Arxiv:1707.03838V3 [Cond-Mat.Str-El] 27 Sep 2017 Point

Higher-Spin Witten Eﬀect and Two-Dimensional Fracton Phases

Michael Pretko Center for Theory of Quantum Matter, University of Colorado, Boulder, CO 80309 (Dated: September 7, 2017) We study the role of “θ terms” in the action for three-dimensional U(1) symmetric tensor gauge theories, describing quantum phases of matter hosting gapless higher-spin gauge modes and gapped subdimensional particle excitations, such as fractons. In Maxwell theory, the θ term is a total derivative which has no effect on the gapless photon, but has two important, closely related consequences: attaching electric charge to magnetic monopoles (the Witten effect) and leading to a Chern-Simons theory on the boundary. We will find that a similar story holds in the higher-spin U(1) gauge theories. These theories admit generalized θ terms which have no effect on the gapless gauge mode, but which bind together electric and magnetic charges (both of which are generally subdimensional) in specific combinations, in a higher-spin manifestation of the Witten effect. We derive the corresponding Witten quantization condition. We find that, as in Maxwell theory, imposing time-reversal invariance restricts θ to certain discrete values. We also find that these new θ terms imply a non-trivial boundary structure. The boundaries host fracton excitations coupled to a tensor U(1) gauge field with a Chern-Simons-like action, in both chiral and non-chiral varieties. These boundary theories open a door to the study of U(1) fracton phases described by tensor Chern-Simons theories, not only on boundaries of three-dimensional systems, but also in strictly two spatial dimensions. We explicitly work through three examples of bulk and boundary theories, the principles of which can be readily extended to arbitrary higher-spin theories.

CONTENTS A Higher Rank Example 16

I. Introduction 1 References 16

II. Lagrangians 2 A. Review of Maxwell Theory 2 I. INTRODUCTION B. Higher-Spin Theories 3 1. Vector Charge Theory 3 Quantum phases of matter, such as spin liquids and 2. Scalar Charge Theory 4 quantum hall systems, are well-described in the language of gauge theory. While most studies have fo- III. θ Terms and the Witten Effect 5 cused on familiar vector gauge theories, recent theo- A. Review of Maxwell Theory 5 retical efforts have established the existence of stable B. Vector Charge Theory 6 three-dimensional phases described by higher rank sym- 1. The Direct Term 6 metric tensor gauge fields. These include both U(1) 2. The Indirect Term 7 tensor gauge theories with gapless higher-spin gauge C. Scalar Charge Theory 7 excitations1–3, and the “generalized lattice gauge theories” of Vijay, Haah, and Fu, which are the natural dis- IV. Two-Dimensional Fracton Phases 9 crete analogue.4,5 These tensor gauge theories are partic- A. Review of Chern-Simons Theory 9 ularly interesting due to the fact that the gauge charges B. Boundary of the Vector Charge Theory 10 necessarily behave as subdimensional particles - excita- 1. The Direct Term 10 tions which are restricted by gauge invariance to move 2. The Indirect Term 11 only within lower-dimensional subspaces of the three- C. Boundary of the Scalar Charge Theory 13 dimensional system. As examples, particles can be re- D. Realization in Two Dimensions 13 stricted to motion within a plane, a line, or even a single arXiv:1707.03838v3 [cond-mat.str-el] 27 Sep 2017 point. In this last case, where the excitation is com- V. Conclusions 14 pletely immobile, the particle is known as a “fracton.” Acknowledgments 15 These exotic new particles have been a topic of intense recent research2–15,17–22, since they sit at the intersec- Appendix A: Mutual Statistics 15 tion of numerous areas of modern theoretical physics, including long-range entanglement4,5,20, quantum error Appendix B: Are θ Terms Topological? 15 correction10,11,21, glassy dynamics6,15,16, and emergent gravity.17 Appendix C: Gauge Fixing 16 In this paper, we will focus on the U(1) higher-spin Review of Maxwell Theory 16 theories, which are in some sense the simplest since they 2 build directly on intuition from the Maxwell theory of tions of Chern-Simons theory. These will include both a conventional U(1) vector gauge field. We therefore chiral and non-chiral varieties. While we first encounter expect many of the interesting stories associated with these phases at the boundary of three-dimensional sys- Maxwell theory to have natural tensor analogues. In tems, they can also occur in strictly two-dimensional sys- particular, we will here investigate the role of θ terms in tems under appropriate conditions, as we will discuss. the action for the higher-spin theories. These terms are This will be the first time the fracton phenomenon has natural analogues of the “E·B” term (conventionally pa- been stably manifested in two dimensions. rameterized by coefficient θ) which can be present in the action for Maxwell theory. The conventional θ term has no effect on the behavior of the gapless photon. However, II. LAGRANGIANS it causes the fundamental magnetic monopole of the system to pick up an electric charge given by θ/2π times In order to discuss the conventional θ term and its the fundamental electric charge. This changes the set generalizations, it is most useful to work in the La- of dyons (particles carrying both electric and magnetic grangian formulation. However, for tensor gauge the- charge) which can occur in the theory, via the Witten ories, essentially all previous work in the condensed quantization condition.23 This shift in the dyon struc- matter literature has taken place in the Hamiltonian ture is often referred to as the “Witten effect,” which has language.1–15,17–22 For gauge theories of this sort, go- received renewed attention in recent years due to its role ing between the two formulations can be a nontrivial in topological insulators and other symmetry protected task. We will therefore begin by discussing how to cor- topological phases.24–26 We will find that a similar story rectly formulate Lagrangians for higher-spin U(1) gauge holds in the higher-spin U(1) gauge theories. The action theories. for each of the higher-spin theories allows for a natural generalization of the θ term, with similar properties. As before, the gapless gauge mode will be untouched, but A. Review of Maxwell Theory the magnetic charges will pick up some amount of electric charge, modifying the dyon structure of the theory. We begin by reviewing how to go between the La- We will generalize Witten’s quantization condition, re- grangian and Hamiltonian formulations of standard lating the θ parameter to the amount of electric charge three-dimensional Maxwell theory. We start from the picked up by the fundamental monopole. The details of Lagrangian formulation, since this is perhaps more fa- the charge attachment will differ depending on the pre- miliar in the field-theoretic context, and use it to derive cise structure of the θ term, and we will find that some the Hamiltonian. We can write the partition function of systems admit multiple different types of θ terms. We the theory in terms of the standard Maxwell action: explicitly work through three rank 2 examples to illus- Z Z trate the general principle, which can then be readily i R d4xL i R d4x 1 F µν F Z = DA e = DA e 4 µν (1) extended to theories of arbitrary rank. µ µ

Another famous feature of the standard θ term is its where the field strength is Fµν = ∂µAν − ∂ν Aµ. (For relationship with Chern-Simons theory. The θ term is ac- readers concerned about issues of gauge fixing, this will tually a total derivative term in the action, which is why be discussed in Appendix C.) More explicitly, in terms it has no effect on the low-energy gauge-mode. Neverthe- of the spatial vector potential A~ and the timelike com- less, this total derivative leads to a Chern-Simons term ponent A0, we can write this as: living on the boundary which will attach electric charge to a monopole passing into the system, causing the Wit- Z Z ~ 3 ten effect.26,27 This teaches us that quantum hall phases Z = DADA0 exp i d xdt (described by Chern-Simons theory) can naturally oc- (2) 1 ˙ 2 1 2 cur on the boundary of certain bulk systems coupled to (A~ − ∇~ A0) − (∇~ × A~) Maxwell theory, such as topological insulators. And for 2 2 systems with an emergent U(1) gauge field, such as U(1) where dots indicate time derivatives. To convert to the spin liquids, the edge can naturally host topologically Hamiltonian formulation, we introduce an auxiliary field ordered phases. The θ term therefore is not only use- E~ , which will eventually play the role of the electric field ful for studying the bulk physics of Maxwell theory. It in the Hamiltonian. Inserting an appropriate Gaussian also provides a novel way of studying the physics of two- ~ dimensional quantum hall systems. In close analogy, we integral over the field E, we can write the path integral will find that the higher-spin θ terms also lead to inter- as: esting two-dimensional phases on the boundary. In this Z Z ~ ~ 3 case, however, these will not be phases which we already Z ∝ DADA0DE exp i d xdt independently understand. Rather, the higher-spin θ (3) ˙ 1 2 1 2 terms open the door to a set of new two-dimensional E~ · (A~ − ∇~ A0) − E − (∇~ × A~) phases hosting fracton excitations, via tensor generaliza- 2 2 3

(where we ignore overall multiplicative constants in the R satisfying eR = 2π. For convenience, we will take partition function). The exponent is now linear in the e = 1 and R = 2π throughout. These charges will play timelike component A0, so after an integration by parts, an important role when we come to discuss the Witten we can integrate A0 out of the theory entirely: effect. Z Z Z ∝ DA~DE~ δ(∇~ · E~ ) exp i d3xdt B. Higher-Spin Theories (4) ˙ 1 1 E~ · A~ − E2 − (∇~ × A~)2 2 2 We now turn our attention towards the higher-spin U(1) gauge theories, which host particles with subdi- We now have a path integral which enforces the rigid mensional behavior, such as fractons. From the Hamil- 1–3 constraint that ∇~ · E~ = 0 everywhere. (Note that the tonian formulations described in earlier work , we will delta function above actually represents a product of show how to transition to a Lagrangian formulation. We delta functions at each point in space.) We can rec- will work through the details for two specific phases ognize this as the appropriate Hamiltonian formulation described by rank 2 symmetric tensor gauge fields to of the path integral. For comparison, a single-particle demonstrate the general principles. By working with path integral can be written in equivalent Lagrangian tensors of the appropriate rank and constraints, these and Hamiltonian forms as: principles can be extended to any other higher-spin theory. Z Z i R dtL i R dt(px˙ −H) Zs.p. ∝ Dx e ∝ DxDp e (5) 1. Vector Charge Theory Along the same lines, we can use Equation 4 to read off the Hamiltonian density of our theory as: We consider a phase described by a rank 2 spatial sym- 1 1 metric tensor gauge field Aij, with a canonical conjugate H = E2 + (∇~ × A~)2 (6) variable E , playing the role of an electric field tensor. 2 2 ij As described in previous work1,2, there are multiple dif- where A~ and E~ are canonical conjugate variables, and ferent stable theories with the same degrees of freedom. the system is subject to the constraint: Each theory can be uniquely specified by its gauge constraint, corresponding to a generalized Gauss’s law. In ∇~ · E~ = 0 (7) the first theory that we will consider, this Gauss’s law takes the form: 1 2 2 We see that we recover precisely the expected 2 (E +B ) ij j behavior of the classical electromagnetic Hamiltonian, ∂iE = ρ (8) along with a constraint expressing the absence of charge for vector charge density ρj. (All indices are spatial, in the pure gauge theory. and repeated indices are summed over.) The particles This is exactly the sort of Hamiltonian structure that carrying this vector charge are notable for obeying an arises in systems described by emergent U(1) gauge extra conservation law which has no analogue in a con- fields, such as U(1) spin liquids. The constraint arises ventional gauge theory. These charges obey the following from some local energetic considerations, such as a spin- two conservation laws: ice rule. The E2 and B2 terms enter as the most rel- Z Z evant gauge-invariant terms. The constraint structure ~ρ = constant ~x × ~ρ = constant (9) of the microscopic Hamiltonian thereby leads directly to the Maxwell Hamiltonian as the effective low-energy where ~x is the spatial coordinate and the integrals run description of the spin liquid phase. One could then per- over three-dimensional space. The first equation is the form the entire analysis above in reverse, demonstrating natural analogue of charge conservation. The second that the Maxwell action is the appropriate low-energy tells us that the angular moment associated with this description of the U(1) spin liquid. charge vector is an independently conserved quantity. Of course, at higher energies, there will also be states This second conservation law has a fairly drastic conse- which violate energetic constraints like a spin-ice rule. ~ ~ quence for the charges: an isolated charge can only move This corresponds to states with ∇ · E = ρ 6= 0, or in in the direction of its charge vector, while transverse mo- other words, states with charges. For the purposes of this tion is strictly forbidden. paper, we will not need to know the detailed dynamics In the low-energy sector, without any charges, we have of these charges. All that we need to know is that these the more restrictive form of Gauss’s law, ∂ Eij = 0, ~ ~ i charges exist, and also are quantized: ∇ · E = ne, for which leads to invariance under the following gauge integer n and fundamental charge e. This quantization is transformation: a result of the fact that the gauge field can generically be compact, Ai ∼ Ai + R, for some compactification radius Aij → Aij + ∂iαj + ∂jαi (10) 4

ij where the vector gauge parameter αi has arbitrary spa- The ﬁeld E can now be integrated out of the problem tial dependence. Including the most relevant terms con- to yield: sistent with the gauge constraint, the low-energy Hamil- Z Z tonian for this phase is given by: Z ∝ DAijDCi exp i (16) 1 ij 1 ij 1 1 1 1 H = E Eij + B Bij (11) (A˙ − ∂ C − ∂ C )2 − BijB 2 2 2 ij 2 i j 2 j i 2 ij where the magnetic tensor takes the form: We can then read oﬀ the Lagrangian of our theory as:

a c bd 1 1 1 1 Bij = iabjcd∂ ∂ A (12) L[A ,C ] = (A˙ − ∂ C − ∂ C )2 − BijB (17) ij i 2 ij 2 i j 2 j i 2 ij ij and we implicitly impose the constraint ∂iE = 0. This We now have a Lagrangian description of a theory with Hamiltonian leads to gapless gauge modes, including a a tensor ﬁeld coupled to a vector ﬁeld, with a larger set spin-2 mode with quadratic dispersion. (There are also of gauge transformations than our original Hamiltonian lower-spin gauge modes, which could be eliminated by formulation. While we originally spoke only in terms of more complicated gauge constraints, if desired.) spatial gauge transformations, we can now identify the We can now take this constrained Hamiltonian struc- full time-dependent gauge transformations of our theory ture and use it to reverse all of the logic from the pre- as: vious section, going to a Lagrangian formulation. From the Hamiltonian, we can write down the partition func- Aij → Aij + ∂iαj + ∂jαi (18) tion as:

Z Z Ci → Ci +α ˙ j (19) ij Z ∝ DAijDEij δ(∂iE ) exp i where the vector gauge parameter α(x, t) now has arbi- (13) 1 1 trary dependence on both space and time. EijA˙ − EijE − BijB ij 2 ij 2 ij where the integral in the exponent is over space and 2. Scalar Charge Theory time (measure suppressed for convenience), and the delta ij function enforces that each component of ∂iE vanishes We now switch to a different rank 2 theory, which is at each point in space. In the case of Maxwell theory, also described by a symmetric tensor gauge field Aij with the gauge constraint was enforced by a Lagrange mul- canonical conjugate Eij, but with a different Gauss’s law, tiplier, which happened to play the role of a timelike given by: component to the gauge field, A . In the present case, 0 ∂ ∂ Eij = ρ (20) we can also introduce a Lagrange multiplier field to han- i j dle the constraint. However, this higher-spin theory does for scalar charge ρ. As in the previous theory, this not possess any relativistic symmetry (the dispersion is Gauss’s law leads to two separate conservation laws: quadratic), and there is no reason to expect the Lagrange Z Z multipliers to behave as timelike components to Aij. We ρ = constant ρ~x = constant (21) therefore abandon relativistic notation and write our La- grange multiplier field as an independent vector variable, C . We can then write our partition function as: The first equation is simply the conventional conserva- i tion of charge, while the second represents the conser- Z Z vation of dipole moment. This extra conservation law Z ∝ DAijDEijDCi exp i forces an isolated charge to be a fracton excitation, un- (14) able to move in any direction, since any motion would 1 1 C ∂ Eij + EijA˙ − EijE − BijB change the dipole moment of the system. j i ij 2 ij 2 ij In the low-energy sector, which is free of charges, we ij have the constraint ∂i∂jE = 0, which leads to invari- Integrating the Ci term by parts and taking advantage ance under the gauge transformation: of the symmetry of Eij, we can write this as: Aij → Aij + ∂i∂jα (22) Z Z Z ∝ DAijDEijDCi exp i for gauge parameter α with arbitrary spatial dependence. The low-energy Hamiltonian consistent with the 1 1 1 1 gauge structure is: Eij(A˙ − ∂ C − ∂ C ) − EijE − BijB ij 2 i j 2 j i 2 ij 2 ij 1 1 (15) H = EijE + BijB (23) 2 ij 2 ij 5

ij where we implicitly impose the constraint ∂i∂jE = 0, III. θ TERMS AND THE WITTEN EFFECT and the magnetic tensor takes the (non-symmetric) form: B = ∂aAb (24) Now that we have Lagrangians in hand, it is time to in- ij iab j vestigate the possibility of θ terms. We will start by giv- (There are other gauge-invariant magnetic tensors which ing a simple treatment of the standard θ term of Maxwell could be formed, such as the one in Equation 12, but theory, which will help us identify the principles neces- the version above has the fewest number of derivatives sary for generalization to the higher-spin theories. and has equal velocities for all of the physical gauge modes, representing a stable fixed point of the renor- malization group.) This Hamiltonian leads to a gapless A. Review of Maxwell Theory spin-2 mode with linear dispersion, along with lower-spin gauge modes. The standard Maxwell action takes the form: Following the same logic as the previous section, we Z 1 write down the partition function as: S = F µν F (32) 4 µν Z Z ij Z ∝ DAijDEij δ(∂i∂jE ) exp i However, this is not the most general action we can write consistent with gauge invariance. We may also consider (25) ij 1 ij 1 ij a θ term (an “E · B” term) of the form: E A˙ ij − E Eij − B Bij 2 2 θ Z S = µνρσF F (33) where the delta function enforces the constraint that θ 32π2 µν ρσ ij ∂i∂jE vanishes at each point in space. For this case, we θ Z ˙ = (A~ − ∇~ A ) · B~ (34) only require one Lagrange multiplier, which we write as 4π2 0 a scalar field φ. We can then write our partition function as: We can immediately conclude that this term will have Z Z no effect on the low-energy gauge mode in the bulk of Z ∝ DAijDEijDφ exp i the system, due to the fact that it is the integral of a total derivative: (26) ij ij ˙ 1 ij 1 ij θ Z −φ∂i∂jE + E Aij − E Eij − B Bij S = ∂ (µνρσA ∂ A ) (35) 2 2 θ 8π2 µ ν ρ σ Integration by parts on the φ term yields: The θ term therefore has no effect on the photon. It will, Z Z however, lead to a Chern-Simons term on a boundary Z ∝ DAijDEijDφ exp i between the system and an external region with θ = 0. (27) Such a term can attach charge and magnetic flux on the ij 1 ij 1 ij boundary, with the net result that a magnetic monopole E (A˙ ij − ∂i∂jφ) − E Eij − B Bij 2 2 brought into the system from the θ = 0 region will pick ij up a specific amount of electric charge. The field E can then be integrated out of the problem We can very easily see this explicitly by introducing to yield: the auxiliary E~ field, as before, after which the action Z Z becomes: Z ∝ DAijDφ exp i Z θ ˙ 1 1 (28) S = (E~ + B~ ) · (A~ − ∇~ A ) − E2 − B2 (36) 1 1 4π2 0 2 2 (A˙ − ∂ ∂ φ)2 − BijB 2 ij i j 2 ij After integrating the A0 term by parts, then integrat- We read off the Lagrangian of the theory as: ing A0 out of the path integral, we see that the gauge constraint now becomes: 1 ˙ 2 1 ij L[Aij, φ] = (Aij − ∂i∂jφ) − B Bij (29) θ 2 2 ∇~ · E~ = − ∇~ · B~ (37) 4π2 We now have a Lagrangian description of a theory with a tensor field coupled to a scalar field, which once again has From the definition of the magnetic field, B~ = ∇×~ A~, and a larger set of time-dependent gauge transformations: the compact nature of the gauge field, Ai ∼ Ai + 2π, the divergence of the magnetic field is quantized as ∇~ · B~ = A → A + ∂ ∂ α (30) ij ij i j 2πg for integer g. The gauge constraint then indicates that any particle carrying magnetic charge g must also φ → φ +α ˙ (31) carry an electric charge given by: where the parameter α(x, t) has arbitrary dependence on θg q = − + n (38) both space and time. 2π 6

the fact that magnetic charge is odd under time-reversal, while electric charge is even, so a time-reversal invariant charge lattice must be symmetric about the q-axis. It can further be shown that θ = π is only time-reversal invariant if the fundamental electric charge is a fermion, due to the eﬀects of particle statistics. These facts have important consequences both for symmetry protected topological phases protected by time-reversal symmetry26 and for time-reversal invariant U(1) spin liquids.27

B. Vector Charge Theory

For the vector charge theory, we derived that the ac- FIG. 1. When θ = 0, the set of possible charges forms a tion takes the form: simple square lattice in the (q, g) plane. (Adapted from Ref- Z 1 1 1 1 S = (A˙ − ∂ C − ∂ C )2 − BijB (39) erence 27.) 2 ij 2 i j 2 j i 2 ij a c bd with magnetic tensor Bij = iabjcd∂ ∂ A . We take Aij to be compact, Aij ∼ Aij + 2π, so the magnetic ﬁeld ij j admits monopoles of the form ∂iB = 2πg for magnetic vector charge gj, which is an integer linear combination of the fundamental charge vectors of the system.28 As in Maxwell theory, we expect there to be some “E · B” term we can write down which attaches electric charge to these magnetic monopoles. However, in this case, it will turn out that there are two diﬀerent types of θ terms that we can consider.

1. The Direct Term

The most natural analogue of an “E · B” term takes FIG. 2. When θ is nonzero, the charge lattice is tilted. The the form of a direct contraction between Eij and Bij: case θ = π is seen above. (Adapted from Reference 27.) θ Z 1 1 S = 1 (A˙ − ∂ C − ∂ C )Bij (40) θ1 4π2 ij 2 i j 2 j i Z for integer n, which is the standard result of the Wit- θ1 iab jcd 23 = (A˙ − ∂ C ) ∂ ∂ A (41) ten effect. (We must allow for the extra integer n, 4π2 ij i j a c bd since a fundamental electric charge can always bind with Some algebra yields that this term is the integral of total the dyon.) This attachment of electric charge to the derivatives, just like the conventional θ term: monopole has the result of “tilting” the charge lattice of Z the system, as seen in Figures 1 and 2. θ1 iab jcd iab jcd Sθ = ∂a(A˙ ij ∂cAbd + Cj ∂i∂cAbd) It is clear that the values of charge in the charge lat- 1 4π2 tice return to their original values at θ = 2π. One might 1 − ∂ (iab∂ A jcd∂ A ) then naively conclude that θ is always a 2π-periodic vari- 2 t a ij c bd able. However, one more subtlety enters in the form of (42) particle statistics. For a system where the fundamen- As such, this term has no effect on the bulk gauge mode. tal electric charge is a fermion, the naive expectation of However, it will contribute a boundary term which can 2π-periodicity is indeed correct. For a bosonic electric result in the binding of charges to monopoles. To see charge, on the other hand, the charge lattice has a dif- the charge attachment explicitly, we combine the θ1 term ferent assignment of statistics at θ = 2π than at θ = 0 with the action of Equation 39. After adding the auxil- (the “statistical Witten effect”). In this case, the system iary Eij field, the total action becomes: turns out to be 4π-periodic in θ.26,27 Z θ It is also important to note that, while θ can a priori S = (Eij + 1 Bij)(A˙ − ∂ C ) 4π2 ij i j take any value, the charge lattice can only be time- (43) 1 1 reversal invariant if θ = 0 or θ = π (or multiples thereof), − EijE − BijB the two cases depicted in Figures 1 and 2. This is due to 2 ij 2 ij 7

After integrating by parts on the Cj term, then integrating Cj out of the path integral, we see that the gauge constraint becomes: θ ∂ Eij = − 1 ∂ Bij (44) i 4π2 i In terms of a particle’s electric charge vector ~q and its magnetic charge vector ~g, this equation tells us that: θ ~q = − 1 ~g + ~n (45) 2π where ~n is any integer combination of the fundamental FIG. 3. The θ2 term serves the purpose of attaching an angular configuration of electric charge (red) to the magnetic vector charges of the theory, representing the possibility charge (blue). of fundamental electric charges binding to the dyon. We therefore see that the θ1 term has the effect of binding together parallel electric and magnetic charge vectors. Like the direct term, some tedious algebra reveals that It is easy to see that, when θ1 = 2π, the set of al- the indirect θ2 term is also the integral of a total deriva- lowed charges will return to itself. Furthermore, there tive: are no concerns associated with statistics in this theory. The particles in this case are 1-dimensional objects, for θ Z S = 2 ∂ (ic`∂ A˙ jkd∂aAb + ∂ C ji`B k) which there is no real distinction between bosons and θ2 8π2 k c `j iab d i j ` fermions. (One can check that there are also no statisti- − ∂ (ik`∂ ∂ A jcd∂aAb ) cal issues with bound states.) We can therefore conclude t k c `j iab d (49) that the θ1 parameter is indeed 2π-periodic. We can then further ask which values of θ1 are allowed if we enforce The gauge mode is therefore unaffected by the presence time-reversal invariance. We note that, taking the con- of a θ term. Repeating the procedure from the previous vention that ~q is even under time reversal29, ~g will be 2 section, introducing the auxiliary Eij and integrating out odd, transforming as ~g → −~g. For the charge lattice to C , we find that the gauge constraint is: be invariant under this transformation, we then require: i

θ1 θ1 ij θ2 jk` i − ~g = ~g + ~n (46) ∂iE = − ∂k∂iB` (50) 2π 2π 8π2 θ In terms of electric and magnetic charge vectors ~q and ⇒ − 1 ~g = ~n (47) π ~g, this implies: for some integer combination ~n of fundamental vector θ ~q = − 2 (∇~ × ~g) + ~n (51) charges. For this to be true, we require θ1 = 0 or θ1 = π, 4π exactly as in the case of the conventional θ term. Thus, just as in the usual U(1) gauge theory, time-reversal in- where ~n is any integer combination of the fundamen- variance restricts the charge lattice of the theory to just tal charge vectors. The physical consequence of this is two possible shapes. that, for every magnetic vector charge, there is a configuration of electric charges winding around perpendicular to it, as seen in Figure 3. In other words, the mag- 2. The Indirect Term netic monopoles have electric angular charge attached, as promised. The θ1 term considered above led to a direct binding The set of allowed charges will return to itself only between the electric vector charges and their magnetic at θ2 = 4π (unlike the 2π repetition of θ1). Further- analogue. However, the vector charge theory also hosts more, the attachment of electric angular charge leaves another class of nontrivial excitations: particles carrying the magnetic vectors as one-dimensional particles, for zero net charge, but a nonzero angular charge moment. which statistics are not meaningful. We therefore con- The angular charge moment ~` of such a particle is also clude that θ2 is a 4π-periodic variable. Similarly, it is a vector quantity. It therefore seems plausible that we straightforward to check that θ2 = 0 and θ2 = 2π are could add a term which binds electric angular moment the only values allowed in the presence of time-reversal to the magnetic vector charge. We can accomplish this invariance. by adding an “indirect” contraction, of the form: θ Z 1 1 S = 2 ik`∂ (A˙ − ∂ C − ∂ C )B j C. Scalar Charge Theory θ2 4π2 k `j 2 ` j 2 j ` i Z (48) θ2 ik` 1 j We now move on to apply the same principles to the = ∂k(A˙ `j − ∂jC`)B 4π2 2 i scalar charge theory. Since there is no self-duality in 8 this case, there is no term which can attach an electric for the charges of the system. Once again, we can see particle to its magnetic analogue. The electric charges this by introducing the auxiliary Eij field, upon which of the theory are scalars, while the magnetic charges are the total action becomes: vectors. The most natural form of charge attachment Z θ in this theory is attaching an electric dipole moment to S = (Eij + Bij)(A˙ − ∂ ∂ φ) 4π2 ij i j a magnetic vector. Indeed, we will find that the direct (55) ij ij 1 1 contraction between E and B accomplishes precisely − EijE − BijB this. 2 ij 2 ij For the scalar charge theory, the action takes the form: After integrating by parts on the φ term and integrating Z 1 1 φ out of the path integral, our gauge constraint becomes: S = (A˙ − ∂ ∂ φ)2 − BijB (52) 2 ij i j 2 ij ij θ ij ∂i∂jE = − ∂i∂jB (56) a b 4π2 where the magnetic tensor is Bij = iab∂ A j, admit- ij j ij ting monopoles of the form ∂iB = 2πg for magnetic In terms of the electric scalar charge ∂i∂jE = q and j ij j vector charge g (an integer linear combination of fun- the magnetic vector charge ∂iB = 2πg , this equation damental charge vectors). Note the distinctly differ- becomes: ent behavior of the electric and magnetic sectors. The ij θ electric tensor E is symmetric and is characterized by q = − ∂ gi + n (57) scalar fracton charges. The magnetic tensor Bij, on the 2π i other hand, is traceless and non-symmetric, character- for integer n. In this case, the θ term has the effect of at- ized by vector charges, which can be shown to behave taching fractonic electric charges to the endpoints of the as 2-dimensional particles, moving only transversely to magnetic charge vectors, as seen in Figure 4, with oppo- their charge vector.2 site sign charges at each end of the vector. Equivalently, Despite the lack of self-duality in the theory, we can we can think of the θ term as binding an electric dipole still write down a natural “E · B” term and investigate to each magnetic vector. We can then see that, if we its consequences: construct a long directed string of such vector charges, θ Z we will only have electric charges at the two endpoints of S = (A˙ − ∂ ∂ φ)iab∂ A j (53) θ 4π2 ij i j a b the string. Since it takes finite energy to create the electric charges, it will be energetically favorable for a group Simple algebra yields that this only contributes a bound- of magnetic vectors to line up into such string arrange- ary term to the action: ments, so as to minimize the number of electric charges present. θ Z 1 S = ∂ ( iabA˙ A j + ∂ φiab∂ A j) We therefore see that, when θ 6= 0, we can more use- θ 4π2 a 2 ij b j i b (54) fully think of the system in terms of a theory of open 1 strings, as opposed to point particles (though there is − ∂ (iab∂ A A j) 2 t a ij b strictly speaking no rigid distinction between the two pictures). And in the limit where the gap to electric so once again this θ term will have no consequences for charges goes to infinity, we can understand the behavior the bulk gauge mode of the system. It will, however, in- of the magnetic charges in terms of a theory of closed troduce boundary physics which will have consequences strings (coupled to a tensor gauge field). If all values of θ are regarded as equally likely, then θ = 0 is a some- what special case where the magnetic charges have no tendency to line up. For any nonzero θ, the magnetic charges will arrange themselves into open string configurations to minimize electric charge. As in the previous theories, we see that θ = 2π corresponds to full integer multiples of charge being attached to the endpoints of magnetic vectors, which takes us back to the original set of allowed charges. However, in this case, we must also worry about particle statistics. Note that a magnetic vector charge has mutual π statistics with a parallel electric dipole in its plane of motion (see the Appendix). Assuming the magnetic vectors and elec- FIG. 4. When θ is nonzero, electric fractons will be attached tric dipoles are both bosons30 (as is the case in the sim- to the endpoints of the magnetic charge vectors. Due to the plest lattice models1,2), then the bound state of the two finite gap to electric charges, it will be energetically favorable will be a fermion. It is then easy to check that, when for the magnetic charges to bind into string-like objects. θ = 2π, the charge lattice has returned to its original 9 shape, but the assignment of statistics is different. Only where B is the magnetic flux through the surface. More at θ = 4π does the system go back to the original charge generally, allowing for charges coupled to the Chern- lattice with the original statistics. We therefore see that Simons field, we will have: θ is 4π-periodic in this system. By similar logic, it is θ straightforward to check that imposing time-reversal in- ρ = − B (61) variance will restrict the θ parameter to only two possible 4π2 distinct values: θ = 0 or θ = 2π. for charge density ρ, which indicates that each 2π flux has −θ/2π electric charge attached. (Or equivalently, each charge has −4π2/θ flux attached.) Note that ρ is IV. TWO-DIMENSIONAL FRACTON PHASES jk 2 a divergence, ρ = ∂j(−θ Ak/4π ), so the usual charge conservation law will hold. Conservation of this charge In the above analysis, we found that θ terms in the on the two-dimensional boundary is a direct consequence action for the higher-spin U(1) gauge theories only lead of the fact that magnetic flux through the boundary to boundary contributions. This is the same situation as cannot change unless a magnetic monopole is passed in Maxwell theory, where a θ term in the bulk leads to a through the surface into the three-dimensional bulk. Chern-Simons term on the boundary of the system. The 1 jk ˙ The second term in the boundary action, 2 AjAk, study of such Chern-Simons terms is important not only determines the canonical structure of the gauge field, for understanding the boundary of three-dimensional dictating that Ax and Ay are canonical conjugates, as U(1) spin liquids, but also for shedding light on quan- is familiar in Chern-Simons theory. The presence of tum hall physics in purely two-dimensional systems. We only one time derivative leads to the vanishing of these therefore wish to study the boundary terms which arise terms upon going to the Hamiltonian formalism, indi- as a consequence of θ terms in the higher-spin U(1) gauge cating that the Hamiltonian itself vanishes within the theories. These terms are obviously important for study- low-energy gauge sector, reflecting the lack of local de- ing the boundary of tensor U(1) spin liquids. But per- grees of freedom of a topological field theory. haps even more importantly, they will teach us how to To verify that the Chern-Simons theory has a finite formulate fundamentally new two-dimensional phases of energy gap, it is instructive to regularize the theory by matter supporting fracton excitations. We will comment including a two-dimensional Maxwell term, after which later on how these phases get around the conventional the action becomes: folklore that fractons do not occur in two-dimensional systems. Z 1 S = ((A˙ − ∂ A )2 − (ij∂ A )2) 2e2 i i 0 i j (62) θ 1 A. Review of Chern-Simons Theory − (A ij∂ A + ijA˙ A ) 4π2 0 i j 2 i j

We first review a perspective on the Chern-Simons Varying the action with respect to Ai, the resulting equa- boundary terms in Maxwell theory with a θ term, so as tion of motion is: to facilitate comparison with the higher-spin case. For 1 θ 1 Maxwell theory, the θ term can be written as: E˙ i + ijE − ij∂ B = 0 (63) e2 4π2 j e2 j Z θ ˙ ijk Sθ = 2 (Ai − ∂iA0) ∂jAk At low energies, the last term is irrelevant, and we obtain 4π an energy gap scaling as ∆ ∼ θe2. As we take e → ∞, Z θ ijk 1 ijk the action approaches pure Chern-Simons theory, and = − ∂i(A0 ∂jAk + A˙ j Ak) (58) 4π2 2 the energy gap approaches infinity. 1 When the Chern-Simons term is regarded as an ef- + ∂ (∂ A ijkA ) 2 t j i k fective action for the physical external electromagnetic gauge field of a two-dimensional system, we can easily For a spatial boundary, taken with normalz ˆ for simplic- derive the Hall response: ity, the boundary action is then: δS θ θ Z hJ ii = − = − ijA˙ = − ijE (64) θ jk 1 jk 2 j 2 j ˙ δAi 4π 4π S∂ = − 2 A0 ∂jAk + AjAk (59) 4π ∂ 2 so the conductivity tensor is: where the integral is over the boundary. This is the usual Chern-Simons theory, expanded out in terms of spatial θ σij = − ij (65) and timelike components. The first term gives us the 4π2 following constraint on the low-energy sector: If we restrict θ to be a multiple of 2π, such that there θ θ are no fractionalized charges in the system, we see that − jk∂ A = − B = 0 (60) 4π2 j k 4π2 the Hall response must be an integer multiple of some 10 fundamental value (1/2π in the conventions taken here), We also note that aij and ∂zaij can be independently representing the integer quantum Hall effect. These cor- varied on the boundary of the system. We therefore respond to the “properly” quantized levels of the Chern- define another independent symmetric tensor field as: Simons theory. For other values of θ, with improperly quantized Chern-Simons terms, we will have both a frac- γij = −∂zaij (a 2d tensor) (72) tional Hall response and fractional charges on the boundary, in a manifestation of the fractional quantum Hall In terms of the three fields aij, λi, and γij, the constraint effect. equations become:

θ1 bi cd 2 ∂i∂cabd = 0 (73) B. Boundary of the Vector Charge Theory 4π

θ Let us now carry these same principles over to a 1 bijc∂ (∂ λ + ∂ λ + γ ) = 0 (74) higher-spin U(1) gauge theory, namely the vector charge 4π2 i c b b c bc theory. We will apply the analysis separately to the direct θ term and the indirect θ term. (The middle term in Equation 74 is identically zero, but 1 2 has been added for later convenience.) In terms of these ﬁelds, the dynamic part of the action (involving time 1. The Direct Term derivatives) becomes: θ Z S = 1 a˙ bijc∂ λ +a ˙ bijdγ + λ˙ bicd∂ a We found that the θ1 term can be written in the fol- dyn 4π2 ij c b ij bd i c bd lowing total derivative form: ∂ θ Z = 1 a˙ bijc(∂ λ + ∂ λ + γ ) θ Z 4π2 ij c b b c bc S = 1 ∂ (A˙ iabjcd∂ A + C iabjcd∂ ∂ A ) ∂ θ1 4π2 a ij c bd j i c bd (75) 1 − ∂ (iab∂ A jcd∂ A ) From the constraints and dynamic terms above, it be- 2 t a ij c bd (66) comes clear that we can eliminate λi entirely and sim- plify the theory by shifting: Taking a spatial boundary with normalz ˆ, the boundary γ → γ − ∂ λ − ∂ λ (76) action becomes: ij ij i j j i Z In terms of the remaining two ﬁelds a and γ , the θ1 ˙ bi jcd bi jcd ij ij S∂ = 2 Aij ∂cAbd + Cj ∂i∂cAbd (67) dynamical action takes the form: 4π ∂ θ Z where we have introduced the 2d Levi-Civita symbol ij 1 bi jc Sdyn[aij, γij] = 2 a˙ ij γbc (77) in the xy-plane. From the second term, we can see that 4π ∂ the Lagrange multiplier Ci imposes several constraints which is subject to the following constraints: on the boundary theory: θ θ 1 bijc∂ ∂ a = 0 (78) 1 bicd∂ ∂ A = 0 (68) 4π2 i j bc 4π2 i c bd

θ1 bi jc θ ∂iγbc = 0 (79) 1 bijc∂ (∂ A − ∂ A ) = 0 (69) 4π2 4π2 i c bz z bc Note that the six independent components of the pair Right away, we can see that this theory is going to be aij, γij are organized into three canonical conjugate a bit more complicated than the conventional Chern- pairs. Since we have three independent constraints on Simons theory. In that case, the Az component and all the system, we see that the pure gauge theory is fully z derivatives were eliminated in the boundary theory. In constrained, with no local degrees of freedom. It should the present case, on the other hand, only Azz is com- be noted that this theory is a tensor generalization of pletely eliminated from the theory, while Aiz (for i 6= z) a mutual Chern-Simons theory and is non-chiral. (The and z derivatives remain. In order to handle this, we first action is invariant under spatial reflections.) break up the three-dimensional tensor Aij as follows: More generally, we can consider charged defects coupled to the gauge field as sources for the constraint equa- aij = Aij for i, j = x, y (a 2d tensor) (70) tions: θ 1 bijc∂ ∂ a = ρ (80) λi = Aiz for i = x, y (a 2d vector) (71) 4π2 i j bc 11

θ1 bi jc j We can use our new Chern-Simons action to derive a ∂iγbc = π (81) 4π2 generalized “Hall” response of the system. If we have a current J ij coupled to the action via J ij a , and simi- for scalar charge density ρ and vector charge density πj. (a) (a) ij From the general arguments regarding Gauss’s laws and larly for γij, then the generalized Hall responses are: conservation laws2, we can see that the scalar charge has ∂S θ θ a conserved dipole moment: hJ ij i = − = 1 bijcγ˙ = 1 bijcE(γ) (86) (a) ∂a 4π2 bc 4π2 bc Z ij d2x (ρ~x) = constant (82)

ij ∂S θ1 bi jc θ1 bi jc (a) This restricts these charges from moving in any direction, hJ i = − = a˙ bc = E (87) (γ) ∂γ 4π2 4π2 bc meaning that the charge ρ represents fracton excitations. ij j Similarly, it is not hard to show that π represents one- Note that these do not represent responses to an ex- dimensional particles, only moving in the direction of ternally applied electric field, but rather to the internal their charge vector. Intuitively, we can understand these emergent tensor field. This is more difficult to control results from the fact that electric vector charges are from an experimental standpoint. Nevertheless, large- bound to magnetic flux on the boundary. Electric charge scale configurations of the emergent tensor field can in vectors in the xy-plane provide 1-dimensional particles principle be set up, perhaps through applying appropri- on the boundary, while charge vectors in the z direction ate strains on the system. In such a scenario, the gen- have no free motion at all on the boundary and give us eralized Hall response should be a measurable quantity. fractons in the boundary theory. Furthermore, when there are no fractional charges in the It should be noted that being in a system with U(1) system, θ1 will be restricted to a multiple of 2π, and the charge behavior, where charge is conserved absolutely, Hall responses will be integer multiples of a fundamental is crucial to the fracton structure. In a discrete charge unit, in close analogy with the integer quantum Hall ef- system, where charge is only conserved modulo some in- fect. Values of θ1 other than these “properly” quantized teger, it is known that these sorts of conservation laws levels will lead to fractional generalized Hall responses 4,5 break down in two dimensions , and the particles be- and also fractional charges. come fully mobile. To verify that there is a finite energy gap, we once again regularize the theory by keeping finite two- 2. The Indirect Term dimensional Maxwell terms for the two gauge fields, after which the action becomes: We found that the θ2 term can be written as: Z 1 1 2 ij k` 2 S = ((˙aij − (∂iCj + ∂jCi)) − ( ∂i∂kaj`) ) θ Z 2e2 2 S = 2 ∂ (ic`∂ A˙ jkd∂aAb + ∂ C ji`B k) θ2 8π2 k c `j iab d i j ` 1 2 jk 2 + ((γ ˙ ij − ∂i∂jφ) − ( ∂jγki) ) ik` jcd a b 2e2 − ∂t( ∂k∂cA`jiab ∂ A d) (88) θ1 bi jc bi jc jc bi + 2 (˙aij γbc + φ ∂i∂jabc + Cj ∂iγbc) 4π Taking a spatial boundary with normalz ˆ, the boundary (83) action becomes:

(We have written a vector charge Maxwell term for aij Z θ2 ic` ˙ jzd a b ji` z and a scalar charge Maxwell term for γij, which is neces- Sθ2 = 2 ( ∂cA`jiab ∂ A d + ∂iCj B` ) 8π ∂ sary in order to match the charge structure of the Chern- (89) Simons theory.) Varying with respect to the gauge fields, We note that, since ∂ C can be varied independently the equations of motion are: z j of Cj on the boundary, the theory now has two extra 1 θ 1 Lagrange multipliers (there is no constraint from ∂zCz). E˙ ij − 1 bijcE(γ) − ikj`∂ ∂ B = 0 (84) e2 (a) 4π2 bc e2 k ` (a) Integrating out all such Lagrange multipliers, our system obeys five total constraints: 1 θ 1 ˙ ij 1 bi jc (a) ik j θ E + Ebc − ∂kB = 0 (85) 2 cd j z z j e2 (γ) 4π2 e2 (γ) ∂c(∂ A − ∂ A ) = 0 (90) 8π2 d d As in conventional Chern-Simons theory, the magnetic terms in the above equations are irrelevant at low ener- 2 θ2 cd z i i z gies, and all gauge modes have a gap scaling as ∆ ∼ θ1e . 2 ∂i∂c(∂ A d − ∂ A d) = 0 (91) As e → ∞, we recover a pure tensor Chern-Simons the- 8π ory, and the gap approaches infinity. Therefore, the tensor Chern-Simons theory describes a stable gapped phase θ 2 jiabcd∂ ∂ ∂ A = 0 (92) of matter. 8π2 i a c bd 12

1 ˙ The second constraint is simply the divergence of the As usual, the equation of motion behaves as e2 E(γ) ∼ first and is therefore redundant. In terms of the decom- θ2E(γ) at low energies, giving a gap which behaves as 2 position of the previous section, the independent con- ∆ ∼ θ2e . straints are: Meanwhile, the other decoupled theory on the bound- ij θ2 cd jd j d ary is a theory of a symmetric tensor a , described by: ∂c(γ + ∂ λ ) = 0 (93) 8π2 Z θ2 ij ab cd S2 = 2 ( ∂aa˙ bi ∂cadj) (103) θ 8π ∂ 2 jiabcd∂ ∂ ∂ a = 0 (94) 8π2 i a c bd and the dynamical piece of the action can be written as: θ 2 jiabcd∂ ∂ ∂ a = 0 (104) Z 2 i a c bd θ2 ij ab cd 8π Sdyn = 2 ( ∂aa˙ bi ∂cadj) 8π ∂ (95) In this case, two of the components of aij come together ij ˙ a a + (γ ˙ ai + ∂aλi)(γ j + ∂ λj) to form one canonical conjugate pair, giving a single degree of freedom. (There is a third independent com- We shift γij → γij − ∂iλj, after which γij becomes a ponent of aij which makes no appearance in either the non-symmetric tensor. After the change of variables, we action or constraint, and can therefore be eliminated.) can write the constraints in the form: The constraint of Equation 104 is enough to fully con- θ2 strain this single degree of freedom, eliminating any lo- ∂aγjb = 0 (96) 8π2 ab cal degrees of freedom. We can also readily check that this theory is chiral. We can introduce a vector-valued θ j 2 ji ab cd 2 jiabcd∂ ∂ ∂ a = 0 (97) charge via ρ = (θ2/8π ) ∂i∂a∂cabd. Due to the 8π2 i a c bd high number of derivatives, these vector charges will be and the dynamical action becomes: fracton excitations. The generalized Hall response of the Z system is given by: θ2 ij ab cd ij a Sdyn = 2 ( ∂aa˙ bi ∂cadj) + γ˙ aiγ j (98) 8π ∂ ij θ2 ic jd ab θ2 ic jd ab hJ(a)i = 2 ∂a∂ca˙ bd = 2 ∂a∂cEbd We can now see that the boundary decomposes into 8π 8π (105) two independent theories. One is a theory of a non- It is important to note that, while all previous Chern- symmetric tensor γij, described by the following action Simons theories were gapped and manifestly stable, the and constraint: theory describing aij is not so lucky. Let us regularize θ Z S = 2 ijγ˙ γa (99) the theory with a Maxwell term, as usual: 1 8π2 ai j ∂ Z 1 ij 1 2 ik j` 2 S = 2 (˙a − (∂iCj + ∂jCi)) − ( ∂i∂jak`) θ2 a jb 2e 2 ab∂ γ = 0 (100) 8π2 θ + 2 (ijab∂ a˙ cd∂ a + C jiabcd∂ ∂ ∂ a ) The four components of γij come together to form two 8π2 a bi c dj j i a c bd independent degrees of freedom, described by a chi- (106) ral generalized Chern-Simons theory. Since we have two constraints on the system, we see that the sys- The equation of motion takes the form: tem is fully constrained. If we introduce charges, via j 2 c jd 1 ij θ2 ic kj ab 1 ik j` ρ = (θ2/8π )cd∂ γ , we will find vector charges that E˙ − ∂c∂aEbk + ∂k∂`B = 0 are fully mobile. (The tensor γij is neither traceless nor e2 8π2 e2 (107) symmetric, so there are no extra conservation laws be- Contracting the above equation with ∂n∂m, and yond conservation of total charge.) We can easily find in jm taking advantage of Gauss’s law, we can also write an the generalized Hall response to be: equation of motion for B as: ij ∂S θ2 j ci θ2 j ci hJ(γ)i = − = 2 cγ˙ = 2 cE(γ) (101) 1 1 ∂γij 8π 8π B¨ + ∂4B = 0 (108) e2 e2 To demonstrate the energy gap, we regularize with a Maxwell term: which is unchanged from the bare Maxwell theory, giving 2 Z 1 1 an ω ∼ k dispersion. Unlike the other tensor Chern- S = (γ ˙ ij − (∂ C + ∂ C ))2 − (ikj`∂ ∂ γ )2 Simons terms we have considered, this term does not 2e2 2 i j j i i j k` produce an energy gap, and therefore cannot stabilize θ + 2 (ijγ˙ γ k + C ∂iγjk) the two-dimensional gauge field. This is related to the 8π2 ik j j ik fact that the Maxwell and Chern-Simons terms have the (102) same number of derivatives, so there is no reason for pure 13

Chern-Simons to dominate at low energies. The Chern- The charges associated with this gauge field are simply Simons theory for aij will have the same confinement the usual Chern-Simons charges, which obey charge con- instability as the pure Maxwell theory1, so this gauge servation, but no other exotic conservation laws. field does not describe a stable two-dimensional phase of The traceless symmetric tensora ˜ij is governed by a matter. This instability will not affect the other gauge chiral generalized Chern-Simons theory: field, γij, which decouples from aij and remains in a θ Z stable deconfined phase. S = bia˜˙ a˜ j (116) dyn 8π2 ij b

C. Boundary of the Scalar Charge Theory θ − bi∂ ∂ a˜ j = 0 (117) 4π2 i j b We next apply similar logic to the scalar charge theory, where we found that the θ term takes the form: We note that the two independent components ofa ˜ij are combined into one canonical conjugate pair. Given θ Z 1 S = ∂ ( iabA˙ A j + ∂ φiab∂ A j) that there is one constraint ona ˜ij, we see that the the- θ 4π2 a 2 ij b j i b ory is fully constrained, and there are no local degrees (109) 1 j of freedom. We can also consider the appropriate gauge iab j − ∂t( ∂aAijAb ) bi 2 charges, ∂i∂ja˜b = ρ. As can readily be checked, this charge has a conserved dipole moment. We therefore For a spatial boundary with normal in thez ˆ direction, have an example of a stable two-dimensional chiral frac- the action becomes: ton phase. Bearing in mind the symmetry ofa ˜ij, we can readily find the generalized Hall response to be: θ Z 1 S = biA˙ A j + ∂ φbi∂ A j (110) ∂ 4π2 2 ij b j i b θ θ ∂ hJ iji = (iba˜˙ j + jba˜˙ i) = (ibE j + jbE i) 8π2 b b 8π2 b b We consider the second term first. The z derivative of (118) φ can be varied independently of φ itself on the bound- If we define a generalized “conductivity” tensor via J ij = ijk` ary and constitutes an independent Lagrange multiplier, σ Ek`, then this tensor will be given by: while the in-plane components of ∂jφ can be integrated θ by parts to yield a second constraint equation. Using σijk` = (ikδj` + jkδi`) (119) the decomposition from the previous sections, the two 8π2 constraints take the form: We can verify the presence of an energy gap by reg- θ ularizing the theory with a Maxwell term. In this case, − bi∂ ∂ a j = 0 (111) 4π2 i j b we include a Maxwell term appropriate to the traceless scalar charge theory:

θ Z 1 1 bi∂ λ = 0 (112) S = ((a˜˙ − (∂ ∂ − δ ∂2)φ)2 − (ij∂ ∂ka˜ )2) 4π2 i b 2e2 ij i j 3 ij j ik Likewise, the dynamical piece of the action can be writ- θ bi ˙ j bi j + ( a˜ija˜b − 2φ ∂i∂ja˜b ) ten as: 8π2 (120) θ Z 1 S = bi(˙a a j + λ˙ λ ) (113) dyn 2 ij b i b 1 ˙ 4π ∂ 2 The corresponding equation of motion behaves as e2 E ∝ θE, giving us an energy gap scaling as ∆ ∼ θe2. This It is noteworthy that, upon separating aij into its trace i Chern-Simons term is therefore capable of fully gapping a i and a traceless symmetric tensora ˜ij, the trace com- the traceless scalar charge theory, yielding a stable two- ponent does not appear in either Sdyn or the constraint dimensional phase of matter. (Note that, alternatively, equations, as can be readily checked. The trace can one might consider adding this same Chern-Simons term therefore be eliminated from the theory entirely, leav- to the traceful scalar charge Maxwell theory, in which ing us with a decoupled theory of a vector field and a case the trace mode will remain gapless and the issue of traceless symmetric tensor field. The vector field λi is stability is unclear. The investigation of such a partially governed by a standard Chern-Simons theory: gapped theory is left as a task for the future.) Z θ bi ˙ Sdyn = 2 λiλb (114) 8π D. Realization in Two Dimensions

θ In this section, we have identified several tensor ana- bi∂ λ = 0 (115) 4π2 i b logues of Chern-Simons theory, most of which are stable 14 theories supporting either fractons or one-dimensional Simons term stabilizes the conventional chiral U(1) spin particles. By looking at the boundaries of other higher- liquid in two dimensions. We therefore see that we have spin theories in three dimensions, we expect to discover a rather narrow path to obtaining fractons in two dimen- many other generalized Chern-Simons theories, with ten- sions: the fractons must be U(1) objects (integer-valued sors of arbitrary rank, which should generically host frac- charge), and the corresponding U(1) gauge theory must ton excitations. be stabilized against confinement by some mechanism. While we have discussed these theories as boundaries The simplest means of stabilization is by adding Chern- of three-dimensional higher-spin systems, we note that Simons-like terms to gap the gauge field, as we have dis- they can also occur in purely two-dimensional systems cussed. We leave as an open question whether or not under appropriate conditions. Most notably, certain val- there are other valid 2d stabilization mechanisms, such ues of θ correspond to a bulk which is identical to the as the addition of appropriate gapless matter fields. θ = 0 system (e.g. when θ1 is a multiple of 2π in the vector charge theory). At these specific values of θ, the bulk is completely irrelevant to the boundary physics, and the generalized Chern-Simons theory is simply a decoupled theory sitting on top of the bulk. This is closely analogous to Maxwell theory, where θ = 2π corresponds to a V. CONCLUSIONS normal two-dimensional integer quantum hall layer de- posited on the boundary.26,27 Similarly, the boundaries of the higher-spin theories at these special values of θ In this paper, we have investigated the role of θ terms host purely two-dimensional fracton theories, in a natu- in the physics of three-dimensional higher-spin U(1) ral generalization of integer quantum hall states. Indeed, gauge theories. As in Maxwell theory, these higher- some of these states can be intuitively understood as the spin θ terms have two primary effects: the attachment mobile particles of the theory (e.g. dipoles in the scalar of electric charge to magnetic monopoles (the Witten charge theory) being put into a quantum Hall state of effect) and the presence of a nontrivial boundary the- their emergent magnetic field. This physical picture will ory. We have worked out the details of the Witten effect be explored further in forthcoming work. for subdimensional particles, relating the θ parameter Values of θ other than these special “integer” val- to the amount of electric charge attached to magnetic ues naturally correspond to generalized fractional quan- monopoles. We have also studied the resulting bound- tum hall systems. In particular, the boundary theory ary theory, which hosts fracton excitations coupled to a will host excitations with a fraction of the fundamen- U(1) gauge field, stabilized by generalized Chern-Simons tal charge of the bulk theory. These fractional theories terms. These boundary theories thereby open the door will also be able to occur in strictly two-dimensional sys- to studying two-dimensional U(1) fracton phases, both tems, modulo issues of symmetry implementation. In as boundary theories and in strictly two dimensions. the story of conventional Chern-Simons theory, the only There remains interesting work to be done, particu- difference between a strictly 2d fractional quantum hall larly with the new two-dimensional phases. We have system and a boundary theory is in the implementation identified some of the basic features characterizing these of symmetries (such as time reversal) on fractionalized phases, such as their generalized Hall responses, but particles, where issues concerning anomalies need to be they surely possess features which still remain to be carefully considered.26,27 If the symmetry is allowed to unlocked. In particular, a purely two-dimensional chi- be anomalous (for example, if an emergent symmetry ral theory may host robust one-dimensional edge modes. like particle-hole symmetry is playing the role of time However, it remains unclear at present how precisely to reversal), then these fractional boundary theories are study such edge modes. It would also be interesting perfectly valid in strictly two dimensions. An entirely to study the role of global symmetries in the theories analogous story should hold in the case of the higher identified here, sorting out which symmetry implemen- rank theories, but we leave the detailed consideration of tations are possible, both with and without anomalies. symmetries and anomalies as a fight for future days. Such a study would be useful, not only for characteriz- The presence of fractons and other subdimensional ing the new two-dimensional phases, but for character- particles in a purely two-dimensional system seems sur- izing the bulk behavior of the higher-spin U(1) gauge prising at first. For discrete fracton models, previ- theories in the presence of enrichment by global symme- ous work has suggested that only three-dimensional (or tries. On the more down-to-earth side of things, it would higher-dimensional) models should be possible.4,5 As for be highly useful to find two-dimensional lattice models U(1) fracton systems, theories involving only Maxwell- which demonstrate the properties of these phases explic- type terms suffer from instabilities to confinement in two itly, to complement the field-theoretic arguments pre- dimensions.1 In the present case, we also have U(1) frac- sented here. In addition, more investigation is required tons, but the two-dimensional gauge modes are gapped regarding how to measure the generalized Hall responses, out via a Chern-Simons-like action. This energy gap re- which will be important for experimental detection of sults in the stabilization of the theory, just as a Chern- these two-dimensional fracton phases. 15

ACKNOWLEDGMENTS APPENDIX B: ARE θ TERMS TOPOLOGICAL?

I would like to acknowledge useful conversations with In this work, we have considered natural generaliza- Kevin Slagle, Claudio Chamon, Rahul Nandkishore, tions of the θ term encountered in the study of Maxwell Mike Hermele, Yizhuang You, Juven Wang, Abhinav theory. In the context of Maxwell theory, we can write Prem, Han Ma, T. Senthil, Liujun Zou, Yahui Zhang, this term in the form: and Yang-Zhi Chou. This work was partially supported Z Z 4 µνρσ by a Simons Investigator Award to T. Senthil at the Mas- Sθ = d x Fµν Fρσ = F ∧ F (122) sachusetts Institute of Technology and partially by the Center for Theory of Quantum Matter at the University where the last step takes advantage of the language of of Colorado Boulder. differential forms. This rewriting makes it manifest that the θ term is topological, in that it does not rely on any special choice of coordinates or on the metric of the system. This ensures that the properties of the θ term, such APPENDIX A: MUTUAL STATISTICS as being a total derivative and giving rise to the Witten effect, do not depend on whether we are considering a In the text, we claimed that, in the scalar charge the- flat or curved spacetime. Correspondingly, the boundary ory, a magnetic vector charge and a parallel coplanar physics arising from the θ term is that of Chern-Simons electric dipole have mutual π statistics (i.e. the wave- theory, which is a topological quantum field theory. function picks up a minus sign when one particle is In the main text, we have studied the generalized θ wound around the other). Assuming that the magnetic terms of the higher rank theories only at the level of flat vector and the electric dipole are both bosons, the mu- space, with all indices raised and lowered by the flat met- ij tual π statistics has the effect of making the bound state ric, δ . But given the formal similarity of these terms of the two into a fermionic object, as is familiar from the to the conventional θ term, it seems reasonable to ask study of two-dimensional anyons. We will now verify the whether or not the new θ terms also possess a topolog- claim of mutual π statistics. ical character which would allow simple generalization It is easiest to see the mutual statistics by considering to curved spaces. Unfortunately, this does not appear winding the dipole around the magnetic vector (though to be the case. First of all, unlike their antisymmetric winding the vector around the dipole should yield the cousins, symmetric tensors do not have a natural formu- same result). As derived from previous work on tensor lation in terms of differential forms, so there is no rea- electromagnetism3, the effective magnetic field seen by son to expect any metric-independent formulation of the a dipole pi takes the form Bi = −p Bij (where the theory. Furthermore, the boundary physics induced by eff j the higher rank θ terms is that of tensor Chern-Simons dipole moment is in units of the minimal dipole). Now theory, hosting fractons and other subdimensional par- let us consider the magnetic field created by a magnetic ticles. Phases with such subdimensional particles tend vector charge ∂ Bij = 2πgj. The effective magnetic field i to have large ground state degeneracies, growing with seen by the dipole due to a magnetic vector charge then the system size4,5. This is clearly incompatible with the obeys: notion of a topological theory, where the physics should depend on the topology, but not the size, of the system i ∂iBeff = −2π(p · g) (121) on which it is defined. Indeed, recent work31 has indi- cated that fracton models have much greater sensitivity When the magnetic vector and electric dipole are parallel to the geometry of the system than a pure topological i and are both unit strength, we see that ∂iBeff = −2π, phase. Fracton phases can even have robust degeneracy meaning that the dipole sees the magnetic vector as a on a topologically trivial manifold, induced purely by the (negative) unit monopole of its effective magnetic field. curvature of the system. In other words, a surface enclosing the magnetic vector Similarly, the higher rank θ terms will likely feel the charge will see a −2π flux of the effective magnetic field effects of geometry in subtle ways, making the exten- for the dipole. The effective magnetic field correspond- sion to curved spaces a nontrivial one. In this work, we ing to this monopole is not necessarily isotropic (it can will need to mostly content ourselves with the study of θ depend on the orientations of pi and gi), but when the terms in approximately flat space, which should be suf- dipole is wound around the vector in their plane of mu- ficient for understanding the realization of these phases tual motion, we can still conclude by symmetry that the in the solid state context. Nevertheless, the behavior enclosed flux (and the corresponding Aharonov-Bohm of these theories in curved space is a question of both phase) is precisely half of the total, −π. Thus, when intrinsic interest and experimental relevance, since one an electric dipole is wound around a magnetic vector can simulate the effects of curvature by putting the sys- within their mutual plane of motion, the wavefunction tem under appropriate strains. We leave the detailed of the system picks up a factor of e−iπ = −1, completing investigation of higher rank phases in curved space to the proof of mutual π statistics. future work, though we can make some conjectures. It 16 is unclear if the higher rank θ terms remain total deriva- We can explicitly take gauge fixing into account by tives in curved space, or if they respond to curvature in adding a resolution of the identity as follows: such a way that changes the bulk physics of the gap- Z α less gauge mode. It is entirely possible that the θ term µ α δG[A ] i R d4x 1 F µν F Z = DA Dα δ(G[A ]) det e 4 µν will somehow modify the dispersion of the gauge mode, δα though we strongly suspect that it remains gapless. We (124) α also expect the Witten effect to carry over unchanged, as where Aµ = Aµ + ∂µα is a gauge-transformed field, and α µ α this arises from a purely local constraint on the electric G[A ] = ∂ Aµ −Λ is a gauge-fixing condition, fixing the and magnetic fields, which should have no dependence longitudinal component of A. This extra determinant on curvature. The validity of these conjectures will need factor leads directly to ghost fields, which play a signif- to be studied in detail in the future. icant role in non-abelian gauge theories. In an abelian gauge theory, however, we have the crucial fact that the determinant, det(δG/δα) = det(∂2), is a constant and can be brought outside the integral. When calculating any expectation values, this constant will cancel out, in- APPENDIX C: GAUGE FIXING dicating that ghost fields are completely decoupled from the physical gauge field. When calculating expectation values via the path integral formulation for a gauge theory, one unfortunate diffi- A Higher Rank Example culty is that the “plain vanilla” sort of path integral (e.g. Equation 1) integrates over all gauge-equivalent configurations of the gauge field, resulting in divergences which We now move on to consider a higher rank theory, need to be tamed. The simplest way to handle this prob- specifically the scalar charge theory. The bare path in- lem is to introduce a gauge-fixing procedure explicitly tegral for this theory takes the form: into the path integral. For standard abelian gauge the- Z R 3 ories, such as Maxwell theory, such gauge-fixing proce- Z = DAijDφei d xdtL (125) dures are fairly mild, merely allowing for some freedom in the choice of the photon propagator. For nonabelian where the Lagrangian was defined in Equation 29. We gauge theories, however, additional complications arise, can similarly add in a resolution of the identity as follows: and one must account for extra “ghost” fields in order to fully decouple the physical gluon from unphysical gauge Z Z = DAijDφDα δ(G[Aα, φα]) (126) degrees of freedom. The higher rank gauge theories considered in this paper have been natural tensor analogues α α δG[A , φ ] R 3 of abelian gauge theories, so one would guess that ghosts det ei d xdtL (127) δα play no role in these theories. We will here check this explicitly and verify that there is no need for ghost fields α α where Aij = Aij + ∂i∂jα and φ = φ +α ˙ are the in the theories considered here. ij ˙ gauge-transformed fields and G = ∂i∂jA + φ − Λ is a gauge-fixing condition which fixes the unphysical components of A and φ. Once again, we take advantage of the crucial fact that the determinant, det(δG/δα) = i 2 2 Review of Maxwell Theory det((∂i∂ ) + (∂t) ) is a constant and can be taken outside of the integral, therefore canceling out of any expectation values. The same principle holds for any of the We begin by reviewing Maxwell theory, in which the other higher rank theories, whose gauge transformations bare path integral takes the form: behave similarly. We can therefore conclude that ghost fields decouple and have no role to play in any of the Z theories considered here, which justifies our neglect of µ i R d4x 1 F µν F Z = DA e 4 µν (123) them throughout this paper.

1 A. Rasmussen, Y.-Z. You, and C. Xu, Stable gapless bose 3 M. Pretko, Generalized electromagnetism of sub- liquid phases without any symmetry. arXiv:1601.08235 dimensional particles. Phys. Rev. B 96, 035119, (2016) arXiv:1606.08857v2 (2017) 2 M. Pretko, Subdimensional particle structure of higher 4 S. Vijay, J. Haah, L. Fu, A new kind of topological quan- rank U(1) spin liquids. Phys. Rev. B 95, 115139, tum order: A dimensional hierarchy of quasiparticles built arXiv:1604.05329v3 (2017) from stationary excitations. Phys. Rev. B 92, 235136, 17

arXiv:1505.02576 (2015) 19 K. Slagle, Y. B. Kim, Fracton topological order 5 S. Vijay, J. Haah, L. Fu, Fracton topological order, gen- from nearest-neighbor two-spin interactions and dualities. eralized lattice gauge theory and duality. Phys. Rev. B 94, arXiv:1704.03870v2 (2017) 235157, arXiv:1603.04442 (2016) 20 B. Shi, Y.-M. Lu, Decipher the nonlocal entanglement 6 C. Chamon, Quantum glassiness in strongly correlated entropy of fracton topological orders. arXiv:1705.09300 clean systems: An example of topological overprotection. (2017) Phys. Rev. Lett. 94 040402, arXiv:cond-mat/0404182v2 21 S. Vijay, L. Fu, A generalization of non-abelian anyons in (2005) three dimensions. arXiv:1706.07070 (2017) 7 S. Bravyi, B. Leemhuis, and B. M. Terhal, Topological 22 G. B. Halász, T. H. Hsieh, L. Balents, Fracton order in an exactly solvable 3D spin model. Ann. Phys., topological phases from strongly coupled spin chains. Vol. 326:4, 839, arXiv:1006.4871 (2010) arXiv:1707.02308 (2017) 8 C. Castelnovo and C. Chamon, Topological quantum 23 E. Witten, Dyons of charge eθ/2π. Phys. Lett. B 86, 3-4 glassiness. Philosophical Magazine Vol. 92, Iss. 1-3, (1979) arXiv:1108.2051v2 (2012) 24 X.-L. Qi, T. L. Hughes, S.-C. Zhang, Topological field the- 9 B. Yoshida, Exotic topological order in fractal spin liquids. ory of time-reversal invariant insulators. Phys. Rev. B 78, Phys. Rev. B 88, 125122, arXiv:1302.6248v4 (2013) 195424, arXiv:0802.3537 (2008) 10 J. Haah, Local stabilizer codes in three dimensions with- 25 G. Rosenberg, M. Franz, Witten effect in a crys- out string logical operators. Phys. Rev. A 83, 042330, talline topological insulator. Phys. Rev. B 82, 035105, arXiv:1101.1962v2 (2011) arXiv:1001.3179v3 (2010) 11 S. Bravyi and J. Haah, Quantum self-correction in the 26 T. Senthil, Symmetry protected topological phases of quan- 3D cubic code model. Phys. Rev. Lett. 111, 200501, tum matter. Annual Review of Condensed Matter Physics arXiv:1112.3252 (2013) 6, 299, arXiv:1405.4015 (2014) 12 D. J. Williamson, Fractal symmetries: Ungauging the cu- 27 C. Wang, T. Senthil, Time-reversal symmetric U(1) bic code. Phys. Rev. B 94, 155128, arXiv:1603.05182v2 quantum spin liquids. Phys. Rev. X 6, 011034, (2016) arXiv:1505.03520v2 (2016) 13 S. Vijay, Isotropic layer construction and phase diagram 28 The vector ~g must be an integer linear combination of the for fracton topological phases. arXiv:1701.00762 (2017) fundamental unit vector charges, which can depend on 14 H. Ma, E. Lake, X. Chen, M. Hermele, Fracton topolog- the underlying lattice. On the cubic lattice, for example, ical order via coupled layers. Phys. Rev. B 95, 245126, ~g must take the form (nx, ny, nz) for integers nx, ny, and arXiv:1701.00747 (2017) nz. For other lattices, different structures of charge lattice 15 A. Prem, J. Haah, R. Nandkishore, Glassy quantum dy- are possible. namics in translation invariant fracton models. Phys. Rev. 29 The other possibility, where q is odd and g is even, is B 95, 155133, arXiv:1702.02952 (2017) simply related by duality. 16 M. Pretko, Finite-temperature screening of U(1) Fractons. 30 One could in principle also consider models where one or Phys. Rev. B 96, 115102, arXiv:1706.01899v2 (2017) the other (or perhaps even both) of the magnetic vectors 17 M. Pretko, Emergent gravity of fractons: Mach’s principle or electric dipoles are fermions. The analysis in this case revisited. Phys. Rev. D 96, 024051, arXiv:1702.07613v3 would be very similar, but likely has a few differences. (2017) 31 K. Slagle, Y. B. Kim, Quantum Field Theory of Fracton 18 T. H. Hsieh, G. B. Halász, Fractons from partons. Topological Order and “Topological” Degeneracy from Ge- arXiv:1703.02973 (2017) ometry. arXiv:1708.04619 (2017)