betaFeedback

WorksPhysics and Astronomy

The Clifford Hierarchy from Clifford Stabilizers in 2D

Abstract

The Bravyi-König theorem limits the constant-depth logical gates of 2D topological Pauli stabilizer codes to be in the Clifford group. Known qubit codes with Clifford stabilizers exceed it by only one level of the Clifford hierarchy, and this was conjectured to hold generally. We show that this bound can be completely bypassed, and that topological codes with Clifford stabilizers in 2D can realize transversal logical gates at any level of the Clifford hierarchy. We present two explicit constructions of codes based on copies of the quantum doubles of D4D_4 and Z2\mathbb{Z}_2, with kk logical qubits and a logical Ck−1ZC^{k-1}Z gate, which is at level kk of the Clifford hierarchy. The physical circuits realizing the gates consist only of physical ZZ or CZCZ gates. The main new ingredient comes from the logical encoding, which depends crucially on non-Abelian group multiplication.

Introduction

Introduction. Transversal logical gates are the simplest fault-tolerant gates, since they do not spread errors within a code block. However, the Bravyi-König (BK) theorem [1] limits them, and more generally all constant-depth logical gates, in topological 2D Pauli stabilizer codes, to be in the Clifford group. In particular, non-Clifford gates that are needed for universal quantum computation require further resources, such as magic states. It was recently shown that non-Abelian codes, whose stabilizers are not simply Pauli operators, can evade this restriction and host non-Clifford constant-depth gates. For example, the twisted Z2\mathbb{Z}_2 gauge theory DIIIω(Z2)D^{\omega}_{\mathrm{III}}(\mathbb{Z}_2), or equivalently D(D4)D(D_4), [2, 3, 4], has Clifford stabilizers and a constant-depth logical T=diag⁡(1,eiπ/4)T = \operatorname{diag}(1,e^{i\pi/4}) gate in 2D [4, 5, 6], and can be implemented [7] and decoded with a high threshold [4, 8]. On the other hand, the quantum doubles D(D4N)D(D_{4N}) host constant-depth logical T1/NT^{1/N} gates [6] at the cost of more complex stabilizers that include non-Clifford gates [9]. However, thus far in all known topological qubit codes in 2D, the constant-depth logical gates lie at most one level above the stabilizers, and this was recently conjectured to be a general bound in [10].

Main Result: Clifford Hierarchy Gates from Clifford Stabilizer Codes. In this Letter we show that Clifford stabilizers in 2D suffice for transversal logical gates in every level of the Clifford hierarchy. We argue this by providing two explicit constructions, which use the quantum doubles of the non-Abelian groups

GA=D4kA−1×Z2,GA×GB=(D4kA−1×Z2)×(D4kB−1×Z2).(1)G_A = D_4^{k_A-1} \times\mathbb{Z}_2, \qquad G_A \times G_B = \left(D_4^{k_A-1} \times\mathbb{Z}_2\right) \times\left(D_4^{k_B-1} \times\mathbb{Z}_2\right). \tag*{(1)}

Here, D4D_4 is the group of symmetries of a square. The parameters kAk_A and kBk_B count the logical qubits of the two factors with k=kA+kBk = k_A + k_B, when considering the code on a square patch with four gapped boundaries (Fig. 1). Both constructions act on kk logical qubits realizing the logical

Ck−1Z=(−1)x1x2⋯xk,(2)C^{k-1} Z = (-1)^{x_1x_2\cdots x_k}, \tag*{(2)}

which lies in level kk of the Clifford hierarchy [11]. Multi-controlled logical gates are central to quantum algorithms and to universal gate sets [12, 13, 14], but in a Pauli stabilizer code they require costly magic states. With our

Figure 1. The quantum double D(G)D(G) on a square spatial patch, with time going up. The four vertical faces are the gapped boundaries labeled by the subgroups KlK_l, KtK_t, KrK_r, Kb⊂GK_b \subset G. A logical state is a configuration of four magnetic anyons with fluxes gl,gt,gr,gbg_l,g_t,g_r,g_b, each ending on a boundary and satisfying glgt=gbgrg_lg_t = g_bg_r. The logical gates act on a spatial slice. In Construction I (light blue), UeU_e is the line of an Abelian charge anyon between the left and right boundaries, and physically a string of Pauli ZZ gates. In Construction II (dark blue), UαU_\alpha multiplies the state by α(gb,gr)/α(gl,gt)\alpha(g_b,g_r)/\alpha(g_l,g_t) where α\alpha is a group 2-cocycle. Both give the transversal logical Ck−1ZC^{k-1}Z, implemented by physical ZZ and CZCZ gates and Clifford stabilizers.

constructions once the quantum double codes for (1) are prepared, (2) is a single layer of one-qubit or two-qubit gates, with stabilizers and physical gates that are all at most Clifford.

How is this possible? In a Pauli stabilizer code a physical Pauli operator that preserves the code space acts on it as a logical Pauli [15]. However, for a code with Clifford stabilizers this need not hold. The logical states are determined from magnetic anyons with fluxes g∈Gg \in G, stretching between the left and right, or top and bottom, boundaries. However, when they cross, non-Abelian group multiplication can force the exponent of a certain group element to be fixed in terms of the others.

Consider a finite group GG with a Z2\mathbb{Z}_2 factor, and denote by mm its magnetic flux. For a suitable choice of GG and boundaries, we show that a vertical flux g1a1g_1^{a_1} that crosses the horizontal fluxes g2a2,…,gkakg_2^{a_2},\ldots,g_k^{a_k} is forced to carry also the Z2\mathbb{Z}_2 flux ma1⋯akm^{a_1\cdots a_k}, as shown in Figure 2: the total flux is g1a1hbma1⋯akg_1^{a_1}h_bm^{a_1\cdots a_k} below the crossings and g1a1htma1⋯akg_1^{a_1}h_tm^{a_1\cdots a_k} above them, where hbh_b and hth_t are group elements fixed by the boundaries and junctions. The exponent of mm is a1⋯aka_1\cdots a_k along the whole vertical path. Then, its corresponding electric (charge) anyon e∈D(Z2)e\in D(\mathbb{Z}_2), which on the lattice is a tensor product of Pauli ZZ gates and acts on a state with mm by a −1-1 phase, will produce (−1)a1⋯ak=Ck−1Z(-1)^{a_1\cdots a_k}=C^{k-1}Z when acting on the state in Figure 2, resulting in a logical controlled-gate.

Figure 2. Action of the charge anyon ee (blue) on the logical state: the anyon ee, which on the lattice is a product of ZZ, acts on the configuration of anyons that realize the logical state, by a sign (−1)a1⋯ak(-1)^{a_1\cdots a_k}, which is the Ck−1ZC^{k-1}Z gate. In the anyon picture, this arises from ee crossing the ma1⋯akm^{a_1\cdots a_k} anyon.

In our first construction, the logical gate U‾e=Ck−1Z\overline{U}_e=C^{k-1}Z is realized by a tensor product of Pauli ZZ operators along a path between two opposite boundaries, while in the second, U‾α=Ck−1Z\overline{U}_\alpha=C^{k-1}Z is obtained by stacking a group 2-cocycle α\alpha as in [6], which is implemented as a layer of physical CZCZ gates on the whole patch. Our setup and the logical gates are illustrated in Fig. 1.

We remark that the non-Clifford nature of the gates does not come from the physical operators implementing them but from the non-Abelian nature of the code, as we explained above. The code stabilizers are Clifford and we expect the constant-depth adaptive protocol of D(D4)D(D_4) [7, 16] to be a natural starting point for its preparation. Then, once our code has been implemented, the logical Ck−1ZC^{k-1}Z gate is a depth-one circuit of Pauli ZZ or CZCZ gates, which can be applied any number of times during a computation, so we expect our protocols to be favorable for computations that require many non-Clifford gates. We will prove in the End Matter that the Clifford level of our gates cannot decrease by a change of logical basis.

Statement of the Main Theorem. Let the stabilizers of a DD-dimensional code lie in C(ℓ)\mathcal{C}^{(\ell)}, where C(1)\mathcal{C}^{(1)} is the Pauli group and V∈C(ℓ+1)V\in\mathcal{C}^{(\ell+1)} if VPV−1∈C(ℓ)VPV^{-1}\in\mathcal{C}^{(\ell)} for all Pauli operators PP [17]. For Pauli stabilizer codes, we have ℓ=1\ell=1 and the BK theorem [1] places the constant-depth logical gates in C(D)\mathcal{C}^{(D)}. All known qubit codes with stabilizers beyond the Pauli group, listed in Table I, have constant-depth logical gates in

C(ℓ+D−1),(3)\mathcal{C}^{(\ell+D-1)}, \tag*{(3)}

which was conjectured to be a general bound in [10].

CodeDDℓ\ellGateLevel of gate
Pauli [1]DD11<<DD
D(D4)D(D_4) [4, 5, 6]2222TT≤3\leq3
D(D4N)D(D_{4N}) [6]222+log⁡2N2+\log_2NT1/NT^{1/N}3+log⁡2N3+\log_2N
I: D(GA)D(G_A)2222Ck−1ZC^{k-1}Zkk
II: D(GA×GB)D(G_A\times G_B)2222Ck−1ZC^{k-1}Zkk

Table 1. Clifford Level ℓ\ell of the stabilizers and of the constant-depth logical gate of topological qubit codes in DD dimensions, with NN a power of two.

We show in this Letter that this conjectured generalized Bravyi-König bound can be completely evaded: transversal logical gates at arbitrary levels of the hierarchy can be obtained, even in the simplest non-Pauli topological setting with D=2D=2 and ℓ=2\ell=2. Our main result is the following:

Theorem 1. Let k≥3k\geq3. The following two codes, realized on an L×LL\times L square lattice patch, are qubit codes with Clifford stabilizers, kk logical qubits and distance LL, with a transversal logical Ck−1ZC^{k-1}Z gate.

Construction I. The quantum double of GAG_A in (1) with kA=kk_A=k and the boundary subgroups (24), with 3k−23k-2 physical qubits on each lattice edge. The gate is realized by a tensor product of LL single-qubit Pauli ZZ gates.

Construction II. The quantum double of GA×GBG_A\times G_B in (1) with kA+kB=kk_A+k_B=k, kA,kB≥1k_A,k_B\geq1, and the boundary subgroups (25), with 3k−43k-4 physical qubits per lattice edge. The gate is realized as a single layer of CZCZ gates, transversal between the two code blocks, D(GA)D(G_A) and D(GB)D(G_B).

Setup. We place the Kitaev quantum double D(G)D(G) [18] on a square, and impose on all its four sides gapped boundary conditions, labeled by subgroups Kl,Kt,Kr,Kb≤GK_l,K_t,K_r,K_b\leq G (and trivial 2-cocycles) [19]. The logical states are ground states of the quantum double Hamiltonian [18, 19]. We will characterize them by four magnetic anyons, each ending on a boundary on which its flux condenses, which meet at a common plaquette. We denote them by ⟨gl,gt,gr,gb⟩\langle g_l,g_t,g_r,g_b\rangle with gl∈Klg_l\in K_l, gt∈Ktg_t\in K_t, gr∈Krg_r\in K_r, gb∈Kbg_b\in K_b which satisfy

glgt=gbgr(4)g_lg_t=g_bg_r \tag*{(4)}

to ensure that the common plaquette has trivial flux (Fig. 1). By choosing a suitable group GG as in (1) and boundary conditions, we can encode any number of logical qubits kk, as we will explain below.

D(D4×Z2)D(D_4\times\mathbb{Z}_2) Code. We begin by discussing the main building block which is the quantum double of the group

GA=⟨r,s⟩×⟨m⟩≅D4×Z2.(5)G_A=\langle r,s\rangle\times\langle m\rangle\cong D_4\times\mathbb{Z}_2. \tag*{(5)}

Here, we present the D4D_4 as the symmetries of a square with rotation rr and reflection ss

D4=⟨r,s∣r4=s2=id, srs=r−1⟩.(6)D_4=\langle r,s\mid r^4=s^2=\mathrm{id},\ srs=r^{-1}\rangle. \tag*{(6)}

while mm generates the Z2\mathbb{Z}_2 factor and we define

t:=r2m.(7)t := r^2m . \tag*{(7)}

All four elements srsr, ss, mm, tt have order two, they satisfy (sr)s=s(sr)r2(sr)s=s(sr)r^2, mt=r2mt=r^2, and r2r^2, mm, tt are central. We place D(GA)D(G_A) on a square with the boundary subgroups

Kl=Kr=⟨sr⟩,Kb=⟨s,m⟩,Kt=⟨s,t⟩,(8)K_l=K_r=\langle sr\rangle,\qquad K_b=\langle s,m\rangle,\qquad K_t=\langle s,t\rangle, \tag*{(8)}

so that the bottom and top boundaries differ only by mm and tt.

This setup encodes two logical qubits, whose basis states are ∣a1,a2⟩\lvert a_1,a_2\rangle with a1,a2∈{0,1}a_1,a_2\in\{0,1\}. The horizontal anyon between the left and right boundaries carries the flux (sr)a2(sr)^{a_2}, and the vertical anyon between the bottom and top boundaries carries sa1s^{a_1} with further factors that are fixed by the boundary conditions when the two anyons cross. For general a1,a2a_1,a_2, the logical state is the following anyon configuration:

Anyon configuration on a square with boundary subgroups

Figure 1 (PDF p. 3).

∣a1,a2⟩(9)\lvert a_1,a_2\rangle \tag*{(9)}

This forces the exponent of mm in gbg_b to be the product a1a2a_1a_2 of the two logical bits. To see this, starting with the general bits a1a_1, a2a_2, A1A_1, for ss, srsr, mm, respectively, we write

gl=gr=(sr)a2,gb=sa1mA1,gt=sa1tA1.(10)g_l=g_r=(sr)^{a_2},\qquad g_b=s^{a_1}m^{A_1},\qquad g_t=s^{a_1}t^{A_1}. \tag*{(10)}

It then follows by straightforward computation that

glgt=(sr)a2sa1tA1=sa1mA1(sr)a2r2(a1a2+A1)=gbgrr2(a1a2+A1),(11)\begin{aligned} g_lg_t&=(sr)^{a_2}s^{a_1}t^{A_1}=s^{a_1}m^{A_1}(sr)^{a_2}r^{2(a_1a_2+A_1)}\\ &=g_bg_rr^{2(a_1a_2+A_1)}, \tag*{(11)} \end{aligned}

so the plaquette stabilizer condition (4) holds if and only if A1=a1a2A_1=a_1a_2. The four logical states are therefore:

a1a_1a2a_2gl=grg_l=g_rgbg_bgtg_tUeU_e
00ididid+1
10idssss+1
01srsridid+1
11srsrsmsmstst−1

Table 1 (PDF p. 3).

The last column is the value of the operator Ue=∏i∈γZm,iU_e=\prod_{i\in\gamma}Z_{m,i} on a horizontal path γ\gamma: at the logical level, U‾e=(−1)a1a2=CZ\overline{U}_e=(-1)^{a_1a_2}=CZ for k=2k=2.

We now explain how quantum doubles of products of these groups are used for our two constructions of C3ZC^3Z and subsequently generalize to any Ck−1ZC^{k-1}Z.

Construction I: C3ZC^3Z from ZZ

To obtain the first gate beyond the putative bound (3) (which is the logical C3ZC^3Z gate) from physical Pauli ZZ gates, we take

GA=D43×Z2(12)G_A=D_4^3\times\mathbb{Z}_2 \tag*{(12)}

with the generators rj,sjr_j,s_j of the three copies of D4D_4, the generator mm of Z2\mathbb{Z}_2, and

t1=r12s2,t2=r22s3,t3=r32m.(13)t_1=r_1^2s_2,\qquad t_2=r_2^2s_3,\qquad t_3=r_3^2m. \tag*{(13)}

The subgroups labeling the four boundaries are

Kl=Kr=⟨s1r1,s2r2,s3r3⟩,Kb=⟨s1,s2,s3,m⟩,Kt=⟨s1,t1,t2,t3⟩,(14)K_l=K_r=\langle s_1r_1,s_2r_2,s_3r_3\rangle,\qquad K_b=\langle s_1,s_2,s_3,m\rangle,\qquad K_t=\langle s_1,t_1,t_2,t_3\rangle, \tag*{(14)}

and the code has four logical qubits a1,…,a4a_1,\ldots,a_4. Note that the subgroups associated to boundaries that share a corner have trivial intersection, and thus there are no short strings. A non-trivial s1s_1 flux now crosses s1r1s_1r_1, s2r2s_2r_2, s3r3s_3r_3, and each crossing constrains one more generator as in (9), so the boundary holonomies are

gl=gr=(s1r1)a2(s2r2)a3(s3r3)a4,gb=s1a1s2a1a2s3a1a2a3ma1a2a3a4,gt=s1a1t1a1a2t2a1a2a3t3a1a2a3a4.(15)\begin{aligned} g_l=g_r&=(s_1r_1)^{a_2}(s_2r_2)^{a_3}(s_3r_3)^{a_4},\\ g_b&=s_1^{a_1}s_2^{a_1a_2}s_3^{a_1a_2a_3}m^{a_1a_2a_3a_4},\\ g_t&=s_1^{a_1}t_1^{a_1a_2}t_2^{a_1a_2a_3}t_3^{a_1a_2a_3a_4}. \tag*{(15)} \end{aligned}

The string of Pauli ZZ gates on the qubits for the Z2\mathbb{Z}_2 factor mm is a −1-1 phase with the exponent of mm in gbg_b,

Ue=∏i∈γZm,i,U‾e=(−1)a1a2a3a4=C3Z,(16)U_e=\prod_{i\in\gamma}Z_{m,i},\qquad\overline{U}_e=(-1)^{a_1a_2a_3a_4}=C^3Z, \tag*{(16)}

which lies in C(4)C^{(4)} and not in C(3)C^{(3)}. So LL single-qubit Pauli gates, on a code with Clifford stabilizers and 10 qubits per edge (Table II), already exceed the putative bound (3).

Construction II: C3ZC^3Z from CZCZ

We take two copies GA×GBG_A\times G_B with GB=GA=D4×Z2G_B=G_A=D_4\times\mathbb{Z}_2. We write the elements as pairs (gA,gB)(g_A,g_B) and the subgroups as products. Since all four boundary subgroups are products, D(GA×GB)D(G_A\times G_B) consists of two decoupled code blocks on the same lattice, D(GA)D(G_A) with the boundary subgroups (8) and D(GB)D(G_B) with the same boundaries rotated by 90 degrees. The boundary subgroups for D(GA×GB)D(G_A\times G_B) are

Subgroup⊂GA×GB\subset G_A\times G_B
KlK_l⟨sr⟩×⟨s,m⟩\langle sr\rangle\times\langle s,m\rangle
KrK_r⟨sr⟩×⟨s,t⟩\langle sr\rangle\times\langle s,t\rangle
KbK_b⟨s,m⟩×⟨sr⟩\langle s,m\rangle\times\langle sr\rangle
KtK_t⟨s,t⟩×⟨sr⟩\langle s,t\rangle\times\langle sr\rangle

Table 2.

and the four logical qubits are labeled by a1,a2a_1,a_2 for GAG_A and b1,b2b_1,b_2 for GBG_B.

Let μA(g)\mu_A(g) and μB(g)\mu_B(g) be the exponents of mm in gAg_A and in gBg_B, which are homomorphisms GA×GB→Z2G_A\times G_B\to\mathbb{Z}_2. In particular, μA(t,id)=μB(id,t)=1\mu_A(t,\mathrm{id})=\mu_B(\mathrm{id},t)=1 since t=r2mt=r^2m, while μA\mu_A and μB\mu_B vanish on srsr. Their cup product is a group 2-cocycle

α(g,h)=(−1)μA(g)μB(h),(17)\alpha(g,h)=(-1)^{\mu_A(g)\mu_B(h)}, \tag*{(17)}

whose class is nontrivial (End Matter). On the four boundaries

α(g,h)=1for all g,h∈Kl, Kt, Kr or Kb,(18)\alpha(g,h)=1\quad\text{for all }g,h\in K_l,\ K_t,\ K_r\text{ or }K_b, \tag*{(18)}

since μA\mu_A vanishes on KlK_l and KrK_r, whose GAG_A factor is ⟨sr⟩\langle s_r\rangle, and μB\mu_B vanishes on KbK_b and KtK_t so the gate UαU_\alpha, constructed as in [6], does not require any boundary terms. With μA(gb)=a1a2\mu_A(g_b)=a_1a_2, μB(gr)=b1b2\mu_B(g_r)=b_1b_2, μA(gl)=μB(gt)=0\mu_A(g_l)=\mu_B(g_t)=0, it evaluates to

U‾α=α(gb,gr)α(gl,gt)=(−1)a1a2b1b2=C3Z.(19)\overline{U}_{\alpha}=\frac{\alpha(g_b,g_r)}{\alpha(g_l,g_t)}=(-1)^{a_1a_2b_1b_2}=C^3Z. \tag*{(19)}

This is the C3ZC^3Z gate for the logical bits a1,a2,b1,b2a_1,a_2,b_1,b_2 of GAG_A and of GBG_B. U‾α\overline{U}_{\alpha} is −1-1 if and only if the flux lines of mAm_A and of mBm_B are both present (Figure 3) and only requires 8 physical qubits per edge (Table II).

Construction II flux lines

Figure 3. Construction II for k=4k=4: the flux lines with μA=1\mu_A=1 (if a1a2=1a_1a_2=1) and μB=1\mu_B=1 (if b1b2=1b_1b_2=1) cross once. Stacking the SPT phase of α\alpha in (17) on the spatial slice counts this crossing, so U‾α=(−1)a1a2b1b2=C3Z\overline{U}_{\alpha}=(-1)^{a_1a_2b_1b_2}=C^3Z.

Constructions I and II for every kk. The codes for larger kk are obtained by extending one factor at a time, starting from the surface code GA=⟨m⟩≅Z2G_A=\langle m\rangle\cong\mathbb{Z}_2 with one logical qubit:

⟨m⟩→⟨r1,s1⟩×⟨m⟩→⟨r1,s1⟩×⟨r2,s2⟩×⟨m⟩→⋯(20)\langle m\rangle\to\langle r_1,s_1\rangle\times\langle m\rangle\to\langle r_1,s_1\rangle\times\langle r_2,s_2\rangle\times\langle m\rangle\to\cdots \tag*{(20)}

In the jj-th step the generator mm of the Z2\mathbb{Z}_2 factor becomes the reflection sjs_j of a new copy ⟨rj,sj⟩\langle r_j,s_j\rangle of D4D_4, and a new Z2\mathbb{Z}_2 with generator mm is added, together with tj:=rj2mt_j:=r_j^2m, which becomes tj=rj+12sj+1t_j=r_{j+1}^2s_{j+1} after the next step. After kA−1k_A-1 steps this gives GAG_A in (1). For Construction II, the second group GBG_B is the same with kBk_B in place of kAk_A. We define the following subgroups

RA=⟨s1r1,…,skA−1rkA−1⟩,SA=⟨s1,…,skA−1,m⟩,TA=⟨s1,t1,…,tkA−1⟩.(21)\begin{aligned} R_A&=\langle s_1r_1,\ldots,s_{k_A-1}r_{k_A-1}\rangle,\\ S_A&=\langle s_1,\ldots,s_{k_A-1},m\rangle,\\ T_A&=\langle s_1,t_1,\ldots,t_{k_A-1}\rangle. \tag*{(21)} \end{aligned}

and likewise for GBG_B [20]. The boundary subgroups generalize (8) for Construction I,

Kl=Kr=RA,Kb=SA,Kt=TA,(22)K_l=K_r=R_A,\qquad K_b=S_A,\qquad K_t=T_A, \tag*{(22)}

and (18) for Construction II,

Kl=RA×SB,Kr=RA×TB,Kb=SA×RB,Kt=TA×RB.(23)\begin{aligned} K_l&=R_A\times S_B,\qquad K_r=R_A\times T_B,\\ K_b&=S_A\times R_B,\qquad K_t=T_A\times R_B. \tag*{(23)} \end{aligned}
kkGateI: GAG_A (qubits per edge)II: GA×GBG_A\times G_B (qubits per edge)
3CCZCCZD42×Z2D_4^2\times\mathbb{Z}_2 (7)(D4×Z2)×Z2(D_4\times\mathbb{Z}_2)\times\mathbb{Z}_2 (5)
4C3ZC^3ZD43×Z2D_4^3\times\mathbb{Z}_2 (10)(D4×Z2)×(D4×Z2)(D_4\times\mathbb{Z}_2)\times(D_4\times\mathbb{Z}_2) (8)
kkCk−1ZC^{k-1}ZD4k−1×Z2D_4^{k-1}\times\mathbb{Z}_2 (3k−23k-2)GA×GBG_A\times G_B (3k−43k-4)

Table 2 (PDF p. 4). Constructions I and II of the logical Ck−1ZC^{k-1}Z gate on kk logical qubits, with the number of physical qubits on each edge of the lattice.

So each step adds sjrjs_jr_j to the left and right boundaries, mm to the bottom boundary and tjt_j to the top boundary. The factor GAG_A encodes kAk_A logical qubits, specified by the exponent a1a_1 of s1s_1 and the exponents aj+1a_{j+1} of sjrjs_jr_j. Similarly to (15), the boundary holonomies are

gl=gr=(s1r1)a2⋯(skA−1rkA−1)akA,gb=s1A0s2A1⋯skA−1AkA−2mAkA−1,gt=s1A0t1A1⋯tkA−1AkA−1.(24)\begin{aligned} g_l=g_r&=(s_1r_1)^{a_2}\cdots(s_{k_A-1}r_{k_A-1})^{a_{k_A}},\\ g_b&=s_1^{A_0}s_2^{A_1}\cdots s_{k_A-1}^{A_{k_A-2}}m^{A_{k_A-1}},\\ g_t&=s_1^{A_0}t_1^{A_1}\cdots t_{k_A-1}^{A_{k_A-1}}. \tag*{(24)} \end{aligned}

with A0=a1A_0=a_1 and Aj=a1a2⋯aj+1A_j=a_1a_2\cdots a_{j+1}. The exponents of gl=grg_l=g_r and the exponent of s1s_1 are the logical bits, while all other exponents are fixed by the condition glgt=gbgrg_lg_t=g_bg_r.

In Construction I, as in (16), the ee anyon acting on the logical state with ma1⋯akm^{a_1\cdots a_k} has the physical and logical gate decompositions:

Ue=∏i∈γZm,i,U_e=\prod_{i\in\gamma}Z_{m,i},
U‾e=(−1)a1a2⋯ak=Ck−1Z.(25)\overline{U}_e=(-1)^{a_1a_2\cdots a_k}=C^{k-1}Z. \tag*{(25)}

In Construction II, with μA\mu_A and μB\mu_B the exponents of mm in gAg_A and in gBg_B, the cocycle is again (17), it still satisfies (18), and the gate is

U‾α=α(gb,gr)α(gl,gt)=(−1)a1⋯akAb1⋯bkB=Ck−1Z.(26)\overline{U}_{\alpha}=\frac{\alpha(g_b,g_r)}{\alpha(g_l,g_t)}=(-1)^{a_1\cdots a_{k_A}b_1\cdots b_{k_B}}=C^{k-1}Z. \tag*{(26)}

Further details and proofs are given in the End Matter, and the groups, the 2-cocycles and the logical gates can also be checked for given kk with the GAP functions in AlgebraTQ\mathrm{TQ}D [21].

Lattice Model and Hamiltonian

We use Kitaev’s quantum double model D(G)D(G) [18] on an L×LL\times L square lattice, with boundaries [19, 22]. Every edge ii carries C[G]\mathbb{C}[G] with basis ∣gi⟩\lvert g_i\rangle, gi∈Gg_i\in G, and the Hamiltonian is

H=−∑vAv−∑pBp−∑i∈∂Bi,Av=1∣Kv∣∑g∈KvAvg,(27)H=-\sum_v A_v-\sum_p B_p-\sum_{i\in\partial}B_i,\qquad A_v=\frac{1}{|K_v|}\sum_{g\in K_v}A_v^g, \tag*{(27)}

where KvK_v is GG in the bulk, or a boundary subgroup on the boundary, and Kv={id}K_v=\{\mathrm{id}\} at the four corners, where adjacent subgroups intersect trivially. The vertex operator AvgA_v^g applies Lg∣h⟩=∣gh⟩L^g\lvert h\rangle=\lvert gh\rangle to the edges that exit vv and Rg∣h⟩=∣hg−1⟩R^g\lvert h\rangle=\lvert hg^{-1}\rangle to those that enter vv, whose circuits we show in Figure 5(a). The projector BpB_p imposes (4) on the plaquette pp, and BiB_i imposes gi∈Kg_i \in K on an edge ii of a boundary with subgroup KK.

Circuits of the stabilizers and logical gates

Figure 5. Circuits of the stabilizers and logical gates: A dot connected to ⊕\oplus is a CNOT gate, with the dot on the control qubit, and two connected dots are a CZ gate. (a) The gates LgL^{g} and RgR^{g} of [6] for the generators rr and ss of a copy of D4D_{4} on its three qubits ρ1,ρ0,σ\rho_{1},\rho_{0},\sigma. (b) The plaquette stabilizer Spr2S_{p}^{r^{2}} in (44) of a copy of D4D_{4} on the qubits of the left and top edges l,tl,t of pp, which acts in the same way on the bottom and right edges b,rb,r. The other plaquette stabilizers (Spr)2(S_{p}^{r})^{2}, (Sps)2(S_{p}^{s})^{2} and Zm(∂p)Z_{m}(\partial p) are products of single-qubit Pauli ZZ gates on the four edges of pp. (c) The gate UeU_{e} of Construction I, a Pauli ZZ gate on the qubit mm of each horizontal edge of a row. (d) The plaquette gate for Uα=∏pMpαU_{\alpha}=\prod_{p}M_{p}^{\alpha} of Construction II in (33): two CZ gates between the qubits mAm_{A} and mBm_{B} of the bottom and right edges and of the left and top edges of pp. Both gate circuits have depth one.

The logical states are created by four magnetic ribbon operators [18, 19, 23] [24]

∣gl,gt,gr,gb⟩‾=∏vAv∏iLξlglLξtgtLξrgrLξbgb⊗∣id⟩i,(28)\overline{|g_l,g_t,g_r,g_b\rangle} = \prod_v A_v \prod_i L_{\xi_l}^{g_l}L_{\xi_t}^{g_t}L_{\xi_r}^{g_r}L_{\xi_b}^{g_b}\otimes|\mathrm{id}\rangle_i , \tag*{(28)}

that meet at a plaquette p∗p_\ast (Figure 4).

The codes on a square lattice

Figure 4. The codes on a square lattice (L=6L = 6). Kl,Kt,Kr,KbK_l,K_t,K_r,K_b label the gapped boundaries given in (22), for Construction I. A logical state is created by four magnetic anyons with fluxes gl,gt,gr,gbg_l,g_t,g_r,g_b and glgt=gbgrg_lg_t = g_bg_r to ensure trivial flux at their common plaquette p∗p_\ast (shaded). The gate UeU_e of Construction I is a string of single-qubit Pauli ZZ gates on the edges of a line between the left and right boundaries (blue). It crosses the flux gbg_b and detects its exponent of mm, and its logical action is the same when crossing gtg_t.

Qubit Realization. Both codes can be realized in terms of physical qubits, where for D(D4)D(D_4), we map a basis state for each lattice edge onto three qubits as ∣rρsσ⟩↦∣ρ1ρ0σ⟩|r^\rho s^\sigma\rangle\mapsto|\rho_1\rho_0\sigma\rangle, where ρ=2ρ1+ρ0\rho= 2\rho_1 + \rho_0 [6]. This gives 3kA−23k_A - 2 physical qubits per edge for GAG_A and 3k−43k - 4 for GA×GBG_A \times G_B. The details of the stabilizer and gate qubit circuits can be found in the End Matter. On the qubits of the jj-th copy of D(D4)D(D_4), the element rjρjsjσjr_j^{\rho_j}s_j^{\sigma_j} with ρj=2ρj,1+ρj,0\rho_j = 2\rho_{j,1}+\rho_{j,0} is (sjrj)ρj,0+σjsjρj,0+σj(s_jr_j)^{\rho_{j,0}+\sigma_j}s_j^{\rho_{j,0}+\sigma_j} up to a power of rj2r_j^2. So the exponents of sjrjs_jr_j and of sjs_j are the bits ρj,0\rho_{j,0} and ρj,0+σj\rho_{j,0}+\sigma_j, while the exponent of mm on each site is stored in its qubit. The lattice realizations of the charge anyons that detect these exponents are, on a path γ\gamma

Zsj(γ)=∏i∈γ(I⊗Z⊗Z)j,i,Z_{s_j}(\gamma)=\prod_{i\in\gamma}(I\otimes Z\otimes Z)_{j,i},
Zsjrj(γ)=∏i∈γ(I⊗Z⊗I)j,i,Z_{s_jr_j}(\gamma)=\prod_{i\in\gamma}(I\otimes Z\otimes I)_{j,i},
Zm(γ)=∏i∈γZm,i.(29)Z_m(\gamma)=\prod_{i\in\gamma}Z_{m,i}. \tag*{(29)}

Denoting by γh\gamma_h a horizontal path between the left and right boundaries and by γv\gamma_v a vertical path between the bottom and top boundaries, the strings Zsj(γh)Z_{s_j}(\gamma_h), Zm(γh)Z_m(\gamma_h) and Zsjrj(γv)Z_{s_jr_j}(\gamma_v) commute with the code stabilizers, and from (24) their action on the logical state ∣a1,…,akA⟩|a_1,\ldots,a_{k_A}\rangle is

Zs1(γh)=(−1)a1=Z‾1,Z_{s_1}(\gamma_h)=(-1)^{a_1}=\overline{Z}_1,
Zsjrj(γv)=(−1)aj+1=Z‾j+1,Z_{s_jr_j}(\gamma_v)=(-1)^{a_{j+1}}=\overline{Z}_{j+1},
Zsj+1(γh)=(−1)a1a2⋯aj+1=CjZ,Z_{s_{j+1}}(\gamma_h)=(-1)^{a_1a_2\cdots a_{j+1}}=C^jZ,
Zm(γh)=(−1)a1a2⋯akA=CkA−1Z,(30)Z_m(\gamma_h)=(-1)^{a_1a_2\cdots a_{k_A}}=C^{k_A-1}Z, \tag*{(30)}

with 1≤j≤kA−11\le j\le k_A-1 in the second line and 1≤j≤kA−21\le j\le k_A-2 in the third line. The first two operators are the logical Z‾\overline{Z} operators of the kAk_A logical qubits, while the last two contain the whole tower CZ,CCZ,…,CkA−1ZCZ,CCZ,\ldots,C^{k_A-1}Z, where CjZC^jZ acts on the first j+1j+1 logical qubits, each as a string of physical Pauli ZZ gates, and the last one is UeU_e in (25).

The logical X‾\overline{X} operators, which flip the logical bits, are instead constructed from the magnetic ribbon operators, similarly to (28). They are not Pauli XX strings, since by (24) flipping a logical bit also changes the exponents A1,…,AkA−1A_1,\ldots,A_{k_A-1} in gbg_b and gtg_t by products of the other logical bits.

Logical Gates. The gate of Construction I is (Figure 5(c))

Ue=∏x=0L−1Zm,(x,y),(31)U_e=\prod_{x=0}^{L-1} Z_{m,(x,y)}, \tag*{(31)}

on the qubits of mm on a horizontal path. It commutes with all stabilizers, since AvgA_v^g applies XX to the qubits of mm of the two edges on the row at vv or leaves them unchanged, and the exponent of mm vanishes on the subgroup RA\mathcal{R}_A of the left and right boundaries.

The gate of Construction II is Uα=∏pMpαU_\alpha=\prod_p M_p^\alpha, where MpαM_p^\alpha evaluates α\alpha on each triangulated plaquette pp as [6]

Mpα∣gtglgrgb⟩=α(gb,gr)α(gl,gt)∣gtglgrgb⟩(32)M_p^\alpha\left|\begin{array}{c}g_t\\ g_l\quad g_r\\ g_b\end{array}\right\rangle = \frac{\alpha(g_b,g_r)}{\alpha(g_l,g_t)} \left|\begin{array}{c}g_t\\ g_l\quad g_r\\ g_b\end{array}\right\rangle \tag*{(32)}

The phase it produces is the same on all configurations of a logical state (End Matter), which gives (26).

By (17), MpαM_p^\alpha only involves the qubits mAm_A and mBm_B, so on the lattice Uα=∏pMpαU_\alpha=\prod_p M_p^\alpha is realized as

Uα=∏pCZ(mA,b),(mB,r)CZ(mA,l),(mB,t),(33)U_\alpha=\prod_p CZ_{(m_A,b),(m_B,r)}CZ_{(m_A,l),(m_B,t)}, \tag*{(33)}

where CZ(mA,b),(mB,r)CZ_{(m_A,b),(m_B,r)} is the CZCZ gate of the qubit mAm_A of the bottom edge bb and the qubit mBm_B of the right edge rr of pp, and likewise for the left and top edges. As the transversal CZCZ gate between two surface codes, it maps every vertex stabilizer to itself times plaquette ZZ checks of the other code block (End Matter), so it preserves the code space. We present its circuit decomposition in Figure 5(d). Finally, U‾e\overline{U}_e and U‾α\overline{U}_\alpha are both Ck−1ZC^{k-1}Z, which is at level kk of the Clifford hierarchy, which completes the proof of our Theorem.

Fault Tolerance and Outlook. We have shown that Clifford stabilizers in 2D suffice for transversal logical gates in every level of the Clifford hierarchy, where the level is set by the non-Abelian group and the boundary conditions. Both our gates have depth one: UeU_e is a product of single-qubit gates, while a single fault in UαU_\alpha affects at most two qubits of one plaquette, one in each code block. In the bulk, our codes are stacks of surface codes D(Z2)D(\mathbb{Z}_2) and of D(D4)D(D_4), which, despite being non-Abelian, can be decoded with a high threshold by the just-in-time decoder [4, 8], and in this setting also the protocol of [25] simplifies substantially. Natural next steps are to extend these decoders to the boundaries and to the code switching, and a threshold analysis of the full protocol.

END MATTER

We now provide some more mathematical and implementation details for our constructions.

The 2-Cocycle. Here, we prove that (17) is a 2-cocycle in a non-trivial cohomology class and provide further details on Construction II of Theorem 1. We denote by μA\mu_A and μB\mu_B the projections of GA×GBG_A \times G_B onto the direct factors ⟨m⟩≅Z2\langle m\rangle\cong\mathbb{Z}_2 of GAG_A and of GBG_B, so modulo 2

μA(gh)=μA(g)+μA(h),μB(gh)=μB(g)+μB(h).(34)\begin{aligned} \mu_A(gh) &= \mu_A(g) + \mu_A(h),\\ \mu_B(gh) &= \mu_B(g) + \mu_B(h). \tag*{(34)} \end{aligned}

By (17) and (34), the two sides of the cocycle condition

α(g1,g2)α(g1g2,g3)=α(g1,g2g3)α(g2,g3)(35)\alpha(g_1,g_2)\alpha(g_1g_2,g_3)=\alpha(g_1,g_2g_3)\alpha(g_2,g_3) \tag*{(35)}

are (−1)(-1) to the powers

μA(g1)μB(g2)+(μA(g1)+μA(g2))μB(g3),μA(g1)(μB(g2)+μB(g3))+μA(g2)μB(g3),(36)\begin{aligned} \mu_A(g_1)\mu_B(g_2) + \bigl(\mu_A(g_1)+\mu_A(g_2)\bigr)\mu_B(g_3),\\ \mu_A(g_1)\bigl(\mu_B(g_2)+\mu_B(g_3)\bigr)+\mu_A(g_2)\mu_B(g_3), \tag*{(36)} \end{aligned}

which are equal. So (35) holds for all g1,g2,g3∈GA×GBg_1,g_2,g_3 \in G_A \times G_B, and since μA(id)=μB(id)=0\mu_A(\mathrm{id})=\mu_B(\mathrm{id})=0, α\alpha is normalized, α(id,g)=α(g,id)=1\alpha(\mathrm{id},g)=\alpha(g,\mathrm{id})=1. For commuting g,hg,h, a coboundary β(g)β(h)/β(gh)\beta(g)\beta(h)/\beta(gh) is symmetric, so the ratio α(g,h)/α(h,g)\alpha(g,h)/\alpha(h,g) depends only on the class of α\alpha, and

α(g,h)α(h,g)=−1+1=−1for g=(m,id), h=(id,m).(37)\frac{\alpha(g,h)}{\alpha(h,g)}=\frac{-1}{+1}=-1\quad\text{for }g=(m,\mathrm{id}),\ h=(\mathrm{id},m). \tag*{(37)}

So the class of α\alpha in H2(GA×GB,U(1))H^2(G_A\times G_B,U(1)) is nontrivial. By (23)–[25], μA∣Kl=μA∣Kr=μB∣Kb=μB∣Kt=0\mu_A|_{K_l}=\mu_A|_{K_r}=\mu_B|_{K_b}=\mu_B|_{K_t}=0, so α\alpha is identically 1 on the boundaries, i.e. (18) for every kA,kB≥1k_A,k_B\geq1.

The Logical Gate. We evaluate the 2-cocycle stacking circuit of [6], for a 2-cocycle α\alpha whose restriction to each boundary subgroup KK is identically one, so all boundary functions βK\beta_K can be taken to be 1. We evaluate it on the representative of (28), as in Fig. 4. Since α\alpha is normalized,

α(id,gl)α(gl,id)=α(gb,id)α(id,gb)=α(id,id)=1,(38)\frac{\alpha(\mathrm{id},g_l)}{\alpha(g_l,\mathrm{id})}=\frac{\alpha(g_b,\mathrm{id})}{\alpha(\mathrm{id},g_b)}=\alpha(\mathrm{id},\mathrm{id})=1, \tag*{(38)}

so only p∗p_* contributes. By (24) for GAG_A, and for GBG_B with the two pairs of opposite boundaries exchanged, its four edges carry

gl=(∏j(sjrj)aj+1,∏jsjBj−1mBkB−1),gr=(∏j(sjrj)aj+1,s1b1∏jtjBj),gb=(∏jsjaj−1mAkA−1,∏j(sjrj)bj+1),gt=(s1a1∏jtjAj,∏j(sjrj)bj+1),(39)\begin{aligned} g_l&=\left(\prod_j(s_jr_j)^{a_{j+1}},\prod_j s_j^{B_{j-1}}m^{B_{k_B-1}}\right),\\ g_r&=\left(\prod_j(s_jr_j)^{a_{j+1}},s_1^{b_1}\prod_j t_j^{B_j}\right),\\ g_b&=\left(\prod_j s_j^{a_{j-1}}m^{A_{k_A-1}},\prod_j(s_jr_j)^{b_{j+1}}\right),\\ g_t&=\left(s_1^{a_1}\prod_j t_j^{A_j},\prod_j(s_jr_j)^{b_{j+1}}\right), \tag*{(39)} \end{aligned}

with Bj=b1b2⋯bj+1B_j=b_1b_2\cdots b_{j+1}. Denoting by A:=AkA−1=a1⋯akAA:=A_{k_A-1}=a_1\cdots a_{k_A}, B:=BkB−1=b1⋯bkBB:=B_{k_B-1}=b_1\cdots b_{k_B} and tkA−1=rkA−12mt_{k_A-1}=r_{k_A-1}^2m, we have

glg_lgtg_tgrg_rgbg_b
μA\mu_A0AA0AA
μB\mu_BBB0BB0

which inserted into (17) gives

α(gb,gr)=(−1)μA(gb) μB(gr)=(−1)AB,α(gl,gt)=(−1)μA(gl) μB(gt)=1,(40)\begin{aligned} \alpha(g_b,g_r)&=(-1)^{\mu_A(g_b)\,\mu_B(g_r)}=(-1)^{AB},\\ \alpha(g_l,g_t)&=(-1)^{\mu_A(g_l)\,\mu_B(g_t)}=1, \tag*{(40)} \end{aligned}

resulting in

U‾α=α(gb,gr)α(gl,gt)=(−1)a1a2⋯akA b1b2⋯bkB=Ck−1Z.(41)\overline{U}_\alpha=\frac{\alpha(g_b,g_r)}{\alpha(g_l,g_t)}=(-1)^{a_1a_2\cdots a_{k_A}\,b_1b_2\cdots b_{k_B}}=C^{k-1}Z. \tag*{(41)}

In (33) the qubit mAm_A of a horizontal (vertical) edge enters only the gate of the plaquette above (to the right of) it, and the qubit mBm_B only that of the plaquette below (to the left of) it, so U‾α\overline{U}_\alpha is a depth-one layer of CZCZ gates on disjoint pairs of qubits. If the qubit mAm_A of the edge ii is paired with the qubit mBm_B of the edge i′i', then

U‾αXmA,iU‾α†=XmA,iZmB,i′,U‾αXmB,i′U‾α†=XmB,i′ZmA,i,(42)\begin{aligned} \overline{U}_\alpha X_{m_A,i}\overline{U}_\alpha^\dagger&=X_{m_A,i}Z_{m_B,i'},\\ \overline{U}_\alpha X_{m_B,i'}\overline{U}_\alpha^\dagger&=X_{m_B,i'}Z_{m_A,i}, \tag*{(42)} \end{aligned}

while U‾α\overline{U}_\alpha commutes with the gates on all other qubits. Since the product of ZmZ_m operators around a plaquette is a stabilizer, U‾α\overline{U}_\alpha preserves the stabilizer group. A similar computation can be performed for the boundaries. Since it preserves the logical space, the phase is independent of the physical state, thus

U‾α∣gl,gt,gr,gb⟩=α(gb,gr)α(gl,gt)∣gl,gt,gr,gb⟩.(43)\overline{U}_\alpha\left|g_l,g_t,g_r,g_b\right\rangle=\frac{\alpha(g_b,g_r)}{\alpha(g_l,g_t)}\left|g_l,g_t,g_r,g_b\right\rangle. \tag*{(43)}

Code Switching. Similarly to [6], we can perform code-switching by condensing the r2r^2 anyon of each D(D4)D(D_4) copy. For the codes of this Letter, this results in D(Z2)D(\mathbb{Z}_2) surface codes of which kk carry the logical qubits. The boundary subgroups meet ⟨r2⟩\langle r^2\rangle trivially, so their images define D(Z2)D(\mathbb{Z}_2) surface codes, one for each of the generators sjrjs_jr_j, sjs_j and mm. On the lattice, this map is realized by measuring all qubits ρj,1\rho_{j,1} in the XX basis and post-selecting the +1+1 outcomes. In the other direction, we initialize the qubits ρj,1\rho_{j,1} in ∣+⟩\lvert+\rangle and project onto the +1+1 eigenstates of the stabilizers of D(GA)D(G_A) or D(GA×GB)D(G_A\times G_B).

Qubit Circuits. As noted in the main text we can implement the circuits on qubits. Each copy ⟨r,s⟩≅D4\langle r,s\rangle\cong D_4 can be realized on three qubits per edge as ∣rρsσ⟩↦∣ρ1ρ0σ⟩\lvert r^{\rho}s^\sigma\rangle\mapsto\lvert\rho_1\rho_0\sigma\rangle, with ρ=2ρ1+ρ0\rho=2\rho_1+\rho_0, and each Z2\mathbb{Z}_2 factor on one qubit. This gives 3kA−23k_A-2 qubits per edge for GAG_A and 3k−43k-4 for GA×GBG_A\times G_B. We denote the three qubits of the jj-th copy of D4D_4 by ρj,1,ρj,0,σj\rho_{j,1},\rho_{j,0},\sigma_j. A subscript ii on an operator on the qubits of one copy, as in (I⊗Z⊗Z)i(\mathbb{I}\otimes\mathbb{Z}\otimes\mathbb{Z})_i, means that it acts on the edge ii, and (I⊗Z⊗Z)(∂p)(\mathbb{I}\otimes\mathbb{Z}\otimes\mathbb{Z})(\partial p) is its product over the four edges l,t,r,bl,t,r,b of the plaquette pp.

On the three qubits, LgL^g and RgR^g for g=r,sg=r,s are the Clifford gates of [6] in Fig. 5(a), and Lm=Rm=XL^m=R^m=X on the qubit of mm. Since Agh=AgAhA^{g h}=A^gA^h, it suffices to take gg among the generators rj,sj,mr_j,s_j,m in the bulk, and among those in (21) on the boundaries.

The plaquette stabilizers are, instead, diagonal. For two commuting operators with eigenvalues ±1\pm1, let @page 6/9 starts_mid=0 ends_mid=1 ... CZ[⋅,⋅]CZ[\cdot,\cdot] be −1-1 if both are −1-1 and +1+1 otherwise, which for two single-qubit Pauli ZZ gates is the usual CZ gate. For each copy of D4D_{4}, omitting the index jj, the plaquette stabilizers are

(Spr)2=(I⊗Z⊗I)(∂p),(Sps)2=(I⊗I⊗Z)(∂p),Spr2=(Z⊗I⊗I)(∂p)×CZ[(I⊗Z⊗Z)l,(I⊗Z⊗I)t]×CZ[(I⊗Z⊗Z)b,(I⊗Z⊗I)r],(44)\begin{aligned} (S_{p}^{r})^{2} &= (\mathbb{I}\otimes Z\otimes\mathbb{I})(\partial p),\\ (S_{p}^{s})^{2} &= (\mathbb{I}\otimes\mathbb{I}\otimes Z)(\partial p),\\ S_{p}^{r^{2}} &= (Z\otimes\mathbb{I}\otimes\mathbb{I})(\partial p)\\ &\quad\times CZ[(\mathbb{I}\otimes Z\otimes Z)_{l},(\mathbb{I}\otimes Z\otimes\mathbb{I})_{t}]\\ &\quad\times CZ[(\mathbb{I}\otimes Z\otimes Z)_{b},(\mathbb{I}\otimes Z\otimes\mathbb{I})_{r}], \tag*{(44)} \end{aligned}

and for the Z2\mathbb{Z}_{2} factor the stabilizer is Zm(∂p)Z_{m}(\partial p). The (Spr)2(S_{p}^{r})^{2} and (Sps)2(S_{p}^{s})^{2} plaquette stabilizers check the bits ρ0\rho_{0} and σ\sigma of the flux, and Spr2S_{p}^{r^{2}} checks the bit ρ1\rho_{1} (Figure 5(b)), where the two CZ gates account for the non-Abelian products glgtg_{l}g_{t} and gbgrg_{b}g_{r}.

On every edge of a boundary, the operator BiB_{i} in (27) acts on a single edge and imposes Pauli ZZ checks that depend on its subgroup:

RA:(I⊗Z⊗Z)j=(Z⊗Z⊗I)j=Zm=1,SA:(Z⊗I⊗I)j=(I⊗Z⊗I)j=1,TA:(I⊗Z⊗I)j=1,(Z⊗I⊗I)j(I⊗I⊗Z)j+1=1,(Z⊗I⊗I)kA−1Zm=1,(45)\begin{aligned} R_{A}:&\quad(\mathbb{I}\otimes Z\otimes Z)_{j}=(Z\otimes Z\otimes\mathbb{I})_{j}=Z_{m}=1,\\ S_{A}:&\quad(Z\otimes\mathbb{I}\otimes\mathbb{I})_{j}=(\mathbb{I}\otimes Z\otimes\mathbb{I})_{j}=1,\\ T_{A}:&\quad(\mathbb{I}\otimes Z\otimes\mathbb{I})_{j}=1,\\ &\quad(Z\otimes\mathbb{I}\otimes\mathbb{I})_{j}(\mathbb{I}\otimes\mathbb{I}\otimes Z)_{j+1}=1,\\ &\quad(Z\otimes\mathbb{I}\otimes\mathbb{I})_{k_{A}-1}Z_{m}=1, \tag*{(45)} \end{aligned}

for all copies jj, with j≤kA−2j\leq k_{A}-2 in the fourth line, and likewise for GBG_{B}. Only the checks of TAT_{A} couple different copies of D4D_{4}: they tie the qubit ρj,1\rho_{j,1} of rj2r_{j}^{2} to the qubit σj+1\sigma_{j+1} of sj+1s_{j+1} in the next copy, or to the qubit of mm.

In a local basis, obtained by acting with a Hadamard gate on the qubit ρj,1\rho_{j,1} of each copy, the stabilizers contain only Pauli XX, CZ and ZZ gates, which is the form of those of Dω(Z23)D^{\omega}(\mathbb{Z}_{2}^{3}) [2, 3, 4]. In both bases, the stabilizers are Clifford operators.

Level of the Gates. We now prove that the Clifford level of our logical gates cannot decrease by performing a change of logical basis. We will use that any two bases are related by a unitary, which preserves the ranks of operators. In general, a change of basis can change levels of the Clifford hierarchy.

The gate Ck−1C^{k-1}-Z lies in C(k)C^{(k)} [11, 17], and in no lower level in any basis.

Proof. We use the following bound: every non-scalar V∈C(j)V\in C^{(j)} on kk qubits has

rank⁡(V−λ)≥2k−j(46)\operatorname{rank}(V-\lambda)\geq2^{k-j} \tag*{(46)}

for every phase λ\lambda. For j=1j=1 it holds since every non-scalar Pauli operator has two eigenvalues of multiplicity 2k−12^{k-1}. We now proceed by induction and assume (46) holds for jj. For V∈C(j+1)V\in C^{(j+1)} and a Pauli operator PP, the commutator VP=VPV−1P−1V_{P}=VPV^{-1}P^{-1} lies in C(j)C^{(j)}, and

VP−1=(V−λ)PV−1P−1+λP(V−1−λ−1)P−1,(47)V_{P}-1=(V-\lambda)PV^{-1}P^{-1}+\lambda P(V^{-1}-\lambda^{-1})P^{-1}, \tag*{(47)}

so rank⁡(VP−1)≤2rank⁡(V−λ)\operatorname{rank}(V_{P}-1)\leq2\operatorname{rank}(V-\lambda). If VPV_{P} is non-scalar for some PP, the bound (46) for VPV_{P} gives rank⁡(V−λ)≥2k−j−1\operatorname{rank}(V-\lambda)\geq2^{k-j-1}. Otherwise VV commutes with every Pauli operator up to a phase, so it is a Pauli operator up to a phase, and the case j=1j=1 applies. Since ranks do not depend on the basis and rank⁡(Ck−1-Z−1)=1<2\operatorname{rank}(C^{k-1}\text{-}Z-1)=1<2, the gate does not lie in C(k−1)C^{(k-1)} in any basis.

Acknowledgments. We thank Dom Williamson for discussions and comments on the draft. This work is supported by the UKRI Frontier Research Grant, underwriting the ERC Advanced Grant “Generalized Symmetries in Quantum Field Theory and Quantum Gravity” (Schafer-Nameki). We acknowledge use of Claude Code (Fable 5.2, Opus 5.5) and ChatGPT Astra-6, for improvement of arguments and writing assistance and Fable 5.2 for initial exploration of this question.

References

  1. [1]S. Bravyi and R. Koenig, “Classification of Topologically Protected Gates for Local Stabilizer Codes,” Phys. Rev. Lett. 110 no. 17, (2013) 170503, arXiv:1206.1609 [quant-ph].DOI
  2. [2]Y. H. Hu, Y. Wan, and Y.-S. Wu, “Twisted quantum double model of topological phases in two dimensions,” Phys. Rev. B 87 no. 12, (2013) 125114, arXiv:1211.3695 [cond-mat.str-el].DOI
  3. [3]B. Yoshida, “Topological phases with generalized global symmetries,” Phys. Rev. B 93 no. 15, (2016) 155131, arXiv:1508.03468 [cond-mat.str-el].DOI
  4. [4]M. Davydova, A. Bauer, J. C. M. de la Fuente, M. Webster, D. J. Williamson, and B. J. Brown, “Universal Fault-Tolerant Quantum Computation in 2D without Getting Tied in Knots,” Phys. Rev. X 16 no. 3, (2026) 031001, arXiv:2503.15751 [quant-ph].DOI
  5. [5]R. Kobayashi, G. Zhu, and P.-S. Hsin, “Clifford Hierarchy Stabilizer Codes: Transversal Non-Clifford Gates and Magic States,” Phys. Rev. Lett. 136 no. 25, (2026) 250802, arXiv:2511.02900 [quant-ph].DOI
  6. [6]A. Warman and S. Schafer-Nameki, “Constant-Depth Clifford-Hierarchy Gates via Non-Abelian Surface Codes,” PRX Quantum 7 no. 3, (2026) 033029, arXiv:2512.13777 [quant-ph].DOI
  7. [7]M. Iqbal et al., “Non-Abelian topological order and anyons on a trapped-ion processor,” Nature 626 no. 7999, (2024) 505–511, arXiv:2305.03766 [quant-ph].DOI
  8. [8]J. C. M. de la Fuente, N. Feldman, J. Eisert, and A. Bauer, “High-threshold decoding of non-Pauli codes for 2D universality,” arXiv:2604.02033 [quant-ph].
  9. [9]Alternative recent approaches for non-Clifford gates via non-Abelian codes include lattice surgery with group surface codes [26–28], non-Abelian quantum LDPC codes [29, 30], measurements of logical Clifford operators by gauging and code surgery [31–33], and computation with gapped boundaries of quantum doubles [34].
  10. [10]C. Fechisin, K. Cooney, M. Raja, V. V. Albert, D. J. Williamson, K. Kawagoe, and S. Musser, “Clifford-hierarchy stabilizer formalism with applications to twisted quantum doubles,” arXiv:2610.00464 [quant-ph].
  11. [11]S. X. Cui, D. Gottesman, and A. Krishna, “Diagonal gates in the Clifford hierarchy,” Phys. Rev. A 95 no. 1, (2017) 012329, arXiv:1608.06596 [quant-ph].DOI
  12. [12]Y. Shi, “Both Toffoli and Controlled-NOT need little help to universal quantum computing,” Quant. Inf. Comput. 3 no. 1, (2003) 084–092, arXiv:quant-ph/0205115.DOI
  13. [13]D. Aharonov, “A Simple Proof that Toffoli and Hadamard are Quantum Universal,” arXiv:quant-ph/0301040.
  14. [14]C. Gidney, “Halving the cost of quantum addition,” Quantum 2 (2018) 74, arXiv:1709.06648 [quant-ph].DOI
  15. [15]D. Gottesman, “Stabilizer codes and quantum error correction,” arXiv:quant-ph/9705052.
  16. [16]N. Tantivasadakarn, R. Verresen, and A. Vishwanath, “Shortest Route to Non-Abelian Topological Order on a Quantum Processor,” Phys. Rev. Lett. 131 no. 6, (2023) 060405, arXiv:2209.03964 [quant-ph].
  17. [17]D. Gottesman and I. L. Chuang, “Demonstrating the viability of universal quantum computation using teleportation and single-qubit operations,” Nature 402 no. 6760, (1999) 390–393, arXiv:quant-ph/9908010.
  18. [18]A. Y. Kitaev, “Fault tolerant quantum computation by anyons,” Annals Phys. 303 (2003) 2–30, arXiv:quant-ph/9707021.
  19. [19]S. Beigi, P. W. Shor, and D. Whalen, “The Quantum Double Model with Boundary: Condensations and Symmetries,” Commun. Math. Phys. 306 no. 3, (2011) 663–694, arXiv:1006.5479 [quant-ph].DOI
  20. [20]For k_A = 1: G_A = ⟨m⟩, R_A = {id}, S_A = T_A = ⟨m⟩.
  21. [21]A. Warman, “AlgebrasTQD: Computing Algebras and Interfaces in Twisted Quantum Doubles,” arXiv:2610.36029 [cond-mat.str-el].arxiv.org/abs/2610.36029
  22. [22]V. V. Albert, D. Aasen, W. Xu, W. Ji, J. Alicea, and J. Preskill, “Spin chains, defects, and quantum wires for the quantum-double edge,” arXiv:2111.12096 [cond-mat.str-el].arxiv.org/abs/2111.12096
  23. [23]H. Bombin and M. A. Martin-Delgado, “A Family of Non-Abelian Kitaev Models on a Lattice: Topological Confinement and Condensation,” Phys. Rev. B 78 (2008) 115421, arXiv:0712.0190 [cond-mat.str-el].DOI
  24. [24]We can use this simplified expression in terms of tensor products of left-multiplication operators on each ribbon ξ because we are acting on identity group elements.
  25. [25]A. Warman, N. Tantivasadakarn, and S. Schafer-Nameki, “Exact Recovery for Non-Abelian Surface Codes,” arXiv:2610.03677 [quant-ph].arxiv.org/abs/2610.03677

Paper details

vlnc:W_jfzswpda
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Fable 5.2, Claude Opus 5.5 (bare), ChatGPT Astra-6

View statement
Looking for related papers…
Contents