Topological superconductivity in a spin-orbit coupled Kondo lattice

Though the first heavy fermion materials were discovered half a century ago[1, 2, 3, 4], they continue to challenge our fundamental understanding of strongly correlated quantum materials. The discovery of topological insulators[5, 6], with helical boundary modes protected by nontrivial topology and time-reversal symmetry, inspired the prediction that analogous topological Kondo insulators, may develop in Kondo systems with strong spin-orbit coupling, such as SmB₆ [7, 8, 9]. On the other hand, the possible existence of topological superconductors in Kondo systems remains elusive, motivating ongoing theoretical and experimental research.

Recently, a heavy-fermion superconductor UTe₂ with an extremely high upper critical field exceeding 65 T has attracted enormous attention[10, 11, 12, 13]. One of the exciting features of this material, is that it may provide a realization of a long-sought chiral topological superconductor [14, 15, 16], as evidenced by the observation of chiral edge states in scanning tunneling experiments [17]. The presence of chiral edge modes implies that the time-reversal symmetry (TRS) must be spontaneously broken, consistent with the early Kerr rotation experiment results [14]. However, the nature of the superconducting transition remains unclear and controversial: in improved samples, the Kerr rotation signal appears to be absent and the superconducting transition seems to be a single second-order transition, which at first sight rules out a TRS superconductor, as the underlying crystal structure allows no multidimensional irreducible Cooper pair representations [18, 19].

In this Letter, we propose a model that describes a novel charge-$e$ topological superconductor sharing a few exotic properties with that of UTe₂. Firstly, it models a superconductor with spontaneous TRS breaking guaranteed by Kramer’s theorem, regardless of the space group of the underlying lattice. Secondly, it is topological and possesses a chiral Majorana edge mode. While our model does not describe UTe₂, the new symmetry class it represents may be of interest as a novel pairing state of non-trivial topology that lies beyond the BCS paradigm. Previous efforts to explore topological superconductivity in the Kondo lattice have relied on the Kitaev Kondo model[20]. Our model is an extension of the Kondo Lattice model considered by Coleman, Panigraphi and Tsvelik (CPT) [21, 22], which describes a phase with charge-$e$ spinor order parameter and gapless Majorana excitation. Unlike the Kitaev-Kondo lattice[20], the CPT model allows the inclusion of Kondo interactions that do not disturb the static gauge degrees of freedom associated with spin fractionalization. Our generalization of the CPT model incorporates spin-orbit coupling, a realistic feature of heavy fermion materials. We also introduce nearest-neighbor Kondo interaction, which allows for a spontaneous chiral (or sublattice) symmetry breaking and the consequential development of chiral edge states. We demonstrate that, for sufficiently large Kondo coupling and nonzero Rashba interaction, at half-filling, the system spontaneously breaks the global $U(1)$ and discrete TRS and chiral symmetries of the original model ($U(1)\times Z_{2}\times Z_{2}$ symmetry). The gapless Majorana band of the original CPT model becomes gapped, acquiring a nontrivial Chern number and forming chiral Majorana edge modes. When the chemical potential is finite, our model describes a TRS-breaking charge-$e$ topological superconductor, the transition of which from the normal state is of single-step.

We consider coupling a Yao-Lee (YL) spin liquid to a conduction sea on the honeycomb lattice described by the Hamiltonian $H=H_{C}+H_{YL}+H_{K}$, where

Here $H_{C}$ describes the conduction electron Hamiltonian in the presence of Rashba spin-orbit interaction where $I$ is the identity matrix, $\vec{\sigma}$ is the Pauli matrix associated with spins, $\langle ij\rangle$ denotes the pair of nearest-neighbor (NN) sites $(i,j)$ where $i\in A,j\in B$, $\vec{d}_{ij}$ represents the vector pointing from site $j$ to $i$, and we have used the shorthand $c_{j}^{\dagger}\equiv(c_{j\uparrow}^{\dagger},c_{j\downarrow}^{\dagger})$. $H_{YL}$ describes the Yao-Lee spin liquid[23], in which $\vec{\sigma}_{i}$ and $\vec{\lambda}_{i}$ are Pauli matrices denoting the spin and orbital degrees of freedom at each site $i$, while $\alpha_{ij}=x,y,z$ denotes the direction of the Ising coupling between neighboring orbitals (see Fig. 1). $H_{K}$ describes the Kondo coupling between the conduction electrons and local moments, where $J_{ij}=J_{1}\delta_{ij}+J_{2}\eta_{\langle ij\rangle}$ where $\eta_{\langle ij\rangle}=1$ when $i$ and $j$ are nearest neighbors and $J_{1,2}\geq 0$ are antiferromagnetic.

It is convenient to make a global gauge transformation for conduction electrons on the $A$ sublattice, $(c_{A}^{\dagger},c_{B}^{\dagger})\rightarrow(ic_{A}^{\dagger},c_{B}^{\dagger})$, so $H_{C}$ becomes

The YL model can be exactly solved using the Majorana representation $\vec{\sigma}_{j}=-i\vec{\chi}_{j}\times\vec{\chi}_{j}$ and $\vec{\lambda}_{j}=-i\vec{b}_{j}\times\vec{b}_{j}$ with the constraint $D_{j}=8i\chi_{j}^{1}\chi_{j}^{2}\chi_{j}^{3}b_{j}^{1}b_{j}^{2}b_{j}^{3}=1$ [24, 25, 23]. Here the Majorana fermions $\chi,b$ obey anticommutation relations $\{\chi_{i}^{\alpha},\chi_{j}^{\beta}\}=\{b_{i}^{\alpha},b_{j}^{\beta}\}=\delta_{ij}\delta^{\alpha\beta}$ and $\{\chi_{i}^{\alpha},b_{j}^{\beta}\}=0$. Equation (2) then becomes

where $u_{ij}=-2ib_{i}^{\alpha_{ij}}b_{j}^{\alpha_{ij}}=\pm 1$ are independent, static $Z_{2}$ gauge fields that commute with the Hamiltonian. Note that unlike a Kitaev Kondo lattice, where the spin-operator creates visons, the Yao-Lee structure of the CPT model means that the spin operator $\vec{S}_{j}$ commutes with the gauge fields so that action of the Kondo interaction does not disturb the static gauge fields. The ground state of Eq. (5) lies in the fluxless gauge sector [26], where the gauge fluxes $W_{p}=\prod_{i,j\in p}u_{ij}=1$ (here we use the convention $i\in A,j\in B$). Provided the flux gap is larger than all other relevant energy scales, we can fix the ground-state gauge to be $u_{ij}=1$ whereupon Eq. (5) becomes

In terms of Majorana fermions $\vec{\chi}$, the Kondo interaction $H_{K}$ becomes

To construct a mean-field solution we need to understand the underlying symmetries. The Hamiltonian $\tilde{H}=\tilde{H}_{C}+H_{YL}+H_{K}$ (4,6,7) has three discrete symmetries:

(1) Time-reversal $\mathcal{\tilde{T}}=\mathcal{TG}$, where $\mathcal{T}c_{j}\mathcal{T}^{-1}=i\sigma_{y}c_{j}$, $\mathcal{T}i\mathcal{T}^{-1}=-i$, $\mathcal{T}\vec{\chi}_{j}\mathcal{T}^{-1}=\vec{\chi}_{j}$ and $\mathcal{G}$ is the global $Z_{2}$ gauge transformation that flips the signs on the B sublattice: $\mathcal{G}c_{A(B)}\mathcal{G}^{-1}=\pm c_{A(B)}$, $\mathcal{G}\vec{\chi}_{A(B)}\mathcal{G}^{-1}=\pm\vec{\chi}_{A(B)}$.

(2) Spin-lattice rotation $\mathcal{C}=\mathcal{RUG}$, where $\mathcal{R}$ and $\mathcal{U}$ rotate the lattice and spin clockwise by $\pi/3$: $\mathcal{U}c_{j}\mathcal{U}^{-1}=\exp({i\frac{\pi}{6}\sigma^{z}})c_{j}$, $\mathcal{U}\vec{\chi}_{j}\mathcal{U}^{-1}=R^{-1}_{z}(\frac{\pi}{3})\vec{\chi}_{j}$, where $R_{z}(\phi)$ is the $SO(3)$ matrix that rotates a vector around $z$-axis by $\phi$; $\mathcal{R}c_{j}\mathcal{R}^{-1}=c_{j^{\prime}}$, $\mathcal{R}\vec{\chi}_{j}\mathcal{R}^{-1}=\vec{\chi}_{j^{\prime}}$, where $\mathbf{r}_{j^{\prime}}=R_{z}(\frac{\pi}{3})\mathbf{r}_{j}$ and $\mathbf{r}_{j}$ is the position vector of site $j$. The additional $Z_{2}$ gauge transformation ${\mathcal{G}}$ in $\mathcal{C}$ is needed to preserve the rotational invariance of the hopping terms in the Hamiltonian.

(3) Sublattice particle-hole, or “chiral” symmetry: $\mathcal{S}$: $\mathcal{S}i\mathcal{S}^{-1}=-i$, $\mathcal{S}c_{A(B)}\mathcal{S}^{-1}=\pm c^{\dagger}_{A(B)}$, and $\mathcal{S}\vec{\chi}_{A(B)}\mathcal{S}^{-1}=\pm\vec{\chi}_{A(B)}$.

With a Hubbard-Stratonovich transformation, the local and nearest-neighbor Kondo interactions become [21, 22]

where $V_{j}$ and $V_{ij}$ are charge-$e$ spinors: stationarity with respect to variations in these spinors imposes the self-consistency conditions:

If $J_{2}=0$, beyond a critical value of $J_{1}$, a uniform spinor condensate develops, in which $V_{j}=W_{1}v_{j}/\sqrt{2}$, where $W_{1}>0$ and we may take $v_{j}=e^{i\alpha}(1,0)^{T}$ [22].

When $J_{2}>0$, a second spinor order parameter $V_{ij}=W_{2}v_{ij}/\sqrt{2}$ ($W_{2}>0$) also develops at sufficiently large $J_{2}$. As we will show below, when $W_{1}$ or $W_{2}$ becomes nonzero, the model describes a superconductor, which has superfluid stiffness proportional to $W_{1}^{2},W_{2}^{2}$. The ansatz for $v_{ij}$ is parametrized by

which is the spin-down spinor along the axis $\hat{n}=(\sin\gamma\cos(\phi+\theta_{ij}),\sin\gamma\sin(\phi+\theta_{ij}),\cos\gamma)$. Here $\theta_{ij}$ is the angle between $\vec{d}_{ij}$ and $-\hat{y}$, as shown in Fig. 2. The ansatz (13) has been chosen so that both $v_{i}$ and $v_{ij}$ transform under the same spin $\frac{1}{2}$ representation under a $60^{\circ}$ spin-lattice rotation: $v_{i}\rightarrow e^{i\pi/6}v_{i}$, $v_{ij}\rightarrow e^{i\pi/6}v_{ij}$.

The low-energy degrees of freedom of the system involve fluctuations in the overall phase $\alpha$, which couple to the external electromagnetic field. Under a spin-lattice rotation $\mathcal{C}$, $e^{i\alpha}\rightarrow e^{i(\alpha+\pi/6)}$ while the relative $U(1)$ phase $\beta$ is unchanged, indicating that $v_{j}$ and $v_{ij}$ transform under the same one-dimensional irreducible representation (see Appendix A for a detailed discussion of the symmetry transformation of the order parameters). We have verified that the energy is minimized for $\beta=0,\pi$. This residual $Z_{2}$ degeneracy originates from spontaenous chiral symmetry breaking: under the chiral operation $\mathcal{S}$, $v_{j}\rightarrow-v_{j}$ while $v_{ij}\rightarrow v_{ij}$. It is this broken chiral symmetry that is responsible for chiral edge states, demonstrated below. The other two degrees of freedom $\phi$ and $\gamma$ are generally gapped and hence do not affect the low-energy physics. Kramers’ theorem implies there are two other states which are the time-reversal partners of the above $Z_{2}$ states, with order parameters given by $v_{j}\rightarrow-i\sigma_{y}v_{j}^{*}$, $v_{ij}\rightarrow i\sigma_{y}v_{ij}^{*}$, leading to 4-fold degenerate $U(1)\times Z_{2}\times Z_{2}$ state, where time-reversal, chiral, and electromagnetic $U(1)$ gauge symmetry are spontaneously broken. From now we only focus on the representative case $v_{j}=(1,0)^{T}$ by gauge-fixing $\alpha=0$ and absorbing the phase factor $e^{i\beta}=\pm 1$ into $W_{2}$ by allowing it to acquire both positive and negative values. In fact, we find the energy is minimized only when $\phi=\pm\pi/2$, allowing us to fix $\phi=-\pi/2$ and $\gamma\in\left[-\pi,\pi\right]$. Without loss of generality, we can take $\lambda_{R}$ in (4) to be positive, noting that the opposite case is obtained by a global spin rotation through 180^∘ around the $z$-axis, under which $\lambda_{R}\rightarrow-\lambda_{R}$, $\gamma\rightarrow-\gamma$ and $W_{2}\rightarrow-W_{2}$.

It is useful[22] to decompose the conduction electrons into a scalar and vector Majorana components, $c^{0}$ and $\vec{c}$ respectively,

The key feature that then emerges, is that when $W_{2}=\lambda_{R}=0$, the vector components $c^{\alpha}_{j}$ ($\alpha=1,2,3$) develop a gap by selectively hybridizing with the Yao-Lee spin liquid, while the scalar components $c^{0}_{j}$ decouple, forming a gapless Majorana sea. The corresponding Hamiltonian is

where $\vec{\tau}$ are the Pauli matrices spanning the sublattice degrees of freedom and the form-factor $A_{\mathbf{k}}=1+e^{-i\mathbf{k}\cdot\mathbf{a_{1}}}+e^{-i\mathbf{k}\cdot\mathbf{a_{2}}}$ (where $\mathbf{a_{1,2}}=(\pm\sqrt{3}/2,3/2)$ are the Bravais lattice vectors) has been rewritten as a vector $\vec{A}_{\mathbf{k}}=-(\text{Im}A_{\mathbf{k}},\text{Re}A_{\mathbf{k}},0)$. Here we have defined Fourier transforms of Majorana fermions $c^{\alpha}$, $\vec{\chi}$

where $\mathbf{R}_{j}$ is the position of the unit cell. The summations run over half of the Brillouin zone (BZ), due to the redundancies $\vec{\chi}_{\mathbf{k}}^{\dagger}=\vec{\chi}_{\mathbf{-k}}$ and $c_{\mathbf{k}}^{\alpha\dagger}=c_{\mathbf{-k}}^{\alpha}$.

The energies of the gapped vector fermions are $\pm E_{\mathbf{k}}^{+}$ and $\pm E_{\mathbf{k}}^{-}$, where

while the energy of the gapless scalars, $c^{0}$ is $E_{\mathbf{k}}^{0}=\pm t|{A}_{\mathbf{k}}|$.

When $\lambda_{R}$ and $W_{2}$ are nonzero, matrix elements develop which link the low-energy subspace $|\mathcal{L}\rangle$ of gapless scalars $c^{0}$ to the high-energy subspace $|\mathcal{H}\rangle$ of gapped vector Majoranas (See Appendix B for the full mean-field Hamiltonian). Using second-order perturbation theory (see Appendix C for detailed calculation), the low-energy effective Hamiltonian near the high-symmetry momenta $K(K^{\prime})=(\pm 4\pi/3\sqrt{3},0)$ is given by

where $v_{F}=3t/2$ is the Fermi velocity, $m=9W_{2}\lambda_{R}\cos(\gamma/2)/W_{1}$ is the mass and we have taken $t=K$ for simplicity. It is important to note that the mass has an opposite sign at different valleys $K,K^{\prime}$, indicating that the low-energy band is topological with a non-trivial Chern number $\mathcal{C}=\text{sgn}(m)$, resembling the Kitaev and Haldane models [25, 27].

As the high-energy bands are topologically trivial and the order parameter carries $U(1)$ electric charge, we conclude that the system is a charge-$e$ topological superconductor, which, at the mean-field level, belongs to the symmetry class D in the Altland-Zirnbauer classification [28, 29]. Due to the bulk-edge correspondence and the fact that the low-energy gapped excitations are Majorana fermions (or complex fermions living in half of the Brillouin zone), there must be a chiral Majorana edge mode, as shown in Fig. 3. For larger $W_{2}$ and $\lambda_{R}$ where the perturbative analysis breaks down, the bulk gap may close and reopen, leading to other topological phases with different Chern numbers. In Fig. 4, we show the $W_{1}$-$W_{2}$ phase diagram for $\gamma=0$ obtained by exact diagonalization of the mean-field Hamiltonian, from which one finds phases with other higher Chern numbers (see Appendix D for phase diagrams at other values of $\gamma$). We note that the phase diagram is symmetric about $W_{2}=0$ and the Chern number changes sign if $W_{2}$ is opposite. This is a consequence of the chiral symmetry, recalling that the chiral operation reverses the chirality (velocity) of the edge modes as well as the sign of $W_{2}$.

The $J_{1}-J_{2}$ phase diagrams are obtained by solving the self-consistent mean-field equations numerically (see Appendix B for the expression of mean-field equations) and depicted in Fig. 5(a)(b). For sufficiently large $J_{1}$ and $J_{2}$, a phase region with nonzero $W_{1}$ and $W_{2}$ exists. Remarkably we find both the $\mathcal{C}=\pm 1$, $\mathcal{C}=\mp 2$ topological superconductor phases, which are separated by a first-order transition with a sudden jump of order parameters.

It is essential to note that the second-order transition from the disordered phase to the topological phase is generally two-step. This can be understood by writing down the Landau theory for the order parameters $W_{1},W_{2}$: $\mathcal{L}=c_{1}W_{1}^{2}+c_{2}W_{2}^{2}+...$, where the $W_{1}W_{2}$ term is absent due to the chiral symmetry. As $W_{1}$ and $W_{2}$ are not coupled to the lowest order, one expects that their condensations are generally independent. To make a single-step second-order transition possible, we introduce a chemical potential term

which explicitly breaks the chiral symmetry for $\mu\neq 0$ and couples $W_{1}$ and $W_{2}$. Consequently, when $\mu\neq 0$, $W_{1}$ and $W_{2}$ condense simultaneously for sufficiently large Kondo coupling, and the 4-fold degenerate state becomes the 2-fold degenerate $U(1)\times Z_{2}$ state, as shown in Fig. 5(c). Remarkably, the topological phases we found earlier are robust against the nonzero chemical potential $\mu$, unless it closes and reopens the gap. Therefore, when $\mu\neq 0$, the TRS-breaking transition to a (possibly topological) superconductor becomes one-step.

We have presented calculations on the CPT Kondo lattice which demonstrate the feasibility of topological superconductivity in the Kondo lattice. Our approach takes advantage of an underlying Yao-Lee spin model to control the gauge fluctuations associated with fractionalization, without the need to invoke a large-N expansion. While our model calculation depends on a specific model, the results exhibit a new class of topological superconductivity that may occur under a wide range of circumstances. The model state we have uncovered involves $Z_{2}$ fractionalized order: the pairing between electrons and a $Z_{2}$ spin liquid with gapless spin excitations. A fascinating aspect of this topological superconductivity, is the spin-1/2 character of the order parameter, for unlike conventional Cooper pairing, the order parameter transforms under a $S=1/2$, double group representation and is consequently subject to Kramer’s theorem, allowing for a spontaneous time-reversal breaking chiral superconductor, developing via a single phase transition, independently of the underlying lattice. We remark that although our mean-field calculations suggest the existence of novel charge-$e$ topological superconductivity, fluctuations beyond mean-field theory may as well lead to other competing phases. Therefore, complementary numerical methods are necessary to justify the existence of the proposed phase, which is beyond the scope of this work and interesting for future studies.

We end by returning to our initial motivation, speculating on whether this novel type of order could develop in heavy fermion materials such as UTe₂. Although our model calculation relies on a pre-existing spin liquid, which is not proved to exist in UTe₂ and hence our model is not directly applicable, it may also be possible to enter this novel superconducting state from a heavy Fermion liquid[30]. As mentioned in the introduction, early measurements on superconducting UTe₂ suggested that the condensate breaks time-reversal symmetry, producing a Kerr rotation, with STM evidence for chiral edge states. More recent measurements, on improved samples have not reproduced the early Kerr rotation results, but chiral edge states are still believed to still be present. A one-stage transition into a chiral topological superconductor is prohibited in orthorhombic Cooper-paired superconductors, as there are no two dimensional triplet representations. It would thus be interesting to revisit the question of time-reversal symmetry breaking in this material, for a single stage transition into a chiral state would be strong evidence for beyond BCS pairing.

This work was supported by the U.S. National Science Foundation grant DMR-1830707. Part of the work by Z.Z. was financially supported by the National Science Foundation, Quantum Leap Challenge Institute for Hybrid Quantum Architectures and Networks Grant No. OMA-2016136. We gratefully acknowledge discussions with Aaditya Panigrahi and Alexei Tsvelik.