Arxiv:2004.00638V3 [Cond-Mat.Str-El] 17 Mar 2021 Tron ﬁlling Νt = ±4 Per Moire´ Unit Cell

Charged Skyrmions and Topological Origin of Superconductivity in Magic Angle Graphene

Eslam Khalaf,1 Shubhayu Chatterjee,2 Nick Bultinck,2, 3 Michael P. Zaletel,2 and Ashvin Vishwanath1, ∗ 1Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA 2Department of Physics, University of California, Berkeley, CA 94720, USA 3Department of Physics, Ghent university, 9000 Gent, Belgium (Dated: March 18, 2021)

Spontaneous symmetry breaking plays a pivotal role in stance takes place in quantum Hall ferromagnets, where topo- many areas of physics, engendering a variety of excitations logical textures of spin in the form of skyrmions acquire a from sound modes in solids to pions in nuclear physics. charge due to the Landau level topology and are found to be While these represent the linear response of the symme- the lowest energy charge excitations [8, 15, 17, 22, 31, 52, 65]. try breaking condensate, equally important are nonlin- On the other hand, finding situations where Cooper pair- ear configurations, solitons, which enjoy exceptional sta- ing occurs between charged topological textures, rather than bility when associated with a topological charge . A no- between electrons, is even harder to come by. For exam- table example is in the Skyrme model of nuclear forces, ple, in the aforementioned quantum Hall ferromagnets where where the neutron and proton are interpreted as topologi- charged topological textures have been experimentally estab- cal solitons of the pion condensate. We argue that similar lished, strong breaking of time-reversal symmetry makes su- models may describe a remarkable new material - magic perconductivity highly unlikely. In fact, obtaining robust su- angle graphene - obtained on twisting and stacking two perconductivity from topological textures typically requires sheets of graphene. When the relative angle between the simultaneously satisfying two conditions: (a) an unbroken sheets is near the magic angle of ∼ 1◦ degree - insulating time-reversal symmetry and (b) the existence of stable low en- behavior resulting from ordering of electrons is observed, ergy charge 2e topological textures. The latter can be achieved which gives way to superconductivity on changing the elec- if the fundamental defects have charge 2e or by pairing charge tron density. Here we propose a unifying description of e defects of the same electric charge. Here, we show that all both the order underlying the insulating state as well as these critera are satisfied in a simple model consisting of two the superconductivity. In our theory, while the symme- time-reversal-related quantum Hall (or flat Chern band) ferro- try breaking condensate leads to the ordered phase, topo- magnets with purely repulsive interactions, coupled via tun- logical solitons in the condensate - skyrmions - are shown neling. Such a model, we point out, captures the essential to be bosons that carry an electric charge 2e. Condensa- physics of magic angle graphene making it a promising can- tion of skyrmions leads to a superconductor whose pair- didate for superconductivity arising from the pairing of topo- ing strength, symmetry and other properties are inferred. logical textures, a mechanism that is fundamentally different More generally, we show how topological textures can mit- from the conventional electron-phonon mechanism. igate Coulomb repulsion to pair electrons and provide a new route to superconductivity. Our mechanism potentially applies to much wider class of systems but crucially Let us start with a brief review of magic angle twisted bi- invokes certain key ingredient such as inversion symmetry layer graphene (MATBG). Two sheets of graphene twisted present in magic angle graphene. We discuss how these relative to one another generate a moire´ pattern and corre- insights not only clarify why certain correlated moire ma- spondingly a reduced Brillouin zone in reciprocal space. Pre- terials do not superconduct, they also point to promising vious work demonstrated that the reconstruction of the elec- new platforms where robust superconductivity is antici- tronic bands by the moire´ lattice leads to minibands with ex- pated. tremely narrow bandwidth near charge neutrality at a magic angle ∼ 1o [5, 25, 29, 57], separated by a band gap from other bands. Recent experiments [8, 9, 30, 63] revealed dramatic I. INTRODUCTION new physics near the same magic angle. Band gaps arising from the moire´ potential are expected and observed at elec- Typically when charge is added to an insulator, electron or arXiv:2004.00638v3 [cond-mat.str-el] 17 Mar 2021 tron filling νT = ±4 per moire´ unit cell. In addition, insula- hole excitations are produced leading to charge conduction. tors at other integer filling including νT = ±2 [8, 30, 63] and Can solitons play the role of charge carriers? This unusual in some experiments also at νT = 0 [30, 55] and at certain scenario can be realized through electrically charged topologi- odd integer fillings have also been observed and are attributed cal textures which have been theoretically proposed in various to the effects of electron-electron interaction. Furthermore, contexts [1, 16, 18, 19, 23, 36, 56], although physical realiza- superconductivity has been repeatedly observed in twisted bi- tions are rare. In fact, the only experimentally established in- layer graphene, though its precise relation to the correlated insulating phase remains to be determined. While early experiments observed the superconductivity in the vicinity of the νT = −2 correlated insulator, subsequently a wider extent of ∗ [email protected] superconductivity has been observed [44, 55]. 2

II. QUANTUM HALL FERROMAGNETISM MODEL OF From valley/sublattice to Chern /pseudospin MAGIC ANGLE GRAPHENE (τ, σ) → (γ, η) γ = (γx, γy, γz) = (σx, σyτz, σzτz) η = (η , η , η ) = (σ τ , σ τ , τ ) We begin by noting an important feature of the nearly flat x y z x x x y z bands near charge neutrality. Electrons in graphene carry both 0 Basis a spin (s =↑, ↓) and a twofold valley degeneracy (τ = K,K ) 0 Valley τz = K/K which would suggest that filling each moire´ lattice site would Sublattice σ = A/B require four electrons. In reality, it takes twice as many elec- z Chern sector γ = σ τ = +/− trons to fill the nearly flat bands, since they consist of two con- z z z Pseudospin η = τ =↑ / ↓ nected bands as shown in Figure 1. Conventionally, one dis- z z ps ps tinguishes the two bands by their kinetic energy but in view of the narrow bandwidth, other linear combinations may be Symmetries preferable. A different choice is the sublattice basis - which Symm (τ, σ) basis (γ, η) basis separates bands on the basis of their weight on the two sites T τxK γxηxK σ = A, B of the honeycomb lattice of monolayer graphene. C2 σxτx ηx iφτz iφηz This basis arises naturally in the ideal chiral limit of Ref. [24] UV (1) e e where the sublattice polarization is complete, but can also be defined at physical parameter values as explained in Ref. [7]. TABLE I: Definition of new variables - Chern Sector and Even for realistic parameters, the sublattice polarization is pseudospin, from the valley and sublattice degrees of substantial, so a reasonable approximation is to ignore sublat- freedom. Action of symmetries on the internal indices are tice off diagonal terms in the density operator [7]. This sublat- shown on the right, in both set of variables. Additionally, the tice polarized (SP) approximation will be assumed throughout independent pseudospin rotations in the Chern sectors are this paper. 1 generated by the ηP± where the projectors P± = 2 (1 ± γz) A remarkable feature of the sublattice basis is that individ- single out a specific Chern sector. ual bands in this basis carry Chern number. This allows for a mapping of MATBG to a pair of time-reversal related copies of quantum-Hall like systems. The presence of flavors then wavefunctions uγη;k(r), which can be conveniently encapsu- relates our problem to quantum Hall ferromagnetism, as ad- lated in the “form” factor expression for the density which can vocated in Ref. [7]. This mapping to a generalized spin-valley be written within the SP approximation as [7]: ferromagnet is broadly consistent with the observation of a X cascade of polarization transitions on varying electron den- n = F (k)eiγΦq (k)Ψ† Ψ (2) sity [62, 66] and it acquires a particularly simple form in the q q γ,k γ,k+q k,γ=± vicinity of half-filling ν = ±2 when the spin degree of freedom can be neglected. In this case, the low energy physics is Note that the form factor is diagonal in the Chern sectors dominated by four flat bands, each band being labelled by a with equal amplitude F and opposite phase Φ for opposite valley and sublattice index. Chern sectors. It follows that nq - and hence the interac- The valley-sublattice resolved flat bands are characterized tion - is symmetric under independent pseudo-spin rotations by Chern number C = ±1. Labelling sublattice (valley) by U+(2) × U−(2) in each C = ± sector. σ (τ), we have C = στ which is opposite for opposite valleys Remarkably, at half-filling the ground state of the interac- due time-reversal symmetry T and also opposite for opposite 1 P tion: HC = 2 q δnqVqδn−q can be determined exactly for sublattices within the same valley due to the combination of any purely repulsive interaction, Vq > 0 ∀q. In this case, two-fold rotation and time-reversal C2T . Thus, the flat bands HC is a sum of positive definite terms, so any state which is can be conveniently studied by introducing the pseudospin annihilated by each δnq |Ωi = 0 is a ground state. This is T T spinors Ψ+ = (cKA, cK0B) and Ψ− = (cKB, cK0A) for satisfied by1 a manifold of C = 0 states, |Ωi, obtained by the ± Chern sectors as shown in Fig. 1. A summary of the two choosing a direction n± in pseudo-spin space for each of the different basis: valley/sublattice and Chern sector/pseudospin ± Chern sectors and “filling” the pseudo-spin bands indepen- and the relationship between them is provided in Table I to- dent of k yielding a pseudospin ferromagnet in each Chern gether with the implementation of each symmetry in both ba- sector. Thus: sis. † The Hamiltonian arises from projecting the Coulomb inter- hΩ|Ψ±ηΨ±|Ωi = n± (3) action V (r) into the flat bands, Z Z 2 † 1 2 0 0 H = d k ΨkhkΨk + d r δn(r)V (r−r )δn(r ) (1) 2 1 Actually, there is another possible set of ground states with Chern number |C| = 2 obtained by filling only one of the two Chern sectors. These where δn(r) is the deviation of the density n(r) = states, however, do not admit non-trivial topological pseudospin textures P † γ=± Ψγ (r)Ψγ (r) from its average value. The Fourier com- and strongly break time-reversal symmetry making them incompatible with ponents of n(r) depend on the detailed structure of the Bloch superconductivity and, as a result, not relevant for our discussion. 3

FIG. 1: Defining Pseudospin: Linear combinations of the nearly flat bands of twisted bilayer graphene (left) give rise to pseudospin bands with opposite Chern numbers. Two states with the same pseudospin that lie in opposite Chern sectors have identical valley wavefunctions but are distinguished by A-B sublattice polarization (red and blue, third column from left). In this figure we have omitted spin. Half filling of this spinless model with net Chern number C = 0, at energies below the interaction scale, correspond to quantum Hall pseudospin ferromagnets in each Chern band, described by unit vectors n±.

corresponds to a tripling of the unit cell on the graphene lattice Now let us reintroduce the single particle dispersion h scale. On the other hand, if n points along the Z direction, the valley charge is conserved and this order corresponds to the whose form is constrained by the pseudo-spin and C2T symmetry, valley Hall state. Anisotropies neglected here which arise on going beyond the SP approximation, prefer n ordering in the x y h(k) = γ hx(k) + γ hy(k) (4) nx, ny plane and select the K-IVC order.

It is important to stress here that the form of h(k) above is the most general form of the dispersion allowed by the symme- III. CHARGED TOPOLOGICAL TEXTURES AND tries of the chiral model and we do not assume the flat band EFFECTIVE SIGMA MODEL limit. Deviation from the chiral limit introduces an extra term proportional to ηzγ0 (other terms are prohibited either by C2T Due to the band topology of the two Chern sectors, pseu- or particle-hole symmetry) which was shown in Ref. [7] to be dospin skyrmions in n± carry electric charge [22, 52] of relatively small. This means that the main effect of deviation ±e. The charge density δρ(r) = e (q+(r) − q−(r)) where from the chiral limit is altering the values of hx,y(k) rather 1 q±(r) = 4π n± · (∂xn± × ∂yn±) is the topological density than introducing new terms in (4). Thus, (4) takes realistic in each Chern sector, which integrates to unity for a single dispersion fully into account. skyrmion. Let us neglect the antiferromagnetic coupling J The dispersion h(k) acts as tunneling between states with between the two Chern sectors for now, postponing discus- the same pseudospin and momentum in opposite Chern sec- sion of its effect to the next section. Then, we find that the tors. Its main effect in the limit of strong interaction is to size of a charge ‘e’ skyrmion in a single Chern sector is deter- introduce a ’superexchange’ term J ∼ h2/U, where U is mined only by the Coulomb repulsion which prefers to spread the typical interaction scale, which antiferromagnetically cou- it out over the entire system. In this case, its energy is given ples the pseudospins in opposite Chern sectors. This reduces only by the elastic contribution ESk = 4πρps, where ρps is the SU(2) × SU(2) symmetry of the interaction Hamiltonian, the pseudospin stiffness associated with the n± vector fields down to a single SU(2). The generation of such superex- [15, 52]. change term is quite generic for any pair of opposite Chern To determine if skyrmions play the role of charge carriers, number tunneling-coupled ferromagnets with SU(2) spin ro- we must compare their energy to that of other charge carri- tation symmetry, as shown in supplemental material, and it ers in the system, in particular the particle-hole excitations. plays a crucial role in the skyrmion superconductivity dis- This question generally depends on system details - in quan- cussed below. It also plays an important role in selecting the tum Hall systems at ν = 1, the energy of a skyrmion-pair, insulating ground state among the quantum Hall ferromag- 8πρps, is smaller than the particle-hole gap, ∆ph by a fac- netic states, favoring those for which the pseudospins in the tor of 2. For MATBG, the ratio: 8πρps/∆ph can be com- two Chern sectors are anti-aligned, n+ = −n− = n. In puted numerically from the self-consistent Hartree-Fock cal- the language of Ref. [7], the resulting state corresponds to the culation leading to the results shown in Fig. 5. For model Kramers-inter valley coherent state (K-IVC) state if n lies in parameters close to the ideal chiral limit, the ratio is close to the XY plane. In this phase the conservation of valley charge the quantum Hall value of 0.5, whereas for realistic parame- is spontaneously broken, (hence inter valley coherence) which ters it is about 0.8 − 0.9 < 1 so that, ignoring anisotropies, 4

FIG. 3: Sigma model parameters: Pseudospin stiffness FIG. 2: Tunneling-induced superexchange J: (Top) Action (upper panel), antiferromagnetic coupling J, and pseudospin of symetries in the Chern and pseudospin basis. (Middle) easy-plane anisotropy λ (lower panel) as a function of the Starting from the two pseudospin ferromagnets that are twist angle θ for different values of the gate screening related by time reversal symmetry, dispersion h acts as distance d for dielectric constant = 9.5. All curves are computed using the analytic expressions provided in the tunneling between C2zT -related states (Bottom) leading to a superexchange term J ∼ h2/U that acts as an supplemental material using the band renormalization antiferromagnetic coupling between pseudospins in opposite scheme of Ref. [26]. Chern sectors. the Berry phase term, the second represents the elastic energy skyrmions are indeed the lowest energy charged excitations. of the non-uniform pseudospin configurations with ρps ∼ 1 While anisotropies are absent in the SP approximation, we meV, and the third term includes the effects of antiferromag- will leave a full calculation of skyrmion energetics in MATBG netic coupling via the ‘super-exchange’ J ∼ 0.5−1 meV. The including the most general kinds of ansiotropy, to future work. chemical potential couples to the charge deviations from the It is sufficient to note that they remain low energy excitations, background which is given by the skyrmion (antiskyrmion) and we will proceed based on the premise that skyrmion exci- topological charge in the + (−) Chern sector. tations are important to the charge physics. We note here that the deviation from the perfect sublat- To derive the effective field theory for skyrmions, we in- tice polarization assumed so far leads to an additional term tegrate out the electrons while retaining the charge degree of λ(n+,xy · n−,xy − n+,z · n−,z) with λ ≈ 0.5 meV for real- freedom in the topological textures. We thus derive an effec- istic MATBG parameters (cf. Fig. 3). This term favors out-of- tive description of MATBG by taking the pseudospin variables plane ferromagnetic coupling and in-plane anferromagnetic in the two Chern sectors n± to be slowly varying fields lead- coupling, thus acting as an easy plane anisotropy which se- ing to (see supplemental material for details) lects the XY pseudospin antiferromagnetic order (this is the Krammers intervalley coherent order discussed in Ref. [7]). The dependence of the sigma model parameter J and λ on the X 1 ρps 2 L[n+, n−] = A[nγ ] · ∂τ nγ + [∇nγ ] angle and interaction parameters is shown in Fig. 3. 2A 2 γ=± M 1 Z + Jn · n − µeδρ + d2r0δρ(r)V (r − r0)δρ(r0) + − 2 IV. SUPERCONDUCTIVITY FROM SKYRMION PAIRING (5) The antiferromagnetic coupling J between opposite Chern 2 where AM is the area of the moire´ unit cell . The first term is sectors leads to the binding of a skyrmion anti-skyrmion pair in the opposite Chern sectors. The net charge of this combined object is Q = e(q+ − q−) = 2e, i.e. the exchange J

√ 2 √ has effectively resulted in an extended Cooper pair. Note the 2 given by 3a (with a = 3a ) 2θ2 CC crucial interplay of anti-ferromagnetic coupling between op- 5

binding energy remains finite for the physically relevant parameter regime which we will assumed in what follows. Superconductivity does not follow from pairing alone. To establish a nonzero superfluid stiffness and transition temperature Tc, the condensing 2e bound state must have a finite effective mass despite the flat-band dispersion of charge e skyrmions. Remarkably, this effective mass can be generated entirely by the Coulomb repulsion through the exchange scale J. This can be understood by noting that the skyrmion and antiskyrmion in opposite Chern sectors feel opposite ef- 2π fective magnetic fields Beff = ~ . This leads to a Lorentz eAM force which tends to pull them apart when they move together. Given that a skyrmion-antiskyrmion pair are bound together by J as shown in Figure 4b, the restoring ”spring” force bal- ances the Lorentz force leading to an effective mass. More precisely, writing the Lagrangian for a a skyrmion and an an- FIG. 4: Pairing of Charged Topological textures: (a) tiskyrmion at R and R : Single charge e skyrmion in one of the Chern sector. + − Pseudospins in n+ and n− not antiferromagnetically aligned eB k L = eff (R˙ ×R −R˙ ×R )− (R −R )2, k = 4πJ in the skyrmion core. (b) skyrmion-antiskyrmion pair in 2 + + − − 2 + − n+-n− binding together due to the antiferromagnetic (6) coupling J which favors local pseudospin anti-alignment and eliminating the relative coordinate R+ − R− in favor of forming a charge 2e object. R++R− the center-of-mass coordinates: Rs = 2 yields L = 2 (eBeff ) ˙ 2 2k Rs from which we can read off the effective mass:

2 2 posite Chern-bands; if we had had J < 0, skyrmions would (eBeff ) π~ Mpair = = 2 (7) bind with skyrmions, leading to a charge-neutral object. Re- k JAM markably, despite the long-range Coulomb interaction, such a bound state will form no matter how small J is, in the ab- and the transition temperature [35], related to the effective sence of other anisotropies. Roughly speaking, an isolated mass and stiffness through: charge e skyrmion in a single Chern sector pays a “Zeeman” energy due to coupling to the uniform ferromagnetic order in 2 νπ~ JAM the opposite sector via the exhange term J. This energy cost kBTc = = ν (8) scales with the size R of the skyrmions as ∼ JR2. As in the 2AM Mpair 2 case of quantum Hall skyrmions [52], the competition with the Coulomb repulsion ∼ U/R will lead to a finite size for where ν is the skyrmion filling fraction. The effective mass such skyrmions and yields an extra energy penalty on top of sets the condensation scale of the composite objects. For the elastic contribution. On the other hand, a pair of antifer- MATBG, JAM ∼ 1 meV leading to the scale Tc ∼ 1 − 5 romagnetically locked charge 2e skyrmions does not pay any K. We will verify these estimates using a field theoretical cal- exchange energy which enables it to evade Coulomb repul- culation of the phase diagram with doping. sion by becoming very large. Hence, the extended nature of the skyrmion allows for a pairing mechanism which evades the Coulomb repulsion while benefiting locally from the anti- V. FIELD THEORY DESCRIPTION OF THE SKYRMION ferromagnetic coupling. SUPERCONDUCTOR AT FINITE CHEMICAL POTENTIAL The inclusion of the easy plane anisotropy λ due to imperfect sublattice polarization affects the above scenario as The antiferromagnetic coupling implies that the n+ and n− follows. First, it leads to the deformation of skyrmions to a pseudospins are antiferromagnetically locked in the ground topologically equivalent configuration of a meron-antimeron state. In addition, the lowest energy charge excitation are also pair to evade the penalty of out-of-plane pseudospin align- bound states: of skyrmions in n+ and of antiskyrmions in ment. Second, it reduces the binding energy of anti- n− where n+ = −n− = n, which can be understood as n- ferromagnetically locked skyrmion-antiskyrmion pair. This skyrmions which carry charge 2e. Thus, we can integrate out can be seen by noting that out-of-plane pseudospin only incurs the massive ferromagnetic fluctuations, n+ +n−, whose mass an energy penalty for such skyrmion pairs but not for individ- is proportional to J. The resulting field theory is then written ual skyrmions (since the energy only has the term n+,zn−,z solely in terms of the SO(3) vector n. Furthermore, we can 2 1 but not n+,z). As a result, this term reduces the binding en- rewrite this in the well known CP representation by writing ergy of the charge 2e skyrmion-antiskyrmion pairs and even- n = z†ηz, where we introduced the bosonic spinon field z = T † tually leads to their unbinding when it is sufficiently large. (z1, z2) which satisfies the constraint z z = 1. [9, 18, 25]. Evidence from a recent numerical study [12] suggests that the The overall phase of z is redundant which leads to the gauge 6

�� integrating out the field aµ by introducing a dual phase vari- φ ■ able which can be identified with the phase of the superfluid �� order parameter ■ ■ ■ Z �� ■ ■ 2 nρSC 2 χSC 2o �� S = d rdτ (∂iφ − 2eAi) + (∂τ φ − 2eAτ ) Δ 2 2 πρ π � ■ �� 3|∆|c 3|∆| / = ρ = , χ = (10) ■ ■ ■ w0 w1 0.27 SC πN SC πcN w0/w1=0.75 �� The phase stiffness ρSC yields the temperature scale for su- πρSC �� perconductivity, using the standard formula Tc = [35]. 2kB θ (°) Let us now turn to producing the superconductor by varying the chemical potential, rather than the coupling g. FIG. 5: Energetics of charged excitations: Ratio of the elastic energy of the 2e skyrmion to the particle-hole band For nonzero chemical potential µ, the phase diagram gap in the nearly chiral (orange) and realistic (purple) limits, can be obtained numerically by solving the mean-field self- with stiffness and gap values extracted from self-consistent consistency equation (see supplemental material) as a func- Hartree-Fock. Below the grey dashed line, the skyrmion tion of g and µ as shown in Fig. 6. Note, a finite value πcΛ energy lies within the particle-hole gap. Note that w0 and w1 |µc| = g (1 − g/gc) is needed to introduce skyrmions in are the interlayer hoppings in AA and AB-stacked regions the ordered phase g < gc. This value reduces to the elastic (see Ref. [24] for details) skyrmion energy 4πρps for g gc which serves as a check on the CPN theory. For larger µ, the skyrmion density ν is obtained numerically. For any finite skyrmion density ν, the † effective field theory takes the form (10) describing a super- field aµ = −iz ∂µz. The topological density is given by the flux of a , and is tied to the charge density. Thus: δρ = conductor whose phase stiffness can be computed numerically µ as shown in Fig. 6. In the dilute limit ν 1 and for g g , − ∂xay −∂y ax = − 2 f [9, 18, 25]. Note, since the flux of c π 2π xy ∆ can be expressed in terms of the doping ν leading to a is tied to a conserved charge, monopole fluctuations which change the flux in units of 2π are disallowed on symmetry 3Λcνg 3JAM ν 3Λgν 1 3ν 1 ρ ≈ = , χ ≈ = grounds, unlike in the usual CP model. Thus the dynamics SC 4πN 2πN SC 4πNc N 16πρ A 1 ps M of this model resembles that of the non-compact CP theory (11) [33], which explicitly disallows monopoles. Interestingly, the superfluid stiffness (charge compressibil- The CP1 action takes the form ity) of the superconductor is inversely proportional to the pseudospin compressibility (psuedoespin stiffness) of the zero Z 3 Λ 2 2eµ doping insulator with the proportionality constant being, up to S[z] = d r |Dµz| + fxy g 2πc numerical prefactors, just the filling ν. This leads to Tc ∼ Jν Z 0 in agreement with Eq. 8. The phase diagram of the chemi- 1 fxy(r) fxy(r ) + d2r0 V (r − r0) (9) 1 2c π π cal potential tuned CP theory with realistic parameters and screened Coulomb interaction should be also be accessible in future studies using the Monte Carlo technique[37]. where Dµ = (∂µ − iaµ). Note, we rescaled the (imaginary) 1 √ time as τ = rz, introduced the cutoff scale Λ = 1/ AM , To fix the pairing symmetry of the superconductor con- c q JAM 2 p sider the symmetries of the operator ∆ that creates flux: the coupling g = , and velocity c = 2JAM ρps. 2ρps Λ ∇ × a = 2π, which corresponds to the Cooper pair creation 1 In the following, we compute the phase diagram of the CP operator. Since flux is left invariant by pseudospin rotation, model at finite doping and interpret the results for MATBG. ∆ is expected to be a pseudospin singlet. Hence the Cooper We use the large-N approximation by extending to N com- T pair operator must be one of: ∆µ(k) = gµ(k)c−kηyγµck plex fields (z1, z2, . . . zN ). We start by reviewing the so- which gives four possible pairings. Reverting to the picture lution in the absence of doping and Coulomb interaction of skyrmion-anti-skyrmion pairing, we note that pairs reside (λ = µ = V = 0) which was discussed in the pioneer- in opposite Chern sectors, which eliminates two of the four ing works of Refs. [5, 9, 25]. In this limit, the model has pairing channels leaving us with ∆x,y ∝ ηyγx,y. Of these, two phases: (i) an ordered phase for g < gc = 4π charac- the first is antisymmetric in internal indices leading to even terized by a finite expectation value of z with aµ Higgsed pairing gx(−k) = gx(k) whereas the second is symmetric and (ii) a disordered phase for g > gc = 4π where the z in internal indices leading to odd pairing gy(−k) = −gy(k). variables are gapped and aµ is free. The gap is given by Further evaluating the rotation quantum number for these pair- |∆| = Λ(1 − gc/g) and the Lagrangian for the gauge field aµ ings [13, 21, 22] (see supplemental material for details), one 1 2 has the standard Maxwell form ∼ |∆| (∂µaν − ∂ν aµ) with sees that the natural zero angular momentum skyrmion corre- an additional Chern-Simons coupling to the background U(1) sponds to ∆x ∝ ηyγx = τy , while the other ∆y ∝ ηyγy = ie electromagnetic field π µνλaµ∂µAλ. The disordered phase τxσz option corresponds to a nonzero (odd) angular momen- actually describes a paired superfluid which can be seen by tum. 7

FIG. 6: Large-N phase diagram for the doped CP1 model: (a) plot of the charge 2e skyrmion filling fraction ν which is directly related to the flux fxy of the gauge field aµ and (b) the gap for the spinon field ∆ which is directly proportional to the superfluid stiffness (cf. Eq. 10) in the superconducting phase plotted as a function of the chemical potential µ and the dimensionless coupling g. The energies µ and ∆ are measured in units of the energy cutoff Λc. The black line represents the N−1 phase boundary between the insulator and the superconductor. In (c), we plot an estimate for superconducting Tc in the CP 3πc|∆| model, computed using the relation: while setting N = 2 and ρps = 1 meV as a function of doping ν for different 2πNkB values of J.

Separately, it is worth noting that the pairing channels ∆x,y are also the ones that correspond to the maximal gap in the presence of the insulating pseudospin antiferromagnetic background. This is readily seen by checking that the corresponding matrices anticommute in the Nambu basis, (see Supple- mentary materials) thus the insulating and superconducting gaps add in quadrature when they are simultaneously present. Furthermore, taking into account the kinetic part of the Hamil- tonian h, in Eq. 4, it is found to anticommute with ∆x, leading to a bigger gap compared to ∆y which commutes with h.

VI. SCHEMATIC PHASE DIAGRAM AND DUALITY BETWEEN THE INSULATOR AND SUPERCONDUCTOR FIG. 7: Schematic phase diagram in the limit of short The large N mean field analysis identified two phases: a screening length. We will be mainly concerned with K-IVC insulator and a superconductor. There is however a chemical potential driven transitions at small g < gc. The third possibility: a phase where both orders coexist. Such dotted blue (red) curve at nonzero T indicates the BKT phase is possible when superconductivity is driven by tuning transition out of the superconductor (KIVC) with quasi-long the chemical potential since the condensation of skyrmions in range order. On increasing g at T = 0 and µ = 0, a direct this case does not disorder the phase of the insulator XY order transition between the KIVC and SC states (deconfined parameter. Thus, for finite N, we expect a phase diagram criticality) is potentially allowed, which then splits into two which schematically has the form of Fig. 7 where the three ordinary quantum critical points (yellow circles) with a different phases can be identified using the expectation values region of co-existence when µ 6= 0. ∗ hzii and hz1 z2i as follows [21, 33, 49]: 8

1. Pure insulator: In the absence of doping, at small g Vortex Gauge Field Charge (nc, nv) Order Parameter ∗ below some critical coupling gc, we have the Higgs phase K-IVC: w1, 2 a˜ (±1, 0) w1w2 = ∆SC hz i= 6 0 ∗ i which, with the expected easy plane anisotropy SC: z1, 2 a (0, ±1) z1 z2 = ∆KIVC (λ > 0), favors equal amplitudes for z1,2, takes the form TABLE II: Summary of properties of K-IVC vortex iθi hzii = r0e . (12) fields(merons) in the first row, charge quantum numbers and relation to superconductor order parameter. Second row, the This is nothing but the K-IVC phase, with K-IVC order pa- corresponding properties for superconducting vortices ∗ 2 i(θ1−θ2) rameter hz1 z2i = r0e . The gapless photon of a is ab- obtained from duality. The superconductor vortex fields is sent due to the Higgs mechanism. This indicates that density equivalent to the CP1 representation of K-IVC order. and current fluctuations are suppressed, implying an electrical insulator. This can be seen by explicitly integrating out a in 2 the Higgs phase which gives: L = (2e) ( ∂ A )2 with 2κ µνλ µ λ interchanged. κ = 16π2r2Λ/g 0 , as expected for a insulating dielectric. On In fact, it is helpful to organize the order parameters into a θ the other hand, fluctuations of in the K-IVC phase lead to a five component object n : n + in = ∆ and n + in = gapless Goldstone mode. i 1 2 IVC 4 5 ∆SC while n3 represents the valley-Hall order. Under the du- 2. Coexistence phase: Next, raising the chemical poten- ality it is readily verified from the fermion transformation that tial µ corresponds to applying a magnetic field to the Higgs n3 ↔ n3, so the duality acts as a rotation in SO(5). In fact condensate. Beyond a critical value of µ, vortices of the combining the SO(3) pseudospin rotations of (n1, n2, n3) Higgs condensate form to accommodate flux. The cheapest and the U(1)c rotation of (n4, n5) with this duality, one ob- such configuration is a simultaneous vortex in both z1 and tains the full SO(5) group. z2, which has a finite energy cost and carries charge 2e (this This duality immediately fixes properties of the supercon- is nothing but the n skyrmion). Condensing these vortices ductor, based on what we know about the K-IVC insulator. kills the Higgs condensate in hzii = 0, which restores a to For example, vortices in the order parameters are interchanged its Coulomb phase. This is the superfluid and the photon of by duality. Here, vortices of K-IVC have ‘cores’ that contain a is simply the superfluid Goldstone mode. Nevertheless the the valley-Hall order n3, giving rise to merons, or two com- K-IVC order parameter is not destroyed since the phase of ponent vortices with opposite direction of n3 order in their both z1, z2 wind around such a vortex, their relative phase ∗ ∗ cores. These carry electric charge ±e for opposite n3 con- z1 z2 remains well defined, so hz1 z2i= 6 0 . This is the coex- figurations of the same vorticity. Under the duality, these are istence phase. The phase transition from the insulator to the mapped to superconducting vortices, which also have a two coexistence phase with doping is in the superfluid-insulator component structure, with n3 core that now determines their transition driven by the chemical potential, with dynamical valley-charge. Denoting the K-IVC and SC vortex creation exponent z = 2 [43]. fields by (w1, w2) and (z1, z2), their properties are summa- 3. Superconductor: Finally, on further raising µ, the ampli- rized in Table II. tude of the KIVC order is reduced, potentially driving a transition when the K-IVC order disappears. Beyond this we simply have the superfluid order since a remains in the Coulomb phase. This transition is expected to be in the 3D XY univer- VII. SO(5) SIGMA MODEL AND WZW TERM sality class. The relationship between the K-IVC insulator and the su- In the previous sections, we have mainly focused on the perconductor can be highlighted by the following duality. We transition into the superconductor starting deep inside the K- start by writing the K-IVC and the superconducting order pa- IVC state by tuning the chemical potential, treated within the † y † † CP1 theory. However, following Refs. [18, 23, 48], it is in- rameters as hcKσcK0σ0 i = ∆KIVCσσσ0 and hcτσcτ 0σ0 i = y iϑ structive to derive an equivalent effective field theory which ∆SCδσ,σ0 τττ 0 , respectively, where ∆IVC = |∆IVC|e and iϕ deals with the insulator and superconductor on equal foot- ∆SC = |∆SC|e are complex numbers. The two are related by the following particle-hole transformation that acts in only ing. Such a field theoretic description is known to include a one of the valleys: topological WZW term. In the following, we will outline this derivation in the context of magic angle graphene including y † 0 the effect of the chemical potential. cK,σ 7→ cK,σ, cK ,σ 7→ −iσσ,σ0 cK0,σ0 (13) To derive universal aspects of the field theory, such as the nˆc ↔ nˆv, ϕ ↔ ϑ (14) presence of a topological term, it is sufficient to adopt a con- This duality transformation exchanges the total U(1)c charge venient starting point where we take a dispersion with Dirac conservation (of total charge nˆc) and the U(1)v valley-charge points for the nearly flat bands with a spontaneously induced conservation (of valley charge nˆv) which is consistent with single particle gap (Dirac mass) that is much smaller than the the fact that the KIVC (SC) order breaks (preserves) U(1)v dispersion. The dispersion can then be linearized close to the 0 but preserves (breaks) U(1)c. Furthermore, this transforma- moire´ Dirac points KM and KM where we expect a gap to tion interchanges the phases ϑ ↔ φ of the K-IVC and Su- be induced via a Dirac mass term. To deal with the insulating perconductor, conjugate to the charges nv and nc which are and superconducting states on equal footing, we introduce the 9

Nambu basis defined as nitude. The latter coupling affects the effective potential, favoring superconductivity at finite doping. We note that both T T † χk = (ψK ,k, ψ 0 ) (15) couplings were considered before in the context of SO(5) the- M KM ,−k ories for cuprates Ref. [15, 64] although the topological term Let us now introduce the Pauli matrices ρx,y,z which act which we turn to next was absent in that context. within the two-dimensional Nambu space in χk. The Hamil- SWZW is the well-known Wess-Zumino-Witten term which tonian now takes the form can be written by introducing an auxillary integration variable as HD = kxγxρz + kyγy + M, (16) Z 1 Z 3i 2 abcde The mass M is a matrix in γ, η, and ρ spaces containing SWZW = du d rdτ na∂unb∂τ nc∂xnd∂yne 4π 0 both the K-IVC and the superconducting order parameters, as (19) well as the valley Hall order which is a part of the antiferro- At µ = 0, this theory has the same form as the theory magnetic manifold. All these correspond to anticommuting describing the transition between an antiferromagnet and a mass terms for the Dirac equation and hence can be written valence-bond-solid [21, 45, 48, 49]. In this case, the vector P5 as: M = i=1 niΓi where the SO(5) order parameter nˆ is nˆ is always either fully in the superconducting (12) phase or defined as (n, Re ∆SC, Im ∆SC). The corresponding orders the insulating (345) phase, i.e. there is no coexistence regime. and the matrices Γi are shown in the Table III. At g = gc, λ = 0, the model has been conjectured to possess an emergent SO(5) symmetry [34, 59], reduced to an SO(4) Order Re ∆IVC Im ∆IVC ∆VH Re ∆SC Im ∆SC symmetry in the easy plane limit. The existence of a WZW nˆ n1 n2 n3 n4 n5 term implies that one cannot simultaneously disorder both the Γ γzηxρz γzηy γzηzρz ηyγxρy ηyγxρx superconductor and the K-IVC/Valley Hall order without inducing either a gapless critical point (deconfined criticality as TABLE III: SO(5) Order parameters and Dirac mass terms shown in Figure 7) or topological order.

We note that the massive Dirac Hamiltonian (16) is invariant under the particle-hole symmetry P = γzρyK (i.e. VIII. ROLE OF SPIN QUANTUM NUMBER {HD, P} = 0) where K is complex conjugation. As a result, it belongs to symmetry class C of the Altland-Zirnbauer clas- The spinless model discussed in this paper is directly rel- sification [3]. The mass term M parametrizes the symplectic Sp(4n) evant to the vicinity of the half-filling insulator (at fillings Grassmanian manifold Sp(2n)×Sp(2n) (our convention here is ν = ±(2 + )), where charge neutrality corresponds to ν = 0. that Sp(2n) constitutes 2n × 2n symplectic matrices). For Achieving an insulator at half-filling requires polarizing a fla- Sp(4) and our case, n = 1 and Sp(2)×Sp(2) is isomorphic to the 4-sphere vor (eg. spin) opening a gap at the Dirac point, presum- parametrized by the 5-dimensional unit vector nˆ. The topol- ably through developing a pseudospin quantum Hall ferro- Sp(4n) magnet (with n = −n ). Spin polarization means that for ogy of the symplectic Grassmanian π = Z + − 4 Sp(2n)×Sp(2n) each valley, K and K0, only one spin species is filled which is what allows for the existence of a WZW term in the action may correspond to a simple spin ferromagnet or a spin-valley [21, 42, 46]. locked state with spin anti-aligned in opposite valleys. The Following the standard procedure by integrating out the main assumption leading to the reduction to a spinless prob- fermions and performing the gradient expansion (see supple- lem is that spin polarization takes place at a higher energy mental material for details), we derive the following effective scale than the scale relevant for superconductivity. This is theory strongly suggested by the cascade picture of Refs. [62, 66] Z which identified a relatively high temperature scale where fla- 2 S = SWZW + dτd rL[ˆn] (17) vor polarization takes place, before the development of insulating or superconducting behavior at lower temperatures. As where L[ˆn] is given by a result, one spin flavor is completely frozen at the scales relevant for superconductivity allowing us to focus on the spinless problem. We note that most elements of our discussion remain ρ˜ 2 χ˜ 2 2 2 2 L[ˆn] = (∇nˆ) + (∂τ nˆ) +(g −gc +2˜χµ )(n −|∆| ) valid even without the assumption of a frozen spin flavor – 2 2 pseudospin skyrmions still carry charge and pair due to an- 2 † µ + λn + 2˜χµ∆ ∂τ ∆ − n · (∂xn × ∂yn) (18) tiferromagnetic coupling – but the ground state manifold has 3 2π a more complicated structure and allows for other competing We see that the chemical potential enters the action in three charged excitations. different places. First it couples to the topological density of It is worth noting that while the spinless model is particle- the vector n representing the insulator. Note that since n now hole symmetric, we do not expect particle-hole symmetry in does not have a fixed length, this term is not quantized. Sec- the real system in the vicinity of half-filling. To see this, ond, the chemical potential couples linearly to the first deriva- note that there are two scenarios for such an insulator shown tive of the superconducting gap and quadratically to its mag- schematically in Fig. 8 where the opposite spin band lies out- 10 side or inside the Dirac gap. In the first scenario (a), elec- any extrinsic attractive pairing mechanism e.g. phonons. In tron (hole) doping at ν = 2(−2) resembles doping a spinless addition, our mechanism explains how the large Coulomb re- version of the charge neutrality state whereas hole (electron) pulsion can be evaded without screening or retardation al- doping resembles doping a spinless version of the full (empty) lowing for the smaller pseudospin antiferromagnetic coupling band structure whereas in the second scenario (b), there is J ∼ h2/U ∼ 1 meV to overcome the larger Coulomb repul- no difference between particle and hole doping at half-filling sion as summarized in Figure 4. and both reduce to a spinless version of the doped state near The role of the coupling J as the source of pairing can ac- charge neutrality. The first scenario is strongly supported by count for the absence of superconductivity in MATBG sam- the quantum oscillation measurements where the Landau fans ples which are aligned with the hexagonal Boron Nitride are only observed at ν = ±(2 + ) but not at ν = ±(2 − ) (hBN) substrate. The latter generates a sublattice potential with a degeneracy that is half that observed at charge neu- which breaks C2T symmetry by inducing an energy gap of trality [9, 30, 63]. Further support of this picture is provided about 15 − 20 meV between the bands polarized on sublat- by the cascade transition picture based on recent compress- tices A and B [6, 50, 65]. This acts as an opposite Zeeman ibility [66] as well as STM [62] data, which suggests that the field in the opposite Chern sectors (a Zeeman field for the vec- ν = ±(2+) state is a flavor polarized version of the one near tor n) which (a) shrinks the charge 2e skyrmions and makes charge neutrality. them energetically less favorable and (b) introduces an energy The pairing symmetry favored by skyrmion condensation detuning between previously resonant bands suppresses the (in the spinless model) is an intervalley singlet, with same tunneling term responsible for the coupling J. This mecha- ˆ † y † sublattice pairing i.e. ∆ = ψστ ττ 0τ ψστ 0 . In order to em- nism is also absent in other moire´ systems lacking C2 sym- bed the spinless model within spinful TBG at ν = 2, we need metry such as twisted double bilayer graphene [10, 28, 51] to polarize a flavor as explained above. If the polarized flavor and ABC graphene on hBN [13] which can also be viewed as is just spin, then we are left with pairing between equal spins. consisting of two quantum Hall systems with opposite Chern On the other hand, if the polarized flavor is more complex, for numbers. In addition to the suppression of J due to bro- example a spin valley locked combination, the correspond- ken C2 symmetry, the direct tunneling h is also strongly sup- ing spin structure of the Cooper pair is also more involved. pressed in these systems since the two opposite Chern sec- A general discussion of the spinful model is contained in the tors reside in opposite valleys and are thus very weakly cou- supplementary material. pled due to valley-charge conservation. Apart from pristine MATBG samples, the other platform that retains the relevant C2T symmetry is twisted trilayer (and multilayer) graphene IX. DISCUSSION with alternating twist as described in Ref. [20]. In addition to having the same symmetries as MATBG, it was shown [20] Let us compare and contrast the present work with previous that the wavefunctions of these systems can be mapped ex- theoretical discussions of skyrmion superconductivity. Start- actly to those of MATBG making them very attractive can- ing with Dirac fermions, [1, 18] pointed out that charge 2e didates for psuedospin antiferromagnetic order and skyrmion skyrmions can arise when interactions spontaneously create mediated superconductivity, and are promising platforms for topological insulators from symmetry breaking. Further, if future theoretical and experimental studies. these skyrmion textures are the lowest energy charge exci- We have noted the key role played by C2 symmetry, which tations, refs. [18, 38, 60] argued that doping could lead to is important to defining a degenerate pusedospin. A natural superconductivity. Here, we have examined a microscopic question that arises is whether disorder, which is expected to model with predominantly repulsive interactions and shown to locally break C2, would play an antagonistic role. However, that it exhibits a related topological ground state with low en- it is important to note that in contrast to the global breaking ergy charge 2e skyrmions. Note, restricting to repulsion is of C2, the symmetry is not broken on average by disorder. necessary to establishing an electronic mechanism of super- For our mechanism, we only need C2 to hold on distances conductivity. To the best of our knowledge, this collection comparable to the moire´ scale. Disorder that preserves C2 of properties has not been established before [67] in a micro- on average and is sufficiently smooth on the moire´ scale is scopic model with repulsive interactions. A distinction from therefore not expected to significantly affect our conclusions. the previous Dirac models is that the 2e skyrmion here is is ac- It is worth emphasizing that the superconducting mecha- tually a composite of a skyrmion anti-skyrmion pair. Further- nism we discussed here is applicable to a much wider class more, our model, is motivated by magic angle graphene which of systems than just magic angle graphene. The key require- seems to naturally incorporate the requisite physics. We also ments are a pair of quantum Hall (or Chern) ferromagnets, re- have calculated the energetics of skyrmions and shown that lated by time reversal symmetry, and coupled to one another in a range of parameters they are the lowest energy charge by tunneling. This leads, via a superexchange process, to an excitations. On doping skyrmions we obtain a superconduc- antiferromagnetic coupling J which is maximized when tun- tor, whose transition temperature we calculate. In our theory neling connects states at the same energy. an essential role is played by J, which induces pairing be- Let us mention a few other promising systems that incor- tween opposite skyrmion textures, and arises from electron porate all these ingredients. First, we have already briefly tunneling between Chern sectors. In particular, we provide mentioned the multilayer graphene platform as described in a scenario where pairing can be achieved in the absence of Ref. [20] where each successive layer is twisted by an al- 11 ternating angle (i.e. θ, −θ, θ . . . ), and where alternating layers, which are at the same angle, are in registry. These systems have shifted magic angles but otherwise retain all essential symmetries and are promising platforms for realizing skyrmion superconductivity. For the simplest alternating angle trilayer case, an additional ingredient is a coexisting Dirac node which intersects the flat bands, but which is not expected to significantly alter the physics due to its rapid dispersion [20]. A second route involves strain induced Landau levels in graphene which have been observed [24], with opposite Chern number in opposite valleys. Here spin plays the role of our pseudospin, although translation symmetry blocks tun- FIG. 8: MATBG close to half-filling ν = 2: Two possible neling between opposite valleys. Activating tunneling with an scenarios for the spin-polarized insulating state at half-filling: appropriate translation symmetry breaking order (eg. a sub- (a) the opposite spin band lies inside the gap of the strate with charge density wave order), can help realize the psuedospin antiferromagnet (taken to be the K-IVC order skyrmion superconductor on doping. Other potential realiza- here) or (b) the opposite spin band lies outside this gap. In tions include any spinful valley-Chern insulator, e.g. twisted both scenarios, doped electrons go into the upper K-IVC double bilayer graphene, where again the crucial additional band. On the other hand, doped holes go into the opposite ingredient absent in current platforms is the presence of a spin band for (a) but into the same spin K-IVC band for (b). source of translation symmetry breaking that opens up inter- We note here that for simplicity, we considered the case valley tunneling. More generally, 2D materials with electronic where flavor polarization corresponds to full spin bands harboring the same fragile topology as magic angle polarization. Another possible flavor polarized state would graphene[2, 39, 40, 54], will be interesting future candidates be the spin-valley locked state for which the same if they can be also brought into the correlated regime. Alter- considerations apply. nately, consider a pair of Haldane models with opposite Chern numbers, with spin degeneracy. Here spins play the role of our pseudospin. We note that such a model, dubbed the ”shift in- Coulomb scale, since the flat band dispersion is mostly gener- sulator” has been recently studied in the context of crystalline ated by the interaction as discussed in [7]. topology [13]. Further, adding repulsive interactions and tun- The discovery of superconductivity in MATBG came as a neling between the opposite Chern sectors should realize the bolt from the blue, and it would be surprising if an explana- physics studied here when placed near quarter filling. Pursu- tion only invoked conventional ingredients. Instead, we being materials realization of these models, such as bilayers of lieve this observation calls for an unusual mechanism like the anomalous quantum Hall ferromagnets, will be an interesting skyrmion pairing one described here, which relies on charac- direction for future work. teristics unique to MATBG. More generally, we have identi- Finally, let us briefly discuss our proposal in light of recent fied a new mechanism for superconductivity from repulsion, experiments on MATBG superconductivity. First, although and it will be interesting to explore in the future a variety of we have chosen to discuss our scenario in terms of doping a other settings that realize the essential ingredients of Chern parent psuedospin-ordered insulator, the presence of the insu- ferromagnets that are tunnel-coupled to their time reversed lating gap is not essential. While an insulating state serves conjugates. as proxy for order, it is by no means necessary, since an or- Note— We would like to draw the reader’s attention to dered metallic phase can also obtain at integer filling, due to Refs. [12, 68] on related topics, that appeared after this wok low-energy electron-hole pockets. The essential features of was first released. the skyrmion pairing mechanism however remain intact, although a theoretical description must now include fermions in addition to the bosonic n degrees of freedom. Hence, our re- X. ACKNOWLEDGMENTS sults are consistent with recent experimental reports of superconductivity in the absence of correlated insulating behavior We thank Shang Liu for a related previous collaboration [44, 55] - an interesting question there is whether psuedospin and S. Sachdev for helpful feedback. SC thanks V. Lohani for order is present in the metallic state near integer filling. We helpful discussions. AV was supported by a Simons Investiga- note that in the calculation of Sec. 4 we find a superconduct- tor award and by the Simons Collaboration on Ultra-Quantum ing phase with doping in the absence of long range psuedospin Matter, which is a grant from the Simons Foundation (651440, order. Thus strictly, the only real requirement is short-ranged AV). EK was supported by a Simons Investigator Fellow- psuedospin order in which skyrmions can be defined. Second, ship, by NSF-DMR 1411343, and by the German National an important energy scale in our theory is J ∼ t2/U which Academy of Sciences Leopoldina through grant LPDS 2018- controls the strength of superconductivity. This may provide a 02 Leopoldina fellowship. SC acknowledges support from the conceptual framework to help design better superconductors. ERC synergy grant UQUAM via E. Altman. MZ was sup- For example, it is helpful to remember that in the interacting ported by the Director, Office of Science, Office of Basic En- problem, not just U but also t increase with increasing the ergy Sciences, Materials Sciences and Engineering Division 12 of the U.S. Department of Energy under contract no. DE- AC02-05-CH11231 (van der Waals heterostructures program, KCWF16)

[1] A. G. Abanov and P. B. Wiegmann. Chiral nonlinear σ models [16] G. Frohlich.¨ On the theory of superconductivity: the one- as models for topological superconductivity. Phys. Rev. Lett., dimensional case. Proc. R. Soc., A223:296, 1954. 86:1319–1322, Feb 2001. [17] M. O. Goerbig. Electronic properties of graphene in a strong [2] Junyeong Ahn, Sungjoon Park, and Bohm-Jung Yang. Fail- magnetic field. Rev. Mod. Phys., 83:1193–1243, Nov 2011. ure of nielsen-ninomiya theorem and fragile topology in two- [18] Tarun Grover and T. Senthil. Topological spin hall states, dimensional systems with space-time inversion symmetry: ap- charged skyrmions, and superconductivity in two dimensions. plication to twisted bilayer graphene at magic angle. arXiv Phys. Rev. Lett., 100:156804, Apr 2008. preprint arXiv:1808.05375, 2018. [19] Chang-Yu Hou, Claudio Chamon, and Christopher Mudry. [3] Alexander Altland and Martin R. Zirnbauer. Nonstandard sym- Electron fractionalization in two-dimensional graphenelike metry classes in mesoscopic normal-superconducting hybrid structures. Phys. Rev. Lett., 98:186809, May 2007. structures. Phys. Rev. B, 55:1142–1161, Jan 1997. [20] Eslam Khalaf, Alex J. Kruchkov, Grigory Tarnopolsky, and [4] I Ya Aref’eva and SI Azakov. Renormalization and phase tran- Ashvin Vishwanath. Magic Angle Hierarchy in Twisted sition in the quantum ofcpn- 1 model (d= 2, 3). Nuclear Physics Graphene Multilayers. arXiv e-prints, page arXiv:1901.10485, B, 162(2):298–310, 1980. Jan 2019. [5] Rafi Bistritzer and Allan H. MacDonald. Moire´ bands in twisted [21] Alexei Kitaev. Periodic table for topological insulators and su- double-layer graphene. Proceedings of the National Academy perconductors. In AIP conference proceedings, volume 1134, of Sciences, 108(30):12233–12237, 2011. pages 22–30. American Institute of Physics, 2009. [6] Nick Bultinck, Shubhayu Chatterjee, and Michael P. Zaletel. [22] Dung-Hai Lee and Charles L. Kane. Boson-vortex-skyrmion Mechanism for anomalous hall ferromagnetism in twisted bi- duality, spin-singlet fractional quantum hall effect, and spin-1/2 layer graphene. Phys. Rev. Lett., 124:166601, Apr 2020. anyon superconductivity. Phys. Rev. Lett., 64:1313–1317, Mar [7] Nick Bultinck, Eslam Khalaf, Shang Liu, Shubhayu Chatterjee, 1990. Ashvin Vishwanath, and Michael P. Zaletel. Ground State and [23] Junhyun Lee and Subir Sachdev. Wess-zumino-witten terms in Hidden Symmetry of Magic Angle Graphene at Even Integer graphene landau levels. Phys. Rev. Lett., 114:226801, Jun 2015. Filling. arXiv e-prints, page arXiv:1911.02045, Nov 2019. [24] N. Levy, S. A. Burke, K. L. Meaker, M. Panlasigui, A. Zettl, [8] Yuan Cao, Valla Fatemi, Ahmet Demir, Shiang Fang, Spencer L F. Guinea, A. H. Castro Neto, and M. F. Crommie. Strain- Tomarken, Jason Y Luo, Javier D Sanchez-Yamagishi, Kenji induced pseudo–magnetic fields greater than 300 tesla in Watanabe, Takashi Taniguchi, Efthimios Kaxiras, et al. Corre- graphene nanobubbles. Science, 329(5991):544–547, 2010. lated insulator behaviour at half-filling in magic-angle graphene [25] Guohong Li, A. Luican, J. M. B. Lopes dos Santos, A. H. Cas- superlattices. Nature, 556(7699):80, 2018. tro Neto, A. Reina, J. Kong, and E. Y. Andrei. Observation [9] Yuan Cao, Valla Fatemi, Shiang Fang, Kenji Watanabe, Takashi of van hove singularities in twisted graphene layers. Nature Taniguchi, Efthimios Kaxiras, and Pablo Jarillo-Herrero. Un- Physics, 6(2):109–113, 2010. conventional superconductivity in magic-angle graphene super- [26] Shang Liu, Eslam Khalaf, Jong Yeon Lee, and Ashvin Vish- lattices. Nature, 556(7699):43, 2018. wanath. Nematic topological semimetal and insulator in magic [10] Yuan Cao, Daniel Rodan-Legrain, Oriol Rubies-Bigorda,` angle bilayer graphene at charge neutrality. arXiv preprint Jeong Min Park, Kenji Watanabe, Takashi Taniguchi, and arXiv:1905.07409, 2019. Pablo Jarillo-Herrero. Electric Field Tunable Correlated States [27] Shang Liu, Ashvin Vishwanath, and Eslam Khalaf. Shift Insu- and Magnetic Phase Transitions in Twisted Bilayer-Bilayer lators: Rotation-Protected Two-Dimensional Topological Crys- Graphene. arXiv e-prints, page arXiv:1903.08596, Mar 2019. talline Insulators. Physical Review X, 2019. [11] Shubhayu Chatterjee, Nick Bultinck, and Michael P. Zaletel. [28] Xiaomeng Liu, Zeyu Hao, Eslam Khalaf, Jong Yeon Lee, Kenji Symmetry breaking and skyrmionic transport in twisted bilayer Watanabe, Takashi Taniguchi, Ashvin Vishwanath, and Philip graphene. Phys. Rev. B, 101:165141, Apr 2020. Kim. Spin-polarized correlated insulator and superconductorin [12] Shubhayu Chatterjee, Matteo Ippoliti, and Michael P. Zale- twisted double bilayer graphene. page arXiv:1903.08130, Mar tel. Skyrmion Superconductivity: DMRG evidence for a 2019. topological route to superconductivity. arXiv e-prints, page [29] J. M. B. Lopes dos Santos, N. M. R. Peres, and A. H. Cas- arXiv:2010.01144, October 2020. tro Neto. Continuum model of the twisted graphene bilayer. [13] Guorui Chen, Aaron L Sharpe, Patrick Gallagher, Ilan T Rosen, Phys. Rev. B, 86:155449, Oct 2012. Eli Fox, Lili Jiang, Bosai Lyu, Hongyuan Li, Kenji Watanabe, [30] Xiaobo Lu, Petr Stepanov, Wei Yang, Ming Xie, Mo- Takashi Taniguchi, et al. Signatures of gate-tunable supercon- hammed Ali Aamir, Ipsita Das, Carles Urgell, Kenji Watanabe, ductivity in trilayer graphene/boron nitride moir\’e superlat- Takashi Taniguchi, Guangyu Zhang, Adrian Bachtold, Allan H. tice. arXiv preprint arXiv:1901.04621, 2019. MacDonald, and Dmitri K. Efetov. Superconductors, orbital [14] A d’Adda, M Luscher,¨ and P Di Vecchia. A 1n expandable magnets, and correlated states in magic angle bilayer graphene. series of non-linear σ models with instantons. Nuclear Physics Nature, 574:653—-657, 2019. B, 146(1):63–76, 1978. [31] M Manfra, B Goldberg, L Pfeiffer, and K West. Skyrmions and [15] Eugene Demler, Werner Hanke, and Shou-Cheng Zhang. the ν= 1 quantum hall ferromagnet. Acta Physica Polonica A, SO(5) theory of antiferromagnetism and superconductivity. 4(92):621–632, 1997. Rev. Mod. Phys., 76:909–974, Nov 2004. [32] K. Moon, H. Mori, Kun Yang, S. M. Girvin, A. H. MacDon- ald, L. Zheng, D. Yoshioka, and Shou-Cheng Zhang. Sponta- 13

neous interlayer coherence in double-layer quantum hall sys- [51] Cheng Shen, Na Li, Shuopei Wang, Yanchong Zhao, Jian tems: Charged vortices and kosterlitz-thouless phase transi- Tang, Jieying Liu, Jinpeng Tian, Yanbang Chu, Kenji Watan- tions. Phys. Rev. B, 51:5138–5170, Feb 1995. abe, Takashi Taniguchi, Rong Yang, Zi Yang Meng, Dongxia [33] Olexei I. Motrunich and Ashvin Vishwanath. Emergent pho- Shi, and Guangyu Zhang. Observation of superconductivity tons and transitions in the O(3) sigma model with hedgehog with Tc onset at 12K in electrically tunable twisted double bi- suppression. Phys. Rev. B, 70:075104, Aug 2004. layer graphene. arXiv e-prints, page arXiv:1903.06952, Mar [34] Adam Nahum, P. Serna, J. T. Chalker, M. Ortuno,˜ and A. M. 2019. Somoza. Emergent so(5) symmetry at the neel´ to valence-bond- [52] S. L. Sondhi, A. Karlhede, S. A. Kivelson, and E. H. Rezayi. solid transition. Phys. Rev. Lett., 115:267203, Dec 2015. Skyrmions and the crossover from the integer to fractional [35] David R. Nelson and J. M. Kosterlitz. Universal jump in the quantum hall effect at small zeeman energies. Phys. Rev. B, superfluid density of two-dimensional superfluids. Phys. Rev. 47:16419–16426, Jun 1993. Lett., 39:1201–1205, Nov 1977. [53] Xue-Yang Song, Yin-Chen He, Ashvin Vishwanath, and Chong [36] Kentaro Nomura, Shinsei Ryu, and Dung-Hai Lee. Field- Wang. From spinon band topology to the symmetry quantum induced kosterlitz-thouless transition in the n = 0 landau level numbers of monopoles in Dirac spin liquids. Technical report. of graphene. Phys. Rev. Lett., 103:216801, Nov 2009. [54] Zhida Song, Zhijun Wang, Wujun Shi, Gang Li, Chen Fang, [37] Peder Notto Galteland, Egor Babaev, and Asle Sudbø. Fluctu- and B. Andrei Bernevig. All magic angles in twisted bilayer ation effects in rotating Bose-Einstein condensates with broken graphene are topological. Phys. Rev. Lett., 123:036401, Jul SU(2) and U(1) × U(1) symmetries in the presence of inter- 2019. component density-density interactions. PHYSICAL REVIEW [55] Petr Stepanov, Ipsita Das, Xiaobo Lu, Ali Fahimniya, Kenji A, 91:13605, 2015. Watanabe, Takashi Taniguchi, Frank HL Koppens, Johannes [38] Giandomenico Palumbo and Mauro Cirio. Skyrmion Super- Lischner, Leonid Levitov, and Dmitri K Efetov. The interplay of fluidity in Two-Dimensional Interacting Fermionic Systems insulating and superconducting orders in magic-angle graphene OPEN. 2015. bilayers. arXiv preprint arXiv:1911.09198, 2019. [39] Hoi Chun Po, Liujun Zou, T. Senthil, and Ashvin Vishwanath. [56] W. P. Su, J. R. Schrieffer, and A. J. Heeger. Soliton excitations Faithful tight-binding models and fragile topology of magic- in polyacetylene. Phys. Rev. B, 22:2099–2111, Aug 1980. angle bilayer graphene. Physical Review B, 99(19), May 2019. [57] E. Suarez´ Morell, J. D. Correa, P. Vargas, M. Pacheco, and [40] Hoi Chun Po, Liujun Zou, Ashvin Vishwanath, and T. Senthil. Z. Barticevic. Flat bands in slightly twisted bilayer graphene: Origin of mott insulating behavior and superconductivity in Tight-binding calculations. Phys. Rev. B, 82:121407, Sep 2010. twisted bilayer graphene. Phys. Rev. X, 8:031089, Sep 2018. [58] Grigory Tarnopolsky, Alex Jura Kruchkov, and Ashvin Vish- [41] A.M. Polyakov. Gauge Fields and Strings. Contemporary con- wanath. Origin of magic angles in twisted bilayer graphene. cepts in physics. Taylor & Francis, 1987. Phys. Rev. Lett., 122:106405, Mar 2019. [42] Shinsei Ryu, Andreas P Schnyder, Akira Furusaki, and Andreas [59] Chong Wang, Adam Nahum, Max A. Metlitski, Cenke Xu, and W W Ludwig. Topological insulators and superconductors: ten- T. Senthil. Deconfined quantum critical points: Symmetries and fold way and dimensional hierarchy. New Journal of Physics, dualities. Phys. Rev. X, 7:031051, Sep 2017. 12(6):065010, jun 2010. [60] Zhenjiu Wang, Yuhai Liu, Toshihiro Sato, Martin Hohenadler, [43] Subir Sachdev. Quantum Phase Transitions. Cambridge Uni- Chong Wang, Wenan Guo, and Fakher F. Assaad. Doping- versity Press, 2 edition, 2011. induced quantum spin Hall insulator to superconductor transi- [44] Yu Saito, Jingyuan Ge, Kenji Watanabe, Takashi Taniguchi, tion. arXiv e-prints, page arXiv:2006.13239, June 2020. and Andrea F Young. Decoupling superconductivity and cor- [61] Edward Witten. Instatons, the quark model, and the 1/n expan- related insulators in twisted bilayer graphene. arXiv preprint sion. Nuclear Physics B, 149(2):285–320, 1979. arXiv:1911.13302, 2019. [62] Dillon Wong, Kevin P. Nuckolls, Myungchul Oh, Biao [45] Anders W. Sandvik. Evidence for deconfined quantum critical- Lian, Yonglong Xie, Sangjun Jeon, Kenji Watanabe, Takashi ity in a two-dimensional heisenberg model with four-spin inter- Taniguchi, B. Andrei Bernevig, and Ali Yazdani. Cas- actions. Physical Review Letters, 98(22), Jun 2007. cade of transitions between the correlated electronic states of [46] Andreas P Schnyder, Shinsei Ryu, Akira Furusaki, and An- magic-angle twisted bilayer graphene. arXiv e-prints, page dreas WW Ludwig. Classification of topological insulators arXiv:1912.06145, December 2019. and superconductors. In AIP Conference Proceedings, volume [63] Matthew Yankowitz, Shaowen Chen, Hryhoriy Polshyn, Yux- 1134, pages 10–21. American Institute of Physics, 2009. uan Zhang, K Watanabe, T Taniguchi, David Graf, Andrea F [47] T. Senthil, Leon Balents, Subir Sachdev, Ashvin Vishwanath, Young, and Cory R Dean. Tuning superconductivity in twisted and Matthew P. A. Fisher. Quantum criticality beyond the bilayer graphene. Science, page 1910, 2019. landau-ginzburg-wilson paradigm. Phys. Rev. B, 70:144407, [64] Shou-Cheng Zhang. A unified theory based on so(5) sym- Oct 2004. metry of superconductivity and antiferromagnetism. Science, [48] T. Senthil and Matthew P. A. Fisher. Competing orders, nonlin- 275(5303):1089–1096, 1997. ear sigma models, and topological terms in quantum magnets. [65] Ya-Hui Zhang, Dan Mao, Yuan Cao, Pablo Jarillo-Herrero, and Phys. Rev. B, 74:064405, Aug 2006. T. Senthil. Nearly flat chern bands in moire´ superlattices. Phys. [49] T. Senthil, Ashvin Vishwanath, Leon Balents, Subir Sachdev, Rev. B, 99:075127, Feb 2019. and Matthew P. A. Fisher. Deconfined quantum critical points. [66] Uri Zondiner, Asaf Rozen, Daniel Rodan-Legrain, Yuan Cao, Science, 303(5663):1490–1494, 2004. Raquel Queiroz, Takashi Taniguchi, Kenji Watanabe, Yuval [50] Aaron L. Sharpe, Eli J. Fox, Arthur W. Barnard, Joe Finney, Oreg, Felix von Oppen, Ady Stern, et al. Cascade of phase Kenji Watanabe, Takashi Taniguchi, M. A. Kastner, and transitions and dirac revivals in magic angle graphene. arXiv David Goldhaber-Gordon. Emergent ferromagnetism near preprint arXiv:1912.06150, 2019. three-quarters filling in twisted bilayer graphene. Science, [67] Stephan Rachel. Interacting topological insulators: a review. 365(6453):605–608, August 2019. Reports on Progress in Physics, volume 81, page 116501, 2018. 1

[68] Maine Christos, Subir Sachdev, and Mathias S. Scheurer. Witten terms in twisted bilayer graphene. Proceedings of the Superconductivity, correlated insulators, and Wess–Zumino– National Academy of Sciences, volume 117, 2020.

Supplemental material: Charged Skyrmions and Topological Origin of Superconductivity in Magic Angle Graphene

I. DERIVATION OF THE EFFECTIVE FIELD THEORY

In this section, we present the derivation of the effective ﬁeld theory describing the low-lying excitations (spin waves and skyrmions) for a pair of tunnel-coupled U(n) quantum hall system related by C2T symmetry. This describes the manifold of low-energy states in TBG close to charge neutrality n = 2 and half-ﬁlling n = 1 [7]. Although we focus in the main text on the spinless (n = 1) case, our derivation will be done for arbitrary n. The Hamiltonian is given by (see Ref. [7] for details)

X 1 X H = h + H , h = c† h(k)c , H = V δρ δρ (S1) eﬀ int k k int 2A q −q q k q X 1 X δρ = ρ − ρ¯ , ρ = c† Λ (k)c , ρ¯ = δ tr Λ (k) (S2) q q q q k q k+q q 2 q,G G k k,G

Here, k denotes a momentum in the ﬁrst Brillouin zone, q denotes a generic (unrestricted) momentum, and G denotes a reciprocal lattice momentum. In the following, we will use the basis of Chern sector and pseudospin deﬁned in the main text given by

γx,y,z = (σx, σyτz, σzτz), ηx,y,z = (σxτx, σxτy, τz) (S3)

In this basis, the Chern number is given by C = γz, the tunneling term is

h(k) = hx(k)γx + hy(k)γy (S4)

The density operator ρq (as well as the average density ρ¯q) can be split into a chiral symmetric piece and a chiral symmetry- breaking piece with the latter being only non-vanishing away from the chiral limit [7, 24]. The main approximation involved in this work is to neglect the effect of the chiral symmetry breaking form factor which was shown to be relatively small in Ref. [7]. The form factor then reduces to the simple expression

iφq (k)γz Λq(k) = Fq(k)e (S5)

We begin by considering translationally symmetric Slater determinant states at half-ﬁlling (which describes charge neutrality for n = 2 and the spinless model for n = 1) which can be described in terms of the projection operator P (k) 1 P (k) = (1 + Q(k)),Q2 = 1, tr Q = 0 (S6) 2

The states which minimize Hint are Slater determinant states characterized by the k-independent order parameter Q commuting 2 with γz. A fixed ground state |Ψ0i is specified by a given Q0 = 1, tr Q0 = 0 and [Q0, γz] = 0. Projected onto the manifold of ground state the effective energy functional has two contributions: (i) the tunneling term h which breaks the U(2n) × U(2n) symmetry down to U(n), (ii) slowly varying fluctuations in the order parameter Q0 that cost small energy.

In the spinless limit n = 1 which we consider in this work, the matrix Q can be written in terms of two unit vectors n± in the ± Chern sectors as

n · η 0 Q = + (S7) 0 n−(r) · η γz =± 2

A. Order parameter ﬂuctuations

The effective action associated with slow variations of the order parameter in space can be derived following Ref. [15] with the generalization to the SU(N) case from Ref. [6]. The manifold of low energy excitations contains the massless spin waves as well as charged topological excitations, skyrmions, which cost a finite energy . In computing the cost of slow variations of the order parameter, we will only consider the effect of Hint since the other terms are smaller corrections. The order parameter fluctuations are parametrized using the generators of the U(2n) × U(2n) symmetry group of the inter- µ 2 action term Hint which we denote by t , µ = 1,..., 4n . These can be obtained from the generators of U(4n) by restricting to those which commute with γz. The effect of the order parameter fluctuations can be taken into account by writing Z −iOˆ X 2 µ † µ |Ψ(w)i = e |Ψ0i, Oˆ = d rw (r)ψ (r)t ψ(r) (S8) µ

Here the sum over µ goes over the generators of U(2n) × U(2n) excluding those which leave the ground state Q0 invariant. The µ latter corresponds to the subgroup U(n) × U(n) × U(n) × U(n). This means that we restrict the sum to {t ,Q0} = 0. ψ(r) is the creation operator for the microscopic electron at point r, it can be written in terms of the operator cα,k which creates a state in the ﬂat band labelled by α (which for the spinless case corresponds to a combined valley-sublattice index α = (τ, σ)) at momentum k as Z X 2 † ψ(r) = φα,k(r)cα,k, d rφα,k(r)φβ,k0 (r) = δαβδk,k0 (S9) α,k

Here, φα,k(r) are the Bloch wavefunctions written in terms of the periodic wavefunctions uα,k(k) as

ik·r φα,k(r) = e uα,k(r), (S10)

Introducing the Fourier transform

µ X iq·r µ w (r) = e wq , (S11) q leads to

ˆ X µ ˆµ ˆµ X † µ O = wq S−q, Sq = cα,kcβ,k+qhuα,k|t |uβ,k+qi (S12) q,µ α,β,k

µ µ µ P µ We can now rewrite the action of t in terms of the corresponding projected operator T deﬁned as t |uα,ki = β Tα,β|uβ,ki leading to

ˆµ X † µ X † µ T µ µ Sq = cα,kcβ,k+q[Λq(k)]α,ηTβ,η = ckΛq(k)[T ] ck+q, [T , γz] = 0, {T ,Q0} = 0 (S13) α,β,η,k k

The expression above assumes U(2n) × U(2n) is a good symmetry and includes only the the symmetric part of the form factor. Although the form (S12) is more reminiscent of the standard derivation for the SU(2) spin skyrmions, we will ﬁnd it more convenient to express Oˆ in the slightly different form

ˆ X ˆ ˆ X † T X µ µ O = Oq, Oq = ckΛq(k)w−qck+q, wq = wq T , [wq, γz] = {wq,Q0} = 0 (S14) q k µ

Thus, wq is a hermitian matrix parametrizing the tangent space to the ground state manifold at Q0. In order to simplify the evaluation of the different terms in the expansion in Oˆ, we will derive a few simple identities below. We will be considering fermion bilinear operators

ˆ X † A = cαAα,βcβ (S15) α,β where we use the hat notation Aˆ for the second quantized operator corresponding to the ﬁrst quantized operator A. Using the 3 relation

† † † † [cαcβ, cγ cδ] = δβγ cαcβ − δαδcγ cβ (S16) we get

[A,ˆ Bˆ] = [\A, B] (S17)

The ground state expectation values can be evaluated using the identities

ˆ T ˆ ˆ T T T T T hAi = Tr P0 A, hABi = Tr ABP0 + Tr AP0 Tr BP0 − Tr AP0 BP0 (S18) where 1 [P ] = hc† c i,P = (1 + Q ) (S19) 0 α,β α β 0 2 0 The traces in the above expressions go over all internal indices including momenta (we will use the symbol “Tr” with capital T to denote traces which involve momentum summation to distinguish them from those involving only flavor summation denoted by “tr”). We notice that whenever A or B commutes with P0, then the first and third terms in (S18) cancel leaving only the second term. Similarly, if A or B anticommute with Q0, then the second and third terms vanish leaving only the first. This can be summarized as  Tr AP T Tr BP T :[A, Q ] = 0 or [B,Q ] = 0  0 0 0 0 ˆ ˆ T hABi = Tr ABP0 : {A, Q0} = 0 or {B,Q0} = 0 (S20)  0 : {AB, Q0} = 0

In our notation, we can write the following

ˆ T ρq = Λq, [Λq]k,k0 = δk0,k+qΛq(k),Oq = Λqw−q (S21)

We note that since the order parameter manifold is isomorphic to the product of two Grassmanian manifolds U(2n)/U(n) × U(n), each of these manifolds admits charged skyrmion textures since π2(U(2n)/U(n) × U(n)) = Z. The skyrmion charge can be evaluated directly using 1 hΨ(w)|δρ |Ψ(w)i = −ih[O,ˆ ρ ]i − h[O,ˆ [O,ˆ ρ ]i (S22) q q 2 q The ﬁrst order term vanishes since

ˆ T h[O, ρq]i = h[O,\Λq]i = Tr P0 OΛq = 0 (S23) which follows from the fact that OΛq anticommutes with Q0. The second order term can be evaluated as

(2) 1 1 T 1 X T δρ = − h[O,ˆ [O,ˆ ρ ]]i = − h[O,\[O, Λ ]]i = Tr P [O, [O, Λ ]] = − Tr(w 0 w 0 Q ) Λ 0 [Λ 0 , Λ ] (S24) q 2 q 2 q 0 q 2 q+q −q 0 q −q −q q q0

0 Since the ﬂuctuations wq are slowly varying in space, we can expand to leading order in small momenta q and q . To simplify the expansion, we note that

2 Λq(k) = 1 + iq · A(k) + O(q ), (S25)

+ − ± where A(k) = diag(A (k), A (k))γz with A (k) denoting the Berry connection for the ± Chern sectors γz = ±1. This leads to the following identity

0 0 3 0 3 0 3 [Λq, Λq0 ] = (q · ∇)(iq · A) − (q · ∇)(iq · A) + O(q ) = i(q ∧ q )(∇ × A) + O(q ) = i(q ∧ q )Ω(k) + O(q ) (S26) where Ω(k) denotes the Berry curvature. 4

Substituting in (S24) yields Z X 0 X i 2 h[O,ˆ [O,ˆ ρ ]i = i (q ∧ q) tr Q w 0 w 0 Ω(k) = d r tr γ Q ∇ w(r) ∧ ∇ w(r) (S27) q 0 −q q+q 2π z 0 r r q0 k

P ± A R 2 ± A where we used the fact that k Ω (k) = (2π)2 BZ d k Ω (k) = ± 2π . Thus, the charge density associated with a skyrmion is i i δρ(r) = − tr γ Q ∇ w(r) ∧ ∇ w(r) = − [tr Q ∇ w (r) ∧ ∇ w (r) − tr Q ∇ w (r) ∧ ∇ w (r)] (S28) 4π z 0 r r 4π 0,+ r + r + 0,− r − r − where we split Q0 and w(r) into ± Chern sectors. Since this expression only depends on gradients of w, we can lift the assumption that w is small used in the derivation and consider a general slowly varying deformation of the order parameter −1 −iw(r) −1 Q(r) = T (r) Q0T (r) with T (r) = e . We can then make the replacement ∇w(r) 7→ i(∇T )T . Substituting in (S28) yields after a sequence of straightforward manipulations i i δρ(r) = tr γ Q(r)∂ Q(r)∂ Q(r) = [tr Q (r)∂ Q (r)∂ Q (r) − tr Q (r)∂ Q (r)∂ Q (r)] (S29) 16π ij z i j 16π ij + i + j + − i − j − This expression says that a skyrmion with topoplogical winding n is associated with charge ±n in the ± Chern sector. In the spinless limit n = 1, we can substitute the expression (S7) in (S29), to get the more familiar expression 1 Z δρ = ∓ d2r n (r) · [∂ n (r) × ∂ n (r)] (S30) ± 8π ij ± i ± j ±

The effective Hamiltonian is obtained as

1 X 1 X H = V ρ ρ − V ρ ρ¯ (S32) 2A q q −q A G G −G q G by separating the density-density part from the background part. To evaluate the leading order term, we need to evaluate

ˆ ˆ ˆ hadOˆ ρqρ−qi = h[O, ρqρ−q]i = h[O,\Λq]Λ−qi + hΛq[O,\Λ−q]i (S33) which follows from the standard identity of commutators

[A, BC] = [A, B]C + B[A, C] ↔ adABC = (adAB)C + B adAC (S34)

Each term can be evaluated using (S18). From (S20), we can see that both terms vanish. In fact, this is the case for the expectation value of any product of operator containing an odd number of Oˆ operators. The second order term is evaluated by evaluating the expectation value

2 2 2 hadOˆ ρqρ−qi = h(adOˆ ρq)ρ−qi + 2h(adOˆ ρq)(adOˆ ρ−q)i + hρqadOˆ ρ−qi (S35) which can be obtained by repeatedly applying (S34). The ﬁrst term yields

2 \ ˆ T T X ˆ ˆ h(adOˆ ρq)ρ−qi = h[O, [O, Λq]]Λ−qi = Tr P0 Λ−q Tr P0 [O, [O, Λq]] = ρ¯−Gh[O, [O, ρG]]i (S36) G which, together with the last term, cancels against the corresponding term in the background (second term in (S32)). The second term is

T X T h(adOˆ ρq)(adOˆ ρ−q)i = h[O,\Λq][O,\Λ−q]i = Tr P0 [O, Λq][O, Λ−q] = Tr(wq0 w−q0 P0) [Λq0 , Λq][Λ−q0 , Λ−q] (S37) q0 5

We notice that, unlike (S24) where both q and q0 were small since the ﬂuctuations are slowly varying in space, this expression only includes w±q0 and as a result we should not assume that q is small. Introducing the gauge invariant combination

0 0 Mk(q, q ) = Λq(k)Λq0 (k + q) − Λq0 (k)Λq(k + q ) (S38)

We can write

1 X 0 0 h(ad ρ )(ad ρ )i = M (q, q )M (−q, −q ) tr w 0 w 0 (S39) Oˆ q Oˆ −q 2 k k −q q k,q0

The corresponding energy is then given by

1 X Z δE(2) = − had2 Hi = ρ d2r tr ∂ w(r)∂ w(r) (S40) 2 Oˆ ij i j i,j=x,y where ρij

1 ∂2 X 0 0 ρij = − 2 VqMk(q, q )Mk(−q, −q ) (S41) 4A ∂q0 ∂q0 q0=0 i j k,q

Since we expect the continuum theory to be rotationally symmetric, we can assume ρxx = ρyy = ρps and ρxy = ρyx = 0 leading to the expresssion

1 2 X 0 0 ρps = − ∇q0 VqMk(q, q )Mk(−q, −q ) (S42) 8A2 q0=0 k,q with the energy given by ρ Z ρ Z δE(2) = ps d2r tr[∇Q(r)]2 = ps d2r tr{[∇Q (r)]2 + [∇Q (r)]2}, (S43) 4 4 + −

One possible simpliﬁcation of (S42) was suggested in [8] by assuming that the magnitude of the form factor Fq(k) decays relatively quickly with increasing |q| and depends weakly on k such that the terms containing the derivative ∇kFq(k) can be ignored. This justiﬁes expanding in small q using Λq(k) = Fq(k)[1 + iq · A(k)] leading to

0 0 Mk(q, q ) = iijqiqjFq(k)Ω(k) (S44) which gives

1 X X ρ = V q2 F (k)2Ω(k)2 (S45) ps 8A2 q q q k in agreement with [8].

The next order term is the fourth order. This term is presumably smaller than the second order term since it has more derivatives. However, it has a part that becomes relevant if the potential V (r) is sufﬁciently long-ranged (weak or no screening limit). To see this, we write

4 4 3 2 2 3 4 hadOˆ ρqρ−qi = h(adOˆ ρq)ρ−qi + 4h(adOˆ ρq)(adOˆ ρ−q)i + 6h(adOˆ ρq)(adOˆ ρ−q)i + 4h(adOˆ ρq)(adOˆ ρ−q)i + hρq(adOˆ ρ−q)i (S46) The ﬁrst and last terms vanish against the background. The second and fourth terms include the product of two operators which each anticommute with Q0 which implies that the ground state expectation value yields a single trace (cf. Eq. S20). As a result, the expression has the form ∼ wq1 wq2 wq3 w−q1−q2−q3 which enables taking the q dependence out of the trace similar to the manipulations leading to (S43). The resulting expression is local in real space and contains at least four derivative terms. This term is obviously smaller than the gradient term (S43) and can be safely neglected. On the other hand, the third term is the product of two even terms under Q0 which yields a product of Traces leading to the form ∼ wq1 w−q−q1 wq2 wq−q2 . In real space, this couples to the gradients of the order parameter at different points as we will see below. This means that such term can be important for sufﬁciently long range interactions. 6

Using (S20), we ﬁnd

2 2 T T h(adOˆ ρq)(adOˆ ρ−q)i = Tr P0 [O, [O, Λq]] Tr P0 [O, [O, Λ−q]] = 4δρqδρ−q (S47) where we used (S24). The corresponding correction to the Hamiltonian is 1 Z δE(4) = d2rd2r0δρ(r)V (r − r0)δρ(r0) (S48) 2

We can now summarize the results of this section as ρ Z 1 Z Z δE = hΨ(w)|H − µδρ|Ψ(w)i = ps d2r tr[∇Q(r)]2 + d2rd2r0δρ(r)V (r − r0)δρ(r0) − µe d2rδρ(r), (S49) 4 2 i i δρ(r) = tr γ Q(r)∂ Q(r)∂ Q(r) = δρ (r) + δρ (r), δρ (r) = ± tr Q (r)∂ Q (r)∂ Q (r) (S50) 16π ij z i j + − ± 16π ij ± i ± j ±

B. Effect of tunneling h

The dispersion term h is off-diagonal in the Chern basis h ∝ γx,y. This means that

n hadOˆ hi = 0 (S51) for any n. That is, the energy does not recerive any corrections to ﬁrst order in h. This is easy to understand since h creates a particle-hole pair between bands with opposite Chern number. Such excitations are massive and cannot be described by slow variations of the order parameter within the manifold of low-energy states. The effect of h can be captured by allowing for massive excitations which take Q0 out of the ground state manifold. In the following, we will restrict ourselves to massive excitations which can be created by h which are given by

−iMˆ ˆ X † |Ψ0i 7→ |Ψ(M)i = e |Ψ0i, M = ckMkck, {Mk, γz} = {Mk,Q0} = 0 (S52) k

The energy associated with these fluctuations can be evaulated by writing a very similar expanstion to (S31) with Oˆ replaced by Mˆ . Similarly, the first order term in the expansion vanishes and the second order term can be expanded as in (S35) with the first and last terms cancelling against the background. The resulting expression then simplifies to

(2) 1 X 1 X 1 X E = − V had2 ρ ρ i = − V Tr P T [M, Λ ][M, Λ ] = V Tr P T [M, Λ ][M, Λ ] M 4A q Mˆ q −q 4A q 0 q −q 4A q 0 q −q q q q 1 X 1 X = V F (k)2 tr P T {M 2 − M M [Λ† (k)]2} = V F (k)2 tr{M 2 − M M [Λ† (k)]2} (S53) 2A q q 0 k k k+q q 4A q q k k k+q q q,k q,k where we used the fact that Mk anticommutes with Q0 in the last line. We can parametrize Mk as

X µ ∗ µ µ µ ν Mk = r (mµ,kγ+ + mµ,kγ−),Q+r Q− = −r , tr r r = δµ,ν (S54) µ

µ µ µ where r are the generators of U(2n) satisfying Q+r Q− = −r (The map M 7→ Q+MQ− is a linear map on the space of 2n × 2n matrices which squares to the identity, so it induces a decomposition on the space of matrices into even and odd µ subspaces under this operator. r are taked to be the generators of the odd subspace). Here, mµ,k are complex functions of k. Substituting in (S53) yields 1 1 (2) X ∗ X 2 2iφq (k) E = m R 0 m 0 ,R 0 = V F (k) {δ 0 − δ 0 e } (S55) M 2 µ,k k,k µ,k k,k A q q k,k k ,k+[q] k,k0,µ q 7 where [q] denotes the component of q in the ﬁrst Brillouin zone. We now compute the leading correction in h given by

(1) ˆ ˆ T X X ∗ ∗ µ Eh = −ih[M, h]i = −i Tr P0 [M, h] = −i tr Q0Mkh(k) = −i (mµ,kzk −mµ,kzk) tr Q+r , zk = hx(k)+ihy(k) k k,µ (S56) where we used the properties of the generators (S54). Thus the effective Hamiltonian for the massive excitations mµ,k is given by

1 X ∗ X ∗ ∗ µ H = m R 0 m 0 − i (m z − m z ) tr Q r (S57) M 2 µ,k k,k µ,k µ,k k µ,k k + k,k0,µ k,µ

We can now integrate out the massive terms to get

X ∗ −1 X µ µ Hh = −2 zk(R )k,k0 zk tr Q+r tr Q+r (S58) k,k0 µ

To evaluate the product of traces summed over the generators rµ, we use the Fierze identity for the generators of U(N) given by

X µ µ tαβtγδ = δαδδβγ (S59) µ which when restricted to generators satisfying (S54) yields

X 1 tr Arµ tr Brµ = [tr AB + tr Q AQ Q BQ − tr Q AQ B − tr AQ BQ ] (S60) 4 + − + − + − + − µ

Substituting in (S58) then yields Z 2 1 X ∗ −1 H = −AJ(tr 1 − tr Q Q ) = AJ tr Q Q + const. = J d r tr Q (r)Q (r),J = z (R ) 0 z (S61) h + − + − + − A k k,k k k,k0

This result means that integrating out the massive modes yields an antiferromagnetic coupling between the + and − Chern sectors which favors Q+ = −Q−.

C. Dynamical term

To obtain the full effective action, we need also to include the dynamical term obtained by allowing the ﬂuctuations Oˆ to be (imaginary) time-dependent 1 1 N N hΨ(w)|∂ |Ψ(w)i = − had2 ∂ i = − h[Oˆ(τ), ∂ Oˆ(τ)]i = − tr P T [wT (τ), ∂ wT (τ)] = − tr Q w ∂ w (τ) (S62) τ 2 Oˆ τ 2 τ 2 0 0 τ 0 2 0 0 τ 0

−1 Writing that Q(r, τ) = T (r, τ)Q0T (r, τ), we notice that this term matches the leading term in the expansion of 1 −1 2 tr T Q0∂τ T . This is in fact the well-known one-dimensional Wess-Zumino term which is actually the unique expression that satisﬁes the required symmetries. Let us write it as

1 X Z β S = dτ tr{T (R)−1Q ∂ T (R) + T −1(R)Q ∂ T (R)} τ 2 + 0,+ τ + − 0,− τ − R 0 Z 1 2 −1 −1 = d rdτ tr{T+(r) Q0,+∂τ T+(r) + T−(r) Q0,−∂τ T−(r)} (S63) 2AM

−1 It is well-known that this term cannot be written in a gauge invariant way in terms of Q = T Q0T . To see this, we notice that the operator T has a gauge freedom T (r, τ) 7→ K(r, τ)T (r, τ) for any matrix K(r, τ) which commutes with Q0. The matrix T can be any matrix in G = S(U(2n) × U(2n)). The matrix Q on the other hand parametrizes the ground state manifold which is given by the coset space G/K where K is the subgroup of G which commutes with Q0 (which is isomorphic to U(n) × U(n) × U(n) × U(n)). All other terms in the action are manifestly gauge invariant since they are written explicitly in terms of Q. The Wess-Zumino term (S63) on the other hand can only be written in a gauge invariant way by introducing an 8 auxiliary dimension which we will not do here. Its transformation under gauge transformations T (r, τ) 7→ K(r, τ)T (r, τ) is given by (for simplicity, consider the + sector and drop the subscript and consider a ﬁxed position r = R0)

Z β Z β 1 −1 1 −1 −1 Sτ = dτ tr T Q0∂τ T 7→ dτ tr T K Q0∂τ KT 2 0 2 0 1 Z β 1 = Sτ + dτ ∂τ tr Q0 ln K = Sτ + tr Q0(ln K[β] − ln K[0]) (S64) 2 0 2

Since Q0 and K commutes, we can simultanuously diagonalize them. The eigenvalues of Q0 are ± whereas the eigenvalues of ln K[β] − ln K[0] is an integer multiple of 4πi (since det K = 1) due to periodic boundary conditions, which implies that the extra term is an integer multiple of 2πi. i i i − φηz θηx φηz We can write this term in the familiar form in the spinless limit by taking Q0 = ηz and T = e 2 e 2 e 2 . Substiting in (S63) yields

i Z β Sτ = − dτ (1 − cos θ)∂τ φ (S65) 2 0

D. Full effective action

The full effective action is given by Z Z 2 1 −1 ρps 2 J 2 1 2 0 0 0 S[Q] = d rdτ tr T Q0∂τ T + tr[∇Q(r)] + tr(Qγx) − µeδρ(r) + d r δρ(r)V (r − r )δρ(r ) 2AM 4 4 2 (S66) In the spinless limit where Q is given by (S7), this leads to the ﬁeld theory (4) in the main text.

E. Deviation from perfect sublattice polarization

We will now brieﬂy investigate the effect of imperfect sublattice polarization. This corresponds to considering the sublattice iΦ˜ q (k)γz off-diagonal part of the form factor which can be generally written as Λ˜ q(k) = γxηzF˜q(k)e . This leads to the following contribution to the action 1 X ˆ ˆ λA λA λ Z E = V hΛ˜ Λ˜ i = tr[Q, γ η ][Q, γ η ]† = − tr(Qγ η )2 + const. = − d2r tr(Q(r)γ η )2 + const. λ 2A q q −q 8 x z x z 4 x z 4 x z q (S67) with λ given by

1 X λ = V F (k)2 (S68) 2A2 q A,q k,q

This reduces to the expression

Eλ = −λ(n+,zn−,z − n+,xy · n−,xy) (S69)

II. EFFECTS OF DYNAMICAL FLUCTUATIONS ON SKYRMIONS

In this section, we discuss the estimate for the effective mass of a skyrmion antiskyrmion pair in a first quantized description of skyrmions. In the main text, we considered the Lagrangian of a skyrmion-antiskyrmion pair with a frozen shape, i.e. where only the location of the skyrmions is a dynamical variable. Here, we critically examine the validity of this assumption and also derive the effective ”spring constant” k used in Eq. (6) of the main text. Our strategy for considering dynamical degrees of freedom of the skyrmion uses a somewhat crude model for physical trans- parency and analytical amenability, where the skyrmion is treated as a magnetic bubble with a nearly circular domain wall. Outside the domain wall, the spins are approximately pointing up whereas inside, they point down, and along the domain wall the spins lie in the x-y plane, i.e, nz(r) = 0 defines the domain wall. Therefore, we can parametrize the domain wall by the 9 following collective coordinates: r(φ) denoting the radius as a function of the azimuthal angle φ, (for a static bubble r = R) and Φ(φ) which denotes the in-plane spin fluctuations on the domain wall. Our goal is to find an effective action for the collective coordinates, and then integrate out the spin fluctuations Φ(φ) to generate the dynamics of r(φ), the harmonics of which corresponds to motion/deformation of the bubble.

Following Ref. 14, we write down the action for a circular domain wall of static radius R, denoting the angular momentum density by g. Z Z κ S = dtL, where L = R dφ L(r, Φ), with L(r, Φ) = −gΦr ˙ − (Φ − Φ )2 (S70) 2 eq where Φeq denotes the equilibrium configuration. As before, our strategy is to integrate over φ first to obtain coupled equations of motion for r(φ) and Φ(φ), and then integrate out Φ(φ) to generate the effective dynamics for r. For this purpose, we decompose the fields into circular harmonics:

∞ ∞ X imφ X imφ r(φ) = R + rm e , Φ(φ) = φ + φ1 + Φm e (S71) m=−∞ m=−∞

In Eq. (S71), rm = r−m denote circular harmonics of r. For example, r0 parametrizes the breathing mode, r±1 = (X ∓ iY )/2 correspond to the center-of-mass motion of the bubble, r±2 parametrize elliptical deformations, and so on. Similarly, Φm = Φ−m, and the shift by constant φ1 is motivated by the observation that in equilibrium, the spins prefer to point at a given angle to 1 ∂r 1 ∂r the domain wall [14]. For example, for Bloch domain walls, φ1 = ±π/2, so we have Φeq(φ) = φ±π/2− r ∂φ ≈ φ±π/2− R ∂φ . However, our results will be insensitive to the the precise value of the angle φ1. Finally, κ ∼ ρpsW parametrizes the spin stiffness, W being the width of the domain wall. Plugging this into the Lagrangian L, we ﬁnd that (upto total time-derivatives):

" ∞ 2# ˙ κ 2 X imrm L = 2πR −gr0Φ0 − Φ + (−g)(r ˙mΦm + r˙mΦm) − κ Φm − (S72) 2 0 R m=1

Now, we can integrate out the (Φm, Φ¯ m) modes to generate the effective action for the skyrmion coordinates:

∞ πRg2 X 2πRg2 L = r˙2 + |r˙ |2 + 2πimg(r˙ r − r˙ r ) eff κ 0 κ m m m m m m=1 M = s R˙ 2 − πg(R × R˙ ) · zˆ + ... (S73) 2

πRg2 where Ms = κ denotes the effective mass of the skyrmion, and the ellipsis denotes the |m| 6= 1 terms which also scale with the skyrmion mass. We see that the skyrmion mass Ms is directly proportional to the linear size R of the skyrmion, and inversely proportional to the stiffness κ (a larger stiffness makes ﬂuctuations expensive and suppresses the kinetic energy term of the skyrmion). Therefore, for a small-size skyrmion or a large spin-stiffness, the dynamical ﬂuctuations are likely to be suppressed. On the contrary, the Berry phase term is independent of the skyrmion-size.

Next, we discuss the effective action for the skyrmion-antiskyrmion pair from opposite Chern sectors. On inclusion of the inertial term Ms in addition to the Berry phase term [23], the action for the pair is given by (with G = 2π~/AM ): Z G M S[R , R ] = dtL , where L = [(R × R˙ ) − (R × R˙ )] · zˆ + s (R˙ 2 + R˙ 2 ) − E(|R − R |) + − sas sas 2 + + − − 2 + − + −

Moving to center of mass coordinates Rs = (R+ + R−)/2 and relative coordinates Rd = R+ − R−, a rewriting of the Lagrangian yields: M L = G(R × R˙ ) · zˆ + MR˙ 2 + R˙ 2 − E(|R |) (S74) sas d s s 4 d d 10

The canonical momenta, and hence the Hamiltonian of the pair, are therefore given by:

∂L ∂L Ms Ps = = 2MR˙ s + G(zˆ × Rd), Pd = = R˙ d (S75) ∂R˙ s ∂R˙ d 2 M H = P · R˙ + P · R˙ − L = M R˙ 2 + s R˙ 2 + E(|R |) sas s s d d s s 4 d d 1 2 1 2 = (Ps − Gzˆ × Rd) + Pd + E(|Rd|) (S76) 4Ms Ms

In order to make further analytical progress, one needs to resort the Ms → 0 limit, which we analyzed in the main text, and is a good approximation when the size of single (charge e) skyrmions is small. In this limit, we have Ps = Gzˆ × Rd and Pd = 0 |Ps| strictly enforced, so that we can replace E(|Rd|) by E G . Therefore, the mass of the pair can be derived by expanding E about Rd = 0: it is given by:

G2 M = (S77) pair E00(0)

What is left now is to determine the pair potential E(|Rd|). For this purpose, we start with a representative KIVC ground state n+(r) = −n−(r) = (1, 0, 0) for the action in Eq. (5) in the main text (at µ = 0); such a ground state would be chosen by anisotropies favoring ordering in the x-y plane that we have neglected in this work for simplicity. We consider skyrmionic (antiskyrmionic) textures in the C = + (-) sector with a relavtive core separation of Rd, such that n±(r) = ±nsk(r ± Rd/2). Here, nsk(r) is a skyrmionic texture with unit winding for a unit-vector ﬁeld in two spatial dimensions, given by: 1 Z n (r) = (cos(Θ(r)) sin(Θ(r)) cos(Φ(r)), sin(Θ(r)) sin(Φ(r))), satisfying d2r n · (∂ n × ∂ n ) = 1 (S78) sk 4π sk x sk y sk

Since the elastic energy of this pair depends only on the winding number, it is independent of the relative core-separation Rd. The exchange energy proportional to J increases, while the Coulomb interaction energy decreases as the charged skyrmions are separated. As a function of separation |Rd|, the change of pair potential is given by: Z Z d2q E(|R |) − E(0) = J d2r (1 + n (r) · n (r)) + V δρ δρ (eiq·Rd − 1) d + − (2π)2 |q| q −q Z Z ∞ 2 1 2 = J d r (1 − nsk(r − Rd/2) · nsk(r + Rd/2)) + dq(δρq) (J0(q|Rd|) − 1) (S79) 4π 0 R 2 iq·r where we have noted that δρq = d r δρ(r)e is only dependent on |q| for cylindrically symmetric charge distributions, and used V|q| = 1/(2|q|) as the Fourier transform of the (unscreened) Coulomb potential for two-dimensional motion. The pair mass, therefore, may simply be computed by Taylor expanding the RHS of Eq. (S79) to second order in Rd, which yields: Z Z ∞ J 2 2 1 2 2 2 4 E(|Rd|) − E(0) = d r(∇nˆ) − dq(δρq) q |Rd| + O(|Rd| ) 4 16π 0 Z Z ∞ 00 J 2 2 1 2 2 =⇒ E (0) = d r(∇nˆ) − dq(δρq) q (S80) 2 8π 0 Next, we compute this correction for the Belavin-Polyakov ansatz [17] that corresponds to Θ(r) = 2 tan−1(R/r), and Φ(r) = φ in the texture deﬁned in Eq. (S78). For this ansatz δρq = e(qR)K1(qR), and it follows that:

45πe2 E00(0) = 4πJ − (S81) 212R3 For large skyrmions, the correction from the Coulomb term is insigniﬁcant and can be neglected, leading to k = 4πJ as quoted in Eq. (6) of the main text. For smaller skyrmions, the Coulomb term can reduce E00(0) substantially leading to an enhancement of the pair effective mass Mpair and a subsequent reduction to Tc.

III. INTEGRATING OUT THE FERROMAGNETIC FLUCTUATIONS

In this section, we explain the procedure to integrate out ferromagnetic spin ﬂuctuations leading to an effective theory. In terms of the matrix order parameter Q, ferromagnetic ﬂuctuations denote deviations from the perfect antiferromagnetic alignment 11

Q+ = −Q− which can be parametrized as

˜−1 i W ˜ − i W ˜ T e 2 Q0e 2 T 0 ˜ Q(W ) = −1 − i W i W , {W, Q0} = 0, (S82) 0 −T˜ e 2 Q˜0e 2 T˜

We assume that both W and its gradients as well as the gradients of T are small. To simplify the notation in the following, we will introduce the variables

−1 −1 Q˜ = T˜ Q˜0T,A˜ µ = (∂µT˜)T˜ (S83)

Since the action is expressed in terms of Q(W ), it is invariant under the gauge transformations

−1 −1 −1 T˜ 7→ KT,W˜ 7→ KWK ,Aµ 7→ KAµK + (∂µK)K , for [K,Q0] = 0 (S84)

Substituting in the different terms in the action, we get ρ ρ ρ ps tr[∂ Q(W )]2 = ps tr[∂ Q˜]2 − ps tr[Q , ∂ W − [A ,W ]]2, (S85) 4 i 2 i 8 0 i i 1 i tr Q (∂ T (W ))T −1(W ) = −i tr Q˜ WA − tr Q˜ ∂ W (S86) 2 0 τ 0 0 2 0 τ J tr(Q(W )γ )2 = J tr W 2 + const. (S87) 4 x i i δρ(r) = ij tr Q∂˜ Q∂˜ Q˜ + ij tr Q˜ ∂ W ∂ W (S88) 8π i j 8π 0 i j We notice that the second term in (S86) is a total derivative which can be ignored since W (β) = W (0). We also notice that the second term in (S88) is a total derivative which vanishes if W is smooth and vanishes quickly enough at inﬁnity (which is simply the statement that smooth deformations cannot change the topological index) so we can also neglect this term. As a result, the action for the ferromagnetic ﬂuctuations W has the form Z Z 2 2 ρps 2 i SW = dτ d r J tr W − tr[Q˜0, ∂iW − [Ai,W ]] − tr Q˜0WA0 (S89) 8 AM

To proceed, we restrict ourselves to the spinless limit n = 1, where we can take Q˜0 = ηz and write

0 w X 1 W = ,A = A + A η ,A = tr A η (S90) w∗ 0 µ 0,µ i,µ i i,µ 2 µ i i=x,y,z

Substituting in (S89) yields Z Z 2 ∗ ∗ 1 ∗ ∗ SW = dτ d r 2Jw w + ρps[(∂i + 2Az,i)w ][(∂i − 2Az,i)w] + (wAxy,0 − w Axy,0) ,Axy,µ = Ax,µ−iAy,µ AM (S91) i φ η We notice that under the gauge transformation generated by K = e 2 z , the w and A variables transform as i w 7→ eiφw, A 7→ A + ∂ φ, A 7→ eiφA (S92) z,µ z,µ 2 µ xy,µ xy,µ which leaves (S91) invariant.

We can now easily integrate the w variable to get Z 1 2 ∗ 2 −1 Seﬀ = 2 dτd rAxy,0[2J − ρps(∂i − 2Az,i) ] Axy,0 (S93) AM Let us now introduce the CP1 representation deﬁned from the 2 × 2 Q matrix via

−1 † T † Q˜± = ±T˜ Q˜0T˜ = ±n · η, n = z ηz, z = (z1, z2), z z = 1 (S94) 12

It is straightforward to verify that

† † † ∗ ˜ z ∂µz z i∂µz σyz † † ∗ T = T ,Aµ = T † ⇒ Az,µ = −z ∂µz, Axy,µ = −iz σy∂µz (S95) iz σy −i∂µz σyz z ∂µz

Introducing the gauge ﬁeld and covariant derivate

† aµ = iz ∂µz, Dµ = ∂µ − iaµ (S96) we can write

† ∗ Az,µ = −iaµ,Axy,µ = −iz σy(Dµz) (S97)

† ∗ where we used the fact that z σyz = 0. Substituting in (S93) yields Z 1 2 T 2 −1 † ∗ Seﬀ = − 2 dτd rz σy(D0z)[2J − ρps(∂i + 2iaµ) ] z σy(D0z) (S98) AM In the following, we will be only interested in terms which has at most two gradients. This means we can neglect the second term in the denominator. The resulting expression can be simpliﬁed by using the identity

† ∗ T † † † † (a σyb )(c σyd) = (a d)(b c) − (a c)(b d) (S99) for any 2D complex vectors a, b, c and d. In addition, we can use the following identities

† † † † z Dµz = (Dµz) z = 0, z DµDν z + (Dµz) Dν z = 0, (S100)

Substituting in (S98) yields Z 1 2 1 2 Seﬀ = 2 dτd r |D0z| (S101) AM 2J The full CP1 action is then given by Eq. 8 in the main text.

IV. LARGE N PHASE DIAGRAM

Our purpose in this appendix is to provide the details for the calculation of the large N phase diagram for the CP1 model whose Lagrangian is repeated here for completeness Z Z 3 Λ 2 2eµ ij 1 2 0 ij 0 lk S[z] = d r |D z| + ∂ a + d r (∂ a ) V (r − r ) (∂ a ) 0 (S102) g µ c 2π i j 2c π i j r π l k r with g and c deﬁned as s JAM 2 p g = , c = 2JAM ρps (S103) 2ρps Λ

A. Zero doping

Let’s ﬁrst review the solution for the case of µ = 0 [5, 9, 18, 25]. In this case, we can start by writing Λ Λ |D z|2 = (|∂ z|2 − a2 ) (S104) g µ g µ µ and decouple the quadratic term in aµ via a Hubbard-Stratonovich transformation as

Λ 2 Z Λ 2 a − (α −2aµαµ) e g µ ∝ dαµe g µ (S105) 13

Upon substituting in (S98), this leads to Λ L = |D z|2, D = ∂ − iα (S106) g µ µ µ µ

† Since αµ enters the ﬁeld theory in the same way as aµ, in the following, we will relabel αµ 7→ aµ. The constraint z z = 1 can be included by writing the integral representation of the delta function

Z † δ(z†z − 1) = dκ eiκ(z z−1) (S107) leading to Λ g L = {|D z|2 + ∆2(z†z − 1)}, ∆2 = iκ (S108) g µ Λ

We now introduce the large N limit by promoting z to an N-component vector in CPN−1 satisfying the constraint z†z = N. The Lagrangian becomes

( N ) Λ X L = z†[−D2 + ∆2]z − ∆2N (S109) g i i i=1 Integrating out the z variables leads to

Λ Z S = N tr ln −D2 + ∆2 − d3r∆2 (S110) g

2 Variation with respect to ∆ and aµ yield Λ = G (∆, r), g 0 Z −iG (∆, r) − 0µν ij d2r0V (r − r0)∂ ∂ a (r0) = 0 (S111) µ π2c ν i j where G0 is the Green’s function at equal points ∂ 0 0 2 2−1 G0(r) = G(r, r),Gµ(r) = G(r, r ) ,G(r, r ) = −D + ∆ r,r0 (S112) ∂rµ r0=r and the derivative in ∂µG0 is taken relative to the first index. Since µ = 0, we can consider saddle points when there is no magnetic field, i.e. ∇ × a = 0, so we can take aµ = 0. In this case, the second saddle point equation is automatically satisfied whereas the first one becomes Z d3k 1 Λ G (∆, r) = G (∆) = = , (S113) 0 0 (2π)3 k2 + ∆2 g

We notice here that the Green’s function at equal points is independent of r. The integral can be regularized as follows

Z d3k 1 1 1 G (∆) = − = {Λ − |∆|} (S114) 0 (2π)3 k2 + ∆2 k2 + Λ2 4π

As a result, we see that a self consistent solution to (S113) is only possible for g > gc = 4π yielding 4π |∆| = Λ 1 − = 4πΛ(g−1 − g−1) (S115) g c 14

This gives a mass for the z ﬁelds which implies a ﬁnite correlation length for the spins, i.e. the phase is spin-disordered. Furthermore, expanding the action in variations in aµ yields [5, 18]

N X S = a (q)Γ (q)a (−q) (S116) 2 µ µν ν q Z d3k 2δ (2k + q )(2k + q ) 1 Γ (q) = µν − µ µ ν ν = (q2δ − q q ) (S117) µν (2π)3 k2 + ∆2 (k2 + ∆2)((k + q)2 + ∆2) 24π|∆| µν µ ν which gives the Maxwell term N L = f 2 , f = ∂ a − ∂ a (S118) Maxwell 96π|∆| µν µν µ ν ν µ

To see how this is related to the phase stiffness of the superconductor, we first include the coupling to the electromagnetic gauge 2ie µνλ field A which takes the form 2π JµAµ where the current operator Jµ is given by Jµ = ∂ν aλ. The Lagrangian for aµ takes the form N 2ie L = J 2 + J A (S119) 48π|∆| µ 2π µ µ where Jµ satisfies the current conservation ∂µJµ = 0 which can be included in the Lagrangian using the delta function representation (S107) leading to N 2ie i N i L = J 2 + J A + φ∂ J = J 2 + J (∂ φ − 2eA ) (S120) 48π|∆| µ 2π µ µ 2π µ µ 48π|∆| µ 2π µ µ µ

We can now integrate out the Jµ variable leading to

3|∆| L = (∂ φ − 2eA )2 (S121) πN µ µ Identifying the phase φ with the superconducting phase, we can write the corresponding action

Z nρ χ o 3|∆|c 3|∆| S = d2rdτ SC (∂ φ − 2eA )2 + SC (∂ φ − 2eA )2 , ρ = χ = (S122) 2 i i 2 τ τ SC πN SC πcN

We note here that it is also possible to access the spin-ordered phase by slightly modifying the procedure above, integrating out the variables z2, . . . , zN but leaving z1. In this case, we can similarly derive a set of saddle points [1, 5] with a saddle point solution ∆ = 0 and g g |z |2 = 1 − G (0) = 1 − (S123) 1 Λ 0 4π so that such solution is only valid for g < gc. This is the spin-ordered phase since aµ ﬁeld is in the Higgs phase.

B. Finite doping µ 6= 0

Let us now reintroduce the chemical potential term µ. Although the chemical potential µ does not enter the saddle point equations (S111), its effect is to make it energetically favorable to introduce a finite magnetic flux of the field aµ given by b = ∇ × a. The spatial distribution of the flux b is determined by solving the second equation in (S111). In the following, we will assume the flux b is spatially uniform. We will show at the end that for parameters of interest in MATBG, this assumption is true to a very good approximation. Since aµ corresponds to the Skyrmion density in the original problem, the b and the cyclotron frequency wc can be written in terms of the filling as

νπ 2 2 b = ∇ × a = = πνΛ wc = 2b = 2πνΛ (S124) AM 15 with ν being the ﬁlling fraction. In the presence of a magnetic ﬁeld, the spectrum of the operator −D2 + ∆2 corresponds to the Landau level spectrum 1 = k2 + w (n + 1/2) + ∆2, with degeneracy νΛ2A (S125) n,kz z c 2

The Green’s function G0(∆, r) is modiﬁed to Z 1 X dkz 1 1 G (∆, ν, r) = G (∆, ν) = νΛ2 − (S126) 0 0 2 2π k2 + 2πνΛ2(n + 1/2) + ∆2 k2 + 2πνΛ2(n + 1/2) + Λ2 n z z where we used the completeness of the wavefunctions within each Landau level to eliminate the r dependence. This integral converges for ∆2 > −πνΛ2 and can be evaluated as

Λr ν 1 1 ∆2 1 1 1 G (∆, ν) = ζ , + − ζ , + , (S127) 0 4 2π 2 2 2πνΛ2 2 2 2πν where ζ(s, q) is the Hurwitz zeta function deﬁned as [16]

∞ X 1 ζ(s, q) = (S128) (q + n)s n=0

2 For ν 6= 0, the saddle point equation (S111) is always satisfied since G0(∆, ν) is unbounded from above (we can choose ∆ to 2 2 be as close as we want to −πνΛ ) with the caveat being that the mass ∆ would have to be negative whenever g < Λ/G0(0, ν). This analysis implies that the purely ordered phase where z’s are gapless and aµ is Higgsed is immediately lost for any finite doping. This arises since doping forces magnetic flux into the system. In principle, we can have a vortex lattice or a coexistence phase as discussed in the main text. However, our current mean field scheme does not allow for any coexistence since non-zero doping immediately implies the loss of spin order z = 0. For the action (S102), the doping is controlled through the chemical potential. The optimal doping at a given chemical potential is obtained by finding the minimum of the action as a function of doping ν. To write the action for finite doping ν, let’s first consider the first term (in the following, we will consider the rescaled action S = NV S˜) 1 S˜ = tr ln−D† D + ∆2 (S129) 1 V µ µ We notice that the second derivative of this action with respect to ∆2 is finite and can be evaluated as

2 ˜ Z ∂ S1 1 X dkz 1 1 X 1 = − νΛ2 = − νΛ2 ∂(∆2)2 2 2π (k2 + ω (n + 1/2) + ∆2)2 8 (ω (n + 1/2) + ∆2)3/2 n z c n c 1 3 1 ∆2 = − √ ζ , + (S130) 16πΛ 2πν 2 2 2πνΛ2 This can be integrated to give a ﬁnite action with the help of two counter terms for the mass and free energy renormalization [1]

r ν 1 1 ∆2 S˜ [∆, ν] = νΛ3 ζ − , + + A(ν)∆2 + C(ν) (S131) 1 2π 2 2 2πνΛ2

2 A(ν) is fixed by the requirement that the first derivative of the action with respect to ∆ is equal to G0(∆, ν) given in (S127) while C(ν) is fixed by the requirement the the free energy vanishes for ∆ = 0 which yields

Λr ν 1 1 1 r ν 1 1 A(ν) = − ζ , + ,C(ν) = −νΛ3 ζ − , (S132) 4 2π 2 2 2πν 2π 2 2

Collecting everything gives ( ) S r ν 1 1 ∆2 1 1 1 1 ∆2 1 1 ∆2 ν2V˜ S˜ = = νζ − , + − ζ , + − νζ − , + − +µν ˜ + 0 NV Λ3 2π 2 2 2πνΛ2 4 2 2 2πν Λ2 2 2 gΛ2 2 (S133) 16 where Z Z 2 V0 2 2 1 2 e d µ V˜0 = ,V0 = Λ d rV (r) = d rV (r) = , µ˜ = (S134) c AM 20AM cΛ

∆2 can be eliminated by solving the saddle point equation

g g r ν 1 1 ∆2 1 1 1 1 = G (∆, ν) = ζ , + − ζ , + , (S135) Λ 0 4 2π 2 2 2πνΛ2 2 2 2πν √ The gap ∆ can be obtained by solving Eq. S135 in the limit g ν 1. Writing

∆2 = Λ2[−πν + ν2g2a2 + O(ν3g4)] (S136)

We then use the asymptotic expressions

x1−s 1 ζ(s, x 1) ≈ x−s + O(x−s+1), ζ(s, x 1) = + x−s + O(x−1−s) (S137) s − 1 2 Substituting in (S135), we ﬁnd 1 a = (S138) 4 which is valid for gν 1, g < gc. The phase boundary between the spin ordered phase and the superconductor can be determined by looking for the point at which the spin ordered phase becomes unstable, i.e. when the derivative of the Free energy relative to the doping is negative. Using (S136) and (S137) we ﬁnd that

∂S˜ 1 π πcΛ g g = − + +µ ˜ = 0 =⇒ µc = − 1 − = −4πρs 1 − (S139) ∂ν ν=0 4 g g gc gc This gives the phase boundary between the spin ordered state and the superconductor. This expression reduces to the energy of a single skyrmion 4πρs in the SU(2) limit when g gc. We can also estimate the stiffness and compressibility of the superconductor similar to the derivation leading to Eq. S122. In general, this requires recomputing (S117) using the propagator at ﬁnite doping. Such evaluation is in general complicated since momentum conservation is lost and we need to use the Landau level wavefunctions. In the following, we will assume we are in the limit ν 1 so that the discrete sum in G0(∆, ν) can be transformed into an integral. This is done by writing

1 X Z d Z d2k νΛ2 f(2πνΛ2n) = f() = f(k2) (S140) 2 4π (2π)2 n √ This allows us to evaluate the integral of (S117) with the main difference being the replacement ∆ 7→ ∆˜ = ∆2 + πνΛ2 leading to 3c 3 ρ ≈ |∆˜ |, χ ≈ |∆˜ |, ν 1 (S141) SC πN SC πNc √ Using the expressions (S136) which are valid for g ν 1, we get

3Λcνg 1 3Jν 3Λgν 1 3ν √ ρSC ≈ = , χSC ≈ = , g ν 1 (S142) 4πN N 2π 4πNc N 16πρsAM Remarkably, the stiffness and compressibility of the superconductor are inversely proportional to the compressibility and the stiffness of the insulator, respectively. At the end, let us revisit the justiﬁcation for assuming a uniform magnetic ﬂux b. To assess this assumption, let us write

b(r) = b + δb(r), b = πνΛ2 (S143)

Substituting in the second equation in (S111), we see that the constant part b drops in the second term. We can estimate the 17

V0 magnitude of the second term to be cΛ2 ij∂jδb(r). The ﬁrst term Gµ has dimensions inverse length square and its magnitude √1 2 can be estimated using the magnetic length lB ∼ Λ ν to be Gµ ∼ Λ ν ∼ b. Thus, we ﬁnd

1 Λc ij∂jδb(r) ∼ b (S144) Λ V0

For parameters relevant to MATBG, we find that Λc ∼ 1 meV while V0 ∼ 30 − 120 meV for the realistic values of the screening parameter d = 10 − 40 nm. Thus, the variation in the magnetic flux on the Moire´ scale is neglegible compared to the value of the flux. This justifies our assumption of uniform flux.

V. PAIRING SYMMETRY FOR THE SKYRMION SUPERCONDUCTOR

In this section, we will discuss the pairing symmetry for the skyrmion superconductor. We will begin by reviewing the different symmetries of MATBG and their implementation in the flat band projected sublattice-polarized basis following Ref. [7]. We will then discuss how these symmetries act on the CP1 variables and correspondingly on the monopole operator generating a 2π flux of the gauge field aµ. Finally, we will compare this to the action of symmetries on the different pairing channels.

A. Action of symmetries in the ﬂat bands

To compute the quantum numbers of the skyrmions under the different symmetries, it is instructive to start by reviewing the different symmetries of TBG. First, let us start by the microscopic symmteries acting on the microscopic annihilation operators fk which is a vector in valley (τ), sublattice (σ), and layer (µ). These can be divided into spatial and local symmetries. The former are given by

π −1 i 3 σz τz −1 −1 C3fkC3 = e fO3k,C2fkC2 = σxτxf−k,MyfkMy = σxµxfmy k (S145) where my(kx, ky) = (kx, −ky) and O3k corresponds to counter clockwise rotation by 2π/3. The local symmetries are

−1 −1 † −1 † † iφτz T fkT = τxf−k, PfkP = σxµyf−k, SfkS = σzfk,UvfkUv = e fk (S146) We notice that the unitary particle hole symmetry P is only a symmetry of the continuum model for small angles and the sublattice S only holds in the so-called chiral limit [24]. Here, following Ref. [7], we will assume all these symmetries are good symmetries which is equivalent to the sublattice polarized approximation discussed in the main text. This approximation relies on the fact that the wavefunctions in the flat bands still have significant sublattice polarization. The combination of S, P, and T symmetries implies the existence of a Z2 unitary symmetry R = iSPT = σyτxµy which anticommutes with τz, the generator of Uv. The resulting problem has an enlarged SU(2) symmetry generated by τz, R = σyτxµy, and iτzR = σyτyµy. This generates the SU(2) pseudospin rotation symmetry considered in the main text. We then project the operator fk to the flat band operator ck which is now labeled by sublattice index σ and valley index τ. In general, the representation of different symmetries depends on the gauge choice for the flat band wavefunctions uσ,τ (k). Once a gauge is fixed for one valley and sublattice, we can fix the gauge in the other valley and sublattice using two of the symmetries which relates different valleys and sublattices. Once this choice is fixed, we no longer have a gauge freedom and the representation of the remaining symmetries will in general have complicated k-dependence. Since we want a simple k- independent representation for the generators of SU(2) pseudospin rotation, we will use T and P symmetries to fix the gauge leading to the simple expressions T = τxK and P = τzσyK. In this gauge, the generators of SU(2) pseudospin rotations are σxτx, σxτy and τz. The remaining symmetries C2, C3, and My will generally have more complicated k-dependent forms in 0 this gauge. To see this, we will find it easier to define C2 = iτzRC2 = σzτzµy which is diagonal in valley and sublattice and satisfies

02 0 0 C2 = 1, [T ,C2] = 0, [R,C2] = 0 (S147) Its action on the ﬂat bands is generally given by

0 iθσ,τ (k) C2uσ,τ (k) = e uσ,τ (−k) (S148) Using (S147) implies

0 iθ(k) C2k = e σ0τ0, θ(−k) = −θ(k) (S149) 18

We note that we cannot set θ(k) to zero everywhere due to the band topology which requires the product of parity eigenvalues eiθ(k) at TRIMs to be −1 [10]. Similarly, C3 symmetry is also diagonal in valley and sublattice so it has the form

iφσ,τ (k) C3uσ,τ (k) = e uσ,τ (O3k) (S150)

0 iφσ,τ (k) Similar to the case of C2, φσ,τ (k) has to be k-dependent so that the product of e at C3-symmetric points is compatible 0 0 with the Chern number [10]. Commutation with C2T , R and C2 implies

iσz τz φ(k) C3k = e , φ(−k) = φ(k), (S151)

Finally, My exchanges sublattice but not valley

iασ,τ (k) Myuσ,τ (k) = e u−σ,τ (myk) (S152)

Commutation with R, C2T , and T implies

iα(k)σz τz Myk = σxe , α(−k) = α(k) (S153)

These symmetries can also be represented in terms of the Chern-pseudospin basis deﬁned in the main text (repeated in Eq. S3 of the supplemental mateiral)

γ = (σx, σyτz, σzτz), η = (σxτx, σxτy, τz) (S154)

We can then summarize the symmetries as follows

0 basis T P S SU(2) generators (τz, R, iτzR) C2 C2 = iτzRC2 C3 My π i σz τz microscopic τxK σxµyK σz (τz, σyτxµy, σyτyµy) σxτx σzτzµy e 3 σxµx iθ(k) iθ(k) iφ(k)σz τz iα(k)σz τz projected (σ, τ) τxK τzσyK σz (τz, τyσx, τxσx) σxτxe e e σxe iθ(k) iθ(k) iφ(k)γz iα(k)γz projected (γ, η) γxηxK γyK γzηz (ηz, ηy, ηx) ηxe e e γxe

B. Action of symmetries on the CP1 variables and the monopole operators

Using the results of table above, we can deduce the action of the different symmetries on the pseudospin vectors n± as well 1 1 as the CP variables z and the corresponding gauge flux fxy = π ij∂iaj. The SU(2) pseudospin rotation acts on the n± as an SO(3) rotation and it acts naturally as an SU(2) rotation on z. For the remaining symmetries, we will find it convenient to multiply them sometimes with a pseudospin rotation such that they either leave the pseudospin invariant and flip its direction. The action of the remaining symmetries is given in the Table below

0 0 ﬁeld T = iηzT = ηyγxK My C2 C3 n±(r) −n∓(r) n∓(myr) n±(−r) n±(O3r) ∗ z(r) z (r) z(myr) z(−r) z(O3r) fxy(r) fxy(r) −fxy(myr) fxy(−r) fxy(O3r)

Here, we have also included the Kramers time-reversal symmetry T 0 discussed in Ref. [7]. The cooper pair creation operator can be associated with the monopole operator ∆ which creates a 2π flux. Since this operator is a scalar, it transforms as a singlet under SU(2) spin rotation [13, 21, 22]. In addition, it transforms as a one-dimensional representation under the group of spatial 0 0 symmetries G generated by {C2,C3,My}. Notice here that we use C2 rather than C2 since only the former commutes with the generators of the SU(2) pseudospin rotation enabling us to decompose the symmetry representation into a spatial part and pseudospin part. 1 Within the CP model described in terms of the variable n = n+ = −n−, we cannot narrow down the properties of the 0 monopole operator further. However, it is possible to determine its angular momentum quantum numbers under C2 and C3 if we go back to the original n± variable and assumed that the monopole operator creates a 2π flux in one Chern sector and a −2π flux in the other sector, related to it by Kramers time-reversal symmetry. Denoting the flux creation operators in the ± Chern 0 0−1 0 sectors by ∆±, we can write ∆− = T ∆+T . Denoting the angular momentum quantum numbers of ∆+ under C2 and C3 19 by L2 and L3, respectively, we find that

0 0−1 iπL2 0 0−1 0 0 0−1 0−1 0 0 0−1 0−1 0 iπL2 0−1 −iπL2 C2∆+C2 = e ∆+ =⇒ C2∆−C2 = C2T ∆+T C2 = T C2∆+C2 T = T e ∆+T = e ∆− 2π 2π 2π −1 i 3 L3 −1 0 0−1 −1 0 −1 0−1 0 i 3 L3 0−1 −i 3 L3 C3∆+C3 = e ∆+ =⇒ C3∆−C3 = C3T ∆+T C3 = T C3∆+C3 T = T e ∆+T = e ∆− (S155)

0 0 Here, we used T commutes with C2 and C3. As a result, the monopole operator ∆ ∼ ∆+∆− has zero angular momentum 0 under both C2 and C3. In summary, ∆ is invariant under the action of SU(2) pseudospin rotations and the group of spatial symmetries G.

C. Transformation properties of the superconducting pairing channels

To relate the skyrmion superconductor to a weak pairing BCS superconductor, we investigate the properties of pairing functions in momentum space under the different symmetries in what follows. We restrict ourselves to pairing between time-reversal related momenta which has the form

T ∆k = ckc−k (S156) which behaves under unitary transformations as

T ck 7→ UkcOk =⇒ ∆k 7→ Uk∆OkU−k (S157)

To behave as a singlet under pseudospin SU(2) rotation ∆k has to be proportional to ηy. This means that the candidate pairing channels are

∆µ(k) = gµ(k)ηyγµ, µ = 0, x, y, z g0,x,z(−k) = g0,x,z(k), gy(−k) = −gy(k) (S158) where we used the fact that the full pairing function is antisymmetric by construction ∆(k)T = −∆(−k). To distinguish these pairing channels, we use the transformation properties under the group of spatial symmetries G. The symmetry transformations properties under C3 and My takes the form

−1 iφ(k)γz iφ(k)γz −1 iα(k)γz iα(k)γz C3∆(k)C3 = e ∆(O3k)e ,My∆(k)My = γxe ∆(myk)e γx (S159)

From this symmetry action, it is immediately obvious that ∆x,y transform as a one-dimensional representation of G. On the other hand, the transformation properties of ∆0,z can be understood by deﬁning instead the pairing channels 1 ± γ ∆ (k) = g (k)η z , g (−k) = g (k) (S160) ± ± y 2 ± ±

The action of C3 and My on ∆± is

−1 ±2iφ(k) −1 ±2iα(k) C3∆±(k)C3 = e ∆±(O3k),My∆±(k)My = e ∆∓(myk) (S161) Since the action of the symmetries is k dependent, it is instructive to focus on high symmetry points. In this regard, we notice that 0 0 the phase φ(k) has to be a multiple of 2π/3 at all C3 symmetric points: Γ, K, and K with φ(K) = φ(K ). Furthermore, due to the Chern number of the bands, the sum φ(K) + φ(K0) + φ(Γ) has to be equal to ±2π/3 modulo 2π [11]. This, in turn implies that either φ(Γ) 6= 0 or φ(K) 6= 0. In the ﬁrst case, ∆±(Γ) has nonzero and opposite angular momentum under C3 and are interchanged by My. Thus, ∆±(Γ) transform as a 2D irrep of G. In the second case, ∆±(K) has opposite angular momentum 0 under C3 and are interchanged by the action of Mx = C2My (which leaves the K invariant). Thus, ∆±(K) transforms as a 2D irrep under G. We conclude from this analysis that ∆±(k) transforms as a 2D irrep under G either at Γ or at K and as a result cannot correspond to the skyrmion superconductor which transforms as a 1D irrep under G. This leaves the two pairing channels ∆x,y discussed in the main text. These can be distinguished further by their behavior 0 under C2

0 0−1 iθ(k) iθ(−k) C2∆(k)C2 = e ∆(−k)e = ∆(−k) (S162)

0 0 Thus, ∆y is odd under C2 whereas ∆x is even under C2. Comparing with the transformation properties of the monopole operator, we deduce that the skyrmion superconductor can only be one of the channels ∆x. It should be emphasized, however, that the 20

1 selection between the channels ∆x,y relies on extra assumptions about the internal structure of the CP monopole operator in terms of the monopole operators in the n± variables.

D. Anticommutation of insulating and superconducting gaps

Here we discuss the relation between superconducting and insulating gaps, in particular we show that the pairing channels ∆x,y are also the ones that correspond to the maximal gap in the presence of the insulating pseudospin antiferromagnetic background. This is readily seen by checking that the corresponding matrices anticommute in the Nambu basis, thus the insulating and superconducting gaps add in quadrature when they are simultaneously present. To check the anticommutation, we perform a particle-hole transformation in the C = −1 sector only given by

T † c+,k 7→ c+,k, c−,k 7→ c−,−kηy (S163)

† Under this transformation, the gap functions for the pseudospin antiferromagnets ckγzηx,y,zck are invariant while the superconducting gap maps to the excitonic insulating gap

† † ∆x(k) + h.c. 7→ ck[γx Re gx(k) + γy Im gx(k)]ck, ∆y(k) + h.c. 7→ ck[γy Re gy(k) − γx Im gy(k)]ck (S164) which manifestly anticommutes with the insulating gap functions implying that the gaps will add in quadrature. Taking into account the kinetic part h, favors the pairing channel ∆x which anticommutes with h relative to ∆y which commutes with h.

VI. THEORY OF THE INTERTWINED INSULATOR AND SUPERCONDUCTOR ORDER PARAMETERS

In the following, our goal is to derive an effective field theory which deals with the superconducting and insulating order parameters on equal footing. Similar to Appendix I, we will perform a general number of flavors n which reveals the general structure of the theory even though in this work we are focusing exclusively on the case n = 1. As a result, we will employ a more general approach than the ones used in Ref. [12] which uses the fact that the mass term has the form of a unit vector in Sd. Instead, our approach is similar to the derivations employed to derive effective theories in the contexts of disordered systems [2–4]. In particular, we will follow the derivation in Ref. [3] very closely. To derive an effective theory for the insulating and superconducting phases it is helpful to go close to the point g = gc where there is a perfect symmetry rotating the two into each other. We consider the weak coupling limit where the order parameter M constitutes a small mass term to the Dirac equation. Although the relation of the parameters of the resulting theory to the microscopic parameters will not be valid for the strong coupling regime where M is a lot larger than the bandwidth, we expect the form of the action, which is fixed by symmetries, to remain valid. To this end, we expand the non-interacting Hamiltonian in 0 the vicinity of the Moire´ Dirac points KM and KM as

X † † 0 HD = vF [ψ (kxγx + kyγy)ψKM ,k − ψ 0 (kxγx + kyγy)ψK ,k] (S165) KM ,k KM ,k M k

0 The relative negative sign is ﬁxed by time-reversal symmetry which has the form T = γxηxK and maps KM to KM . To deal with the insulator and superconductor on equal footing, we introduce the Nabmu basis deﬁned as

T T † χk = (ψK ,k, ψ 0 ) (S166) M KM ,−k This form is inspired by the fact that the superconducting pairing takes place between states at opposite momenta and opposite Moire´ valleys. We can now introduce the Pauli matrices ρx,y,z which act in the Nambu space and rewrite the Dirac Hamiltonian (S165) as

X † HD = vF χk(kxγxρz + kyγy)χk (S167) k

By construction, the Nambu Hamiltonian HD is invariant under the particle-hole transformation P = γzρyK which satisﬁes P2 = −1 (class C). The analysis can be further simpliﬁed by introducing the Pauli matrices

α = (γxρz, γy, γzρz), β = (γyρy, γyρx, ρz) (S168) 21

In this basis, the Dirac Hamiltonian has the form HD ∝ kxαx + kyαy and P = αxβyK. The most general mass term for the Dirac Hamiltonian takes the form M˜ = αzM with

∗ ∗ αxβyM˜ αxβy = −M˜ =⇒ βyM βy = +M (S169)

If we impose in addition the condition M 2 = 1, this means that M parametrizes the symplectic Grassmanian manifold Sp(4n)/Sp(2n) × Sp(2n) which for n = 1 is isomorphic to S4. In this case, we can write M explicitly in terms of the vector n represting the insulating phases and the complex number ∆ representing the superconductor or alternatively an SO(5) vector n as 5 n · η ∆ηy X ˆ M = ∗ T = niΓi, Γ = (ηx, ηyβz, ηz, ηyβx, ηyβy) (S170) ∆ ηy n · η β i=1

The full massive Dirac action for for χ becomes Z 2 † S = dτd rχ [∂τ − iαx∂x − iαy∂y − µβz + αzmM]χ (S171)

† where we have included the chemical potential µ. We now deﬁne χ¯ = χ αz to get Z 2 ¯ ˜ S = dτd rψ[˜αν ∂ν − µα˜zβz + mM]ψ, α˜ = αz(α0, iαx, iαy) = (αz, −αy, αx) (S172)

We can now integrate out the fermions to get an effective action of the matrix ﬁeld M leading to

Seﬀ = − Tr ln[˜αν ∂ν − µα˜zβz + mM] (S173)

This expression is UV-divergent and needs to be regularized before any further manipulation. In the following, we follow Refs. [2–4] by subtracting off the following action

S0 = − lim Tr ln[˜αν ∂ν − µα˜zβz + M] (S174) →0+

In the limit → 0+, the M-dependence of this action drops and it contributes an inessential constant. On the other hand, for 2 large momenta, this term cancels the UV-divergence of Seﬀ . We can now use the condition M = 1 to write

M(x, τ) = U †(x, τ)ΛU(x, τ), Λ2 = 1, tr Λ = 0 (S175) where Λ is just a constant matrix which can then be substituted in the gradient expansion. Since the action is now UV-ﬁnite, we can use the cyclic property of the trace to get

† † † Seﬀ = − Tr ln[˜αν ∂ν +α ˜ν Aν − µα˜zR + mΛ] − S0,Aν = U∂ν U = −∂ν UU ,R = UβzU (S176)

We can now expand the action in powers of µ and the gradients Aµ

∞ X 1 n S = − Tr ln[G−1(m) − Σ] − S = − Tr ln G−1 − S + Tr(G (m)Σ) (S177) eﬀ 0 0 0 0 n 0 n=1 with G0 and Σ given by

1 −ikν α˜ν + mΛ ˜ ˜ G0(k) = = 2 2 , Σ =α ˜ν Aν , Aν = Aν − µ δν,0R (S178) ikν α˜ν + mΛ k + m The quadratic term in the action can be written as

1 X 1 n o S = m2 trα ˜ α˜ tr ΛA˜ (q)ΛA˜ (−q) − k (k + q ) trα ˜ α˜ α˜ α˜ tr A˜ (q)A˜ (−q) 2 2 (k2 + m2)((k + q)2 + m2) µ ν µ ν ρ λ λ µ ρ ν λ µ ν k,q (S179) In the long wavelength limit, we can set q = 0. In this case, we can use the symmetry of the integral under k 7→ −k and the 22

2 rotation symmetry to reduces kρkλ = δρλk /3. In addition, the traces over products of α˜ can be evaluated as X trα ˜µα˜ν = 2δµν , trα ˜µα˜ρα˜ν α˜ρ = −2δµν (S180) ρ

Evaluating the momentum integrals in S2 leads to V Z 1 1 mV mV S = dkk2 m2 tr ΛA˜ ΛA˜ + k2 tr A˜ A˜ = tr[(ΛA˜ )2 − A˜2 ] = tr[Λ, A˜ ]2 (S181) 2 2π2 (k2 + m2)2 ν ν 3 ν ν 8π ν ν 16π ν

Here, in evaluating the momentum integral, we have subtracted off a term from S0 to cancel the UV divergence of the second term. The result can be written in terms of the mass matrix M as Z ρ Z ρ S = d2rdτ ν tr[Λ, A˜ ]2 = d2rdτ ν tr(D M)2, D M = ∂ M − µδ [β ,M] (S182) 2 8 ν 8 ν ν ν ν,0 z where m m ρi =ρ ˜ = , ρτ =χ ˜ = 2 (S183) 2π 2πvF For n = 1, this can be written explicitly by substituting (S170) leading to

Z ρ˜ χ˜ S = d2rdτ [(∇n)2 + |∇∆|2] + [(∂ n)2 + |∂ ∆|2] + 2˜χµ∆∗∂ ∆ − 2˜χµ2|∆|2 (S184) 2 2 2 τ τ τ

The last term implies that the chemical potential always disfavors the insulator. We notice that Aµ is of the same order as the momentum qµ. Thus, the third order term has two contributions. First there is the term ∼ A3

1 X 1 n 3 o S 3 = m trα ˜ α˜ α˜ tr ΛA˜ ΛA˜ ΛA˜ − 3k k m trα ˜ α˜ α˜ α˜ α˜ tr ΛA˜ A˜ A˜ 3,A 3 (k2 + m2)3 µ ν λ µ ν λ ρ δ µ ρ ν δ λ µ ν λ k 2iµνλm X 1 n o = m2 tr ΛA˜ ΛA˜ ΛA˜ + k2 tr ΛA˜ A˜ A˜ (S185) 3 (k2 + m2)3 µ ν λ µ ν λ k

Here, both momentum integrals are UV ﬁnite and there are no corrections coming from S0 and we have used

trα ˜µα˜ν α˜λ = 2iµνλ (S186)

Evaluating the momentum integrals yields Z i 2 S 3 = d rdτ {tr ΛA˜ ΛA˜ ΛA˜ + 3 tr ΛA˜ A˜ A˜ } (S187) 3,A 48π µνλ µ ν λ µ ν λ

Second, there is the term proportional to qA2 which is obtained from the second order term in the gradient expansion of the action

im X qλ V µνλ X S 2 = − trα ˜ α˜ α˜ tr ΛA˜ (q)A˜ (−q) = − q tr ΛA˜ (q)A˜ (−q) 3,qA 2 (k2 + m2)2 µ λ ν µ ν 8π λ µ ν k,q q i Z = − µνλ d2rdτ tr ΛA˜ ∂ A˜ (S188) 8π µ λ ν

† The part not containing µ is obtained by replacing A˜µ by Aµ. Using ∂ν Aµ = AµAν − (∂µ∂ν U)U , it can be written as i Z S = d2rdτ {tr ΛA ΛA ΛA − 3 tr ΛA A A } (S189) 3 48π µνλ µ ν λ µ ν λ

This part has the form of a Chern-Simons term in the non-Abelian gauge ﬁeld Aµ and it is not manifestly gauge invariant. In fact, under a gauge transformation U 7→ KU, with [K, Λ] = 0, it changes by a total derivative + a topological contribution associated with large gauge transformations which cancels against a corresponding contribution from the regulator S0 [19, 20] 23

(see Ref. [3] for a detailed discussion). As a result, this term is only gauge invariant on a manifold without a boundary and it cannot be written explicitly in terms of M without introducing an extra dimension. To write it in the more familiar form, we promote Aµ to depend on the extra variable u and notice that 3 ∂ [tr ΛA ΛA ΛA −3 tr ΛA A A ] = −3 [tr ΛA ΛA ΛA A −tr ΛA A A A ] = tr M∂ M∂ M∂ M∂ M µνλρ ρ µ ν λ µ ν λ µνλρ µ ν λ ρ µ ν λ ρ 8 µνλρ µ ν λ ρ (S190) which means that we can write Z 1 Z i 2 S3,0 = du d rdτµνλρ tr M∂µM∂ν M∂λM∂ρM (S191) 128π 0

For the spinless model where M = naΓa with a = 0,..., 4, this reduces to

Z 1 Z i 2 S3,0 = du d rdτµνλρabcdena∂µnb∂ν nc∂λnd∂ρne 32π 0 Z 1 Z 3 2 = 2πi 2 du d rdτabcdena∂unb∂τ nc∂xnd∂yne (S192) 8π 0 which is the well-known WZW term. Finally, we can extract the linear term in µ in S3 which is given by iµ Z S = d2rdτ [tr ΛA ΛA ΛR − tr ΛA A R − tr ΛA RA − tr ΛRA A ] (S193) 3,µ 16π ij i j i j i j i j

Now we notice that ∂µR = [R,Aµ] which means that the term ij tr ΛAiAjR is a total derivative since

ij∂j tr ΛAiR = ij tr Λ(AiAjR + AiRAj − AiAjR) = ij tr ΛAiRAj (S194)

As a result, we can rewrite (S193) as iµ Z iµ Z S = d2rdτ tr RΛ[Λ,A ][Λ,A ] = d2rdτ tr β M∂ M∂ M (S195) 3,µ 16π ij i j 16π ij z i j For n = 1, we can substitute (S170) in the above expression yielding µ Z S = d2rdτ n · (∂ n × ∂ n) (S196) 3,µ 4π ij i j Combining all the terms yields the effective action (18) in the main text.

[S1] Ian Affleck and Edouard Brezin.´ Superconducting abrikosov transition in the large-n limit. Nuclear Physics B, 257:451 – 473, 1985. [S2] Alexander Altland and Dmitry Bagrets. Effective field theory of the disordered weyl semimetal. Phys. Rev. Lett., 114:257201, Jun 2015. [S3] Alexander Altland and Dmitry Bagrets. Theory of the strongly disordered weyl semimetal. Phys. Rev. B, 93:075113, Feb 2016. [S4] Alexander Altland, B.D. Simons, and M.R. Zirnbauer. Theories of low-energy quasi-particle states in disordered d-wave superconductors. Physics Reports, 359(4):283 – 354, 2002. [S5] I Ya Aref’eva and SI Azakov. Renormalization and phase transition in the quantum ofcpn- 1 model (d= 2, 3). Nuclear Physics B, 162(2):298–310, 1980. [S6] D. P. Arovas, A. Karlhede, and D. Lilliehoök.¨ SU(n) quantum hall skyrmions. Phys. Rev. B, 59:13147–13150, May 1999. [S7] Nick Bultinck, Eslam Khalaf, Shang Liu, Shubhayu Chatterjee, Ashvin Vishwanath, and Michael P. Zaletel. Ground State and Hidden Symmetry of Magic Angle Graphene at Even Integer Filling. arXiv e-prints, page arXiv:1911.02045, Nov 2019. [S8] Shubhayu Chatterjee, Nick Bultinck, and Michael P. Zaletel. Symmetry breaking and skyrmionic transport in twisted bilayer graphene. Phys. Rev. B, 101:165141, Apr 2020. [S9] A d’Adda, M Luscher,¨ and P Di Vecchia. A 1n expandable series of non-linear σ models with instantons. Nuclear Physics B, 146(1):63– 76, 1978. [S10] Chen Fang, Matthew J. Gilbert, and B. Andrei Bernevig. Bulk topological invariants in noninteracting point group symmetric insulators. Phys. Rev. B, 86:115112, Sep 2012. [S11] Chen Fang, Matthew J Gilbert, and B Andrei Bernevig. Bulk topological invariants in noninteracting point group symmetric insulators. PHYSICAL REVIEW B, 86:115112, 2012. [S12] Pavan Hosur, Shinsei Ryu, and Ashvin Vishwanath. Chiral topological insulators, superconductors, and other competing orders in three 24

dimensions. Phys. Rev. B, 81:045120, Jan 2010. [S13] Shang Liu, Ashvin Vishwanath, and Eslam Khalaf. Shift Insulators: Rotation-Protected Two-Dimensional Topological Crystalline Insulators. Physical Review X, 2019. [S14] Imam Makhfudz, Benjamin Kruger,¨ and Oleg Tchernyshyov. Inertia and chiral edge modes of a skyrmion magnetic bubble. Phys. Rev. Lett., 109:217201, Nov 2012. [S15] K. Moon, H. Mori, Kun Yang, S. M. Girvin, A. H. MacDonald, L. Zheng, D. Yoshioka, and Shou-Cheng Zhang. Spontaneous interlayer coherence in double-layer quantum hall systems: Charged vortices and kosterlitz-thouless phase transitions. Phys. Rev. B, 51:5138– 5170, Feb 1995. [S16] Frank WJ Olver, Daniel W Lozier, Ronald F Boisvert, and Charles W Clark. NIST handbook of mathematical functions hardback and CD-ROM. Cambridge university press, 2010. [S17] Alexander M. Polyakov and A. A. Belavin. Metastable States of Two-Dimensional Isotropic Ferromagnets. JETP Lett., 22:245–248, 1975. [Pisma Zh. Eksp. Teor. Fiz.22,503(1975)]. [S18] A.M. Polyakov. Gauge Fields and Strings. Contemporary concepts in physics. Taylor & Francis, 1987. [S19] A. N. Redlich. Gauge noninvariance and parity nonconservation of three-dimensional fermions. Phys. Rev. Lett., 52:18–21, Jan 1984. [S20] A. N. Redlich. Parity violation and gauge noninvariance of the effective gauge ﬁeld action in three dimensions. Phys. Rev. D, 29:2366– 2374, May 1984. [S21] T. Senthil, Leon Balents, Subir Sachdev, Ashvin Vishwanath, and Matthew P. A. Fisher. Quantum criticality beyond the landau- ginzburg-wilson paradigm. Phys. Rev. B, 70:144407, Oct 2004. [S22] Xue-Yang Song, Yin-Chen He, Ashvin Vishwanath, and Chong Wang. From spinon band topology to the symmetry quantum numbers of monopoles in Dirac spin liquids. Technical report. [S23] Michael Stone. Magnus force on skyrmions in ferromagnets and quantum hall systems. Phys. Rev. B, 53:16573–16578, Jun 1996. [S24] Grigory Tarnopolsky, Alex Jura Kruchkov, and Ashvin Vishwanath. Origin of magic angles in twisted bilayer graphene. Phys. Rev. Lett., 122:106405, Mar 2019. [S25] Edward Witten. Instatons, the quark model, and the 1/n expansion. Nuclear Physics B, 149(2):285–320, 1979.