe number of minimum energy proton configurations inIh and Ic and, in fact, leads to an exponentially largenumber of nearly degenerate proton configurations. Li-nus Pauling had estimated in 19358 the number of ice-rule fulfilling ground states in water ice and the resultinglow-temperature residual entropy S0, finding close agree-ment with the experimental value being determined atabout the same time by Giauque and co-workers9. Paul-ing’s reasoning for estimating S0 can be adapted to An-derson’s model in a number of closely related ways3–5.One finds S0 ∼ (NkB/2) ln(3/2), where N is the num-ber of magnetic sites on the pyrochlore lattice and kB

is the Boltzmann constant. Pauling’s estimate is accu-rate within a few percent from the more precise estimate,both for Ih

10 and the Anderson model (or, equivalently,Ic)11.

Anderson’s antiferromagnetic (AF) model with Isingspins pointing along the ±z direction (as in Fig. 1) isunrealistic since there is no reason for the spins to pre-fer the z axis over the x or y axes in a system such as

the pyrochlore lattice which has cubic symmetry. Con-sequently, that model did not much attract the atten-tion of theorists and experimentalists investigating realmagnetic materials for a long time. This changed withthe 1997 discovery by Harris, Bramwell and co-workersof frustrated ferromagnetism in the insulating Ho2Ti2O7

pyrochlore oxide12,13.

FIG. 1: The pyrochlore lattice with Ising magnetic momentson the lattice sites with a global z axis anisotropy. The mag-netic lattice can be described as a face-centered cubic spacelattice with either an “up” or “down” primitive basis. Thefigure shows a spin configuration fulfilling “2-up”/“2-down”ice rules on each tetrahedron.

In Ho2Ti2O7, the magnetic Ho3+ ions reside on a py-rochlore lattice while Ti4+ is non-magnetic. Despiteoverall ferromagnetic (FM) interactions characterizedby a positive Curie-Weiss temperature, θCW ≈ +2 K,Ho2Ti2O7 was found to not develop conventional long-range magnetic order down to at least 50 mK12,13. To ra-tionalize this surprising behaviour, Harris and co-workersput forward the following argument. Because of thestrong local crystal electric field of trigonal symmetryacting at the magnetic sites, each Ho3+ magnetic mo-ment is forced to point strictly “in” or “out” of the twotetrahedra joined by that site, a direction that is alongthe pertinent cubic [111] threefold symmetry axis thatpasses through the middle of the two site-joined tetra-hedra (see Fig. 2). This large magnetic anisotropy al-lows one to describe the orientation of a Ho3+ mag-netic moment by an effective classical Ising spin Szii ,with the zi quantization direction now the local [111]direction at site i. For a single tetrahedron with such a[111] Ising spin at each of the four corners of a tetra-hedron, and interacting with a nearest-neighbour FMcoupling, the minimum energy is one of six states withtwo spins pointing “in” and two spins pointing “out”.The “up”/“down” directions of Anderson’s AF modelin Eqn. (1) now become the “in”/“out” directions ofthe frustrated FM Ising system14,15. This point can bemade clearer by considering the following simple model

4

of interacting microscopic angular momenta Ji writtenas HFM = −Jex

∑〈i,j〉 Ji · Jj , where Jex > 0 is ferromag-

netic. With Ji constrained to point along the local zi[111] Ising direction, we have Ji = 2〈Jzi〉Szi zi with 〈Jzi〉the magnitude of the angular momentum Ji of the rare-earth ion. Since zi · zj = −1/3 on the pyrochlore lattice,

we recover the model of Eqn. (1) with J‖ = 4Jex3 〈Jzi〉2.

The ferromagnetically coupled spins forced to point alonglocal 〈111〉 directions thus become equivalent to the frus-trated AF Ising model of Anderson in Eq. (1).

However, unlike the AF model, the FM model withlocal [111] Ising spins is physical since the crystal fieldresponsible for the effective Ising nature of the magneticmoments is compatible with the cubic symmetry of thesystem (Fig. 2). On the other hand, as in the AF model,the number of ground states in this frustrated Ising fer-romagnet model is exponentially large in the system size,resulting in an extensive residual entropy, again given ap-proximately by the Pauling entropy S0. In accord withthis prediction, a number of magnetic specific heat mea-surements, and thus magnetic entropy, have confirmedthat Ho2Ti2O7

16 and Dy2Ti2O717, along with the Sn-

variants, Ho2Sn2O718 and Dy2Sn2O7

19 indeed possess alow-temperature residual entropy consistent with S0.192

Recent developments in high-pressure materials synthesishave allowed people to make Ho2Ge2O7 and Dy2Ge2O7,with these also displaying a residual entropy of S0

21.Apart from holmium and dysprosium pyrochlores, thematerial Cd2Er2Se4, spinel with magnetic erbium of py-rochlore sites, also exhibits a residual entropy S0.22

FIG. 2: The oxygen environment around a magnetic ion inR2M2O7 materials with rare earth sites shown in purple. Onedistorted cube of oxygen ions is shown with the axial [111]oxygens in red and the six transverse oxygens in beige. Thesix transverse oxygen ions are equidistant from the centralrare earth site (∼ 2.5 A), lying further from the rare earthsite than the axial oxygen ions which are at ∼ 2.2 A.

Perhaps at least as interesting as the experimental de-termination of an S0 residual entropy in these systems is

the observation, originally made by Harris et al.12, thatthe “in”/“out” Ising spins in the frustrated pyrochloreIsing ferromagnets can be seen to physically represent,or ‘map onto’, the hydrogen bonding or, more precisely,the proton displacement with respect to the midpoint be-tween two O2− ions in water ice3–5. This observation ledHarris, Bramwell and co-workers to coin the name spinice for these systems.

While the frustrated nearest-neighbour FM Isingmodel helped rationalize the thermodynamic and mag-netic properties of spin ice compounds12, it was never-theless initially found rather puzzling why these materialsshould be described by such a simple model or even pos-sess a residual low-temperature magnetic entropy equalto S0

23. While the resolution of this paradox for classi-cal spin ice may at first hand appear peripheral to thetopic of this review, it will prove of significant impor-tance and rich in physical insight when we later discussthe low-energy excitations in quantum spin ice.

In the Ho and Dy based spin ice compounds, the mag-netic moment µ of the Ho3+ and Dy3+ ions is of theorder of µ ∼ 10 µB

24. Considering a nearest-neighbourdistance rnn ∼ 3.6 A in (Ho,Dy)2(Ti,Sn,Ge)2O7 spinices, one finds the scale for the magnetostatic dipolarinteractions among nearest neighbours to be D ∼ 1.4K23,24. We thus have D ∼ |θCW| in these systems anddipolar interactions must therefore be considered care-fully at the outset when discussing spin ice physics inreal materials23. Crucially, magnetostatic dipole-dipoleinteractions display two key features that would seemto make them antagonistic to the formation of a de-generate low-temperature state with extensive entropy.Firstly, they are very long-ranged, decaying as 1/|rij |3with the separation distance |rij | between magnetic mo-ments. Secondly, they strongly couple the direction of themagnetic moments, µi and µj , at position ri and rj , re-spectively, with the relative position vector rij ≡ rj − ri.Both properties would naively appear to dramaticallyconstrain admissible minimum energy orientations of themoments well beyond the ice rule imposed by the nearest-neighbour model of Eqn. (1). Interestingly, one findsthat dipolar interactions between local 〈111〉 Ising mo-ments truncated at the nearest-neighbour distance areferromagnetic-like, as in the Harris et al. frustrated Isingmodel12,14,15. This dipolar spin ice model (DSIM)23, andits subsequent generalization, including additional short-range Ising exchange interactions25,26, has proven highlysuccessful at explaining quantitatively a wide variety ofbehaviours displayed by spin ice compounds. Perhapsmost noteworthy about the DSIM is the observation, firstmade through Monte Carlo simulations23, that the long-range part of the dipolar interactions beyond the nearestneighbour distance appears “self-screened”27. That is,they only lead to a transition to long-range order at atemperature Tc ∼ 0.07D/kB

28,29. This critical temper-ature is much lower than the temperature TSI ∼ D/kB

at which the system crosses over from the trivial param-agnetic state to the spin ice regime, characterized by a

5

fulfilment of the ice rules, and marked by a broad peak inthe magnetic specific heat17,23.193 This behaviour origi-nates from a remarkable feature of the anisotropic natureof the dipolar interactions on the pyrochlore lattice23,27.Namely, on this lattice, magnetostatic dipole-dipole in-teractions differ very little, and only at short distance,from a model Hamiltonian with nearest neighbour inter-actions with the same ground state degeneracy as thatprescribed by the ice rules30. Of the two formulationsexplaining that phenomenology30–32, the one that we dis-cuss next is the most relevant to the topic of this review.

Perhaps the formulation that has been most simplyable to capture in one sweep the key features of the DSIMis the so-called dumbbell model of Castelnovo, Moessnerand Sondhi31. In essence, the dumbbell model is a wayof visualizing a multipole expansion about diamond lat-tice sites. It takes for a start the point-like magneticdipoles of the DSIM and fattens them up into a rod withtwo magnetic charges ±qm centered on the two tetrahe-dra connected by a [111] Ising spin. The centers of thetetrahedra form a diamond lattice with lattice spacingad. The magnetic charge qm is chosen so that, giventhe diamond center-to-center distance ad, one recoversthe original magnetic dipole moment with µ ≡ qmad.The original model of point-like dipoles has now been re-cast as a model of magnetic charges interacting thougha “magnetic Coulomb potential”, the latter model hav-ing, by the construction µ ≡ qmad, the same 1/r3 dipolarfar-field as the DSIM. Since in this model the magneticcharges, or “monopoles”, from the four dumbbells at thecenter of a given tetrahedron overlap, a regularization ofthe Coulomb potential is introduced31. The strength ofthe regularization potential for overlapping monopoles isadjusted so as to correctly recover the interaction energyfor nearest-neighbour dipoles along with the contributioncoming from a nearest-neighbour exchange31,32.

As a result of these constructions, the dumbbell Hamil-tonian Hdb can thus be expressed in terms of the netcharge Qα =

∑i∈α = 0,±2qm and ±4qm at the center of

the α’th tetrahedron, with

Hdb =µ0

4π

∑α>β

QαQβrαβ

+v0

2

∑α

Qα2, (2)

where the onsite term v0 is determined from the condi-tion that this model should reproduce the spin flip energyof the dipolar spin ice model. From this formulation, itis clear that in the limit v0 → ∞, the ground states ofthe system have Qα = 0 for all α. These are precisely thetwo-in/two-out ice-rule obeying states. For finite v0, thelow-temperature state of the system is ultimately deter-mined by the competition between the self-energy cost ofa charge and the energy gain of a specific arrangement ofpositive and negative magnetic charges on the diamondlattice. This dumbbell model describes rather accuratelythe DSIM, in particular the long-range 1/r3 nature of itsdipolar part. Most importantly, the dumbbell model hasthe same quasi-degenerate ice-rule obeying states as thenearest-neighbour frustrated Ising model of Harris et al.

and displays, in particular, a residual low-temperaturePauling entropy S0.

Noting that the condition of vanishing charge Qα = 0is equivalent to the “2-in”/“2-out” spin configurationon each tetrahedron, we may introduce a coarse-grained“spin field” which we denote as B for a reason that willbecome clear in the next section. Then the statementthat Qα = 0 is akin to stating that the “spin field”has zero divergence, ∇ · B = 0. When there is no lo-cal source or sink of the coarse-grained “spin field” inclassical spin ice, the magnetic structure factor, whichcan be measured experimentally via neutron scattering,exhibit singularities in reciprocal space at nuclear Braggpoints33,34. The intensity profile around these singulari-ties has a characteristic “bow-tie” form in planes throughthe singular points which have come to be known as“pinch-points”34,35.

For a reason that will be expanded upon in Section 3,one refers to the Qα 6= 0 charges, sources and sinks ofthe spin field B, as gauge charges. The beauty of thedipolar spin ice is that, thanks to the underlying dipolarinteractions, the gauge charge is also a magnetic charge,or “monopole” that is a sink or source of the local mag-netization field,31, since each spin comes along with itsmagnetic moment µ. So one perspective on the effectivelow-energy theory of dipolar spin ice is that of a mag-netic Coulomb gas, with charges on a diamond latticeand in the grand canonical ensemble36,37. The flippingof an individual Ising spin thus corresponds to the nu-cleation of two magnetic charges of opposite sign whichcan then thermally diffuse while interacting with an ef-fective emergent Coulomb potential. In short, at low-temperature and at low-energy, the long-distance physicsof dipolar spin ice is equivalent to a magnetic formulationof Gauss’s law in which the elementary excitations inter-act with 1/rαβ “magnetic” Coulomb potential as givenby Eqn. (2). As we shall see in Section 3, the presentmagnetic charges correspond to the classical limit of themagnetic charge of the U(1) description of the quantumspin liquid. In the present case of the classical dipo-lar spin ice, these are a remarkable manifestation of the“fractionalization” of the individual elementary dipolemoment (spin) flip excitation and, because of the 1/rαβnature of their mutual Coulomb interaction, the work toseparate two charges of opposite sign is finite and thecharges are said to be “deconfined”31,32.

The paper31, highlighting the existence of monopoleexcitations of energetic origin in the dipolar spin ice ma-terials, has led to a number of experiments aimed at prob-ing their effects on various magnetic properties. Becausethis chapter in the history of spin ices fits neatly intoa discussion of the experimental probes of exotic exci-tations, we briefly describe a cross-section of this work.We direct the reader to recent reviews for a more ex-tensive discussion32,34. For the purpose of this review,it suffices to say that, while the description of the low-energy excitations in dipolar spin ice model in terms ofmonopoles is most likely correct, few experiments can

6

be said to have positively exposed monopoles in clas-sical spin ice compounds. Some first generation experi-ments using muon spin relaxation38 and relaxation of thebulk magnetization39 have later been subject to a con-troversy in the former case40–42 and the interpretationcritiqued in the latter43. The temperature dependenceof the relaxation time extracted from magnetic AC sus-ceptibility has been partially rationalized in terms of dif-fusive motion of monopoles36,37, but such a descriptionbecomes less compelling deep in the spin ice regime43,either because of complexities arising from sample qual-ity and/or disorder issues43 or yet other aspects of thereal materials that have not been incorporated into themagnetic Coulomb gas theoretical framework36,37. Fromanother perspective, we note that the assignment of thetemperature dependence of the width of the neutronscattering lineshape near the so-called “pinch points”in the Ho2Ti2O7 material35 to the thermal nucleationof monopoles with separations of O(102) A is not evi-dent. The reason for that is that the lowest tempera-ture considered in Ref. [35] was barely below TSI ∼ 1.9K at which this material enters the spin ice state44.194

The monopole-based description of the neutron scatter-ing data of Dy2Ti2O7 in a magnetic field along the [100]direction is, at this time, perhaps the one that most com-pellingly endorses the picture of fractionalized monopolesin these materials46. While not direct evidence for theirexistence, the field-driven first order metamagnetic tran-sition for a field along the [111] direction is nicely andsimply rationalised in terms of a crystallisation of posi-tive and negative monopoles31,47.

In view of the prospect of ultimately identifying quan-tum spin ice candidate materials characterized by an-other gapped and fractionalized excitation, in additionto the magnetic monopole, as well as an emerging gaugeboson (“photon”), new experimental methodologies withunambiguous signatures for these various excitationsneed to be developed. Such techniques could be firstbenchmarked on classical dipolar spin ices and perhapsproved successful in achieving an explicit demonstra-tion of the existence of the aforementioned “monopoles”.Having reviewed the topic of classical spin ices, with a fo-cus on the description of their low temperature behaviourin terms of fractionalized monopoles, we now move onto

the heart of this review – the topic of quantum spin ice.

B. Naming Conventions

The reader who wishes to study the original litera-ture on classical and quantum spin ice will come acrossa number of different naming conventions for the emer-gent particles and fields. We give here a quick guide tothe main conventions and, at the same time, fix our own.As stated above, in classical spin ice, the ice constrainton each tetrahedron can be thought of as a divergencefree condition on a coarse-grained field. In the litera-ture one can find the various naming conventions givenin the top panel of Table I. Point-like defects resultingfrom the local breaking of the ice constraint are sourcesof the coarse-grained field. Since the emergent chargeshave a physical magnetic response, we call them mag-netic charges or monopoles in common with the classicalspin ice literature31,32.

A second kind of gapped excitation appears in quan-tum spin ice which we call a vison. The nature of this ex-citation can be seen most clearly from the compact gaugetheory discussed in Section III B. In this effective fieldtheory48, the spin ice constraint appears as Gauss’ lawin an electric field defined on links of the diamond lattice.The effective field theory has both electric and magneticdegrees of freedom. In the convention of Hamiltonianlattice gauge theory, the vison excitation is a source ofmagnetic flux and is typically called a magnetic monopolein that community. At this point, confusion might eas-ily arise. We have tried to keep our notation consistentwith that, so far, predominantly employed by classicalspin ice researchers and, for this reason, have swappedthe convention − the vison appears as a source of elec-tric field while the magnetic monopole keeps its identityas the emergent charge familiar from classical spin ice31.We keep ourselves, however, from calling the vison anelectric charge because, while the magnetic monopole isindeed a source of the physical magnetic field (or, morecorrectly the macroscopic field H), the vison is a merelya source of the fictitious electric field in the lattice gaugetheory.

The vison borrows its name from the literature onquantum spin liquids where, to our knowledge, it firstappeared to describe fluxes in Z2 gauge theory51. Thisliterature would also naturally assign the term spinon (orfractionalized spin) to the charges that coherently hopon the diamond lattice sites and which decohere into themagnetic monopoles of classical spin ice at finite temper-ature.

Figure 3 shows how the excitations are organised byenergy scale in quantum spin ice.

III. QUANTUM SPIN ICE

Quantum spin ice is a type of U(1) quantum spin liq-uid which might be observed in certain pyrochlore mag-nets. A U(1) spin liquid in three dimensions is a collectiveparamagnetic phase of matter with fractionalized excita-tions at low energy that are gapless, with linear dispersionω ∼ |k| and with two transverse polarizations. In short,these excitations behave like particles of light. From thestandpoint of modern relativistic quantum field theory,physicists regard gauge invariance, and hence electro-magnetic radiation, as being the inevitable consequenceof having a quantum theory of relativistic massless spinone particles52. In the present condensed matter context,

7

Coarse-grained field present also in classical limit

Magnetic field B or H [31]

Polarization P [33]

Spin field S [30]

Electric field e [48]

Flipped spin defects/ Emergent charges on diamond lattice

Magnetic monopole/charge [31] + spin ice literature since 2008

Electric charge [48] from gauge theory literature

Spinon [49] from quantum spin liquid literature

Gapped topological defects

Vison [50] and quantum spin liquid literature

Magnetic monopole [48] from gauge theory literature

TABLE I: Naming conventions for excitations and fields in quantum spin ice. Our conventions are highlighted.

FIG. 3: Schematic of the spectrum of excitations in quantumspin ice including the approximate energy scales and differentnaming conventions.

the reasoning is turned around: the quantum spin liquidhas an emergent low energy gauge redundancy so thatlocalised magnetic excitations of spin one, which have nopreferred axis because there is no spontaneous symmetrybreaking, lose one polarization and behave instead likeparticles of light. In contrast to magnons in long-rangemagnetically ordered phases, which have two polarisa-tions because one direction is fixed by the broken sym-metry, the fact of having two polarizations of photon ex-citations, whether fundamental or emergent, is enforcedby gauge invariance. In this section, we shall see in a littlemore detail how magnetic interactions may give rise to avariant of ordinary quantum electrodynamics. We thendescribe various properties of these exotic phases and re-view some of the ways in which they might be probedexperimentally in real quantum magnets. Next, we con-

sider the naturalness of quantum spin ice models anddiscuss from a very general perspective, the prospects ofseeing quantum liquids of this type in real materials. Weconclude with a short section (Section III F) mentioningother possible condensed matter realizations of U(1) liq-uids as well as putting quantum spin ice into the broadercontext of understanding quantum spin liquid phases.

A. From a spin model to loops

We begin by returning to classical spin ice because itis, in some sense to be made more precise later, the pre-cursor state to the quantum spin liquid state of quantumspin ice systems. Also, it will give us a classical exam-ple in which a U(1) gauge redundancy appears, or reallyemerges, at low energies in a magnet. The key to makinga spin ice is to frustrate an Ising model by putting it ona pyrochlore lattice (see Fig. 2). As discussed earlier inSection II A, in real magnets, the Ising spins interact inspin space as though they were pointing along the local〈111〉 directions. The interactions in a classical nearest-neighbour spin ice (CSI) model are described by the sameHamiltonian as in Eqn. (1) that we rewrite here:

HCSI = J‖∑〈ij〉

Szi Szj . (3)

To emphasize something we have already mentioned: thisclassical Hamiltonian has a hugely degenerate groundstate composed of spin configurations fulfilling the “icerule” of two spins pointing in and two pointing out ofeach tetrahedron as illustrated in Fig. 2 and the toppanel of Fig. 6. We denote the Hilbert space of icestates as I. The spectrum of states has a gap of 4J‖to flipped spin defects. The ice rule can be formulatedas∑a Sa · za = 0 (where the sum runs over all the sub-

lattice sites a of a tetrahedron) for each tetrahedral ele-ment of the pyrochlore lattice. This condition is a zerodivergence condition on a lattice34 which may be coarse-

8

grained to ∇ · B = 0 where the “magnetic field” B is acoarse-grained analogue of the spin field Sa on the lat-tice. Since any vector field can be decomposed into thesum of two fields with, respectively, zero divergence andzero curl, in order to obtain thermodynamic quantitieswithin the restricted (∇ · B = 0) manifold of spin icestates, one must average solely over the circulation of B.In dramatic contrast to conventional long-range orderedmagnets at low energies, in spin ices, this coarse-grainedcirculation is unconstrained and runs over a number ofstates that scales as exp(αV ) in the volume V of the sys-tem. We can look at the divergence-free constraint asan emergent gauge invariance, since one may introducea vector potential A such that B = ∇ × A and couldcarry out gauge transformations on A that would leavethe divergence-free condition invariant. The divergence-free condition is thus a characteristic of the ground statesof Eqn. (3). At finite but low temperatures, this condi-tion is weakly violated by the thermal excitation of spinflip defects (i.e. the “monopoles” of the classical dipolarspin ice). As the electrostatic analogy suggests, these de-fects behave like sources of B and, at temperatures wheresuch effective charges are dilute, spin ice should behavemuch like a dilute plasma described in the grand canoni-cal ensemble. This physics becomes richer still when theunderlying microscopic magnetic moments, µa ∝ Sa, in-teract through a long-range dipolar coupling − hence thereview of dipolar spin ices in Section II A. In particular,as discussed in Section II A, the dipolar interaction aboutthe spin ice background fractionalizes into an energeticCoulomb interaction between defects in a background oftetrahedra satisfying the spin ice rule31.

FIG. 4: Closed hexagonal loop on the pyrochlore lattice andon the diamond lattice. The figure shows a segment of a [111]kagome plane in the pyrochlore lattice showing six tetrahedrawith the diamond lattice sites (at the centres of the tetrahe-dra) and the hexagonal loop connecting the diamond sites (inred).

The gauge invariance of classical spin ice turns out to

be crucial to the quantum case to which we now turn. Wenow allow for the presence of (perturbative) “transverse”nearest-neighbour exchange couplings in addition to the“longitudinal” (Ising) exchange part defined by Eqn. (3).Our only requirement is that the transverse couplingsshould have a characteristic energy scale J⊥ J‖ so thatthere is little mixing of the ice rule states with cantedspin states away from the local [111] Ising direction. Weshall discuss in Section IV A the most general nearest-neighbour symmetry-allowed anisotropic Hamiltonian onthe pyrochlore lattice that does not commute with HCSI

and thus causes quantum dynamics. For now, we con-sider a minimal spin model that contains quantum dy-namics within a spin ice state and which is a sort of localXXZ model with transverse coupling J⊥.

HQSI ≡ HCSI+H⊥ = HCSI−J⊥∑〈ij〉

(S+i S−j +S−i S

+j ) (4)

We shall comment, in Section III E, on the conditionsunder which real materials may exhibit such a J⊥ J‖separation of energy scales.

The reason for considering such a separation of scalesis that we require the manifold of classical spin ice statesI, which form a reference classical spin liquid2, to bethe background on which quantum fluctuations can actperturbatively. This allows one to carry out perturba-tion theory in the transverse couplings − the zeroth or-der states being the whole manifold I of degenerate icestates. The lowest order terms derived from a canonicalperturbation theory that preserve the ice rule constraintare ring exchange terms that live on the hexagonal loopson the pyrochlore lattice (Fig. 4). Up to a numerical pref-actor, the effective low energy Hamiltonian that describesquantum fluctuations within I is

Hring ∼J3⊥J2‖

∑h∈7

S+h,1S

−h,2S

+h,3S

−h,4S

+h,5S

−h,6 + h.c., (5)

where the sum is taken over the set of all hexagonalplaquettes in the pyrochlore lattice labelled by 7. Itturns out that the sign of the ring exchange couplingJring ∼ J3

⊥/J2‖ is not important since one can unitarily

transform one sign to the other48195. The U(1) liquidphase of the quantum spin ice model arises from this ef-fective Hamiltonian. The ring exchange term has a localU(1) gauge invariance: one can perform a rotation of thespin coordinate frame about the local [111] z axis on agiven tetrahedron

∏i∈t exp(iθSzi ), where the product is

taken over sites i on pyrochlore tetrahedron t, which maycapture 0 or 2 vertices of a hexagonal ring exchange op-erator. This rotation leaves the effective ring exchangeHamiltonian of Eq. (5) unchanged. The notation U(1)refers to the fact that the local transformation that leavesthe Hamiltonian invariant is an element of the U(1) groupof 1 × 1 unitary matrices (complex phases). More gen-erally, to all orders in the perturbation expansion in J⊥,there is a local gauge invariance of this sort.

9

Pyrochlore XXZ Spin Model

Ring Exchange Model

Fully-packed Loop Model on Diamond Lattice

Frustrated U(1) Gauge Theory

Hard core bosons on pyrochlore with nearest neighbour repulsion

Effective Field Theory

Add Diagonal Terms to Tune to RK Point

Simulation

Simulation

U(1) LiquidInsight into excitations

Deconfined Phase = Unfrustrated Gauge Theory

Perturbation Theory

...mapped to...

...mapped to...

...mapped to...

FIG. 5: Figure showing the conceptual relationships betweenvarious models mentioned in the review and which have beeninstrumental to understanding the U(1) liquid phase in py-rochlore lattice magnets.

Now that we have motivated the existence of domi-nant ring exchange terms in certain effective low energyHamiltonian models of pyrochlore magnets, we seek tounderstand the resulting low energy phase. This discus-sion will take a while and involve ramified ideas. Theproblem of understanding the ring exchange model hasbeen tackled in several different ways that are summa-rized in Figure 5. In particular, it is helpful to recognizethe ring exchange model as a type of quantum dimermodel on the diamond lattice as illustrated in Fig. 6.We refer the reader to Refs. [48,54,55] and to Fig. 6for further details of the mapping of the dimer model.The main idea is that the spin ice states correspond toso-called “loop coverings” of the diamond lattice wheredimers on the links of the diamond lattice are placedend-to-end. In this representation, the spin ice rule isequivalent to the constraint that every diamond site hasexactly two dimers connected to it. The pyrochlore ringexchange Hamiltonian of Eq. (5) can be written in a formthat symbolizes the quantum dynamics of these dimers

(6)

so the V ∼ O(J3⊥/J

2‖ ) effect of the hexagonal ring ex-

change is to cause dimers to resonate around hexagonalplaquettes on the diamond lattice.

It is possible to gain some insight into this model byadding the number operator for flippable plaquettes, HN,to HDimer. The HN term, which alone maximizes or mini-mizes the number of flippable hexagons depending on thesign of the coupling and allows one to tune the theory ofquantum dimers to an exactly solvable Rokhsar-Kivelson(RK) point48,56, is given by:

(7)

At the RK point V = µ, one may write down theexact ground state wavefunction and obtain informa-

FIG. 6: Figure showing how the loop manifold for the dimermodel is constructed from the spin ice states. The figure atthe top shows one particular spin ice configuration on a pairof tetrahedra on the pyrochlore lattice and the lower figureshows the corresponding dimer links of the diamond latticeconnecting the centres of the pyrochlore tetrahedra. The spinconfiguration at the top is mapped to the dimer configuration(blue/yellow rods) in the lower figure as follows. The diamondlattice is bipartite so that alternating pyrochlore tetrahedracan be labelled A and B. Suppose the left tetrahedron is an Atetrahedron. The rule to make a loop configuration is to laya dimer along a diamond lattice link when a moment pointsinto an A tetrahedron (or out of a B tetrahedron). Evidently,the ice states lead to dimer configurations where exactly twodimers meet at each diamond lattice site. The set of suchstates are acted on by the quantum dimer Hamiltonian H =HDimer +HN given in the main text.

tion about the excitations either through a single modeapproximation48,50,56 or by computing certain correla-tion functions in this model numerically exactly usingclassical Monte Carlo. The underlying insight in the lat-ter procedure is that the ground state wavefunction ofthe RK model is an equal weight superposition of differ-ent “loop coverings” which can be sampled using MonteCarlo at infinite temperature57,58. The result is that thelow energy spectrum is gapless with a k2 dispersion pre-cisely at the RK point48,50. This and other aspects ofthe RK point can be captured using an effective fieldtheory which also allows one to infer the phase diagramin Fig. 7 of the model H = HDimer + HN

48,50. On oneside of the RK point, µ > V , perturbation theory tellsus that the ground state immediately enters into a long-range ordered crystalline dimer phase. On the opposite

10

side, we expect the number of flippable plaquettes to bemaximized when µ is sufficiently large and negative −producing another state with long-range order which hasbeen named “squiggle state” for the intertwined loops ofdimers characterizing the phase54. Directly away fromthe RK point, for µ . V where the resonating plaquetteterm HDimer is important, a liquid state should persistwithin some window of couplings (µ/V )c < (µ/V ) < 1.The presence of a linearly dispersing photon-like low-energy excitation mode within this window can be in-ferred from an effective field theory48,50. Specifically, onefinds that, while the dispersion is strictly quadratic atthe RK gapless point, the dispersion becomes linear forµ/V < 1. This can be rationalized on the basis of an ef-fective noncompact field theory: a term (∇×A)2 whichvanishes at the RK point is a relevant perturbation whichimmediately drives the system into the U(1) phase awayfrom the RK point48,50.

FIG. 7: Schematic phase diagram of the fully packed loopquantum dimer model described in the main text.

More recently, concrete evidence for the presence of aU(1) phase has come from numerics. Fortunately, thedimer Hamiltonian, H = HDimer + HN of Eqns. (6) and(7) has no sign problem and can be studied using quan-tum Monte Carlo (QMC). In the guise of a hard coreboson model on a diamond lattice with large nearest-neighbour repulsion, the dimer model also arises at lowenergies. Such a model has been simulated using QMCimplemented using the stochastic series expansion (SSE)method59. These simulations find a state with no super-fluid order and no Bragg peaks in the structure factor,and there thus appears to be a liquid phase down to thelowest temperatures accessed by the simulations. In a re-cent zero temperature Monte Carlo study, Shannon andco-workers mapped out the whole phase diagram of thedimer model54. In particular, the authors of Ref. [54]found further evidence for a gapless mode with lineardispersion persisting across a finite µ/V window betweenthe exactly solvable RK point µ/V = 1 and µ/V ≈ −1/2.This includes the crucial “physical” µ = 0 point corre-sponding to the effective ring exchange model Eqn. (6)

derived via perturbation theory starting from the originalspin Hamiltonian. The conclusion is that there is goodevidence that the spin model with Hamiltonian Eqn. (5)acting within the ice manifold has a quantum spin liquidground state.

B. From loops to a gauge theory

One can gain a great deal of physical insight into thequantum spin ice model by making a set of transforma-tions from the dimer Hamiltonian Eqns. (6) and (7) toobtain a lattice gauge theory48,60,61. These transforma-tions require two main steps: the first is to enlarge theHilbert space of the model by allowing the link dimer oc-cupation numbers to take any integer value on each linkL, not only zero and one. The associated dimer occupa-tion operator is nL. The original physical subspace of Sz

eigenvalues, SzL = ±1/2, is a subset of SzL = nL−1/2 witheigenvalues of nL running over all integers.196 To recoverthe original physical subspace (Sz = ±1/2), which corre-sponds to the aforementioned hardcore dimer constraint,one introduces a “soft” constraint with tunable couplingthrough an extra term in the Hamiltonian of the form

HConstraint = U∑

L

(nL −

1

2

)2

. (8)

There is also a requirement for a second constraint which,taken together with the constraint on the occupationnumbers on a link, imposes the ice-rule condition. Thissecond constraint takes the form

QI ≡∑

L∈Diamondsites

nL = 2 (9)

where the sum runs over the four links connected to agiven diamond site I. The dimer constraint introducedin the previous section is that exactly two dimers con-nect to each diamond site. Now, the “soft” constraintof Eq. (8) together with QI = 2 enforce the ice-rule pat-tern of dimers. The constraint QI = 2 can be thought ofas an analogue of Gauss’ law. To see this, we assign anorientation to the diamond lattice links using the bipar-titeness of that lattice as follows. The A sites are thosedefined as “UP” tetrahedra in Fig. 1 and the B sites cor-responding to the “DOWN” tetrahedra. We let the linkshave orientation towards the A sites and away from the Bsites. Having done this, we introduce so-called orientedlink variables defined through

BLA→B= +

(nL −

1

2

)(10)

which we call magnetic fields (since they are related tothe orientation of the physical microscopic magnetic mo-ments) and with the sign reversed when the orientationis reversed (BLA→B

= −BLB→A). Now, the constraint on

11

QI in Eq. (9) is ∑L

BL = 0, (11)

which is recognizably Gauss’ law discretized on a lattice.The ring exchange term of Eqn. (5) gets modified when

we enlarge the Hilbert space from the dimer model bymoving over to the occupation number operators nL. Theconjugate variables to nL are phases φL and the opera-tors exp(±iφL) raise (+) and lower (−) the occupationnumbers on links. In the next step, to make the corre-spondence with a U(1) gauge theory, we give these phasesan orientation on the links (as described above) and re-name them φL → AL. With these steps, we obtain theHamiltonian48,60,61

HGauge = U∑

L∈Links

B2L −K

∑P

cos

∑L∈7

AL

(12)

where the sum over AL is an oriented sum around ahexagon and we call the flux variable through a givenhexagonal plaquette P the electric flux EP. This electricflux is not to be confused with the physical (or funda-mental) electric field which would enter the theory whenconsidering, for example, the dielectric response of thesystem62.

In summary, the first term in the right hand side of thisHamiltonian is there to impose the constraint on the (di-amond lattice) link occupation number nL so that it as-sumes only values 0 and 1. The second term in Eqn. (12)is the ring exchange of Eqn. (6), but now written ex-plicitly in terms of a vector potential AL. The model isconsistent when Gauss’ law, Eqn. (11), is satisfied. Thisis a quantum theory because the field components sat-isfy canonical commutation relations [BL, AL′ ] = −iδLL′ .Taken together, these ingredients constitute a version ofquantum electromagnetism on the lattice with BL as themagnetic field components and the cosine in the righthand term of the Hamiltonian (Eq. 12) as the latticeanalogue of electric flux. Schematically, this route fromthe dimer model to a gauge theory is outlined in Fig. 8.

We note that the vector potential AL is defined mod-ulo 2π in contrast to ordinary electromagnetism whereit is defined over all real numbers. The effective low en-ergy gauge theory of quantum spin ice is therefore a com-pact gauge theory with gauge group U(1), the group ofphase rotations. The origin of the compactness is thediscrete spectrum of magnetic field states on each linkwhich itself comes from the discrete dimer constraint.The compactness of the U(1) gauge theory has one cru-cial consequence which sets it apart from ordinary elec-tromagnetism, namely, that there is a novel charge-likeexcitation in the theory − the vison − with no corre-spondence to the magnetic monopoles of classical spinice. This lowest energy gapped excitation plays (see Fig-ure 3) a crucial role in determining the phases of thelattice gauge theory.

ei

ei

2

1

0

+1

+2

U ! 1

DimerEmpty link

Magnetic field on links

ei

ei

FIG. 8: Figure illustrating the connection between the gaugetheory degrees of freedom and the quantum dimer states.Each link of the diamond lattice in the gauge theory hasa tower of states, labelled by integer nL connected by rais-ing and lowering operators e±iφL where φL is an operatorwith continuous spectrum.The nL and φL on each link areunoriented. Then they are assigned an orientation they be-come, respectively, magnetic fields BL and vector potentialsAL. When U in Eq. () is taken to be large, all magnetic fieldstates except for BL = 0,±1 are gapped out. The low-lyingmagnetic field states coincide with the allowed dimer stateswith occupation number nL = 0 and nL = 1 on each link.

Having reviewed the essential steps leading to the con-struction of a gauge theory for quantum spin ice systems,we discuss the phases of this theory. We will end upwith the conclusion that the quantum liquid phase of thedimer model of the previous section can be thought of asthe deconfined phase of the compact U(1) gauge theory.This means that the compactness of the gauge theoryis unimportant in the low energy limit or, equivalently,the fluctuations of the electric flux are small. In the de-confined phase, we may then expand the cosine of theHamiltonian Eqn. (12), omitting all but the lowest ordernontrivial contribution. The Hamiltonian is then

HGauge,Deconfined = U∑

L∈Links

B2L +K

∑Plaq.

E2P (13)

where EP is, slightly unconventionally (see Section II B),the circulation of AL around plaquette P

EP =∑

L∈7AL.

This Hamiltonian is recognizably a discretized form of theHamiltonian describing electromagnetism in the absenceof charges. This means that the excitations at low energyare photons − with linear dispersion and two transversepolarizations.

Before we discuss in more detail the phases ofEqn. (12), we take the opportunity to make the followingobservation. The discretized compact electrodynamicsthat we have discussed above differs from the standard

12

Abelian lattice gauge theory that has been studied tradi-tionally by lattice gauge theorists in one crucially impor-tant respect. Namely, at strong coupling U/K →∞, theHamiltonian of Eqn. (12), has nL = 0, 1 within its vac-uum (constrained by Gauss’ law) corresponding to thespin ice configurations or superpositions of these states.Typically, pure Abelian gauge theories at strong couplinghave a trivial vacuum with nL = 0. In the literature,gauge theories with a trivial vacuum have come to becalled “even”, or unfrustrated, gauge theories to contrastthem with gauge theories such as Eqn. (12) which arecalled “odd” or frustrated gauge theories60. The natureof the vacuum states at strong coupling makes a dra-matic difference to the nature of the phase diagram as afunction of the coupling K/U .

As discussed above, the unfrustrated gauge theory(with a trivial vacuum at large U) is known to have twophases in the 3 + 1 dimensional case of interest here dueto the well known work of Refs. [63–67]. When U/K issmall, there is a gapless photon excitation and a Coulomblaw between test charges (both monopoles and visons) in-serted into the system. This is the deconfined, Maxwellor Coulomb phase. In the opposite limit, with largeU/K, the theory is confining, meaning that the photonis gapped out and particles with opposite electric chargesare bound together by a potential which grows linearly inthe separation of the charges. There is a critical (U/K)value at which the theory undergoes a transition betweena confined and deconfined phase63–65.

The frustrated theory, in contrast, maps to the µ/V =0 dimer model for large U/K. The evidence from numer-ical simulations is that the dimer model is in a decon-fined phase when the ring exchange term is the only termpresent in the Hamiltonian (i.e. at the µ/V = 0 of thephase diagram shown in Fig. 7)54,59. When U/K is smallin the frustrated gauge theory, the frustration should notbe important and, once again, we expect the theory tobe deconfined. In summary, it appears that confinementmay be completely absent in the frustrated gauge theory,though, to our knowledge, this has not been establishedbeyond the heuristic arguments given here. Since thedeconfined phase of a U(1) gauge theory exhibits uni-versal features, it should not matter whether the dimermodel (in its deconfined phase) maps to a frustrated orunfrustrated gauge theory − the excitations below theenergy scale of the hard core dimer violating fields (setby U) in the frustrated gauge theory can be inferredfrom the unfrustrated gauge theory about which muchis known63,64,64,66,67. In the next section, we borrowinsights from the deconfined phase of the unfrustratedgauge theory to say something about the phenomenologyof the deconfined phase of the quantum dimer model or,by extension via the series of mappings reviewed above,quantum spin ice.

C. Quantum liquid: excitations andphenomenology

The perspective on the quantum spin ice problemgained by thinking about the deconfined phase of U(1)lattice gauge theory allows us to infer the nature of theexcitations in the dimer model. At energies below thescale of the ring exchange Jring in the deconfined phase,where fluctuations of the gauge field AL are small, thecompactness of the theory should not matter, and werecover familiar noncompact electromagnetism with gap-less photon excitations with a pair of transverse compo-nents.

At much higher energies, on the scale of the Ising ex-change term J‖ in the original spin model, there aregapped magnetic charge excitations (the “monopoles” ofSection II A) corresponding to spin flips out of the man-ifold of spin ice states. Both in classical and in quan-tum spin ice, single spin flips correspond to a breaking ofthe divergence-free condition on the magnetic field, BL,and hence the creation of magnetic field sources or mag-netic charges. Via successive spin flips, it is possible toseparate these charges. Whereas in classical spin ice aneffective magnetostatic interaction arises between thesecharges owing to thermal averaging solely over circula-tions of the field BL, in quantum spin ice these magneticcharges interact both with the emergent magnetic andelectric fields.197

At intermediate energies, a third type of excitation ispresent which arises from the compactness of the gaugefield. These are the aforementioned visons. For read-ers familiar with Z2 spin liquids, it might be helpful topoint out that the magnetic charges in U(1) spin liquidsare the analogues of so-called gapped Z2 flux excitationsalso called visons in these systems51. One way of see-ing that these should be present is as follows. The loopsof dimers in the quantum dimer model introduced abovemay be interpreted as magnetic field strings within thelanguage of the gauge theory. While the dimer modelis defined to be a theory of closed strings, excitationsout of the spin ice manifold may occur in the originalspin model breaking these strings and lead to the mag-netic charges (monopoles). Importantly, there are nowalso electric loops in the theory. These, unlike the mag-netic strings, are not imposed by a kinematic constraintof Eqn. (9). Instead, they arise from the dynamics of thedimer model – the resonating hexagonal plaquettes formloops of electric fluxes. When these flux strings break, thestring endpoints form new sources: electric charges whichare gapped in the liquid phase with a gap of the order ofJring. This picture of visons appearing at ends of broken(electric) strings is true also for the vison excitations inZ2 spin liquids. The visons are massive particle-like ex-citations with a net charge. Given the pyrochlore latticestructure, they can be thought of as hopping on a sec-ond diamond lattice displaced from the original diamondlattice on which the magnetic charges hop by half of oneelementary cubic cell in each coordinate direction. 198

13

Having noted the types of excitations and the hier-archy of energy scales in quantum spin ice, we turn toplausible experimental signatures of a magnet with a lowenergy U(1) phase. At the highest temperatures, themagnet is a featureless paramagnet. Upon cooling, thematerial enters a classical spin ice regime with a tem-perature dependent density of monopoles and a Paulingresidual entropy. This state of matter is well-known toexhibit distinctive dipolar spin-spin correlations whichshow up as pinch points in the neutron scattering crosssection34. The quantum dimer model at infinite tempera-ture is nothing other than classical spin ice and, therefore,this model does not capture the crossover into the truehigh temperature paramagnetic regime of the underlyingspin model55.

At the lowest temperatures, within the U(1) phase,there should be quite distinctive experimental signatures.While there should be no magnetic Bragg peaks, inelasticneutron scattering can in principle probe the linearly dis-persing photon modes as recently worked out in Ref. [55]since, like magnons, these excitations carry spin one.Our calculation of the expected inelastic scattering pat-tern (based on that in Ref. [55)] between high symmetrypoints in the Brillouin zone is shown in Fig. 9(a) wherethe most energetic modes appear on the scale of the ringexchange coupling. The energy integrated scattering atzero temperature is presented in Fig. 9(b). Thermal fluc-tuations cause the photons to decohere and the neutroncross section crosses over smoothly into the pinch pointscattering of classical spin ice55.

At intermediate temperatures, the phenomenology is,to date, not completely clear. The T 3 (thermal radiation)law in the heat capacity should break down as visonsand magnetic charges become thermally nucleated andthe residual entropy recovers from zero, at zero temper-ature, to the Pauling entropy, in the classical ice regime.These effective charges will also lead to diffuse neutronscattering as is well-known in classical spin ice33,45.

Also, the analogy with the compact U(1) gauge theorysuggests that thermal fluctuations should give the photona small mass and the otherwise Coulomb-like electric andmagnetic charges should become screened.199. Whereasthe thermal occupation of photon modes is sufficient toobserve the crossover between quantum and classical ice,and although the quantum liquid at zero temperatureis adiabatically connected to the trivial paramagnet, amore featured scenario is possible. In particular, a formof gauge mean field theory for quantum spin ice produces,at finite temperature, a truly novel behavior: a first ordertransition between the high temperature paramagnet andthe quantum spin ice state72.

D. Stability of the U(1) liquid

One of the remarkable features of the U(1) spin liq-uid is that it is stable to all local perturbations48,73,74.This is surprising for at least two reasons. One is simply

that the phase is gapless and there is therefore an a pri-ori danger that some perturbations may open a gap. Asecond reason is that the gauge invariance of the latticemodel is not exact but is an emergent property at lowenergies. One can see by power-counting that all gaugenoninvariant perturbations to the Maxwell action are rel-evant in the renormalization group (RG) sense in (3+1)dimensions52 so one might expect that the U(1) liquidwould not survive such perturbations.200 Crucially, forthe search of this exotic state of matter among real ma-terials, it turns out that this fear is likely not justified aswe now discuss.

Because the U(1) gauge theory comes from a micro-scopic spin model on a lattice, the gauge group is compactso gauge non-invariant terms such as M2A2 are not al-lowed in the theory. Instead, the gauge non-invariant per-turbations can only appear through terms like exp(iA)which are magnetic monopole hopping operators, or fromterms that hop visons. But then, since both matterfield (vison and magnetic) excitations are gapped, thesegauge non-invariant operators cannot affect the low en-ergy physics − the U(1) liquid is stable to all gauge non-invariant perturbations − while all gauge invariant per-turbations are irrelevant in the RG sense. There is somework putting these arguments on a rigorous footing usingthe idea of quasi-adiabatic continuity. The idea, roughlystated, is to switch on a gauge non-invariant perturbationto the U(1) liquid with small coupling s which has theeffect of taking the ground state |Ψ(s = 0)〉 into |Ψ(s)〉.Then one transforms all the operators O in the theoryin tandem with the switching on of the perturbation insuch a way as to (i) preserve the locality of the opera-tors and (ii) so that the expectation values of operatorsO(s) computed using the |Ψ(s)〉 and the dressed opera-tors are the same as those computed using |Ψ(s = 0)〉and the undressed operators O, up to corrections thatvanish in the thermodynamic limit. The authors of Ref.[75] were able to show that, although the bare Hamilto-nian has small gauge non-invariant terms, the generatorsof gauge transformations are dressed after the addition ofthese perturbations to the bare model in such a way thata local gauge invariance survives in the dressed model.Although unproven, this is thought to preserve also thegaplessness of the theory.

E. Naturalness of the U(1) liquid

In the previous section, we explained that the U(1)liquid has the remarkable property of being stable to alllocal perturbations. This means that if the U(1) liquid isknown to occur at some point in the space of all possiblecouplings, then one can vary the couplings in any direc-tion in the space of couplings by a finite amount with theground state remaining in this phase (unless our originalpoint is at a phase boundary).

While it might be possible to engineer a quantum sys-tem, perhaps a trapped cold atom system, that enters

14

the U(1) phase, in the immediate future the most likelycandidate systems that might host such a phase are cer-tain pyrochlore magnets. Assuming this to be the case,the stability of the U(1) liquid is of marginal relevancecompared to the broader issue of whether the space ofcouplings explored by real materials accommodates theU(1) phase over a significant region in this space. Inshort, it is useful to know whether the U(1) liquid is toofinely tuned to be observed in at least one of the set ofavailable materials. If a U(1) liquid were discovered wecould, with reasonable confidence, pronounce that suchphases are natural in parameter space. Since a decon-fined U(1) liquid phase has not been found in any of thepyrochlore magnets (at the time of writing), the ques-tion is worth asking from the perspective of microscopicmodels because it has a bearing on the discoverabilityof such phases and may provide some guidance to theexperimental search.

To date, the naturalness of quantum spin liquids ingeneral is largely an open question2. For the partic-ular problem of the naturalness of the Coulomb phasein pyrochlore magnets, the location of the U(1) liquidphase within the space of nearest-neighbour anistropicexchange couplings has been partially mapped outwithin a form of gauge mean-field theory for even elec-tron (Kramers) magnetic ions49 and odd electron (non-Kramers) ions53. The zero temperature gauge mean-fieldtheory (gMFT) employed by the authors of these papersis a variant of slave particle mean field theory which, inthis case, involves the formally exact step of splitting theanisotropic exchange Hamiltonian Eqn. (15) in SectionIV A below into gauge field degrees of freedom defined onlinks of the diamond lattice and new boson fields definedon the centres of diamond sites. The latter bosonic fieldsare referred to as spinons in that paper and as magneticmonopoles here (see Section II B on naming conventions).One decouples the resulting interacting theory and solvesself-consistently for the expectation values of the gaugefield and monopole fields. Whereas, the frustrated gaugetheory discussed in Section III B describes the physics ofthe quantum spin liquid within the spin ice manifold, theaim of the gauge mean-field theory is to capture furtheraspects of the physics of the full anisotropic spin modelon the pyrochlore. This is the reason why, in additionto a compact gauge theory, one has couplings to elec-trically charged matter fields and one can expect Higgsphases, in addition to possible deconfined and confinedphases. Higgs phases have the property that the pho-ton is gapped out by the condensation of the bosonicmatter fields; in some sense they are superconductingphases. The criteria for distinguishing different phasesof the resulting gauge theory − deconfined, confined andHiggs phases − are reminiscent of those in works on themean-field theory of lattice gauge theories (see for exam-ple, Refs. [76,77]). The principal difference between thetypes of theories considered in that early work comparedto those arising from the pyrochlore spin models is thatthe former have explicit gauge kinetic terms in the action

whereas, in the latter, there are only matter-gauge cou-plings − the gauge kinetic terms being solely generatedby the dynamics of the monopole fields. Secondly, thephase diagram of the gauge theory can be interpreted inthe language of the microscopic spin/magnetic degrees offreedom. Specifically, the confined phase corresponds tosome ordered spin ice phase while Higgs phases are long-range ordered magnetic phases in which the momentshave nonzero expectation value perpendicular to the lo-cal Ising directions. Phase diagrams produced by solvingthe gMFT are shown in Fig. 10.

The phase diagrams arising from the gauge mean-fieldtheory show the U(1) liquid phase surviving out to cou-plings J⊥/J‖ ∼ O(10−1) away from the classical spin ice

point (where only J‖ 6= 0)49,53. Supposing that the mean-field theory correctly captures the U(1) phase boundary,we next turn to the naturalness of materials with ex-change couplings satisfying J⊥/J‖ ∼ O(10−1). In otherwords, we consider the physics that leads to XXZ-likemodels in pyrochlore magnets conceptually akin to thatof Eqn. (4) with a dominant Ising term HCSI.

The crystal field, in tandem with spin-orbit cou-pling, which are responsible for the single ion magneticanisotropy, become increasingly important for magneticions further down in the periodic table. This trend coin-cides with a reduction of the typical exchange couplings.Ising magnetism protected by the largest anisotropy gapsrelative to the interactions is expected to occur amongthe rare earth or actinide magnets. Indeed, among therare-earth pyrochlores, typical crystal field anisotropygaps are of O(102) K78,79 while, thanks to the trigonalsymmetry, the single ion ground states are often dou-blets with effective exchange couplings of order O(1) Kbetween effective spin-1/2 degrees of freedom.

The crystal field ground state doublet of “would benon-interacting” non-Kramers ions is strictly Ising-like.In these systems, effective transverse exchange couplingsbetween the low energy effective spin one-half momentsmay arise in two ways: via multipolar couplings80,81 andvia mixing with excited crystal field levels82,83. The lat-ter effect is suppressed by powers of 1/∆, where ∆ isthe energy gap to the first excited crystal field states,while multipolar couplings of superexchange origin aresuppressed by powers of the charge transfer gap. In suchnon-Kramers materials, Ising models with weak trans-verse couplings are therefore expected to be natural, aswe discuss in Section IV A. Pragmatically speaking, thisis borne out by the existence of the spin ice materialswhich are very well described by a classical dipolar Isingspin ice model23,26,84. While the energy scale J⊥/J‖ issufficiently small that the effective ring exchange termmay be typically negligible, a number of rare-earth py-rochlores are coming to light that exhibit some quantumdynamics85,86. We return to this topic in Section IV B.It is of considerable interest to establish the nature ofthe low temperature magnetism in these materials. Un-fortunately, the degeneracy of non-Kramers crystal fielddoublets being accidental, it is sensitive to perturbations

15

that need not be time-reversal symmetry invariant. It isthen important to address the size of likely ring exchangeterms compared to crystal field degeneracy-breaking per-turbations.

The crystal field ground state doublets of Kramers ionsare, by comparison, robust to time reversal invariant per-turbations and there are no a priori symmetry constraintson the strength of the relative strength of the effec-tive anisotropic exchange couplings (such as J‖ and J⊥).Such materials are thus expected to afford an explorationof the full space of symmetry-allowed nearest-neighbourcouplings. We would then expect to commonly findJ⊥/J‖ ∼ O(1) among Kramers ions. The recently de-

termined couplings in Yb2Ti2O787–89 and Er2Ti2O7

90,91

bear out this expectation. The gMFT described aboveyields a region of parameter space in which the U(1) liq-uid lives where the transverse terms are not necessarilymuch smaller than J‖, indicating that the U(1) liquidmay not be unnatural in Kramers rare earth pyrochlores.Indeed, there is extensive on-going work exploring the na-ture of the low temperature phase of Yb2Ti2O7

87–89,92–94,a matter we return to in the Section IV B 3.

The perturbation theory deployed to find the ring ex-change Hamiltonian earlier in Section III A does nothingto suppress Ising couplings beyond nearest neighbour25,26

which may be of superexchange origin or from the long-range dipolar interaction23,26, the latter being typicallylarge among rare-earth and actinide magnetic ions. SuchIsing couplings tend to lift the degeneracy of the icestates28,29 and the stability of the U(1) liquid ultimatelyboils down, roughly, to a comparison of the energy scalesof ring exchange and of the further neighbour Ising cou-plings of the form Jzzij (rij)S

zii S

zjj . Ref. [55] includes a

study of the effect of the third neighbour Ising couplingon quantum spin ice finding that it drives long range or-der above some threshold which is O(1) times the ringexchange coupling. It would be interesting to considerfurther the role of couplings beyond nearest neighbouron the U(1) phase.

In summary, the lanthanide (4f) and actinide (5f) py-rochlore magnets offer a tantalising opportunity to dis-cover quantum spin ice phases. For Kramers magnetsamong these materials, it is possible to satisfy all the cri-teria necessary to see quantum spin ice physics and themain difficulty to be overcome is the large space of pa-rameters that such materials can, in principle, explore. Innon-Kramers magnets, the pessimistic perspective thatsuch materials should be of no interest for exotic quan-tum states of matter, because of their large angular mo-mentum J and large single ion anisotropy is not correct.Thanks to deep low-energy Ising doublets, the underlyinghigh-energy microscopic Hamiltonian can, once projectedin the low-energy Hilbert space spanned by the doublets,be described by an effective spin-1/2 Hamiltonian allow-ing for significant quantum dynamics. Indeed, a micro-scopic calculation for a toy-model of Tb2Ti2O7 makesthis point crisp82,83. Unlike Kramers magnets, crystalfield degeneracies in non-Kramers magnets is sensitive

to being broken by disorder and sample quality will bea particularly important issue. Unfortunately, short ofhaving ab-initio calculations guidance, it is a matter ofluck finding the right material that falls in a regime ofinteractions where a U(1) spin liquid is realized, or someother novel quantum states49. We discuss this topic inSection IV A.

F. A broader perspective on quantum spin ice

In this review, we have concentrated on the possibil-ity that the deconfined phase of a U(1) gauge theorycan arise in certain pyrochlore magnets. This discussionwould be incomplete without widening the scope a lit-tle by mentioning some other models with emergent lowenergy U(1) liquid phases.

A natural place to begin is the work of Baskaran andAnderson95 and Affleck and Marston96 who used slavebosons to study certain quantum spin models in 2 + 1dimensions. Within this approach, the spins are frac-tionalized into bosons with an accompanying U(1) gaugeredundancy. The pure U(1) gauge theory is compact andis confining in (1 + 1) and (2 + 1) dimensions: the gaugeboson is gapped out and the matter fields to which it iscoupled are bound into states of zero gauge charge63–65.This argument implies that the analogue of quantum spinice in two dimensional magnets is in a long-range or-dered magnetic phase which has been explored in variousworks97–100. The implication of confinement for the slaveboson procedure is that the fractionalization of spins isnot a correct description of the physics in the cases stud-ied in Refs. [95,96].

Fluctuations of the gauge field in slave particle descrip-tions of quantum spin models can be suppressed by tak-ing the SU(2) magnet to SU(N) in the large N limitleading to deconfinement even in two dimensional mag-nets. In this limit, slave particle mean field theory isa controlled approximation74,101,102. A promising direc-tion, motivated by slave fermion treatments of quantummagnets, comes from recent work showing that fraction-alization is possible in 2+1 dimensions either when U(1)gauge fields are coupled to a number n of different typesof fermion with a Dirac dispersion103,104 (where n neednot be very large) or in the presence of a Fermi surface105.The physical significance of these results, and perhaps abroader lesson, is that a deconfined U(1) liquid phasemay arise in condensed matter models of real materialswith two dimensional magnetism despite the fact that theminimal 2 + 1 gauge-only U(1) theory is confining63–65.

In 3 + 1 dimensions, as we have described in preced-ing sections, the U(1) gauge theory can have a decon-fined phase. One of the earliest examples of a bosonicmodel with an emergent low energy electromagnetismis described in Ref. [73] which argues that certain com-pact but non gauge invariant theories can exhibit de-confined emergent electromagnetism at low energies. Ina condensed matter context, apart from the quantum

16

spin ice state of the pyrochlore XXZ model, a U(1)liquid is expected also for the partially magnetized py-rochlore magnet with three of the four tetrahedral sublat-tice spins pointing along an applied field and the remain-ing spins anti-aligned along the field106,107. The result-ing spin model, in common with the XXZ model in zerofield, maps to a dimer model which has been extensivelystudied50,106–109. Apart from quantum magnets, there istheoretical and numerical evidence for the emergence ofa gapless photon in models of exciton condensates110–112,in a rotor model on a cubic lattice113,114 and there is asuggestion that protons in conventional water ice mightbe strongly correlated and, through quantum mechanicaltunneling of the protons, also display a deconfined U(1)phase61. There is also a proposed way of simulating arotor model with emerging photons at low energies in asystem of trapped cold atoms on a pyrochlore lattice115.

The quantum U(1) liquids can be viewed as string-netcondensate phases as made explicit in a three dimen-sional rotor model of Levin and Wen116. String-net con-densation is a framework within which a class of achiralgauge theories can be understood116,117. The idea, statedbriefly, is that these phases can be obtained by specify-ing a lattice model with elementary bosonic degrees offreedom and a local Hamiltonian which condenses closedloops of different types into the ground state of a bosonicmodel. The Hamiltonian may also contain terms whichgive the condensed strings a tension which breaks thedegeneracy of the resulting equal weight superposition ofstring states. Excitations of the string net condensateare the end points of closed strings and correspond tocharges of some gauge theory. This framework offers away to generalize the toric code of Kitaev118 by writ-ing down an infinite set of exactly solvable models withHamiltonians consisting of mutually commuting terms.

Instabilities of the U(1) liquid in ordered magneticphases due to the condensation of visons (which aregapped in the deconfined phase) has been studied inseveral works107,119,120. We also note that a three-dimensional U(1) gauge theory coupled to a bosonic mat-ter field has been found useful to describe transitions outof classical spin ice121, through the condensation of thematter field, to give Higgs phases. In this context, theHiggs phases are magnetically ordered phases within thespin ice manifold.

IV. A MATERIALS PERSPECTIVE

A. General considerations

We discussed in Section III how the spin liquid stateof quantum spin ice (QSI) materials originates from the(perturbative) anisotropic interactions between effectivespins one-half away from the frustrated classical Isinglimit. Central to this story was the point that the modelof Eqn. (4) generates the crucial ring exchange Hamilto-nian of Eqn. (5) that drives classical spin ice into the

U(1) spin liquid state. Yet, we did not explain howEqn. (4) arises from the microscopic interactions betweenthe magnetic ions. We explore in this subsection wherethe effective spin-1/2 model comes from and how it isamended when considering real quantum spin ice candi-date pyrochlore materials such as the ones discussed inSubsection IV B.

In the search for QSI materials, we are de facto seekingsystems with strong effective Ising-like anisotropy suchthat the order zero of the effective spin Hamiltonian isgiven by HCSI in Eqn. (3). Such leading HCSI inter-actions may arise from the single-ion anisotropy, as in(Ho,Dy,Tb)2Ti2O7 or, it may be inherited from the in-teractions between the ions themselves, even though thesingle-ion anisotropy of a would-be isolated ion may notbe Ising-like, as is the case for Yb2Ti2O7

87–89. We arethus looking for materials that possess spin-orbit interac-tions that are stronger than the crystal field interactionand where the orbital angular momentum of the unpairedelectrons is not quenched. In practice, one expects tofind such a situation most prevalently among rare-earth(lanthanide, 4f or actinide, 5f) elements. However, it isperhaps not ruled out that some materials based on 3d,4d or 5d elements may eventually be found to displaysome of the classical or quantum spin ice phenomenol-ogy described above. For example, Co2+ magnetic ionsoften exhibit a strong Ising-like anisotropy, as foundin the quasi-one-dimensional Ising-like antiferromagnetCsCoBr3 compound122. In that context, one may notethat the GeCo2F4 spinel, in which the magnetic Co2+

ions reside on a (pyrochlore) lattice of corner-sharingtetrahedra, has been reported to display a coexistence ofspin ice, exchange and orbital frustration123. In the fol-lowing, we restrict ourselves to insulating magnetic rare-earth pyrochlores of the form R2M2O7

124 where R3+ is atrivalent 4f rare-earth element and M is a non-magnetictetravalent ion such as Ti4+, Sn4+ or Zr4+. Indeed, allmaterials for which QSI phenomenology has so far beeninvoked are based on rare earth pyrochlores.

The hierarchy of energy scales at play in rare-earth ionsmake them well suited for a material exploration of QSIphysics. In these systems, the spin-orbit interaction islarger than the crystal field energies, but not strongerthan the intra-atomic electronic energy scale. Conse-quently, the spin-orbit interaction acts within the statesdefined by Hund’s rules and leads to a 2J+1 degenerateionic ground state of spectroscopic notation 2S+1LJ whereJ=L+S if the 4f electron shell is more than half-filled andJ=L-S otherwise. Note that S here is the electronic spin,not the pseudospin-1/2 S of the Hamiltonian in Eqn. (4)and Eqn. (15) below. The effect of crystal-field pertur-bations (originating from the surrounding ligands) is tolift the degeneracy of the 2S+1LJ isolated ionic groundstate. For the R2M2O7 pyrochlores with Fd3m spacegroup, odd-numbered electron (i.e. Kramers) ions (e.g.Dy, Er, Yb) have a magnetic ground state doublet andso do even-number electron (i.e. non-Kramers) (Pr, Tband Ho) ions124, but not Tm3+ that has a non-magnetic

17

singlet125. In other words, the symmetry is sufficientlylow for Kramers ions to cause them to have solely a dou-blet crystal field ground state, while the symmetry is stillsufficiently high for non-Kramers ions for them (exceptTm3+) to have an accidental magnetic doublet groundstate.

At this point, armed with the knowledge about thesingle-ion crystal field states79, one can in, in principle,start to consider the inter-ionic interactions, Hint, andconstruct the pertinent microscopic (“UV”) Hamiltonianfrom which realistic amendments of Eqn. (4) would re-sult when perturbing the crystal field ground state withHint. It is here that the problem gets complicated – espe-cially when compared with, say 3d transition metal ionsystems. In these 3d systems, the angular momentumis typically quenched, the spin-orbit interaction is smalland one is often dealing with a relatively simple spin-onlyexchange Hamiltonian of the form JijSi · Sj . In systemwith 4f elements, except for Gd3+, most of the angularmomentum is provided by the angular momentum part ofthe atomic electrons and is not quenched. Furthermore,the relevant unfilled 4f orbitals are buried rather deep in-side the inner part of the ion, orbital overlap is reduced,and direct exchange or superexchange do not have the op-portunity to completely dominate in Hint. Consequently,the ion-ion interactions among 4f systems end up havinga multitude of origins: direct classical electric and mag-netic multipole interactions, electric and magnetic mul-tipole interactions arising from direct exchange, superex-change electric and magnetic multipole interactions andlattice-mediated electric multipole interactions126. Themicroscopic couplings Ωij , between ions i and j defin-ing Hint are thus of a very high degree of complexity.The microscopic Hamiltonian Hint can in principle bewritten in terms of Stevens equivalent operators O(Ji).However, such a determination or parametrization, eitherexperimentally or theoretically, of the pertinent (tensor)couplings Ωij between O(Ji) and O(Jj) (we have omit-ted here all angular momentum components that woulddefine the various components of Ωij) is enormously dif-ficult to say the least. On the theoretical front, it is evenquestionable whether estimates of Ωij accurate to within100% of the true couplings could be achieved.

Yet, things are not as hopeless as it appears. Formost cases of interest, we have a situation where theΩij couplings are often of the order of 10−2 K – 10−1

K or so, and are therefore typically small compared tothe gap ∆ ∼ 101 − 102 K between the magnetic crys-tal field ground doublet and the first excited crystal fieldstate(s). This means that the two pairs of crystal fielddoublet wavefunctions for two interacting ions i and j getvery weakly admixed with the excited crystal field statesvia the action of Hint. Out of the (2J + 1)N crystalfield states, where N is the number of ions, one can con-sider only the subspace, Ψ, spanned by the 2N individualcrystal field ground states. One can then consider one ofthe many variants of degenerate perturbation theory toderive an effective S = 1/2 “pseudospin” Hamiltonian,

Heff, 12, describing the perturbed energies and eigenstates

within Ψ. Thanks to the relative high symmetry of thepyrochlore lattice, the projection of Hint into Ψ gives fornearest-neighbours of interacting ions a relatively simpleform for the effective nearest-neighbour interactions be-tween the pseudospins Si and Sj for ions i and j. Theseeffective interactions, Heff, 12

, are the ones that we seek

to spell out the additional symmetry-allowed couplingsin HQSI in Eqn. (4). Several notation conventions haveappeared over the past two years for Heff, 12

87,127,128,130.

Here, we adopt the one introduced in Ref. [87] as it re-lated most directly with Eqn. 4:

Heff, 12(14)

=∑〈i,j〉J‖Szi Szj − J±(S+

i S−j + S−i S

+j ) + J±±[γijS

+i S

+j

+ γ∗ijS−i S−j ] + Jz±[(Szi (ζijS

+j + ζ∗i,jS

−j ) + i↔ j].

(15)

In Eqn. (15), 〈i, j〉 refers to nearest-neighbour sites of thepyrochlore lattice, γij is a 4×4 complex unimodular ma-trix, and ζ = −γ∗87,90. In this formulation, the effectiveSi = 1/2 spins are now expressed in terms of a local triadof orthogonal unit vectors xi, yi and zi with zi along thelocal “Ising” [111] direction, with ± referring to the twoorthogonal complex directions xi ± iyi. As before, wehave HCSI = J‖

∑〈i,j〉 S

zi S

zj , the classical (Ising) term

responsible for the spin ice degeneracy at the nearest-neighbour level. The three other terms (proportional toJ±, J±± and Jz±) are all the extra nearest-neighbourterms allowed by symmetry on the pyrochlore130, thatdo not commute with HCSI, and hence cause quantumdynamics within the classical spin ice manifold. Onemay ask what is the physical origin (or content) of theterms proportional to J±± and Jz± in Heff, 12

. A simple

perspective on this matter goes as follows. One may con-sider the following four nearest-neighbour interactions onthe pyrochlore lattice128: (i) an Ising interaction J‖Szi S

zj

as in Eqn. (15), (ii) an isotropic interaction of the formJisoSi · Sj , (iii) a pseudo-dipolar “exchange” interac-tion that has the same “trigonometric form” as mag-netostatic dipole-dipole, Jpd(Si · Sj − 3 (Si · rij rij · Sj)and, finally, (iv) a Dzyaloshinskii-Moriya interaction of

the form JDM(dij · Si × Sj)129. The set of interactions(J‖, Jiso, Jpd, JDM) can be linearly transformed into theset (J‖, J±, J±±, Jz±).

We are in the very early days of the systematic ex-perimental and theoretical investigation of quantum spinices. On the theoretical front, the immediate questionis to determine what are the possible zero-temperaturephases that Heff, 12

displays. Using a form of gauge mean-

field theory (gMFT), this question has been tackled forsystems with Kramers ions49 as well as non-Kramersions53 (non-Kramers ions must have Jz± = 053). Re-cent work has extended this approach to investigate thenonzero temperature phase diagram of this model72. Ex-amples of phase diagrams produced using this method are

18

shown in Fig. 10. The top two panels show phases ex-pected through a section of the available space of nearest-neighbour couplings for odd electron magnetic ions. Ofthe two exotic phases in the top two panels of this fig-ure, one is the quantum spin ice (QSI) phase which isthe topic of this review. This appears, as we expectfrom perturbation theory in the vicinity of the classi-cal spin ice point. The second (CFM) phase has co-existing symmetry broken long-range order, gapless pho-ton excitations and gapped magnetic monopoles. TheFM and AFM phases are respectively the six-fold degen-erate ferromagnetic state with net moment along one ofthe the 〈001〉 directions and an antiferromagnetic phasewith moments perpendicular to the local 〈111〉 direc-tions. For non-Kramers ions, we refer to the bottompanel of Fig. 10 which shows the quantum spin ice andtwo quadrupolar phases. One of the long-range orderedphases is ferroquadrupolar (FQ) and the other antifer-roquadrupolar (AFQ). The AFQ phase, in the languageof effective pseudospin-1/2, is a coplanar antiferromag-netic phase while the FQ is the maximally polarized fer-romagnet with moments perpendicular to the local Isingdirections53.

For what concerns us here with this review, in terms ofthe existence of a U(1) spin liquid, it probably suffices tosay that as long as J±, either positive or negative, is theleading perturbation beyond J‖, while being sufficientlylarger than J±± and Jz±, one finds a finite region overwhich the U(1) spin liquid exists. Magnetic rare-earthions often possess a sizeable magnetic dipole moment. Anobvious question is how the phase diagram of Refs. [49,53,72] is modified by long-range magnetostatic dipole-dipoleinteractions. Also, away from the U(1) spin liquid phaseof the phase diagram of Ref. [49,53], one may expect toencounter more complicated long-range ordered phasesat nonzero ordering wavevector28,131,132.

We end this subsection by a brief discussion about howthe coupling parameters (J‖, J±, J±±, Jz±) may be de-termined experimentally – a necessary task if one wantsto rationalize the behaviour of real materials in relationto the theoretical phase diagrams49,53. First, we makea comment about how the complexity of the microscopicinteractions Hint ∼ ΩijO(Ji)O(Jj) was swept aside whenHint was projected onto Ψ. As long as |Ωij | ∆, onecan calculate any correlation functions involving the ob-servable Ji angular momentum operators via calculationof correlation functions involving the effective Si spins.In such a case, the theory is quantitative and character-ized by the four (J‖, J±, J±±, Jz±) couplings (to be deter-mined by experiments) along with the magnetic dipole-dipole interaction and the matrix elements of Ji withinthe non-interacting ground state doublet (involving theso-called single-ion g-tensor)79. However, if Hint signifi-cantly admixes excited crystal field states into the groundstate doublet, observables involving Ji are no longer triv-ially related to the pseudospins Si, and the theory isno longer quantitative if the microscopic Ωij couplingsare not known. We return to this in the Subsection

IV B 1 when discussing the Tb2(Ti,Sn)2O7 compounds.The situation is conceptually similar to the problem ofcalculating the staggered magnetization, M†, of the sim-ple square lattice one-band Hubbard model at half-fillingwhen recast as an effective spin-1/2 Hamiltonian. Awayfrom the Heisenberg limit, in which the hopping t is in-finitely small compared to the Hubbard U (i.e. t/U 1),M† is a function of t/U , and is no longer determinedby the textbook formula of the thermodynamic averageof the staggered z component of localized spin opera-tors, Szi , given by M† =

∑i(−1)i〈Szi 〉133. The problem

becomes even more complicated when the microscopicmodel is that of an extended Hubbard model134. In orderto calculate the staggered magnetization with the effec-tive spin-1/2 model the corresponding operator must bedefined in the microscopic (electronic) Hubbard theoryfirst, and then canonically transformed in the effectivelow-energy (spin-1/2) theory133,134.

Following the initial realization that anisotropic ex-change may be of importance in R2M2O7 materials130,a number of experimental studies, mostly using elasticand inelastic neutron scattering, have been targetted todetermine the value of these couplings. These studiesfall in two categories. A first category has assumed thatthe nearest-neighbour part of the interactions Hint be-tween Ji angular momentum operators are bilinear andof anisotropic nature127,135–138. The other group con-sists of studies that do not make this assumption87,90,139

but work, instead, with a model with anisotropic ex-change between pseudospin-1/2 degrees of freedom as inEqn. (15). Ultimately, from the discussion above regard-ing the need to transform observables, one concludes thatif the crystal field gap ∆ is not very large compare to theΩij interactions in Hint, the theory and the ultimate de-scription of the data is not on on a very strong footinganyway from the word go. This is particularly the casefor Tb2Ti2O7

82,83. In the other extreme, for example inYb2Ti2O7, the gap ∆ ∼ 600 K79 is so large that onecan employ either a pseudospin-1/2 representation87,139,or a model with bilinear couplings between the Ji

127,138.In that case the anisotropic bilinear exchange couplings,Ωij , between Ji and Jj are more or less an exact inverselinear transformation of the couplings between the Sipseudospins127. The same argument probably applies toEr2Ti2O7 where a spin-1/2 model90 and one with Ji−Jjcouplings128,136 can be used for zero field of for magneticfields less than a few tesla, though we note that this ma-terial is no quantum spin ice candidate. In the case ofYb2Ti2O7, a recent paper89 has discussed the pitfalls ofsome of the analysis and models employed in a numberof works127,135–137.

Staying with systems that can be described by aneffective spin-1/2 model as in Eqn. (15), two neutronscattering methods have been used to determine theJe ≡ (J‖, J±, J±±, Jz±) couplings. One method is tofit the energy integrated paramagnetic scattering, S(q),obtained at sufficiently high temperature, and compareit with the corresponding S(q) obtained via mean-field

19

theory140,141. If wanting to neglect the long-range dipo-lar interactions, one can also fit the experimental S(q) atsufficiently high temperature with that calculated on thebasis of an approximation using finite clusters142. Suchan approach could be formally rationalised on the basis ofthe so-called numerical linked cluster method (NLC)88,89.It is probably fair to say that mean-field methods havenot yet successfully yielded accurate values for the Jecouplings for any of the R2Ti2O7 materials considered.The main technical difficulty is that one is required toperform experiments in a temperature regime which is atleast 5 to 10 times higher than the mean-field transitiontemperature (Tmf

c ) of the underlying spin-1/2 Hamilto-nian in order for these methods to be quantitative. Atsuch high relative temperatures, the intensity modulationof the experimental S(q) can be quite weak and difficultto fit. Previous works on Yb2Ti2O7

127,139 have consid-ered temperature that are actually below Tmf

c and thereported values of Je couplings are, consequentially,quite inaccurate89. A work on Er2Ti2O7 that comparesthe experimental S(q), obtained at a temperature rea-sonably high compared to the true critical temperature,with that obtained from calculations on a single tetrahe-dron, may be less subject to this concern142.

The other method that is currently enjoying some pop-ularity employs inelastic neutron scattering to probe theexcitations in the field-polarized paramagnetic state87,90.By fitting the dispersion and intensity of these excitationscompared to those calculated from the model (15), sup-plemented by a magnetic Zeeman field term, one can ob-tained the four couplings Je. This approach has beenapplied to determine (fit) the Je for both Yb2Ti2O7

87

and Er2Ti2O790. In these fits, the magnetostatic dipole-

dipole interactions have only been considered at thenearest-neighbour level, and their nonzero value are thusimplicitly folded in the fitted Je values. There is noth-ing in principle that would prevent performing an analy-sis of these in-field excitations that would include the truelong-range dipolar interactions143,144 and thus determinethe “real” Je nearest-neighbour values. In view of thefact that the current Heff model in Eq. (15) neglects ef-fective exchange couplings between the pseudospin-1/2beyond nearest-neighbours, which have been found inthe classical dipolar spin ice Dy2Ti2O7 to be about 10%of the nearest-neighbour J‖26, it is indeed perhaps jus-tifiable to ignore dipolar interactions beyond nearest-neighbours altogether. In that context, it is interest-ing to note that recent calculations that make use of asort of series expansion method based on the numeri-cal linked-cluster (NLC) expansion, has shown that thespecific heat88 and the magnetic field and temperaturedependence of the magnetization89 of Yb2Ti2O7 is welldescribed using the Je determined by the aforemen-tioned fit to inelastic neutron scattering data87.

B. Candidate materials

In this section we briefly discuss four materials amongthe R2Ti2O7 familly that may be candidates for display-ing some of the quantum spin ice (QSI) phenomenol-ogy. These are Tb2Ti2O7, Pr2M2O7 (M=Sn,Zr) andYb2Ti2O7. While these compounds display a numberof attributes that warrant discussing their exotic ther-modynamic properties in the context of QSI physics, itis fair to say that, at the time of writing, that there isno definitive evidence that any one of them displays aquantum spin ice state.

1. Tb2Ti2O7

This was the first material for which the name quan-tum spin ice was coined82. Upon cooling, Tb2Ti2O7

starts to develop magnetic correlations at a temperatureof 20 K or so, but most experimental studies have sofar failed to observe long-range order down to the lowesttemperature145,146. Some early reports found signs forslow dynamics below ∼ 300 mK146. suggesting some kindof spin-freezing/spin-glassy phenomena147. With over-all antiferromagnetic interactions, indicated by a neg-ative Curie-Weiss temperature θCW, one would naivelyexpect this non-Kramers Ising system to display a non-frustrated long-range ordered ground state with all “up”tetrahedra in Fig. 1 having their four spins pointing “in”and all “down” tetrahedra having spins pointing “out”,or vice-versa14,15. By considering an Ising dipolar spinice model with competing nearest-neighbour antiferro-magnetic (J‖ < 0 in Eqn. (3)) and long-range dipolar in-teractions, Ref. [23] found a critical temperature around1 K, in total disagreement with experiments145. Furtherevidence that such an Ising model was too simple forTb2Ti2O7 came from neutron scattering experiments148.These found a broad region in reciprocal space near thepoint q = 002 with high scattering intensity146,148 whichis inconsistent with what is naively expected for an Isingmodel149. Subsequent theoretical work141,149 found thatallowing for the magnetic moments to fluctuate trans-verse to their local [111] Ising direction could lead to ahigh scattering intensity at q = 002. These early observa-tions, along with the fact that such broad q = 002 inten-sity remains down to a temperature of 50 mK146 made itclear that one or more mechanisms had to be consideredto generate non-Ising fluctuations and response down tothe lowest temperature.

The classical spin ice compounds Ho2Ti2O7 andDy2Ti2O7 have a large gap ∆ of order of 300 K betweentheir crystal field ground state doublet and their first ex-cited doublet. As discussed in Section IV A, this featureis at the origin of an Ising description of these systems.In contrast, Tb2Ti2O7 has ∆ ∼ 18 K150. Calculationsof the dynamical structure factor S(q, ω) in the param-agnetic phase that employs the random phase approxi-mation (RPA) method141 make a strong point that this

20

small gap ∆ allows for a significant admixing betweenthe crystal field states that is induced by the interactionsamong the Ji angular momenta through superexchangeand long-range dipolar interaction. These effects are ren-dered even more significant since it appears that, at thelevel of an Ising model description that ignores excitedcrystal field states23, Tb2Ti2O7 is near a boundary be-tween an “all-in/all-out” long range ordered phase and a“2-in”/“2-out” spin ice state23. In other words, sinceprojected interactions Hint in the crystal field groundstate puts the material near a phase boundary betweendistinct classical ground states, corrections beyond thisprojection that involve the details of the Ji − Jj inter-actions, Ωij , and the excited crystal field states must berevisited and incorporated into the effective low-energyHamiltonian.

Considering a simple model of nearest-neighborisotropic exchange Ji · Jj between the Ji angular mo-menta as well as long-range dipole-dipole interactions,Refs. [82,83] found, via second order perturbation the-ory calculations, that the original dipolar Ising spin icemodel description of Tb2Ti2O7 is significantly modi-fied. Rather, an early form of an effective low-energypseudospin-1/2 model similar to that of Eqn. (15), sup-plemented by further anisotropic exchange terms as wellas long-range dipolar interactions, was obtained83. Onemay thus consider that Tb2Ti2O7 is one system for whicha quantum spin ice Hamiltonian of the form such as inEqn. (15) is well motivated. The lack of long-range orderin Tb2Ti2O7 is thus, perhaps, intriguingly related to theU(1) spin liquid of Eqn. (15)82.

Over the past couple of years, however, the situationregarding the phase of Tb2Ti2O7 below a temperatureof 2 K has become perhaps ever more cloudy. The oldevidence151 for a sizeable magneto-elastic response in thissystem has resurfaced152–155. Some authors have pro-posed that, possibly related to this lattice effect, thecrystal field ground state of Tb3+ is split into two sin-glets separated by an energy gap of order of 2 K, andthat this is the principal reason why this system does notorder156,157. Interestingly, the closely related Tb2Sn2O7

compound develops long-range order at 0.87 K into a q =0 long-range ordered version of a “2-in”/“2-out” spin icestate, but with the magnetic moments canted away fromthe strict 〈111〉 Ising directions158. Tb2Sn2O7 has alsobeen proposed to have such a split crystal field grounddoublet159. But, with the suggestion that it has strongerinteractions than its Tb2Ti2O7 cousin, Tb2Sn2O7 is ar-gued to overcome the formation of a trivial non-magneticsinglet ground state and to develop long-range order. Thesuggestion that there exist an inhibiting doublet splitingmechanism leading to a singlet-singlet gap as large as 2K in Tb2Ti2O7 is currently being debated160.

Some recent neutron scattering work on Tb2Ti2O7

finds evidence for pinch-points suggesting the presence of“2-in”/“2-out” spin ice -like correlations (see discussionin Section II A)161. Even more recent inelastic neutronscattering studies have identified some clear and reason-

ably intense features at the q = 12

12

12 reciprocal space

point lattice point suggesting the development of nontriv-ial magnetic correlations at the lowest temperature andwhich have been referred to as “antiferromagnetic spinice correlations”157,162–164. This, along with the obser-vation of elastic magnetic scattering below an energy of0.05 meV∼ 0.5 K, may be viewed as inconsistent with theabove non-magnetic singlet ground state scenario156,157.Finally, it has become clear over the past two years thatthere are significant sample-to-sample variations in thethermodynamic properties exhibited among single crys-tals of Tb2Ti2O7

160,164,165. This may ultimately be thecherry on the cake in terms of the plethora of phenomenaTb2Ti2O7 displays. The sample-to-sample variabilitymay be endorsing a picture that this compound is natu-rally located near the vicinity of a transition between two(or more) competing states. There is probably sufficientevidence in place suggesting that there are spin-ice likecorrelations and transverse fluctuations of the angularmomenta in Tb2Ti2O7 so that a quantum spin ice pictureis not, at this time, ruled out. However, there is definitelymore to the story and there are numerous hints thatthe lattice degrees of freedom are not inert bystandersin Tb2Ti2O7, and that magneto-elastic couplings shouldprobably be considered carefully. In that context, thepossibility of magneto-elastic interactions and the con-current existence of symmetry related quadrupolar-likeinteractions naturally brings up the question: “what isthe role of quadrupolar-like interactions in even electron(non-Kramers) magnetic ion systems in modifying thesimplest (Ising magnet) description of these systems?”This question has been explored with the quantum spinice candidates, Pr2M2O7 (M=Sn,Zr), that we next dis-cuss.

2. Pr2Sn2O7 & Pr2Zr2O7

It might be said that the modern research era in frus-trated quantum spin systems was, at least partially, trig-gered by Anderson’s 1987 Science paper1. In this pa-per, Anderson noted that geometrical frustration mightnaturally lead to exotic quantum states of magneticmatter, including resonating valence bond (RVB) statesfrom which unconventional superconductivity may arise.Unfortunately, if the field of condensed matter physicshas long been experiencing a drought in the abundanceof (quantum) spin liquid candidate materials166, thescarcity of highly frustrated magnetic materials that areat the verge of a Mott-insulator transition, or displaysimultaneously frustrated localized magnetic momentsalong with itinerant electrons, may remind one of theMartian atmosphere. From that perspective, frustrationand development of superconductivity in organic Mott in-sulators is more than a curiosity167. In that context, it isperhaps not surprising that the discovery of spin-ice like“2-in”/“2-out” correlations, signalling geometrical frus-tration for would-be isolated non-Kramers Pr3+ ions, in

21

the metallic Pr2Ir2O7 pyrochlore compound168,169 has at-tracted a fair amount of interest170–172. Before attackingthe complexities of Pr2Ir2O7, such as the resistivity min-imum as a function of temperature (Kondo-like effect)168

and anomalous Hall effect169, one may wonder whether,even in insulating Pr-based compounds, there might ex-ist unusual properties that may be of relevance to thephysics of metallic Pr2Ir2O7.

The pyrochlore form of Pr2Ti2O7 does not exist124 atambient temperature and pressure. However, the insu-lating and magnetic Pr2Sn2O7 and Pr2Zr2O7 compoundsdo exist and both form a regular pyrochlore structure.In these two materials, the Pr3+ non-Kramers ions pos-sess a magnetic crystal field Ising doublet ground state,as the Ho2Ti2O7 spin ice and the above paradoxicalTb2Ti2O7. Pr2Sn2O7 is, unfortunately, not amenable tosingle-crystal growth using modern image furnace meth-ods, but Pr2Zr2O7 has been grown successfully. Inter-estingly, neither material appears to develop long-rangeorder down to temperatures of the order of 50 mK.AC susceptibility measurements found a spin freezingin Pr2Sn2O7 below a temperature of order of 100 − 200mK173. Powder neutron diffraction on Pr2Sn2O7 revealshort range correlations and, surprisingly, it has a resid-ual low-temperature entropy larger than the Pauling en-tropy S0 found in classical spin ices85. Inelastic neutronscattering suggests that the low-temperature state of thissystem remains dynamic down to at least 200 mK85.

Early work on Pr2Zr2O7174 found a negative Curie-

Weiss temperature of −0.55 K (determined below a tem-perature of 10 K), indicating effective antiferromagneticinteractions between the magnetic moments. AC mag-netic susceptibility measurements did not find evidencefor a transition to long-range order down to 80 mK. How-ever, some frequency dependence of the AC susceptibilitywas observed below 0.3 K, indicating some form of spinfreezing. More recent work on single crystals of Pr2Zr2O7

report evidence of spin ice -like correlations and quantumfluctuations86. Heat capacity and magnetic susceptibil-ity measurements show no sign of long-range order downto 50 mK. The wave vector dependence of quasi-elasticneutron scattering at 100 mK shares some similaritieswith one of a classical Ising spin ice, including pinch-points. These results are interpreted as an indication of“2-in”/“2-out” ice rule being satisfied over a time scaleset by the instrumental energy resolution. Quite interest-ing, and in sharp contrast with classical spin ices wherealmost no inelastic response is observed175, inelastic scat-tering in Pr2Zr2O7 with an energy transfer of 0.25 meVdoes not show pinch points. This suggests that there arefluctuations operating which break the ice rule.

In summary, it appears that the insulatingPr2(Sn,Zr)2O7 compounds develop significant cor-relations at temperatures below approximately 1 K, butdo not develop true long range order nor do they appearto behave like the conventional Ho2Ti2O7 and Dy2Ti2O7

classical dipolar spin ices. As in these latter systems,the lowest excited crystal field levels in Pr2(Sn,Zr)2O7

are at high energy, with a gap ∆ ∼ O(102) K abovethe ground doublet85. Consequently, the virtual crystalfield fluctuation mechanism induced by the interactionsbetween the magnetic moments, and proposed to beat play in Tb2Ti2O7 because of its small gap ∆ 18K82,83, is likely not significant in these two Pr-basedcompounds. It has been suggested that the microscopicHint interactions in these Pr-based compounds containstrong multipolar interactions between the Ji angularmomenta operators81,176 and that these introducequantum fluctuations that “melt” the low temperatureclassical spin ice state that would have developed intheir absence. The idea, although not quite presentedin this form in Refs. [81,176] is that high multipolarinteractions between the Ji operators, which involvehigher powers of Ji than simpler bilinear exchange-likecouplings Ωuvij J

ui J

vj , where u, v are cartesian compo-

nents, have large matrix elements between the two statesforming the crystal field doublet ground state of Pr3+

in Pr2(Sn,Zr)2O7. At the end of the day, the projectionof the derived complex microscopic inter-ionic Pr-Printeractions onto the crystal field ground doublet leadsto transition matrix elements between the two states,|ψ±〉 that make up the doublet, these bringing aboutquantum dynamics between |ψ+〉 and |ψ−〉. An effectivepseudospin-1/2 Hamiltonian, which describes thatphysics, can then be constructed and found to be of theform of Heff, 12

, but with Jz± = 0 since degenerate states

of non-Kramers ions have vanishing matrix elementsof time-odd operators such as J. We thus reach theinteresting conclusion that Pr-based quantum spin icecandidates are rather attractive from the perspective ofsystematic experimental and theoretical investigations:(i) the effect of the excited crystal field states canprobably be safely ignored, (ii) the magnetic momentµ ∼ 3 µB means that dipolar interactions are 10 timesweaker than in the 10 µB Ho2Ti2O7 and Dy2Ti2O7

classical spin ices materials23,24 and can be neglected asa first approximation down to about 0.1 K. and, beingnon-Kramers ions, their effective pseudospin-1/2 Hamil-tonian consists of only three couplings (J‖, J±, J±±),making the theoretical description of these materialsmore sober in terms of number of free parameters.

The systematic theoretical53,72,81,176 and experimen-tal exploration of quantum spin ice physics in Pr-basedmaterials has now begun in earnest85,86 and one may ex-pect exciting results to emerge from future studies. Thatsaid, given the difficulty involved in synthesizing cleanSn-based and Zr-based pyrochlore oxides124, one could,or perhaps even will, always worry about the effect thatweak/dilute non-symmetry-invariant perturbations mayhave on the Pr-based materials given that the magneticcrystal field ground state is not protected by the Kramerstheorem. The above discussion about Tb2Ti2O7 andPr2(Sn,Zr)2O7 allow us to rationalize the naturalnessof the next class of materials candidates for the studyof quantum spin ice phenomenology. We desire materi-als with large energy gaps ∆ between their crystal field

22

ground state and their first excited energy level. Ideally,they should have small magnetic dipole moments (sayless than 3 µB) so that dipolar interactions may be ne-glected, at least initially. Were it not for the concernof disorder breaking their crystal field degeneracy, non-Kramers ions would seem appealing candidates becausethe single ion dipolar doublets are Ising-like. The crystalfield doublets of Kramers ions are stable against smalllocal crystal field deformations that may be caused byimperfect sample quality but the low energy anisotropyis not constrained to be Ising-like and, indeed, rare earthKramers ions are known that span the range from Ising toHeisenberg to XY spins. While the extent of the availableparameter space might disfavour Kramers magnets as po-tential quantum spin ice candidates, it certainly does notrule them out and in fact, Yb2Ti2O7, which we discussnext, is attracting much current interest in this respect.

3. Yb2Ti2O7

Going back to some of the very earliest experimen-tal studies of magnetic pyrochlore oxides, Blote andco-workers had observed in Yb2Ti2O7 a broad specificheat bump at a temperature of about 2 K, followedat lower temperature by a sharp specific heat peak ata critical temperature of Tc ∼ 0.214 K suggesting atransition to long-range order177. It was not until un-til the late 1990s and early 2000s that this compoundwas reinvestigated24,178, with the previously observed177

sharp specific heat transition confirmed. The work ofRef. [178] reported data from 170Yb Mossbauer andmuon spin relaxation (muSR) measurements revealinga rapid collapse of the spin fluctuation rate just aboveTc. However, powder neutron diffraction did not findsigns of long-range magnetic order below Tc and muSRfound a temperature-independent muon spin depolariza-tion rate below Tc which was interpreted as a quan-tum fluctuation regime. The work of Ref. [178] pro-vided strong evidence that the Yb3+ ion in Yb2Ti2O7

should be viewed as an XY system (i.e. with g tensorcomponents g⊥ > g‖), meaning that the magnetic mo-ments have their largest magnetic response perpendicularto the local [111] direction. A subsequent single-crystalneutron scattering study reported evidence for ferrimag-netic order179, but this was soon contested by neutrondepolarization measurements180. In Refs. [181] and182,neutron scattering measurements revealed the develop-ment of rods of scattering intensity along the 〈111〉 di-rections, which was interpreted as the presence of quasi-two-dimensional spin correlations, an interesting and un-usual phenomenon, assuming this interpretation to becorrect, for a three-dimensional cubic system. Theserods of scattering were found to be present at a tem-perature as high as 1.4 K127. The application of a mag-netic field as low as 0.5 Tesla along the [110] directionwas found to induce a polarized three-dimensional orderaccompanied by spin waves182. Polarized neutron scat-

tering measurements found, through an analysis of theneutron spin-flip ratio135,136, a non-monotonic tempera-ture evolution of the component of the local spin suscep-tibility, χloc, parallel to the local [111] direction. Suchbehaviour is surprising for an XY system with g⊥ > g‖,and the behavior observed for χloc provided compellingevidence for strongly anisotropic effective exchanges atplay in Yb2Ti2O7, with a very strong effective Ising ex-change J‖ term as in Eqn. (15).

As discussed in Section IV A, severalworks87,127,135–137,139 have endeavoured to deter-mine the strength of the interactions in Yb2Ti2O7. Itappears that the effective coupling between pseudospins1/2 are in fact strongly anisotropic, with the largest onebeing indeed J‖. Among all values having been reported,it appears that the exchange parameters determined inRef. [87] describe the bulk thermodynamic properties ofthe material reasonably well, at least down to 0.7K88,89.Quite recently, the previously debated180 report ofa long-range ferrimagnetic order179 in Yb2Ti2O7 hasbeen reconfirmed139. In this ferrimagnetic state, themagnetic moment are found to be predominantlyaligned along one of the six 〈100〉 cubic directions, butslightly splayed away from complete alignment, hencethe label ferrimagnetic state. In this context, it isworth mentioning a very recent paper on the closelyrelated Yb2Sn2O7 material183. In this latter work,specific heat, 170Yb Mossbauer, neutron diffraction andmuon spin relaxation measurements on powder samplesfind a first order transition at 0.15 K to a state thatMossbauer and neutron diffraction suggest to be theabove ferrimagnetic order, referred to as long-range“splayed ferromagnetic” order. The situation about thenature of the low-temperature state of Yb2Ti2O7 is thusrather confusing. It may be potentially illuminating toknow that there is significant sample-to-sample variabil-ity among single crystal samples as inferred from thesharpness of the Tc ∼ 0.2 K specific heat peak94,184–186.A recent extensive structural study investigation186 hasidentified at least one origin of these variations: singlecrystals grown via the floating zone technique show,compared to sintered powder samples, that up to 2.3 %of the non-magnetic Ti4+ sites get replaced by magneticby Yb3+. Such “stuffing” of the transition metal ionsite Yb3+ ions would introduce random exchange bondsand local lattice deformations and these may be at theorigin of the mechanism affecting the stability of themagnetic ground state of a would-be structurally perfectYb2Ti2O7 material. Finally, we note that new and veryrecent muon spin relaxation results find no evidence forthe development of static order in either powder or singlecrystal samples of Yb2Ti2O7 and, unlike in Ref [178] findno rapid collapse of the Yb3+ spin fluctuation rate uponapproaching the transition at ∼ 0.2 K from above94.

The overall situation with Yb2Ti2O7 is thus as fol-lows. The effective exchange interactions are stronglyanisotropic. On the basis of the determined87 andreasonably well validated88,89 exchange couplings, sim-

23

ple mean-field theory87, classical ground state energyminimisation187, and more sophisticated gauge mean-field theory calculations49,72 predict a conventional long-range ferrimagnetic order with the spins slightly splayedaway from from the six cubic 〈111〉 directions as recentlyreported for Yb2Sn2O7

183 and characterized by negligi-ble quantum fluctuations87. On the basis of these samecalculations, Yb2Ti2O7 is predicted to be located deeplyin this semi-classical splayed ferromagnetic state, awayfrom any phase transition boundaries with other conven-tional classical long-range ordered phases187, or with theU(1) quantum spin liquid phase or yet with the uncon-ventional quantum Coulomb ferromagnetic (CFM) phasethat Refs. [49,72] predict. It therefore seems likely thatneither Yb2Ti2O7 nor Yb2Sn2O7

183 are good realizationsof the sought U(1) liquid in a quantum spin ice set-ting. That said, with a tendency towards a broken dis-crete symmetry state (i.e. ferrimagnetic order along oneof 〈100〉 directions) characterized by gapped excitationsthroughout the Brillouin zone, it is rather unclear whythis material should be so sensitive to dilute disorder suchas the one generated by Yb3+ stuffing on Ti4+ sites186.One might then expect that if the observed amount of dis-order and strength was greater than some critical value,that the resulting random frustration would then firstdrive the system into a semi-classical spin glass state. Weare not aware of experimental studies having reported re-sults suggesting a spin glass state in Yb2Ti2O7.

Let us end by stating that further experimental studiesof Yb-based pyrochlores to search for either the U(1) liq-uid or the CFM phase are certainly most warranted. Per-haps variants such as Yb2Ge2O7, Yb2Zr2O7 or Yb2Hf2O7

could display interesting properties. As discussed in Sec-tion 3, we would expect a material finding itself in theU(1) quantum spin liquid state to be robust for a finitevariation in the microscopic interaction parameters. Inthis context, it may be worth noting that the disorderedYb2GaSbO7 material does not display a sharp specificheat peak near 0.2 K with its uniform susceptibility be-low 10 K characterized by an antiferromagnetic Curie-Weiss temperature, θCW ∼ −2.3 K, unlike Yb2Ti2O7

for which θCW is ferromagnetic-like with θCW ∼ +0.4K177. Notwithstanding the random spin coupling causedby the GaSb randomness on the transition metal ion site,a new generation of experiments (e.g AC and DC suscep-tibility, neutron, Mossbauer, muSR measurements) onYb2GaSbO7 may prove interesting.

We conclude by saying that the three materials dis-cussed in this section, Tb2Ti2O7, Pr2(Sn,Zr)2O7 andYb2Ti2O7 are all described to some extent by the type ofeffective spin-1/2 Hamiltonian of Eqn. (15) from whichone obtains an exotic gapless U(1) spin liquid statewith gapped electric and magnetic excitations, in addi-tion to gapless photons. However, none of them have(yet) been found to be a clear realization of such astate. The wide variety of materials in the family ofR2M2O7 pyrochlore oxides124 offer the possibility thatone or more of these compounds may eventually prove

compelling candidates to support the exotic physics ofthe U(1) spin liquid. In such a case, the experimen-tal and theoretical lessons learned while investigating(Tb,Pr,Yb)2(Ti,Zr,Sn)2O7 will undoubtedly prove use-ful.

It would also perhaps be interesting to investigate inmore detail the AR2S4 and AR2Se4 (A=Cd,Mg) chalco-genide spinels in which the R3+ rare-earth ion sits on apyrochlore lattice22,188. In the context of spin ice-like sys-tems, CdDy2Se4 likely has XY-like Dy3+ ions and couldprove particularly interesting187.

V. CONCLUSION

In the foregoing sections, we have given an account ofa theoretical proposal that effective spin-1/2 pyrochloremagnets, close to the Ising limit, may have a spin liquidground state with gapless photon-like excitations − theso-called quantum spin ice state. On the purely theoreti-cal front, there is strong evidence both from effective fieldtheory and numerics that quantum fluctuations acting onthe set of spin ice states can lead to just such a quantumspin liquid ground state. We can, moreover, make strongstatements about the experimental observables in quan-tum spin ice. For example, unlike the case with manyspin liquids, the gapless excitations carry spin one andcouple directly to neutrons and would therefore be re-solved as a sharp band of scattering intensity revealingthe linear dispersion of the emergent photon.

Quantum spin ice exhibits a hierarchy of energy scales.The crossover between the quantum spin ice state andclassical spin ice is controlled by the magnitude of thehexagonal ring exchange. At higher scales, there is agap to the creation of spin ice defects − the magneticmonopoles of Ref. [31]. The schematic recipe for makinga quantum spin ice is that the material exchange shouldbe dominated by an Ising coupling either coming purelyfrom the balance of exchange parameters or assisted bythe presence of an Ising-like single-ion anisotropy. Thereshould be weaker transverse exchange couplings whichgenerate the effective ring exchange. For realistic val-ues of the couplings in real materials, the hexagonal ringexchange can vary over many orders of magnitude but,inevitably, in rare-earth magnets it will commonly be 10to 100 times smaller than J‖. If this is the case, then it isimportant to consider the effect of competing couplingson the quantum spin ice state. The success of the ex-perimental search for quantum spin ice is contingent onwhether the spin liquid is natural in the space of avail-able couplings within an effective spin-1/2 model: whichconsists of four linearly independent nearest neighbourexchange, the long-range dipole coupling, further neigh-bour exchange, higher order ring exchange and, poten-tially, couplings to non-magnetic degrees of freedom.

On the theoretical side, recent work has mapped outthe phase diagram in the presence of the symmetry-allowed nearest neighbour exchange couplings using a

24

type of gauge mean-field theory that allows one tostudy directly the fractionalized gauge theory degreesof freedom as well as conventional magnetically orderedphases49,72. From an experimental point of view, thefinding from this study is encouraging: the quantumspin liquid ground state lives in a significant region inthe space of parameters. It would be interesting to lookat the situation beyond mean field theory and to con-sider further the phenomenology of quantum spin ice asit crosses over into classical spin ice. One issue is howone might probe directly the gapped excitations in thequantum spin ice state and their classical analogues athigher temperature.

On the experimental side, we have discussed in somedetail four materials among the rare-earth pyrochloreswhich meet the simple criteria of having an Isinganisotropy with weaker transverse fluctuations of dif-ferent microscopic origin. Unfortunately, while all fourmaterials are associated with potentially very interest-ing open questions relating to the effect of disorder(Yb2Ti2O7 and Tb2Ti2O7) and the precise nature of thelow temperature state (Tb2Ti2O7, Yb2Ti2O7 and thePr3+ based materials), none appears to exhibit cleanlythe quantum spin ice phase that is the main topic of thisreview. The existence of these materials does, however,demonstrate the naturalness of magnets in the vicinityof classical spin ice with added quantum fluctuations de-scribed, at least partially, by a model such as in Eqn. (15).In addition, the last few years have seen considerable the-

oretical progress in developing a quantitative understand-ing of this class of materials. Given that many magnetsamong the pyrochlore rare-earths remain to be investi-gated - a number have been mentioned in the preced-ing pages - and given our, now, reasonable mature un-derstanding of the broad series of pyrochlore magnets,the next few years should see similar rapid progress inmapping out the experimental parameter space for thesematerials. Optimistically, within a few years, we will un-derstand that the materials discussed in this review areskirting around the edge of a real and significant regionexhibiting spin liquid ground states and we will also havediscovered examples of real pyrochlore materials that dolive in that region and are the gapless spin liquids adver-tised in the title of this review.

VI. ACKNOWLEDGEMENTS

We thank Zhihao Hao, Roderich Moessner, Frank Poll-mann, Pedro Ribeiro and Arnab Sen for useful discus-sions. This work is supported in part by the NSERCof Canada and the Canada Research Chair program(M.J.P.G., Tier 1) and by the Perimeter Institute forTheoretical Physics. Research at the Perimeter Instituteis supported by the Government of Canada through In-dustry Canada and by the Province of Ontario throughthe Ministry of Economic Development & Innovation.

1 Anderson P W 1987 Science 235 11962 Balents L 2010 Nature 464 1993 Bramwell S J and Gingras M J P 2001 Science 294 14954 Bramwell S T Gingras M J P Holdsworth P C W 2004

“Spin Ice”, Frustrated Spin Systems, ed. Diep H, WorldScientific

5 Gingras M J P 2011 “Spin ice”, In Introduction to Frus-trated Magnetism, ed. Lacroix C, Mendels P Mila F, 293-330. New York: Springer

6 Anderson P W 1956 Phys. Rev. 102 10087 Bernal D Fowler R H 1933 J. Chem. Phys. 1 5158 Pauling L. 1935 J. Am. Chem. Soc. 57 26809 Giauque W F Ashley M F 1933 Phys. Rev. 43 81; Giauque

W F Stout J W 1936 J. Am. Chem. Soc. 58 114410 Nagle J F 1966 J. Math Phys. 7 148411 Singh R R P Oitmaa J 2012 Phys. Rev. B 85 14441412 Harris M J et al. 1997 Phys. Rev. Lett. 79 255413 Harris M J 1998 J. Mag. Magn. Mater. 177 75714 Bramwell S T Harris M J 1998 J. Phys. Condens. Matter

10 L21515 Moessner 1998 Phys. Rev. B 57 R558716 Cornelius A L Gardner J S 2001 Phys. Rev. B 64 06040617 Ramirez A P 1999 Nature 399 33318 Prando G et al. 2009 J. Phys. Conf. Ser. 145 01203319 Ke X et al. 2007 Phys. Rev. B 76 21441320 Pomaranski D et al. 2013 Nature Physics 9 35321 Zhou H D et al. 2012 Phys. Rev. Lett. 108 20720622 Lago J et al. 2010 Phys. Rev. Lett. 104 247203

23 den Hertog B C Gingras M J P 2000 Phys. Rev. Lett. 843430

24 Siddharthan R et al. 1999 Phys. Rev. Lett. 83 185425 Ruff J P C Melko R G Gingras M J P 2005 Phys. Rev.

Lett. 95 09720226 Yavors’kii T et al. 2008 Phys. Rev. Lett. 101 03720427 Gingras M J P den Hertog B C 2001 Can. J. Phys. 79

133928 Melko R G den Hertog B C Gingras M J P 2001 Phys.

Rev. Lett. 87 06720329 Melko R G Gingras M J P 2004 J. Phys.:Condens. Matter

16 R127730 Isakov S V Moessner R Sondhi S L 2005 Phys. Rev. Lett.

95 21720131 Castelnovo C Moessner R and Sondhi S L 2008 Nature

451 4232 Castelnovo C Moessner R Sondhi S L 2012 Annu. Rev.

Condens. Matter Phys. 3 3533 Henley C L 2005 Phys. Rev. B 71 01442434 Henley C L 2010 Annu. Rev. Condens. Matter Phys. 1

17935 Fennell T et al. 2009 Science 326 41536 Jaubert L D C Holdsworth P C W 2009 Nat. Phys. 5 25837 Jaubert L D C Holdsworth P C W 2011 J. Phys.: Con-

dens. Matter 23 16422238 Bramwell S T et al. 2009 Nature 461 95639 Giblin S R et al. 2010 Nat. Phys., 7 25240 Dunsiger S R et al. 2011 Phys. Rev. Lett. 107 207207

25

41 Blundell S J 2012, Phys. Rev. Lett. 108 14760142 Sala G et al. 2012 Phys. Rev. Lett. 108 21720343 Revell H M et al. 2013 Nat. Phys. 9 3444 Bramwell S T et al. 2001 Phys. Rev. Lett. 87 04720545 Sen A Moessner R and Sondhi S L 2013 Phys. Rev. Lett.

110 10720246 Morris D J P et al. 2009 Science 326 41147 Kadowaki H et al. J. Phys. Soc. Jpn. 2009 78 10370648 Hermele M Fisher M P A Balents L 2004 Phys. Rev. B

69 06440449 Savary L and Balents L 2012 Phys. Rev. Lett. 108 03720250 Moessner R Sondhi S L 2003 Phys. Rev. B 68 18451251 Senthil T and Fisher M P A 2000 Phys. Rev. B 62 785052 Weinberg S W 2005 Quantum Theory of Field Volume

One, CUP53 Lee S Onoda S Balents L 2012 Phys. Rev. B 86 10441254 Shannon N et al. 2012 Phys. Rev. Lett. 108 06720455 Benton O Sikora O Shannon N 2012 Phys. Rev. B 86

07515456 Rokhsar D S and Kivelson S A 1998 Phys. Rev. Lett. 61

237657 Henley C L 2004 J. Phys.:Condens. Matter 16 S89158 Lauchli A M Capponi S and Assaad F F 2008 J. Stat.

Mech. P0101059 Banerjee A et al. 2008 Phys. Rev. Lett. 100 04720860 ‘Field Theories of Condensed Matter Systems’, E. Frad-

kin, Addison-Wesley Pub. Co., (1991).61 Castro Neto A H Pujol P and Fradkin E 2006 Phys. Rev.

B 74 02430262 Saito M, Higashinaka R and Maeno Y 2005 Phys. Rev. B

72 14442263 ‘Gauge Fields and Strings’, A. M. Polyakov, Harwood

Academic Publishers, (1987)64 Polyakov A M 1975 Phys. Lett. B 59 8265 Polyakov A M 1977 Nucl. Phys. B 120 42966 Banks T Myerson R Kogut J 1977 Nucl. Phys. B 129 49367 Guth A H 1980 Phys. Rev. D 21 229168 Georgi H 1975 Particles and Fields Ed. C. Carlson, AIP

Conf. Proc. No. 23 (AIP, New York, 1975); Fritzsch H andMinkowski P 1975 Ann. Phys. 93 193

69 Milstead D and Weinberg E J Particle Data Book J. Phys.G 37 075021 (2010).

70 Polyakov A M 1978 Phys. Lett. B 72 47771 Susskind L 1979 Phys. Rev. D 10 261072 Savary L and Balents L 2013 Phys. Rev. B 87 20513073 Foerster D Nielsen H B and Ninomiya M 1980 Phys. Lett.

B 94 13574 Wen X-G 2004 Quantum Field Theory of Many-Body Sys-

tems, OUP75 Hastings M B and Wen X G 2005 Phys. Rev. B 72 04514176 Balian R Drouffe J M and Itzykson C 1974 Phys. Rev. D

10 337677 Drouffe J M 1982 Phys. Lett. B 105 4678 Rosenkranz et al. 2000 J. Appl. Phys. 87 591479 Bertin A et al. 2012 J. Phys.:Condens. Matter 24 25600380 Onoda S 2011 J. Phys.:Conf. Ser. 320 01206581 Onoda S Tanaka Y 2011 Phys. Rev. B 83 09441182 Molavian H R Gingras M J P and Canals B 2007 Phys.

Rev. Lett. 98 15720483 Molavian H R McClarty P A and Gingras M J P

arXiv:0912.295784 Lin T et al. arXiv:1303.724085 Zhou H D et al. 2008 Phys. Rev. Lett. 101, 22720486 Kimura K et al. 2013 Nature Comm. 4 1934

87 Ross K A et al. 2011 Phys. Rev. X 1 02100288 Applegate R et al. 2012 Phys. Rev. Lett. 109 09720589 Hayre N R et al. 2013 Phys. Rev. B 87 18442390 Savary L et al. 2012 Phys. Rev. Lett. 109 16720191 Oitmaa J et al. arXiv:1305.293592 Chang L-J et al. 2012 Nature Communications 3 99293 Dalmas de Reotier P et al. 2012 Phys. Rev. B 86 10442494 D’Ortenzio R M et al. arXiv:1303.3850 and references

therein.95 Baskaran G and Anderson P W 1988 Phys. Rev. B 37 58096 Affleck I and Marston J B 1988 Phys. Rev. B 37 377497 Chakravarty S et al. 2002 Phys Rev. B 66 22450598 Shannon N, Misguich G and Penc K 2004 Phys. Rev. B

69 220403(R)99 Syljuasen O F and S. Chakravarty S 2006 Phys. Rev. Lett.

96 147004 (2006)100 Chern C-H and Nagaosa N arXiv:1301:4744101 Wen X-G 2002 Phys. Rev. Lett. 88 011602102 Wen X-G 2002 Phys. Rev. B 65 165113103 Hermele M et al. 2004 Phys. Rev. B 70 214437104 Nogueira F S Kleinert H 2005 Phys. Rev. Lett. 95 176406105 Lee S-S 2008 Phys. Rev. B 78 085129106 Bergman D L 2006 Phys. Rev. Lett. 96 097207 [erratum

: 2006 Phys. Rev. Lett. 97 139906]107 Bergman D L Fiete G A and Balents L 2006 Phys. Rev.

B 73 134402108 Sikora O 2009 Phys. Rev. Lett. 103 247001109 Sikora O 2011 Phys. Rev. B 84 115129110 Lee S-S Lee P A 2005 Phys. Rev. B 72 235104111 Lee S-S Lee P A 2006 Phys. Rev. B 74 115101112 Lee S-S Lee P A 2006 Phys. Rev. B 74 035107113 Motrunich O I and Senthil T 2002 Phys. Rev. Lett. 89

277004114 Motrunich O I and Senthil T 2005 Phys. Rev. B 71 125102115 Tewari S Scarola V W Senthil T Das Sarma S 2006 Phys.

Rev. Lett. 97 200401116 Levin M and Wen X-G 2006 Phys. Rev. B 73 035122117 Levin M and Wen X-G 2005 Phys. Rev. B 71 045110118 Kitaev A Y 2003 Ann. Phys. 303 2119 Hermele M Senthil T Fisher M P A 2005 Phys. Rev. B

72 104404120 Alicea J 2008 Phys. Rev. B 78 035126121 Powell S 2011 Phys.Rev.B 84 094437122 Yelon W B, Cox D E and Eibschutz 1975 Phys. Rev. B

12 5007123 Tomiyasu K 2011 Phys. Rev. B 84, 054405124 Gardner J S et al. 2010 Rev. Mod. Phys. 82 53125 Zinkin M P et al. 1996 J. Phys.: Condens. Matter 8 193126 Santini P et al. 2009 Rev. Mod. Phys. 81, 807127 Thompson J D et al. 2011 Phys. Rev. Lett. 106, 187202128 McClarty et al. 2009 J. Phys.: Conf. Ser. 14 012032129 Canals B Elhajal M and Lacroix C 2008 Phys. Rev. B 78

214431130 Curnoe S H 2008 Phys. Rev. B 78 094418131 McClarty P A Stasiak P Gingras M J P arXiv:1011.6346132 Javanparast B et al. arXiv:1310.5146133 Delannoy J.-Y. et al. 2005 Phys. Rev. B 72, 115114134 Delannoy J.-Y. et al. 2009 Phys. Rev. B 79 235130135 Cao H B et al. 2009 Phys. Rev. Lett. 103, 056402136 Cao H B et al. 2009 J. Phys.: Condens. Matter 21, 492202137 Malkin B Z 2010 J. Phys.: Condens. Matter 22 276003138 Thompson J D et al. 2011 J. Phys. Condens. Matter 23,

164219139 Chang L.-J. et al. 2012 Nat. Comm. 3, 992

26

140 Enjalran M and Gingras M J P 2004 Phys. Rev. B 70,174426

141 Kao Y.-J. et al. 2003 Phys. Rev. B 68, 172407142 Dalmas de Reotier et al. P 2012 Phys. Rev. B 86 104424143 Del Maestro A G and Gingras M J P 2004 J. Phys.: Con-

dens. Matter 16, 3339144 Del Maestro A G and Gingras M J P 2007 Phys. Rev. B

76, 064418145 Gardner J S et al. 1999 Phys. Rev. Lett. 82, 1012146 Gardner J S et al. 2003 Phys. Rev. V 68, 180401147 Luo G et al. 2001 Physics Letters A 291 306148 Gardner J S et al. 2001 Phys. Rev. B 64, 224416149 Enjalran M and Gingras M J P 2004 Phys. Rev. B 70,

174426150 Gingras M J P et al. 2000 Phys. Rev. B 62, 6496151 Mamsurova L G et al. 1986 JETP Lett. 43, 755152 Mirebeau I et al. 2002 Nature 420, 54153 Ruff J P C et al. 2007 Phys. Rev. Lett. 99, 237202154 Malkin B Z et al. 2010 J. Phys.: Condens. Matter 22

276003155 Fennell T et al. arXiv:1305:5405156 Bonville P et al. 2011 Phys. Rev. B 84 184409157 Petit S. et al. 2012 Phys. Rev. B 86, 174403158 Mirebeau I et al. 2005 Phys. Rev. Lett. 94, 246402159 Petit S et al. 2012 Phys. Rev. B 85, 054428160 Gaulin B D et al. 2011 Phys. Rev. B 84, 140402161 Fennell T 2012 Phys. Rev. Lett. 109, 017201162 Fritsch K et al. 2013 Phys. Rev. B 87 094410163 Guitteny S 2013 Phys. Rev. Lett. 111 087201164 Taniguchi T et al. 2013 Phys. Rev. 87, 060408(R)165 Chapuis Y 2009 Ph.D. thesis, Universite Joseph Fourier166 Lee P A 2008 Science 321, 1306167 Powell B J and McKenzie R H (2011) Rep. Prog. Phys.

74 056501168 Nakatsuji S et al. 2006 Phys. Rev. Lett. 96, 087204 (2006)169 Machida Y et al. 2010 Nature 210 463170 Flint R and Senthil T 2013 Phys. Rev. B 87 125147. See

references therein171 Rau J G and Kee H-Y arXiv:1306.5749 and references

therein172 Lee S et al. arXiv:1305.0827 and references therein173 Matsuhira K et al. 2004 J. Mag. Magn. Mater. 272-276,

e981-e982174 Matsuhira K et al. 2009 J. Phys.: Conf. Ser. 145, 012031175 Clancy J P et al. 2009 Phys. Rev. B 79, 014408176 Onoda S Tanaka Y 2010 Phys. Rev. Lett. 105 047201177 Blote H W J et al. 1969 Physica 43, 549178 Hodges J A et al. 2002 Phys. Rev. Lett. 88, 077204179 Yasui Y et al. 2003 J. Phys. Soc. Japan 72, 3014180 Gardner J S et al. (2004) Phys. Rev. B 70, 180404181 Bonville P et al. 2004 Hyperfine Interactions 156/157

103182 Ross K A et al. 2009 Phys. Rev. Lett. 103, 227202183 Yaouanc A et al. 2013 Phys. Rev. Lett. 110 127207184 Yaouanc A et al. 2011 Phys. Rev. B 84, 172408185 Ross K A et al. 2011 Phys. Rev. B 84 174442186 Ross K A et al. 2012 Phys. Rev. B 86, 174424187 Wong A W C et al. 2013 Phys. Rev. B 88 144402188 Lau G C et al. 2005 Phys. Rev. B 72 054411189 The precise definition of a quantum spin liquid is some-

thing that has drifted as people’s understanding of themhas grown. In their original incarnation, quantum spinliquids were magnets where zero point fluctuations aboutordered states are of the order of the total spin moment

so that the ground state is a superposition of states suchthat each spin has no net moment. Later, Anderson1 pro-posed that liquid phases of paired spin singlets or valencebonds could form in frustrated magnets and, it was this,resonating valence bond (RVB) idea that set into motionthe intense period of research into quantum spin liquidsthat continues to the present day.

190 Common water ice, Ih, has a hexagonal structure, whilethe pyrochlore has cubic symmetry. The Ising pyrochloreproblem is equivalent to cubic ice, Ic, and not Ih. Thisdifference does not modify the second ice-rule analogy orthe connection between the statistical mechanics problemof proton coordination in water ice and that of the spinarrangement in Anderson’s model or spin ice below

191 The first rule states that there should be only one pro-ton per O2−−O2− bond. This rule has no equivalent inAnderson’s model and in spin ice discussed below3,4

192 Recent efforts to explore the characteristic timescalesfrom a.c. susceptibility within the correlated paramag-netic regime of Dy2Ti2O7 have led to experiments measur-ing the heat capacity on similar time scales. These showthe residual entropy falling significantly below the Paulingestimate20.

193 This argument omits the nearest neighbour exchange in-teraction J . For positive ferromagnetic J the crossovertemperature TSI ∼ J/3 + 5D/33–5,23.

194 The existence of a pinch point at T TSI, deep in theparamagnetic regime, is presumably due to the pinch-point singularity arising directly from the anisotropic 1/r3

dipolar interaction45.195 The transformation that changes the sign of the ring

exchange changes also the background flux through thehexagonal plaquettes. This does not change the low en-ergy physics. However, in the gMFT49,53 described laterin this article, the background π flux affects the dispersionof spinon excitations which, in turn, can affect the phasediagram in the presence of competing interactions.

196 This is not to say that we have increased the effectivespin. Instead, the Hilbert space has been enlarged to thatof a set of U(1) quantum rotors.

197 In this paragraph we refer to nearest neighbour classicalspin ice in which there is an emergent Coulomb potentialbetween monopoles which is of entropic origin. In dipolarspin ice, there is an effective Coulomb potential arisingdirectly from the dipolar interaction.

198 It is worth pointing out that magnetic monopoles in highenergy physics have a similar origin to the vison excita-tion discussed here. In brief, gauge theories where elec-tromagnetism arises from a compact U(1) subgroup of alarger compact gauge group have gapped magnetic chargeexcitations which are generalisations of the visons ob-tained here. In the Standard Model they do not appearbecause the electromagnetic gauge group is non compact.However, magnetic monopoles appear when the StandardModel is embedded in a theory with a larger symmetryincluding the famous SO(10) Grand Unified Theory of thestrong, weak and electromagnetic interactions68. One ofthe features of compact U(1) gauge theory is that charges(the visons and monopoles) are naturally quantised. Thisfeature assumes some importance in fundamental physicswhere the quantisation of charges is something that onewould like to explain. Within the aforementioned GUT,the appearance of monopoles goes hand-in-hand with thequantisation of the fundamental charges (among other im-

27

portant features). Such GUTs have inspired experimentalsearches for magnetic charges69.

199 At finite temperature, the temporal direction of the 3 +1 dimensional gauge theory is wrapped onto a circle ofradius 2πβ where β is the inverse temperature. At hightemperatures the radius of the circle is small so that the3 + 1 dimensional gauge theory may be viewed as a 2 + 1dimensional gauge theory at zero temperature. In threedimensions, the Wilson loop expectation value exhibitsan area law63. Then, the same argument that implies thatthe 2 + 1 dimensional gauge theory has a gapped photoncan be used to see that the four dimensional gauge theorydoes as well at high temperatures70,71

200 The deconfined phase comes with a natural cutoff – themagnitude of the transverse terms in the spin Hamiltonian

– below which we expect that we do not have to worryabout gauge invariant interactions beyond the Maxwellaction. One can see that this is the case by studying thekinds of gauge invariant terms that one can add to thelow energy effective theory. The aptly named irrelevantoperators in the theory are suppressed by powers of the(large) cutoff at long wavelengths so that their effect atsufficiently low energies is negligible. The relevant oper-ators are not suppressed in this manner. Power countingis a simple criterion to assess how couplings flow under achange of scale. Specifically, the irrelevant operators havecouplings with dimensions of some negative power of theenergy scale where the dimension is determined directlyfrom the action.

28

(a)Inelastic Scattering

(b)Energy Integrated. Zero T

(c)Energy Integrated. T = 10ca−10

FIG. 9: Characteristic neutron scattering patterns observedfor quantum spin ice. Panel (a) shows the characteristic in-elastic scattering from photon excitations. The reciprocalspace path is taken along straight lines between high symme-try points in the Brillouin zone. Panels (b) and (c) show theunpolarized energy integrated scattering for photon scatteringfor zero temperature and high temperature (compared to thering exchange) respectively. The high temperature plot is sim-ilar to the scattering expected for classical spin ice. The tem-perature is measured in units of c/a0 where c =

√UKa0~−1

is the velocity of the linearly dispersing excitations and a0 isthe cubic unit cell edge length for the pyrochlore lattice. Thedynamical structure factor and unpolarized energy-integratedscattering were computed from formulae in Ref. [55].

29

FM

QSIAFM

CFM

J±/Jk

Jz±

/Jk

(a)Kramers

FM

QSIAFM

CFM

J±/Jk

Jz±

/Jk

(b)Kramers. Corrected.

QSIAFQ

Non-coplanar FQ

J±/Jk

J±

±/J

k

(c)Non-Kramers

FIG. 10: Zero temperature mean field phase diagrams takenthrough sections in the space of symmetry-allowed nearestneighbour exchange couplings on the pyrochlore lattice. Themean field theory (gMFT), described in the main text, cancapture both conventionally ordered and quantum spin liq-uid phases. The top panel (for odd electron magnets) showsa pair of exotic phases emerging from the classical spin icepoint J± = 0 and Jz± = 0 - the quantum spin ice (QSI)and Coulombic ferromagnet (CFM). The gMFT does notcapture the perturbatively exact phase (in small parameterJz±/Jzz)

49,72. Panel (b) is a schematic plot to show the likelyeffect of correcting the gMFT to account for the perturbativeresult. Panel (c) is a phase diagram for even electron systemsshowing the quantum spin ice phase and two quadrupolarphases53.