betaFeedback

WorksPhysics and Astronomy

Phase-Tunable Liouvillian Topology: Dual Winding Structures and Skin Effects

Abstract

Liouvillian skin effects are usually associated with dissipation whose rates favor unidirectional flow. In this work, we show that directional edge localization can emerge even when population exchange between sites is exactly balanced, with a single tunable phase parameter αα deciding their direction and a dual winding structure deciding their nature. Although this number-conserving bath is known for eigenstate selection, we demonstrate here that its phase controls a steady current, which Hamiltonian hopping converts to directional edge localization. Further, we study the spectral winding structure by independently twisting the left (ket) and right (bra) vector indices of the density matrix, revealing two distinct windings whose relationship is dictated by an antiunitary mirror symmetry. Nearest-neighbor jump operators break this symmetry: both windings coincide, giving rise to a common spectral winding W0(α)=±1W_0(α)=\pm1, producing a unipolar skin effect in which the density accumulates at a specific edge. However, next nearest neighbor jump preserve it. The two windings are forced to be opposite, and the ket and bra are pushed to opposite edges, giving a bipolar skin effect of coherences present in eigenmodes, without edge localization of the density in the steady state. Remarkably, in a dimerized chain the phase-controlled density skin effect gives the steady-state zero modes a size-independent weight at the skin edge, accompanied by a kink at the topological transition. In addition, the relaxation time grows linearly with the system size, however at specific phases αα where the jump operators are rendered Hermitian, edge-localized eigenmodes with skin character coexist with diffusive relaxation.

Introduction

Open quantum systems coupled to Markovian baths evolve under the Lindblad master equation, whose generator, the Liouvillian superoperator, is non-Hermitian [1, 2, 3]. The complex Liouvillian spectrum can exhibit point-gap topology [4, 5, 6, 7, 8, 9, 10], with nonzero spectral winding giving rise to the Liouvillian skin effect (LSE) under open boundary conditions, characterized by the accumulation of a macroscopic number of eigenmodes at a boundary [11, 12, 13, 14, 15, 16, 17, 18, 19, 20]. The LSE also reshapes dissipative dynamics, most strikingly through relaxation times that diverge with system size despite an open Liouvillian gap [12].

In the mechanisms studied so far, the direction of the LSE is intrinsically encoded in the dissipative processes. Asymmetric incoherent hopping (γR≠γL\gamma_R \ne\gamma_L) induces preferential particle transport toward one boundary [12, 14], while unidirectional transfer between the sublattices of a chiral chain imprints the band winding onto the Liouvillian spectrum [19]. Gain and loss produce a non-reciprocal damping matrix [11] and two-body loss yields an LSE only in combination with an inversion-breaking Hamiltonian [15]. In each case the bath itself distinguishes left from right. Skin effects without asymmetric hopping are known for short-time approximated effective non-Hermitian Hamiltonians, where on site gain and loss act on bands with spin-momentum locking [21], however for Liouvillians the push has always been supplied by the bath. Could a bath with exactly balanced population exchange still produce a skin effect, and if yes then what selects its direction.

A natural setting for this question is a bath that has been studied extensively for a different purpose. The following number conserving jump operator Oj=(cj†+eiαcj+1†)(cj−eiαcj+l)O_j=(c_j^\dagger+e^{i\alpha}c_{j+1}^\dagger)(c_j-e^{i\alpha}c_{j+l}) removes a particle from one superposition of the sites jj and j+lj+l and returns it into another. For α=0\alpha=0, the jump operator corresponds to reservoir-engineering schemes that lock the relative phase between neighboring sites [22, 23, 24] and enable dissipative preparation of topological states [25, 26], while its effects in interacting chains were investigated in Ref. [27]. With a tunable phase it acts as a phase-selective filter in energy: it drives quasiperiodic chains into extended or localized steady states [28], prepares the edge states of a topological chain when it acts within one sublattice [29], and reshapes the relaxation and localization caused by boundary dissipation [30] and Mpemba-type effects [31, 32]. In all these studies, the phase determines the occupation of eigenstates. However, the real-space consequences of this phase remain unexplored from the perspective of Liouvillian topology. Phase-dependent jumps are known to produce Liouvillian skin effects for linear loss and gain baths [11]. To our knowledge, this question has not been examined for a number-conserving bath whose population exchange is exactly balanced and whose phase acts solely on the coherence between two sites.

Characterizing such an effect raises a second question. The Liouvillian acts on the density matrix from two sides at once, and a boundary twist can be applied to either side on its own. This method gives two separate winding numbers instead of one, windings of the Liouvillian’s spectrum, and their effects on the density and on the coherences may be distinct. The invariants used for the LSE so far give only one such number. Either the winding in the quasimomentum transfer KK, where the two sides left (ket) and right (bra) are equally affected [19] or a twist applied to one side alone [15]; where two windings have been compared, they belong to a Lindbladian and its postselected non-Hermitian effective Hamiltonian [33]. Twisting both sides independently defines instead a pair of windings (νK,νB)(\nu_K,\nu_B).

In this work we address both questions for a dimerized chain [34, 35] coupled to bond jumps of range l=1l=1 and l=2l=2. Several interesting features emerge. First, every jump exchanges population symmetrically on their own, but their two parts interfere through the bond coherences. Together with the coherent hopping, this interference drives a steady current around a periodic boundary, which an open boundary turns into edge localization. Second, for nearest-neighbor bonds, this directional push is captured by a Liouvillian winding W0(α)=±1W_0(\alpha)=\pm1, which belongs to the bath and is therefore the same in the topological and trivial phases of HH, unlike in Ref. [19]. Next, an antiunitary mirror-chiral symmetry decides whether νK\nu_K and νB\nu_B must move together. Nearest-neighbor jumps break it, and the density accumulates. Next-nearest-neighbor jumps preserve it, the ket and bra are forced to opposite edges instead, a bipolar skin effect. Furthermore, where the density accumulates, the zero-mode weight switches on with a kink at topological transition. Finally, the accumulation also makes the relaxation time grow linearly with system size [12], while at the Hermitian points relaxation stays diffusive even though the eigenmodes are edge localized. More broadly, our results demonstrate that directional dissipative transport can emerge from interference between coherent and incoherent processes, without requiring asymmetric rates. They further reveal that two indices of the density matrix can exhibit distinct point-gap windings [6, 7, 15], and that symmetry-protected bipolar localization [6, 36, 37] has a Liouvillian counterpart manifested in the coherences.

The remainder of this paper is organized as follows: Section II introduces the model, the selection of eigenstates (Sec. II A), and the skin effect with its dual windings (Sec. II B). Sections III and IV discuss the current, the zero modes, and the relaxation, § 1.4 a realization, and Sec. VI the outlook; derivations can be found in the Supplementary Material.

Model and the dissipator

We consider a particle hopping on a one-dimensional lattice of NN sites (NN even), labeled i=0,…,N−1i=0,\ldots,N-1, with sublattice AA on even and BB on odd sites, described by the Hamiltonian

H=∑i=0N−2ti(ci†ci+1+h.c.),ti={t1,i even,t2,i odd,(1)H=\sum_{i=0}^{N-2}t_i(c_i^\dagger c_{i+1}+\mathrm{h.c.}),\qquad t_i=\begin{cases} t_1,&i\ \mathrm{even},\\ t_2,&i\ \mathrm{odd}, \end{cases} \tag*{(1)}

where ci†c_i^\dagger and cic_i create and annihilate a particle on site ii. For t1<t2t_1<t_2 the off-diagonal Bloch block h(k)=t1+t2e−ikh(k)=t_1+t_2e^{-ik} winds once around the origin, binding one zero mode to each edge (topological phase) and for t1>t2t_1>t_2 it does not wind (trivial phase) [34, 35, 38]. Our results use only that HH is real, bipartite, and mirror symmetric namely,

ΓHΓ−1=−H,PHP−1=H,H∗=H,(2)\Gamma H\Gamma^{-1}=-H,\qquad PHP^{-1}=H,\qquad H^*=H, \tag*{(2)}

where the sublattice sign Γ\Gamma maps to ci↦(−1)icic_i\mapsto(-1)^ic_i and the selected bond mirror PP maps to ci↦cN−1−ic_i\mapsto c_{N-1-i}. According to the Born-Markov approximation [3, 39], the density matrix obeys the Lindblad master equation [1, 2]

ρ˙=Lαρ=−i[H,ρ]+γ∑j(OjρOj†−12{Oj†Oj,ρ}),(3)\dot{\rho}=\mathcal{L}_{\alpha}\rho=-i[H,\rho]+\gamma\sum_j\left(O_j\rho O_j^\dagger-\frac{1}{2}\{O_j^\dagger O_j,\rho\}\right), \tag*{(3)}

with the number-conserving bond jumps

Oj=(cj†+zcj+l†)(cj−zcj+l),z=eiα,(4)O_j=(c_j^\dagger+zc_{j+l}^\dagger)(c_j-zc_{j+l}),\qquad z=e^{i\alpha}, \tag*{(4)}

acting on every bond (j,j+l)(j,j+l), j=0,…,N−1−lj=0,\ldots,N-1-l. All bonds are oriented along the chain, j→j+lj\to j+l; this orientation, together with the phase α\alpha distinguishes the two ends of density matrices. We study l=1l=1, where each bond joins sites of different sublattices, and l=2l=2, where both sites of a bond lie on one sublattice (Figure 1(a)). We restrict our analysis to single-particle sector, since OjO_j conserves particle number.

Figure 1. (a) Hoppings t1t_1 (thin) and t2t_2 (thick); jumps act on bonds of range l=1l=1 (A–B) or l=2l=2 (A–A, B–B). (b) A jump at rate γ\gamma empties vjv_j and refills uju_j; arrows give the phase of each amplitude relative to site jj.

Balanced Current — Each jump factorizes into one annihilation and one creation parts of bond,

Oj=uj†vj,uj†=cj†+zcj+l†,vj=cj−zcj+l.(5)O_j=u_j^\dagger v_j,\qquad u_j^\dagger=c_j^\dagger+zc_{j+l}^\dagger,\qquad v_j=c_j-zc_{j+l}. \tag*{(5)}

So a jump removes a particle from vjv_j and returns it into $u_j {xref}Fig. 1(b)`, with {uj,uj†}={vj,vj†}=2\{u_j,u_j^\dagger\}=\{v_j,v_j^\dagger\}=2. Expanding Eq. (4) in the site operators separates the jump into an on-site and a transfer part,

Oj=OjD+OjT,OjD=nj−z2nj+l,O_j=O_j^D+O_j^T,\qquad O_j^D=n_j-z^2n_{j+l},
OjT=z(cj+l†cj−cj†cj+l).(6)O_j^T=z(c_{j+l}^\dagger c_j-c_j^\dagger c_{j+l}). \tag*{(6)}

with ni=ci†cin_i = c_i^\dagger c_i. In OjTO_j^T, the two hopping amplitudes differ only in sign while retaining equal magnitudes for an arbitrary α\alpha. This symmetry is preserved in the population dynamics of njn_j, for which the bath induces only symmetric particle exchange across an individual bond, irrespective of the coherences (see Sec. S2 of the Supplementary Material),

n˙j∣bond=γ(nj+l−nj)=−n˙j+l∣bond.(7)\dot{n}_j\big|_{\mathrm{bond}} = \gamma(n_{j+l}-n_j) = -\dot{n}_{j+l}\big|_{\mathrm{bond}}. \tag*{(7)}

By itself, the bath thus only flattens the density. This is the essential difference from incoherent hopping γRcj+1†cj\sqrt{\gamma_R}c_{j+1}^\dagger c_j and γLcj†cj+1\sqrt{\gamma_L}c_j^\dagger c_{j+1}, where the ratio γR/γL\gamma_R/\gamma_L usually sets the direction [12].

Hermitian points, interference, and symmetry — A bath that only flattens populations might be expected to leave the fully mixed state ρ=I/N\rho= I/N untouched. However, it does not. Evaluating the dissipator on II shows that it pumps energy into every bond,

Lα(I)=2γ(z+z∗)∑j(cj†cj+l+h.c.).(8)\mathcal{L}_{\alpha}(I) = 2\gamma(z+z^*)\sum_j(c_j^\dagger c_{j+l}+\mathrm{h.c.}). \tag*{(8)}

and this pump switches off only at the Hermitian points α=π/2+nπ\alpha= \pi/2+n\pi, where Oj†=OjO_j^\dagger= O_j and the identity is left invariant, ρss=I/N\rho_{\mathrm{ss}} = I/N (Sec. S2). Taken alone, each of ODO^D and OTO^T generates unital dynamics with ρss=I/N\rho_{\mathrm{ss}} = I/N for every HH and α\alpha, so every α\alpha-dependent feature of ρss\rho_{\mathrm{ss}} stems from their interference,

D[OD+OT]=D[OD]+D[OT]+X,\mathcal{D}[O^D+O^T] = \mathcal{D}[O^D]+\mathcal{D}[O^T]+\mathcal{X},
X(ρ)=ODρOT†+OTρOD†−12{OD†OT+OT†OD,ρ}.(9)\mathcal{X}(\rho)=O^D\rho O^{T\dagger}+O^T\rho O^{D\dagger} -\frac{1}{2}\{O^{D\dagger}O^T+O^{T\dagger}O^D,\rho\}. \tag*{(9)}

The interference requires coherent hopping to act on and for H=0H=0 the steady-state density is symmetric at every α\alpha (Sec. S2).

We now build the symmetry that governs this by reflecting the chain about its center, applying the sublattice sign, and complex-conjugating in the site basis, A(ρ)=PΓρ∗ΓPA(\rho)=P\Gamma\rho^*\Gamma P. The coherent dynamics remains invariant for any HH satisfying Eq. (2), since both conjugation and the sublattice transformation reverses the hopping, so the two cancel. Mirror reflection and complex conjugation each send α→−α\alpha\to-\alpha, so together they leave the phase of the jumps unchanged, and only the sublattice sign still acts on it. There its effect depends on where the bond sits: for l=2l=2 both sites of a bond share one sublattice and Γ\Gamma leaves the jump alone, so

ALαA−1=Lα(l=2).(10)A\mathcal{L}_{\alpha}A^{-1}=\mathcal{L}_{\alpha}\qquad(l=2). \tag*{(10)}

a symmetry at every α\alpha. For l=1l=1 the two sites of a bond lie on opposite sublattices, and Γ\Gamma shifts the phase instead, ΓOj(α)Γ−1=Oj(α+π)\Gamma O_j(\alpha)\Gamma^{-1}=O_j(\alpha+\pi) (Sec. S2), so that

ALαA−1=Lα+π(l=1),(11)A\mathcal{L}_{\alpha}A^{-1}=\mathcal{L}_{\alpha+\pi}\qquad(l=1), \tag*{(11)}

which relates two baths that push toward opposite edges rather than leaving Lα\mathcal{L}_{\alpha} fixed. Since mirror reflection alone already maps Lα\mathcal{L}_{\alpha} onto L−α\mathcal{L}_{-\alpha}, the phase α\alpha is the only parameter in the model that breaks inversion, and hence the only one that can select which edge the bath pushes toward.

Dissipative selection in the energy basis

The bath acts as a phase-selective pump in energy, a property that underlies its use for eigenstate selection [28, 29]. A jump on the bond jj removes the particle from the orbital ψj\psi_j, with amplitudes (1,−z∗)(1,-z^*) on the sites (j,j+l)(j,j+l), and returns it into ψj\psi_j, with amplitudes (1,z)(1,z). A single particle state ∑iψici†∣vac⟩\sum_i\psi_i c_i^\dagger|{\rm vac}\rangle is emptied with an amplitude

vj∑iψici†∣vac⟩=(ψj−zψj+l)∣vac⟩.(12)v_j\sum_i\psi_i c_i^\dagger|{\rm vac}\rangle=(\psi_j-z\psi_{j+l})|{\rm vac}\rangle. \tag*{(12)}

The two orbitals coincide at the Hermitian points, where a jump returns the particle to the orbital it came from. Further, they are orthogonal at α=0,π\alpha=0,\pi, where the transfer between them is complete and in between it is partial.

For l=1l=1 the two orbitals carry the bond energies ∓tjcos⁡α\mp t_j\cos\alpha, so every jump raises the energy by 2tjcos⁡α2t_j\cos\alpha, the pump of Eq. (8). The bath therefore fills the upper band for cos⁡α>0\cos\alpha>0 and the lower band for cos⁡α<0\cos\alpha<0, most strongly at α=0\alpha=0 and π\pi, where the refilled orbital is the in-phase or the out-of-phase combination of two neighbors. For l=2l=2 the jump compares amplitudes two sites apart on one sublattice. At α=0\alpha=0 it spares states with ψj+2=ψj\psi_{j+2}=\psi_j, which on a ring are the outer band edges E=±(t1+t2)E=\pm(t_1+t_2), and at α=π\alpha=\pi states with ψj+2=−ψj\psi_{j+2}=-\psi_j, the gap edges E=±∣t2−t1∣E=\pm|t_2-t_1|; on the other hand, the steady state favors the states closest to these patterns. The zero modes share the sign pattern favored at α=π\alpha=\pi but not its magnitude, and are emptied only weakly (Sec. S2C).

The symmetry (Eq. (11)) makes this useful for any HH obeying Eq. (2). Let dEd_E denote the eigenmode of HH at energy EE, and pE=⟨dE†dE⟩p_E=\langle d_E^\dagger d_E\rangle its steady-state population. Since the product of the mirror and the sublattice sign (we call PΓP\Gamma), maps dEd_E onto ±d−E\pm d_{-E}, these populations obey (Sec. S2)

{pE(α)=p−E(α),l=2,pE(α+π)=p−E(α),l=1.(13)\begin{cases} p_E(\alpha) = p_{-E}(\alpha), & l = 2,\\ p_E(\alpha+ \pi) = p_{-E}(\alpha), & l = 1. \tag*{(13)} \end{cases}

For l=2l = 2 both bands are always equally filled and α\alpha selects only the magnitude, i.e., ∣E∣|E|; for l=1l = 1 the shift α→α+π\alpha\to\alpha+ \pi exchanges the bands, as it later reverses the skin effect. Figure 2 shows both. For l=1l = 1 the weight appears in the upper (lower) band at α=0\alpha= 0 (π\pi), and is equipartitioned at Hermitian point. For l=2l = 2 it gathers at the outer band edges at α=0\alpha= 0 and around E=0E = 0 at α=π\alpha= \pi, where the in-gap zero-mode pair holds well above its equipartition weight. [29].

Steady-state density matrices in the energy eigenbasis

Figure 2. The bath phase selects a band for l=1l = 1 and an energy magnitude for l=2l = 2. Steady-state ∣ρpq∣|\rho_{pq}| in the eigenbasis of HH (ascending energy; zero modes at p,q=25,26p,q = 25,26), t1=1.0t_1 = 1.0, t2=1.5t_2 = 1.5, γ=1\gamma= 1, N=50N = 50. (a1)–(a3) l=1l = 1 and (b1)–(b3) l=2l = 2 at α=0,π/2,π\alpha= 0,\pi/2,\pi.

Liouvillian skin effect

Equation (7) excludes any directional push from the populations alone. In this section we show that the push nevertheless appears, first in the density for l=1l = 1, where a winding in the quasimomentum transfer fixes its direction (Sec. II B 1), and then in the two indices of ρ\rho separately, where l=1l = 1 and l=2l = 2 behave in opposite ways (Sec. II B 2).

Density skin effect for l=1 and its spectral winding

On a translationally invariant ring the Liouvillian couples the blocks ρ(k,k′)\rho(k,k') only within sectors of fixed quasimomentum transfer K=k′−kK = k' - k, which plays the role of the Bloch momentum of an effective one-dimensional Liouvillian, while kk labels its internal degrees of freedom [19]. For l=1l = 1 each sector takes the closed form

L(K)=diag⁡[L(K,k1),…,L(K,kL)]+NL⊗M0intra+S[NL⊗M~0(K)]S−1,(14)\mathcal{L}(K) = \operatorname{diag}[\mathcal{L}(K,k_1),\ldots,\mathcal{L}(K,k_L)] + N_L \otimes M_0^{\mathrm{intra}} + S[N_L \otimes\widetilde{M}_0(K)]S^{-1}, \tag*{(14)}

where L(K,k)\mathcal{L}(K,k) acts within the block ρ(k,k+K)\rho(k,k+K), NLN_L is the L×LL \times L all-ones matrix (L=N/2L = N/2), and M0intraM_0^{\mathrm{intra}} and M~0(K)\widetilde{M}_0(K) carry the jumps of the intracell and intercell bonds (Sec. S1). Its determinant D(K,λ)=det⁡[L(K)−λ]D(K,\lambda) = \det[\mathcal{L}(K) - \lambda] reduces to blocks of fixed size and is evaluated in closed form at every α\alpha.

The steady state sits at λ=0−\lambda= 0^-, and the topology of the band that supports it is read from the spectral winding about a reference just below,

W0=12πi∮dK ∂Kln⁡det⁡[L(K)−0−].(15)W_0 = \frac{1}{2\pi i}\oint dK\,\partial_K \ln\det[\mathcal{L}(K) - 0^-]. \tag*{(15)}

Nonzero spectral winding implies skin localization [6, 7, 8], so the steady state should pile up along the diagonal of ρss\rho_{\mathrm{ss}} whenever W0≠0W_0 \ne0, at the right edge for W0>0W_0 > 0; W0W_0 counts how often D(K,0−)/∣D∣D(K,0^{-})/|D| encircles the origin as KK sweeps the Brillouin zone Figure 3(c). Because α\alpha is the only element of the model that breaks inversion and Eq. (11) ties it to α+π\alpha+\pi,

Steady-state density skin effect and spectral winding

Figure 3. The bath phase sets the direction of the l=1l = 1 density skin effect, and the KK-sector winding counts it. (a1)–(a4) Steady-state ∣ρij∣|\rho_{ij}| in the site basis (sites labeled 1,…,N1,\ldots,N; N=50N = 50, t1=1.5t_1 = 1.5, t2=2.5t_2 = 2.5, γ=1\gamma= 1) at α=0\alpha= 0 and π\pi (identical), π/4\pi/4, π/2\pi/2, and 5π/45\pi/4. (b) W0(α)W_0(\alpha) from Eqs. (14) and (15) for L=10L = 10 unit cells, t1=0.5t_1 = 0.5, t2=1t_2 = 1, γ=0.1\gamma= 0.1. Squares mark the Hermitian points α=π/2+nπ\alpha= \pi/2 + n\pi, where the evaluated value (dots) is not an invariant. (c) Normalized determinant D(K,0−)/∣D∣D(K,0^-)/|D| over the Brillouin zone at α=π/4\alpha= \pi/4, colored by K/2πK/2\pi, encircling the origin once.

W0(−α)=−W0(α),W0(α+π)=−W0(α).(16)W_0(-\alpha)=-W_0(\alpha), \qquad W_0(\alpha+\pi)=-W_0(\alpha). \tag*{(16)}

Evaluating Eq. (15) gives

W0(α)={+1,0<α<π,−1,π<α<2π,α≠π2+nπ,0,α=0, π,(17)W_0(\alpha)= \begin{cases} +1, & 0<\alpha<\pi,\\ -1, & \pi<\alpha<2\pi,\quad\alpha\ne\frac{\pi}{2}+n\pi,\\ 0, & \alpha=0,\ \pi, \end{cases} \tag*{(17)}

Figure 3(b) depicts within each half-period no smooth change of α\alpha can reverse the push. At the Hermitian points the dynamics is unital and the steady-state branch touches the reference point, so the value returned there [dots in Figure 3(b)] carries no protected meaning, consistent with the featureless steady state.

Figure 3(a) confirms the expected localization. The steady state piles up at the right edge for 0<α<π0<\alpha<\pi and at the left edge for π<α<2π\pi<\alpha<2\pi, over a decay length of about two unit cells that does not change with NN, which identifies a bulk skin effect rather than the tail of an edge state. The profile is mirror symmetric at α=0,π\alpha=0,\pi, is mirror reflected under α→α+π\alpha\to\alpha+\pi, and is flat at α=π/2+nπ\alpha=\pi/2+n\pi (Sec. S2), following W0W_0 throughout. The push therefore needs the interference term (9) and coherent hopping for it to act on, together with broken inversion (α≠0,π\alpha\ne0,\pi) and non-unitality (α≠π/2+nπ\alpha\ne\pi/2+n\pi).

W0W_0 counts the displacement of the ket and the bra together, as a single density. Since ρ\rho carries two indices, we next twist them separately.

Density versus coherence: windings of ket and bra

The closed form (14) is specific to the l=1l=1 bonds, and W0W_0 in any case counts only the joint displacement of the two indices. A twisted ring resolves them and requires no unit cell [15]: every term of L\mathcal{L} that moves a particle forward across one chosen bond, cp†cqc_p^\dagger c_q with p=N−1p=N-1 and q=0q=0, is multiplied by eiθe^{i\theta}. The twist is a flux through the ring, and a nonzero winding of the spectrum under it means that the generator carries the twisted index around the ring in one direction; on an open chain the same transport piles that index at one end. Because the particle number is conserved, the operators acting on the ket and those acting on the bra of ρ\rho can be twisted separately (Sec. S1 C), which defines a ket winding νK\nu_K, a bra winding νB\nu_B, and, twisting both together, a density winding νC\nu_C,

νX(Λ)=12πi∫02πdθ ∂θln⁡det⁡[LX(θ)−Λ],(18)\nu_X(\Lambda)=\frac{1}{2\pi i}\int_0^{2\pi}d\theta\,\partial_\theta\ln\det\left[\mathcal{L}_X(\theta)-\Lambda\right], \tag*{(18)}

where X=K,B,CX=K,B,C labels the twisted index and positive values correspond to the right edge.

One relation holds for any number-conserving Lindbladian (Sec. S2). Hermiticity preservation means that conjugating ρ\rho exchanges the two twists with reversed sign, so

νB(Λ∗)=νK(Λ).(19)\nu_B(\Lambda^*)=\nu_K(\Lambda). \tag*{(19)}

On the real axis, the steady state and fast-relaxing modes have equal windings, so ket and bra move together, localizing the density. Opposite motion occurs at complex-conjugate eigenvalues, where the ket and bra windings are opposite.

For l=1l=1 a twist of 2π2\pi on either index moves every KK sector of Eq. (14) by one grid step, so that at the steady-state reference (Sec. S1)

νK(0−)=νB(0−)=W0,νC(0−)=2W0.(20)\nu_K(0^-)=\nu_B(0^-)=W_0,\qquad\nu_C(0^-)=2W_0. \tag*{(20)}

The closed-form W0W_0 is thus the winding of the density, with ket and bra pushed together. Remarkably, this invariant belongs to the bath rather than to the band topology of the bare Hamiltonian. With jumps on every bond, a shift of the unit cell by one site maps the ring Liouvillian for (t1,t2)(t_1,t_2) onto that for (t2,t1)(t_2,t_1), so that W0W_0 and the bulk skin effect are identical in the topological and trivial phases; the two open chains differ only at their ends.

This should be contrasted with Ref. [19], where jumps that anticommute with the sublattice operator let the band winding of the Hamiltonian set W0W_0, while jumps that commute with it leave W0W_0 to the dissipative channels alone. The l=1l=1 bond jump belongs to neither class: its on-site part commutes with Γ\Gamma and its transfer part anticommutes, and it is precisely the interference of the two that produces the skin effect.

The same push is already visible in the simplest sector of the dynamics. The coherence between one particle and the empty chain, ρi,vac=⟨vac∣ciρ∣vac⟩\rho_{i,\mathrm{vac}}=\langle\mathrm{vac}|c_i\rho|\mathrm{vac}\rangle, forms a closed block of the full Liouvillian generated by G=−iH−(γ/2)∑jOj†OjG=-iH-(\gamma/2)\sum_j O_j^\dagger O_j; the jump term is present but acts on nothing in this block (Sec. S3). Twisting its single index yields one closed loop per band. Figure 4 confirms the l=1l=1 picture. At α=π/4\alpha=\pi/4 every edge-localized eigenmode of the density block has ket and bra at the same edge, on a single region with νK>0\nu_K>0, and the vacuum sector shows one loop with νK=+1\nu_K=+1 whose open-chain modes sit at the right edge. At α=π\alpha=\pi, where every winding vanishes, the loops collapse and no eigenmode is edge localized.

Spectral winding and edge localization for ket and bra

Figure 4. For l=1l=1 the bath pushes ket and bra to the same edge, and the skin effect appears in the density. Topological phase (t1=1.5t_1=1.5, t2=2.5t_2=2.5, γ=1\gamma=1 and N=30N=30). (a) α=π\alpha=\pi: no edge-localized modes. (b) α=π/4\alpha=\pi/4: every edge-localized mode has ket and bra on the right edge (purple), on a single region with νK>0\nu_K>0. (c),(d) Vacuum-sector loops: two arcs with νK=0\nu_K=0 at α=π\alpha=\pi, one loop with νK=+1\nu_K=+1 at α=π/4\alpha=\pi/4, whose open-chain modes sit at the right edge.

Chirality changes the constraint. For a Hamiltonian satisfying Eq. (2) and jumps confined to one sublattice, as for l=2l=2, the symmetry AA holds at fixed α\alpha and leaves every twist unchanged. Combined with Hermitian conjugation it becomes the linear map ρ↦PΓρTΓP\rho\mapsto P\Gamma\rho^{T}\Gamma P, which sends the ket twist onto the reversed bra twist and the density twist onto itself reversed, so that at every reference point

νB(Λ)=−νK(Λ),νC(Λ)=0(l=2).(21)\nu_B(\Lambda)=-\nu_K(\Lambda), \qquad\nu_C(\Lambda)=0 \qquad(l=2). \tag*{(21)}

The content of Eq. (21) is read off from the two factors of AA: conjugation exchanges the roles of ket and bra, and the mirror exchanges the two ends, so whatever the bath does to the ket at one edge it does to the bra at the other. Together with Eq. (19) this gives νK=νB=0\nu_K=\nu_B=0 on the real axis: the l=2l=2 bath has no winding at the steady-state reference and hence no directional skin effect of the density, and the opposite windings it does allow are pushed off the axis. Equation (21) also identifies where the asymmetry between the two indices originates. The sublattice sign treats ket and bra alike; what distinguishes them is the preservation of Hermiticity, Eq. (19), while Γ\Gamma only undoes the sign change of HH under conjugation, which is possible at fixed α\alpha only when each jump preserves chirality. For l=1l=1 the same operation shifts α\alpha by π\pi, so it relates two baths that push opposite ways and leaves the density free to move.

The relation (21) has a direct consequence for the open-chain eigenmodes, each of which is a table over the ket index ii and the bra index jj. Since the ket winding at an eigenvalue λ\lambda predicts the edge at which the ket of the corresponding eigenmode localizes, and likewise for the bra, Eq. (21) allows only crossed corners for l=2l=2: the ket at one end and the bra at the other. Each such eigenmode is a coherence between the two ends rather than a population at either, its partner at λ∗\lambda^* has ket and bra exchanged, and its diagonal remains delocalized. The arrangement resembles the Z2\mathbb{Z}_2 skin effect, in which a symmetry places the two members of a partner pair at opposite ends [6] and here the two partners are the ket and the bra of a single eigenmode, paired by A\mathcal{A} together with Hermiticity preservation.

Figure 5 confirms this picture. At α=π/4\alpha=\pi/4 [Figure 5(b)] all edge-localized eigenmodes occupy the crossed corners, in equal numbers, and the ket side follows the sign of νK\nu_K at the eigenvalue. Figures 5(e) and 5(f) display a conjugate pair with ket and bra at opposite ends. At α=π\alpha=\pi [Figure 5(a)], where all windings vanish, no eigenmode is edge localized. These are genuine skin modes: a weak coupling between the two ends shifts their eigenvalues by an amount that grows exponentially with NN, and their left eigenvectors occupy the mirror corner, (Sec. S3). In the vacuum sector the two band loops wind in opposite directions for 0<α<π0<\alpha<\pi [Figure 5(d)], with modes at the left and right edges, and both collapse into arcs at α=π\alpha=\pi [Figure 5(c)], whose signs the outer arcs in Figure 5(b) inherit. The vacuum loops need less symmetry than Eq. (21) (Sec. S3).

Figure 5. For l=2l=2 the bath pushes ket and bra to opposite edges: a bipolar skin effect of coherences. Topological phase (t1=1.5t_1=1.5, t2=2.5t_2=2.5), γ=1\gamma=1. (a),(b) Spectra of the density block on the open chain (N=30N=30) at α=π\alpha=\pi and π/4\pi/4. Points are colored by the corner in which the eigenmode localizes (blue: ket left, bra right; red: ket right, bra left; gray: not edge localized). Shading is the sign of the ket winding νK\nu_K on a 12-site twisted ring (red: νK>0\nu_K>0, ket pushed right; blue: νK<0\nu_K<0, ket pushed left). (c),(d) Loops of the vacuum sector on a twisted ring at α=π\alpha=\pi and π/4\pi/4; arrows give the sense of rotation and xˉ\bar{x} the mean position of the corresponding open-chain modes. (e),(f) ∣ρij∣|\rho_{ij}| of the conjugate eigenmodes at λ=−4.53±2.49i\lambda=-4.53\pm2.49i: ket at the left and bra at the right edge, and the reverse.

The full dynamics, jumps included, does not transfer this bipolar structure to the steady state, which sits on the real axis where Eq. (21) forces both windings to vanish. Each jump refills a bond orbital and erases the band polarization built up by the push, so that the end-to-end coherence is absent from ρss\rho_{\mathrm{ss}}, and its residual imprint, a mirror-symmetric enhancement of short-range coherence at both edges, fades roughly as N−1/2N^{-1/2}. The bipolar skin effect is thus a property of the decaying modes and of transient coherence: a coherence prepared in the bulk spreads to both ends equally before it decays (Sec. S3).

Results and Analysis

The skin effect determines the spatial accumulation of the steady-state weight. We next investigate whether the zero modes of the bare Hamiltonian inherit this accumulation (Sec. III A) and how it is maintained against the bath-induced tendency toward a uniform distribution (Sec. III B).

Zero modes: skin-edge filling

The weight the steady state keeps on the zero modes is PZM=⟨Π⟩P_{\mathrm{ZM}}=\langle\Pi\rangle, with Π=∑mdm†dm\Pi=\sum_m d_m^\dagger d_m the number operator built from the two eigenstates of HH closest to E=0E=0. Since Π\Pi commutes with HH, the Hamiltonian cannot change this weight on its own, so PZMP_{\mathrm{ZM}} can only drift because of the jumps (Sec. S2). Each jump empties one bond orbital, vjv_j, and refills another, uj†u_j^\dagger. How much of this exchange reaches the zero modes is set by how much zero-mode character each orbital carries. We write wj=∑m∣{dm,uj†}∣2w_j=\sum_m|\{d_m,u_j^\dagger\}|^2 for the share carried by the refilled orbital uj†u_j^\dagger, and the emptied orbital vjv_j enters through its own population ⟨vj†vj⟩\langle v_j^\dagger v_j\rangle. With this notation,

dPZMdt=γ∑j[wj⟨vj†vj⟩−⟨{uj†vj,Π}⟩],\frac{dP_{\mathrm{ZM}}}{dt}=\gamma\sum_j\left[w_j\langle v_j^\dagger v_j\rangle-\langle\{u_j^\dagger v_j,\Pi\}\rangle\right],
wj=∑m∣{dm,uj†}∣2.(22)w_j=\sum_m|\{d_m,u_j^\dagger\}|^2. \tag*{(22)}

The first term feeds the zero modes at a rate set by wjw_j, and the second drains them whenever the emptied orbital itself has zero-mode weight. Since the zero modes decay away from the edges, wjw_j and the overlap with Π\Pi are both significant only on the few bonds nearest the two ends, so PZMP_{\mathrm{ZM}} tracks how much steady-state weight sits on those end bonds, accurately what the skin effect controls. For nearest-neighbor jumps the zero mode at the accumulation edge should therefore gain weight, and the one at the far edge should lose it.

Figure 6(a) confirms this behavior at α=π/4\alpha=\pi/4. The zero-mode occupation exceeds its equipartition value 2/N2/N across nearly two decades in γ\gamma, with weight concentrated entirely in the zero mode at the skin edge. It remains essentially size-independent (approximately 0.09 at γ=1\gamma=1 for N=50N=50–170), while 2/N2/N decreases with increasing NN. The maximum at intermediate γ\gamma reflects a competition between coherent and dissipative dynamics: weak dissipation allows the Hamiltonian to distribute the particle nearly uniformly, whereas strong dissipation drives the system toward a jump-determined state that no longer distinguishes the edges of HH from the bulk. By contrast, the two lowest states of the trivial chain, which are not edge-bound, remain below equipartition throughout.

Weight of zero modes and lowest-energy states as a function of bath parameters

Figure 6. The bath fills the zero mode at the skin edge, and the filling switches on at t1=t2t_1=t_2. l=1l=1, α=π/4\alpha=\pi/4, N=50N=50. (a) PZMP_{\mathrm{ZM}} (topological, t1=1.5t_1=1.5, t2=2.5t_2=2.5), PlowP_{\mathrm{low}} (trivial, t1=1.5t_1=1.5, t2=1.0t_2=1.0), and 2/N2/N versus γ\gamma. (b) Weight of the two lowest-∣E∣|E| states versus t2/t1t_2/t_1 at t1=1.0t_1=1.0.

The behavior reverses across the transition, as shown in Figure 6(b). For t2>t1t_2>t_1, the projected states are localized at the two ends, allowing the edge bonds to couple strongly to them and their occupation to increase with γ\gamma. In contrast, for t2<t1t_2<t_1, the two lowest states extend across the chain, yielding overlaps of order 1/N1/N with individual bond orbitals and a progressive depletion of their occupation as γ\gamma increases. Consequently, the curves for different γ\gamma intersect at t1=t2t_1=t_2, and in the thermodynamic limit,

lim⁡N→∞Plow=0(t2≤t1),lim⁡N→∞PZM>0(t2>t1),(23)\begin{aligned} \lim_{N\to\infty} P_{\mathrm{low}} &= 0 \qquad(t_2\le t_1),\\ \lim_{N\to\infty} P_{\mathrm{ZM}} &> 0 \qquad(t_2>t_1), \tag*{(23)} \end{aligned}

with finite-size effects rounding the resulting kink. Thus, topology determines the existence of edge states, whereas the bath controls their occupation. Remarkably, although the nearest-neighbor bath breaks the chirality protecting these states and it preferentially populates the zero mode at the skin edge while depleting its equally protected counterpart. For next-nearest-neighbor bonds, the symmetry AA enforces equal occupation of both edges, preventing a finite share from surviving in either zero mode (Sec. S2).

Coherent current

The bath only evens out the populations of neighboring sites, so weight cannot stay gathered near one edge unless something carries it there. The only other process is coherent hopping. Under HH, a particle moves from site jj to site j+1j+1 with the current

Jj=2tjIm⁡ρj,j+1,ρj,j+1=⟨cj+1†cj⟩,(24)J_j=2t_j\operatorname{Im}\rho_{j,j+1},\qquad\rho_{j,j+1}=\langle c^\dagger_{j+1}c_j\rangle, \tag*{(24)}

which is set by the coherence ρj,j+1\rho_{j,j+1} between the two sites. It is nonzero only when the amplitudes on the two sites differ in phase, and the bath phase α\alpha enters precisely through such phase differences between the sites of a bond. In an open-chain steady state no weight leaves through the ends, so the current must cancel the eveningout by the bath,

Jj={γ(nj+1−nj),l=1,γ(nj+1+nj+2−nj−1−nj),l=2,(25)J_j = \begin{cases} \gamma(n_{j+1}-n_j), & l=1,\\ \gamma(n_{j+1}+n_{j+2}-n_{j-1}-n_j), & l=2, \end{cases} \tag*{(25)}

with njn_j the steady-state population of site jj and terms on nonexistent sites omitted (Sec. S2). The bath moves weight from the more crowded site to the less crowded one, and the current carries it back.

Equation (25) links the coherences to the populations, and through them to the zero modes. For l=1l=1, adding it up from the bulk to an end shows that the extra population at the end equals the current collected along the way divided by γ\gamma. The zero modes of HH sit on the last few sites and are fed by the population there (Sec. III A). The zero mode at an edge should therefore be filled by the current flowing into that edge, and only when the steady state carries coherence. The zero modes cannot carry this current themselves, since each lives on a single sublattice and has no coherence between neighboring sites.

Current as the origin of nearest-neighbor jumps accumulation.- On a ring, the steady-state current is uniform across all bonds. Since the bath-induced contributions γ(nj−nj+1)\gamma(n_j-n_{j+1}) cancel within each unit cell, the net current,

J=12(2t1Im⁡ρAB+2t2Im⁡ρBA′),(26)J=\frac{1}{2}\left(2t_1\operatorname{Im}\rho_{AB}+2t_2\operatorname{Im}\rho_{BA'}\right), \tag*{(26)}

arises entirely from coherent hopping, mediated by bond coherences imprinted by the bath phase α\alpha, where ρAB\rho_{AB} and ρBA′\rho_{BA'} denote the intracell and intercell coherences, respectively. Both mirror reflection, which maps α→−α\alpha\to-\alpha, and the symmetry A\mathcal{A}, which shifts α→α+π\alpha\to\alpha+\pi, reverse the current. Consequently, JJ vanishes at α=0,π\alpha=0,\pi, where inversion symmetry is restored, and at the Hermitian points, where the steady state reduces to the identity and all coherences vanish. Upon disconnecting the ring, the circulating current is interrupted and particles accumulate toward the edge favored by the current. Consistently, the current direction matches the open-chain accumulation direction for all tested values of α\alpha, irrespective of the bare Hamiltonian’s topological phase.

On the open chain the current is appreciable only near the edge, where the populations change from site to site [Figure 7(a)]. Away from the last bond or two it is nearly the same in the topological and trivial chains, as expected since the winding W0W_0 belongs to the bath. The same current fills the zero mode at the skin edge. The population that the current gathers at the end is what the end bonds draw on to feed the zero mode, so as α\alpha is varied the zero-mode weight above equipartition should follow the current into that edge. Figure 7(b) and Figure 7(c) confirm this. The two curves change sign at nearly the same phases, peak at similar phases, and vanish at the Hermitian point, where the steady state is the identity and carries no coherence. They are similar in shape but not identical. Panel (c) is panel (b) shifted by π\pi, since A\mathcal{A} relates α\alpha and α+π\alpha+\pi and thereby exchanges the two edges.

Five-panel plots of hopping current, zero-mode occupation, and current into the edges

Figure 7. The coherent current holds the pile-up and tracks the zero modes, as their cause for l=1l=1 and as their consequence for l=2l=2. (a) Hopping current 2tjIm⁡ρj,j+12t_j\operatorname{Im}\rho_{j,j+1} versus bond position for l=1l=1 at α=π/4\alpha=\pi/4, topological (solid) and trivial (dashed) phases; gray bands mark the edges. (b),(c) Zero mode at the skin edge (excess over equipartition) and current into that edge versus α\alpha, each scaled to its maximum (line colors as in the legends): right edge on (0,π)(0,\pi), left edge on (π,2π)(\pi,2\pi). (d) Same as (a) for l=2l=2 at α=0.9π\alpha=0.9\pi: the current flows into both edges. (e) l=2l=2: zero-mode pair (excess over equipartition) and current into the edges versus α\alpha; shading marks where the zero modes are filled and the current flows in, or emptied and it flows out. t1=1.0t_1=1.0, t2=1.5t_2=1.5 (topological) and t1=1.5t_1=1.5, t2=1.0t_2=1.0 (trivial), γ=1\gamma=1, N=50N=50.

The filling of the zero mode is thus a consequence of the current driven by the bath phase, and not a property of the zero modes alone.

Current as a consequence for next-nearest neighbor jumps.—For l=2l=2 the symmetry A\mathcal{A} holds at every α\alpha, so the two ends of the chain are equivalent. A current that flows toward the right at the right end must flow toward the left at the left end. It therefore runs into both ends or away from both, and no current can circulate around a ring [Figure 7(d)]. Here the zero modes are not filled by a current. They are filled directly by the energy selection of Sec. II A, which favors their sign pattern around α=π\alpha=\pi. Once filled, they leave the last sites more populated than their neighbors, the bath tries to even this out, and a coherent current from the bulk has to run into the ends to keep it in place. The current into the edges then follows the zero-mode excess [Figure 7(e)]. It flows inward for π/2<α<3π/2\pi/2<\alpha<3\pi/2, where the zero modes hold more than equipartition, and outward elsewhere, equally at both ends. The trivial chain has no zero modes and carries almost no current at its edges.

Relaxation dynamics

We finally ask how the density skin effect shows up in the approach to the steady state, restricting the discussion to l=1l=1. Following Haga et al. [12], we start a particle at the left end, ρ(0)=c0†∣vac⟩⟨vac∣c0\rho(0)=c_0^\dagger\lvert\mathrm{vac}\rangle\langle\mathrm{vac}\rvert c_0, monitor the density at the right end, and define τ\tau as the latest time at which its deviation from the steady value exceeds e−1e^{-1} of the initial deviation. In the skin regime the steady state sits at the right end, so this protocol starts the particle as far as possible from where it ends up. The asymptotic decay is set by the Liouvillian gap Δ=∣Re⁡λ1∣\Delta=\lvert\operatorname{Re}\lambda_1\rvert, but the gap alone does not fix τ\tau. If the slowest right and left eigenmodes are localized at opposite edges with length ξ\xi, their overlap is exponentially small, a left-edge initial state needs an exponentially large amplitude in the slowest mode, and this amplitude must decay at rate Δ\Delta [12],

τ∼Δ−1(1+N/ξ).(27)\tau\sim\Delta^{-1}(1+N/\xi). \tag*{(27)}

Here ξ\xi is the localization length of the steady state, read off from the exponential decay of the density away from the skin edge. The argument uses only the fact that the steady state and the slowest modes are localized at an edge, not how that localization arises. It therefore applies to our bath, whose transfer amplitudes have equal magnitude in both directions, exactly as it does to the asymmetric hopping of Ref. [12]: once the interference of Eq. (9) has produced a skin effect, the relaxation follows the same law.

In the skin regime the gap stays finite, Δ→Δ∞(α)\Delta\to\Delta_{\infty}(\alpha), and ξ(α)\xi(\alpha) is finite and independent of NN (about 3.8 sites at α=π/4\alpha= \pi/4), so Eq. (27) predicts relaxation linear in size,

τ≃Nξ(α)Δ∞(α).(28)\tau\simeq\frac{N}{\xi(\alpha)\Delta_{\infty}(\alpha)}. \tag*{(28)}

Figure 8(a) and Figure 8(b) confirm this: Δ\Delta saturates, τ\tau grows linearly, and τΔ∞ξ/N≈0.2\tau\Delta_{\infty}\xi/N \approx0.2 for all skin-regime values of α\alpha shown (ξ\xi measured in sites). At α=0\alpha= 0 and π\pi there is no edge accumulation, the gap closes as Δ∝N−2\Delta\propto N^{-2} and τ≃Δ−1∝N2\tau\simeq\Delta^{-1} \propto N^2, which is Eq. (27) with N/ξ→0N/\xi\to0 (diffusive spreading). The dependence on α\alpha [Figure 9(a) and Figure 9(b)] is symmetric under α→π−α\alpha\to\pi- \alpha, which follows from combining A\mathcal{A} with the mirror. Relaxation is fastest near α=π/4\alpha= \pi/4, where the relaxation velocity N/τ∼ξΔ∞N/\tau\sim\xi\Delta_{\infty} is largest. Because τ\tau is linear in NN there and quadratic at α=π/2\alpha= \pi/2, the Hermitian point is already about three times slower than α=π/4\alpha= \pi/4 at N=80N = 80, and the ratio keeps growing with NN.

Relaxation gap and relaxation time versus system size

Figure 8. Relaxation is linear in size only in the skin regime. l=1l=1, t1=1.0t_1=1.0, t2=1.5t_2=1.5, γ=1\gamma=1. (a) Gap Δ\Delta and (b) relaxation time τ\tau versus NN; black lines indicate N−2N^{-2}, NN, and N2N^2.

Plots of the relaxation gap and time versus phase

Figure 9. The relaxation is symmetric about α=π/2\alpha= \pi/2 and fastest near α=π/4\alpha= \pi/4 and 3π/43\pi/4. (a) Δ\Delta and (b) τ\tau versus α\alpha for N=40N = 40 and 80; other parameters as in Figure 8.

Hermitian points.—The points α=π/2+nπ\alpha= \pi/2 + n\pi are an exception to Eq. (28). Here the jumps are Hermitian, Oj†=OjO_j^\dagger= O_j, and the open-chain eigenmodes are still edge localized, with skin character. However, they occur in mirror pairs: the adjoint Liouvillian, which propagates the left eigenmodes, is the mirror image of the complex-conjugated Liouvillian (Sec. S2),

L†(X)=P[L(PX∗P)]∗P(α=π/2+nπ),(29)\mathcal{L}^{\dagger}(X)=P\left[\mathcal{L}\left(PX^{*}P\right)\right]^{*}P \qquad(\alpha= \pi/2+n\pi), \tag*{(29)}

so the left eigenmode belonging to each right eigenmode sits at the opposite edge, as in any skin effect. The right eigenmodes themselves stay localized at the same edge as for neighboring α\alpha. What the Hermitian point removes

is the localization of the steady state, not that of the decaying modes.

A Hermitian bath has a more basic consequence. Hermitian jumps always leave the identity invariant, L(I)=0\mathcal{L}(\mathbb{I}) = 0, for any HH, so the steady state is the featureless mixture ρss=I/N\rho_{\mathrm{ss}} = \mathbb{I}/N [Eq. (8)]. The argument of Haga et al. requires the steady state and the slowest decaying mode to sit at an edge: only then does an initial state at the opposite edge need an exponentially large amplitude in the slowest mode. At the Hermitian points both are extended, so N/ξ→0N/\xi\to0 in Eq. (27) and the relaxation is set by the gap alone, τ≃Δ−1\tau\simeq\Delta^{-1}. Because Δ∝N−2\Delta\propto N^{-2}, the relaxation is diffusive, τ∝N2\tau\propto N^2 [Figure 8]. Edge-localized eigenmodes therefore do not by themselves produce skin-type relaxation; what matters is whether the steady state is localized, and a Hermitian bath always leaves it unitary.

Proposed experimental realization

In order to connect the theory with a concrete setting, we outline how the bond jump Eq. (4) can be engineered and how its consequences can be detected. From a microscopic point of view, the jump factorizes into two steps [Figure 10(a)]: a coherent drive that transfers the particle from the bond mode vjv_j to a short-lived auxiliary mode aja_j, and a decay of aja_j that returns it into uju_j,

Proposed realization of the bond jump and dimerized optical lattice

Figure 10. Proposed realization. (a) A drive couples the bond mode vjv_j (amplitudes Ω\Omega and −Ωeiα-\Omega e^{i\alpha} on the sites jj and j+lj+l) to a short-lived auxiliary mode aja_j, which decays at rate κ\kappa into uju_j (amplitudes 1 and eiαe^{i\alpha}); for κ≫Ω\kappa\gg\Omega this realizes OjO_j with γ∝Ω2/κ\gamma\propto\Omega^2/\kappa. (b) Dimerized optical lattice with an auxiliary level for every bond (l=1l = 1); the steady-state density njn_j piles up at the right edge for 0<α<π0 < \alpha< \pi.

Hd=Ω∑j(aj†vj+h.c.),Lj=κ uj†aj,(30)H_d=\Omega\sum_j(a_j^\dagger v_j+\mathrm{h.c.}),\qquad L_j=\sqrt{\kappa}\,u_j^\dagger a_j, \tag*{(30)}

with vjv_j and uj†u_j^\dagger of Eq. (5). For κ≫Ω\kappa\gg\Omega, adiabatic elimination of the auxiliary modes [40] yields Eq. (4) with γ∝Ω2/κ\gamma\propto\Omega^2/\kappa. This is the structure of reservoir-engineering schemes for cold atoms [22, 23, 41]: the auxiliary mode is an excited level held in a state-dependent lattice shifted by half a period, the drive is a laser whose phase pattern sets the relative phase in vjv_j, and the decay proceeds by spontaneous emission or by emission into a reservoir gas. Explicit cold-atom schemes for these bond jumps with l=1l=1 and l=2l=2 have been proposed in Ref. [31], and related proposals exist for superconducting resonator arrays [42].

In such schemes, isotropic emission refills the in-phase combination, which corresponds to eiα=1e^{i\alpha}=1 in uj†u_j^\dagger. A general phase α\alpha is nevertheless accessible. The single-site gauge transformation cj→eiαj/lcjc_j\rightarrow e^{i\alpha j/l}c_j maps the jump Eq. (4) onto

Oj↦(cj†+cj+l†)(cj−e2iαcj+l),(31)O_j\mapsto(c_j^\dagger+c_{j+l}^\dagger)(c_j-e^{2i\alpha}c_{j+l}), \tag*{(31)}

at the expense of a uniform Peierls phase α/l\alpha/l per lattice spacing in HH. Since this transformation is diagonal in the site basis, the open-chain densities are unchanged. An in-phase refill, a drive with relative phase π+2α\pi+2\alpha between the bond sites, and a synthetic gauge field realized by laser-assisted tunneling [43, 44] therefore reproduce the model at arbitrary α\alpha. Dimerized optical lattices of the required form have been realized and characterized topologically [45].

The steady-state signatures discussed above are accessible with site-resolved imaging in quantum gas microscopes [46, 47]: the density profile reveals the pile-up and its reversal under α→α+π\alpha\rightarrow\alpha+\pi, and the coherent current can be measured with the double-well projection technique of Ref. [48]. Since the dynamics is restricted to a single particle, the particle statistics play no role, and arrays of qubits with a single excitation offer an alternative platform [49, 50, 51]. There, the vacuum-sector coherence of Sec. II B 2 becomes a local observable, ρi,vac=⟨σi−⟩\rho_{i,\mathrm{vac}}=\langle\sigma_i^-\rangle, so that the opposite windings of the two vacuum loops for l=2l=2 can be probed by single-qubit tomography.

Summary and outlook

We have shown that the number-conserving jump operators with a tunable phase generate a Liouvillian skin effect in dimerized chain, even if their exchange of population is balanced. The push originates from the interference between the on-site and transfer parts of each jump operators and is fixed by the phase α\alpha through the spectral winding W0(α)W_0(\alpha), which we evaluated from the closed-form KK-sector Liouvillian L(K)\mathcal{L}(K) for the nearest-neighbor bonds and which belongs to the bath rather than to the band topology of the bare Hamiltonian. Resolving the winding into the two indices of the density matrix revealed two distinct skin effects, decided by a single antiunitary mirror-chiral symmetry. For l=1l=1 the jumps operators break it, the ket and bra are pushed together, producing a skin effect of the density. For l=2l=2 the jumps preserve however it, the ket and bra are pinned to opposite edges, producing a bipolar skin effect of coherences that resides in the fastest relaxing eigen modes. The density skin effect is sustained by a coherent current, fills the zero mode at the skin edge with a size-independent weight that switches on at t1=t2t_1=t_2, and makes relaxation linear in system size except at the Hermitian points.

Several questions remain open like an analytic expression for W0(α)W_0(\alpha) beyond its symmetry constraints, Eq. (16), is lacking. The necessary ingredients identified here, interference, coherent hopping, broken inversion, and non-unitality, may serve as a starting point for a general criterion for skin effects induced by balanced jumps [52]. The relation between the pair (νK,νB)(\nu_K,\nu_B) and the localization corner of each eigenmode is supported numerically but not yet proven. Finally, the jumps are quadratic in the fermion operators, so that the Lindbladian is not quadratic and the many-particle problem lies beyond the Gaussian frameworks of [53, 54] and the fate of the two skin effects at finite filling remains to be explored.

References

Liouvillian windings: the KK sector and ket and bra twists

We first construct the Liouvillian of the periodic l=1l=1 chain in the sector of fixed quasimomentum transfer, following [19]. We then define the twisted ring of [15] separately for the ket and the bra index, which applies to both ranges, and show that the two constructions agree at the steady-state reference.

The KK sector for l=1l=1

Conventions. The ring has LL unit cells m=1,…,Lm=1,\ldots,L with sites (Am,Bm)(A_m,B_m). The two l=1l=1 jumps of cell mm are

Omintra=(cA,m†+zcB,m†)(cA,m−zcB,m),Ominter=(cB,m†+zcA,m+1†)(cB,m−zcA,m+1).(32)O_m^{\mathrm{intra}}=\left(c_{A,m}^{\dagger}+z c_{B,m}^{\dagger}\right)\left(c_{A,m}-z c_{B,m}\right),\qquad O_m^{\mathrm{inter}}=\left(c_{B,m}^{\dagger}+z c_{A,m+1}^{\dagger}\right)\left(c_{B,m}-z c_{A,m+1}\right). \tag*{(32)}

With cσ,m=L−1/2∑keikmcσ,kc_{\sigma,m}=L^{-1/2}\sum_k e^{ikm}c_{\sigma,k} (σ=A,B\sigma=A,B) and kn=2πn/Lk_n=2\pi n/L, the Bloch Hamiltonian is

H(k)=(0h(k)h∗(k)0),h(k)=t1+t2e−ik.(33)H(k)=\begin{pmatrix} 0 & h(k) \\ h^{*}(k) & 0 \end{pmatrix},\qquad h(k)=t_1+t_2e^{-ik}. \tag*{(33)}

We collect the coherences ρσσ′(k,k′)=⟨cσ′,k′†cσ,k⟩\rho_{\sigma\sigma'}(k,k')=\langle c_{\sigma',k'}^{\dagger}c_{\sigma,k}\rangle into 2×22\times2 sublattice blocks ρ(k,k′)\rho(k,k'), vectorized in the order (AA,AB,BA,BB)(AA,AB,BA,BB) with vec⁡(X⊗Y)=(X⊗YT)vec⁡ρ\operatorname{vec}(X\otimes Y)=(X\otimes Y^{T})\operatorname{vec}\rho. A translation by one cell multiplies ρ(k,k′)\rho(k,k') by ei(k′−k)e^{i(k'-k)} and leaves L\mathcal{L} invariant, so L\mathcal{L} is block diagonal in K=k′−kK=k'-k. Within a sector, blocks are labeled by the internal momentum kk (with k′=k+Kk'=k+K). We write L(K,k)\mathcal{L}(K,k) for the part of L\mathcal{L} that acts within one block, while the jump terms couple different blocks of the same sector. Operators such as O=∑σσ′cσ†(O)σσ′cσ′O=\sum_{\sigma\sigma'}c_{\sigma}^{\dagger}(O)_{\sigma\sigma'}c_{\sigma'} are represented by their coefficient matrices.

Hamiltonian and loss terms. The Hamiltonian part of the block (k,k′)(k,k') is

LH(k,k′)=−i[H(k)⊗I−I⊗HT(k′)]=(0ih∗(k′)−ih(k)0ih(k′)00−ih(k)−ih∗(k)00ih∗(k′)0−ih∗(k)ih(k′)0).(34)\mathcal{L}_{H}(k,k')=-i\left[H(k)\otimes\mathbb{I}-\mathbb{I}\otimes H^{T}(k')\right] = \begin{pmatrix} 0 & ih^{*}(k') & -ih(k) & 0 \\ ih(k') & 0 & 0 & -ih(k) \\ -ih^{*}(k) & 0 & 0 & ih^{*}(k') \\ 0 & -ih^{*}(k) & ih(k') & 0 \end{pmatrix}. \tag*{(34)}

The intracell jump carries no momentum dependence, and its coefficient matrix and that of O†OO^{\dagger}O are

Ointra=(1−zz−z2),Ointra†Ointra=2(1−z−z∗1).(35)O^{\mathrm{intra}}=\begin{pmatrix}1&-z\\z&-z^2\end{pmatrix},\qquad O^{\mathrm{intra}\dagger}O^{\mathrm{intra}}=2\begin{pmatrix}1&-z\\-z^{*}&1\end{pmatrix}. \tag*{(35)}

and its loss term is −γ2[O†O⊗I+I⊗(O†O)T]-\frac{\gamma}{2}\left[O^{\dagger}O\otimes\mathbb{I}+\mathbb{I}\otimes(O^{\dagger}O)^{T}\right]. The intercell jump at equal input and output momentum is

OI(k)=(−z2ze−ik−zeik1),OI†OI(k)=2(1−z∗e−ik−zeik1).(36)O_I(k)=\begin{pmatrix}-z^2&ze^{-ik}\\-ze^{ik}&1\end{pmatrix},\qquad O_I^{\dagger}O_I(k)=2\begin{pmatrix}1&-z^{*}e^{-ik}\\-ze^{ik}&1\end{pmatrix}. \tag*{(36)}

with loss term −γ2[OI†(k)OI(k)⊗I+I⊗(OI†(k′)OI(k′))T]-\frac{\gamma}{2}\left[O_I^\dagger(k)O_I(k)\otimes I+I\otimes\left(O_I^\dagger(k')O_I(k')\right)^{\mathrm T}\right]. The diagonal block is

L(K,k)=LH(k,k+K)+Llossintra+Llossinter(k,k+K).(37)\mathcal{L}(K,k)=\mathcal{L}_H(k,k+K)+\mathcal{L}_{\mathrm{loss}}^{\mathrm{intra}}+\mathcal{L}_{\mathrm{loss}}^{\mathrm{inter}}(k,k+K). \tag*{(37)}

Jump (recycling) terms. Because OintraO^{\mathrm{intra}} is local and momentum independent, ∑mOmρOm†\sum_m O_m\rho O_m^\dagger couples every pair of blocks of a sector with the same matrix,

M0intra=γLO⊗O∗=γL(1−z∗−z1z∗−(z∗)2−1z∗z−1−z2z1−z∗−z1).(38)M_0^{\mathrm{intra}}=\frac{\gamma}{L}O\otimes O^*=\frac{\gamma}{L} \begin{pmatrix} 1 & -z^* & -z & 1\\ z^* & -(z^*)^2 & -1 & z^*\\ z & -1 & -z^2 & z\\ 1 & -z^* & -z & 1 \end{pmatrix}. \tag*{(38)}

A jump removes the particle from one Bloch pair and refills a local bond orbital, which contains every momentum. For the intercell bond, Ominter=cB,m†cB,m−zcB,m†cA,m+1+zcA,m+1†cB,m−z2cA,m+1†cA,m+1O_m^{\mathrm{inter}}=c_{B,m}^\dagger c_{B,m}-z c_{B,m}^\dagger c_{A,m+1}+z c_{A,m+1}^\dagger c_{B,m}-z^2c_{A,m+1}^\dagger c_{A,m+1} in real space, and in momentum space

Ominter=1L∑k,qe−i(k−q)mck†OI(k,q)cq,OI(k,q)=(−z2e−i(k−q)ze−ik−zeiq1),(39)O_m^{\mathrm{inter}}=\frac{1}{L}\sum_{k,q}e^{-i(k-q)m}c_k^\dagger O_I(k,q)c_q,\qquad O_I(k,q)= \begin{pmatrix} -z^2e^{-i(k-q)} & ze^{-ik}\\ -ze^{iq} & 1 \end{pmatrix}, \tag*{(39)}

with OI(k,k)≡OI(k)O_I(k,k)\equiv O_I(k). The sum over mm conserves only KK, so the block (k,k+K)(k,k+K) receives a contribution from every block (q,q+K)(q,q+K) of the sector,

γLOI(k,q)⊗OI∗(k+K,q+K)=SkM~0(K)Sq−1,(40)\frac{\gamma}{L}O_I(k,q)\otimes O_I^*(k+K,q+K)=S_k\widetilde{M}_0(K)S_q^{-1}, \tag*{(40)}

with gauge factors Sk=diag⁡(1,e−ik,eik,1)S_k=\operatorname{diag}(1,e^{-ik},e^{ik},1) and

M~0(K)=γL(1−zeiK−z∗eiKze−iK−z2−e−iKzz∗−eiK−(z∗)2z∗eiKe−iK−z−z∗e−iK1).(41)\widetilde{M}_0(K)=\frac{\gamma}{L} \begin{pmatrix} 1 & -ze^{iK} & -z^* & e^{iK}\\ ze^{-iK} & -z^2 & -e^{-iK} & z\\ z^* & -e^{iK} & -(z^*)^2 & z^*e^{iK}\\ e^{-iK} & -z & -z^*e^{-iK} & 1 \end{pmatrix}. \tag*{(41)}

Closed form, determinant, and W0

Collecting the pieces gives Eq. (14) of the main text, with S=diag⁡[Sk1,…,SkL]S=\operatorname{diag}[S_{k_1},\ldots,S_{k_L}]; the k=qk=q parts of the jump terms sit on the diagonal of NLN_L. To evaluate the determinant we write L(K)−λ=A+UVT\mathcal{L}(K)-\lambda=A+UV^{\mathrm T} with

A=⨁n[L(K,kn)−λ],U=[1L⊗I4, S(1L⊗I4)],VT=[1LT⊗M0intra(1LT⊗M~0)S−1],(42)A=\bigoplus_n\left[\mathcal{L}(K,k_n)-\lambda\right],\qquad U=\left[\mathbf{1}_L\otimes I_4,\ S(\mathbf{1}_L\otimes I_4)\right],\qquad V^{\mathrm T}=\left[ \begin{array}{c} \mathbf{1}_L^{\mathrm T}\otimes M_0^{\mathrm{intra}}\\ (\mathbf{1}_L^{\mathrm T}\otimes\widetilde{M}_0)S^{-1} \end{array} \right], \tag*{(42)}

where 1L\mathbf{1}_L is the all-ones vector. With Rn=[L(K,kn)−λ]−1R_n=[\mathcal{L}(K,k_n)-\lambda]^{-1} the matrix determinant lemma gives

D(K,λ)=∏ndet⁡[L(K,kn)−λ]det⁡[I8+Q(K,λ)],Q=∑n(M0intraRnM0intraRnSknM~0Skn−1RnM~0Skn−1RnSkn).(43)D(K,\lambda)=\prod_n\det\left[\mathcal{L}(K,k_n)-\lambda\right]\det\left[I_8+Q(K,\lambda)\right],\qquad Q=\sum_n \begin{pmatrix} M_0^{\mathrm{intra}}R_n & M_0^{\mathrm{intra}}R_nS_{k_n}\\ \widetilde{M}_0S_{k_n}^{-1}R_n & \widetilde{M}_0S_{k_n}^{-1}R_nS_{k_n} \end{pmatrix}. \tag*{(43)}

which reduces the determinant of the 4L×4L4L\times4L sector matrix to LL determinants of size 4 and a single one of size 8.

The winding. On a discrete grid, Eq. (15) reads

W0=12π∑mArg⁡D(Km+1,0−)D(Km,0−).(44)W_0=\frac{1}{2\pi}\sum_m\operatorname{Arg}\frac{D(K_{m+1},0^-)}{D(K_m,0^-)}. \tag*{(44)}

with 0−0^- a reference point just below zero. Close to the Hermitian points the steady-state branch passes this reference tangentially, so the grid must resolve the neighborhood of K=0K=0. The result is Eq. (17) of the main text, with W0>0W_0>0 corresponding to the right edge. At α=π/2+nπ\alpha=\pi/2+n\pi the evaluated value is reversed [Figure 3(b)]. Since the spectrum touches 0−0^- at these points and the value depends on the reference offset, it cannot be regarded as an invariant.

Symmetry constraints in KK space. The mirror reverses both momenta, so it maps Lα(K)\mathcal{L}_\alpha(K) unitarily onto L−α(−K)\mathcal{L}_{-\alpha}(-K). The symmetry A\mathcal{A} of Eq. (11) combines the mirror with complex conjugation, which also reverses both momenta, and with the sublattice sign, which leaves them unchanged; it maps Lα(K)\mathcal{L}_{\alpha}(K) antiunitarily onto Lα+π(K)\mathcal{L}_{\alpha+\pi}(K). Both maps preserve the kk grid. For real EE,

D−α(−K,E)=Dα(K,E),Dα+π(K,E)=Dα(K,E)∗.(45)D_{-\alpha}(-K,E)=D_{\alpha}(K,E),\qquad D_{\alpha+\pi}(K,E)=D_{\alpha}(K,E)^{*}. \tag*{(45)}

Running KK backward reverses the winding, and so does complex conjugation. Hence W0(−α)=−W0(α)W_0(-\alpha)=-W_0(\alpha) and W0(α+π)=−W0(α)W_0(\alpha+\pi)=-W_0(\alpha), Eq. (16); with 2π2\pi periodicity, W0(0)=W0(π)=0W_0(0)=W_0(\pi)=0.

Ket and bra twists

Following Ref. [15], we close the chain into a ring of NN sites and twist the boundary. Let p=N−1p=N-1 and q=0q=0 be the sites on either side of the boundary bond. In every operator X∈{H,Oj,Oj†Oj}X\in\{H,O_j,O_j^{\dagger}O_j\}, the terms that move a particle forward across the boundary, cq†cpc_q^{\dagger}c_p, are multiplied by eiθe^{i\theta}, and their partners cp†cqc_p^{\dagger}c_q by e−iθe^{-i\theta}; we denote the result by XθX_{\theta}. For l=2l=2 the bath bonds (N−2,0)(N-2,0) and (N−1,1)(N-1,1) also cross the boundary and are twisted in the same way. In the Lindblad equation (3) the operators to the left of ρ\rho act on the ket and those to the right on the bra (Table S1); twisting them separately gives

TermIndex that movesPhase
HρH_{\rho}, OρO†O_{\rho}O^{\dagger}, O†OρO^{\dagger}O_{\rho}ket, p→qp \to qcq†cpc_{q}^{\dagger}c_{p} in XX: ×eiθK\times e^{i\theta_K}
ρH\rho H, ρO†O\rho O^{\dagger}Obra, p→qp \to qcp†cqc_{p}^{\dagger}c_{q} in XX: ×eiθB\times e^{i\theta_B}
OρO†O\rho O^{\dagger}bra, p→qp \to qcp†cqc_{p}^{\dagger}c_{q} in O†O^{\dagger}: ×eiθB\times e^{i\theta_B}

Table S1. Terms that carry the twist phases (p=N−1p=N-1, q=0q=0; X=HX=H, OjO_j, or Oj†OjO_j^{\dagger}O_j). Hermitian partners carry the conjugate phase.

L(θK,θB)ρ=−i(HθKρ−ρH−θB)+γ∑jOj,θKρ(Oj,−θB)†−γ2∑j[(Oj†Oj)θKρ+ρ(Oj†Oj)−θB].(46)\mathcal{L}(\theta_K,\theta_B)\rho=-i\left(H_{\theta_K}\rho-\rho H_{-\theta_B}\right)+\gamma\sum_j O_{j,\theta_K}\rho\left(O_{j,-\theta_B}\right)^{\dagger}-\frac{\gamma}{2}\sum_j\left[(O_j^{\dagger}O_j)_{\theta_K}\rho+\rho(O_j^{\dagger}O_j)_{-\theta_B}\right]. \tag*{(46)}

The ket winding νK\nu_K uses the loop (θ,0)(\theta,0), the bra winding νB\nu_B the loop (0,θ)(0,\theta), and the density winding νC\nu_C the loop (θ,θ)(\theta,\theta), all for θ\theta from 00 to 2π2\pi, as in Eq. (18). On the bra side the twist carries −θB-\theta_B, because in ρX\rho X the bra index is the first index of XX, so that the term moving the bra index forward across the boundary is cp†cqc_p^{\dagger}c_q, the transpose of the ket term. With this convention νK\nu_K and νB\nu_B both count displacement to the right; placing +θB+\theta_B instead would reverse every bra winding and turn Eq. (21) into νB=+νK\nu_B=+\nu_K. At θK=θB=0\theta_K=\theta_B=0, Eq. (46) is the ordinary ring Liouvillian.

Counting. A winding counts eigenvalue paths encircling Λ\Lambda counterclockwise minus clockwise. With the phase on the forward hop, ν>0\nu>0 means accumulation at the right edge. Only signs are quoted, because magnitudes grow with the ring size. The ket-only twist reproduces the winding of Ref. [15].

Relation of the twists to the KK sector. The relation W0=νK(0−)=νB(0−)=νC(0−)/2W_0=\nu_K(0^-)=\nu_B(0^-)=\nu_C(0^-)/2, Eq. (20), follows in five steps. (1) A phase eiθe^{i\theta} on the boundary ket hops is equivalent to eiθ/Ne^{i\theta/N} on every ket hop, since the transformation cj→eiθj/Ncjc_j\to e^{i\theta j/N}c_j applied to the ket side alone is a similarity transformation of L\mathcal{L} and does not change det⁡[L(θ)−Λ]\det[\mathcal{L}(\theta)-\Lambda]. (2) A phase on every ket hop shifts every ket momentum by δ=θ/L\delta=\theta/L while the bra momenta stay put, so that each pair (k,k′)(k,k') moves to K+δK+\delta and the whole sector slides along the KK axis. (3) At θ=2π\theta=2\pi one has δ=2π/L\delta=2\pi/L, so that every ket momentum lands on the next grid point and every sector KmK_m has moved to its neighbor. The spectrum is thus mapped onto itself with the sectors relabeled. (4) det⁡[L(θ)−Λ]\det[\mathcal{L}(\theta)-\Lambda] is the product of the sector determinants, so its total phase change is the sum over mm of the change along Km→Km+1K_m\to K_{m+1}, which summed over all mm is one trip around the KK circle, i.e., the sum defining W0W_0. Hence νK(0−)=W0\nu_K(0^-)=W_0. The twist supplies the continuous path between grid points that the discrete sum samples, and a refined grid is therefore needed near K=0K=0. (5) On the real axis Hermiticity preservation, Eq. (19), gives νB(0−)=νK(0−)=W0\nu_B(0^-)=\nu_K(0^-)=W_0. Twisting both indices by θ\theta by two steps and gives νC(0−)=2W0\nu_C(0^-)=2W_0, while a twist of the ket forward and the bra backward is a gauge transformation that leaves KK fixed and gives zero.

The case l=2l=2. The ring is still invariant under translation by one unit cell, so KK remains conserved, but each jump couples the same sublattice in neighboring cells, and the closed form (14) does not apply as written. More importantly, W0W_0 counts the joint displacement of ket and bra, Eq. (20), and for l=2l=2 it vanishes by Eqs. (19) and (21). The bipolar behavior appears only at complex reference points and only when the two indices are twisted separately. For real Λ\Lambda, det⁡[LK(θ)−Λ]\det[\mathcal{L}_K(\theta)-\Lambda] is real at every θ\theta, and the windings at complex references come in conjugate pairs of opposite sign (Sec. S2).

Symmetry and balance

The symmetry statements of the main text rest on two ingredients. The first is the antiunitary map A\mathcal{A} of Eq. (11), assembled from the mirror PP (ci→cN−1−ic_i \to c_{N-1-i}), the sublattice sign Γ\Gamma (ci→(−1)icic_i \to(-1)^i c_i), and complex conjugation in the site basis. The second is Hermiticity preservation, which every Lindbladian has. We first derive the balance relations that follow from a single jump, and then show how these maps act on the windings and on the steady state.

Single-jump identities and balance relations

Write the jump on the bond (j,j+l)(j,j+l) as O=u†vO=u^\dagger v, with u†=cj†+zcj+l†u^\dagger=c_j^\dagger+z c_{j+l}^\dagger, v=cj−zcj+lv=c_j-z c_{j+l}, and {u,u†}={v,v†}=2\{u,u^\dagger\}=\{v,v^\dagger\}=2. In the single-particle sector O†O=2v†vO^\dagger O=2v^\dagger v and OO†=2u†uOO^\dagger=2u^\dagger u, so that

[O,O†]=2(u†u−v†v)=2(z+z∗)(cj†cj+l+cj+l†cj)(47)[O,O^\dagger]=2(u^\dagger u-v^\dagger v)=2(z+z^*)(c_j^\dagger c_{j+l}+c_{j+l}^\dagger c_j) \tag*{(47)}

Summed over bonds, Lα(I)=γ∑j[Oj,Oj†]\mathcal{L}_{\alpha}(\mathbb{I})=\gamma\sum_j[O_j,O_j^\dagger] is Eq. (8). It vanishes only for u∝vu\propto v, that is, for z∗=−zz^*=-z or α=π/2+nπ\alpha=\pi/2+n\pi, where the jump is Hermitian. Each part of Eq. (6) is normal on its own, since ODO^D is built from commuting number operators and OTOT†=OT†OT=nj+nj+lO^T O^{T\dagger}=O^{T\dagger}O^T=n_j+n_{j+l}, and either part alone therefore gives a unital bath. For the full jump [O,O†]=[OD,OT†]+[OT,OD†][O,O^\dagger]=[O^D,O^{T\dagger}]+[O^T,O^{D\dagger}], and only the cross terms of Eq. (9) remain.

Populations. The bond adds γ⟨O†njO−12{O†O,nj}⟩\gamma\langle O^\dagger n_jO-\frac{1}{2}\{O^\dagger O,n_j\}\rangle to n˙j\dot n_j. With O†njO=∣{cj,u†}∣2v†v=v†vO^\dagger n_jO=|\{c_j,u^\dagger\}|^2v^\dagger v=v^\dagger v and 12{O†O,nj}={v†v,nj}\frac{1}{2}\{O^\dagger O,n_j\}=\{v^\dagger v,n_j\},

⟨v†v⟩=nj+nj+l−2Re⁡(z⟨cj†cj+l⟩),⟨{v†v,nj}⟩=2nj−2Re⁡(z⟨cj†cj+l⟩).(48)\langle v^\dagger v\rangle=n_j+n_{j+l}-2\operatorname{Re}(z\langle c_j^\dagger c_{j+l}\rangle),\qquad\langle\{v^\dagger v,n_j\}\rangle=2n_j-2\operatorname{Re}(z\langle c_j^\dagger c_{j+l}\rangle). \tag*{(48)}

The coherence drops out of the difference, which leaves γ(nj+l−nj)\gamma(n_{j+l}-n_j), Eq. (7), and the same calculation applies to site j+lj+l.

Current. A bath bond (a,b)(a,b) carries the flow γ(na−nb)\gamma(n_a-n_b) from aa to bb, and the Hamiltonian changes the occupation of site jj at the rate Jj−1−JjJ_{j-1}-J_j, with JjJ_j from Eq. (24). Nothing crosses the cut to the left of site 0, so adding the stationary site equations from the left sets the net flow across every cut to zero. The cut between jj and j+1j+1 is crossed by one Hamiltonian bond together with one bath bond for l=1l=1, or with the two bath bonds (j−1,j+1)(j-1,j+1) and (j,j+2)(j,j+2) for l=2l=2, which gives the two lines of Eq. (25). On a ring the net flow is uniform instead. For l=1l=1 the occupations are also uniform on each sublattice, the bath flows on the intracell and intercell bonds cancel, and the average over one cell is Eq. (26).

Zero modes. Because [H,Π]=0[H,\Pi]=0, only the jumps change ⟨Π⟩\langle\Pi\rangle. In the single-particle sector Oj†ΠOj=wjvj†vjO_j^\dagger\Pi O_j=w_jv_j^\dagger v_j, with wj=⟨vac∣ujΠuj†∣vac⟩=∑m∣{dm,uj†}∣2w_j=\langle\mathrm{vac}|u_j\Pi u_j^\dagger|\mathrm{vac}\rangle=\sum_m|\{d_m,u_j^\dagger\}|^2, and 12{Oj†Oj,Π}={vj†vj,Π}\frac{1}{2}\{O_j^\dagger O_j,\Pi\}=\{v_j^\dagger v_j,\Pi\}. Together these give Eq. (22).

Windings under the mirror, Hermiticity, and A

Three rules are used below. (R1) A linear map U\mathcal{U} with UL(θK,θB)U−1=L(θK′,θB′)\mathcal{U}\mathcal{L}(\theta_K,\theta_B)\mathcal{U}^{-1}=\mathcal{L}(\theta'_K,\theta'_B) leaves the determinant unchanged, and with it the winding along corresponding loops. (R2) An antilinear map conjugates the determinant, sends Λ→Λ∗\Lambda\to\Lambda^*, and reverses the sense of rotation. (R3) Running a loop backward reverses its winding.

Hermiticity. Under J(ρ)=ρ†\mathcal{J}(\rho)=\rho^\dagger, the operators acting on the ket and on the bra trade places; for the Hamiltonian part, [−i(HθKρ†−ρ†H−θB)]†=−i(H−θBρ−ρHθK)[-i(H_{\theta_K}\rho^\dagger-\rho^\dagger H_{-\theta_B})]^\dagger=-i(H_{-\theta_B}\rho-\rho H_{\theta_K}), and the jump terms follow suit. Hence

JL(θK,θB)J−1=L(−θB,−θK).(49)\mathcal{J}\mathcal{L}(\theta_K,\theta_B)\mathcal{J}^{-1}=\mathcal{L}(-\theta_B,-\theta_K). \tag*{(49)}

The ket loop becomes the bra loop run backward, and (R2) with (R3) yields Eq. (19) for any number-conserving Lindbladian.

Mirror. PP maps the bonds that cross the boundary onto themselves with their direction reversed, so PHθP−1=H−θPH_{\theta}P^{-1}=H_{-\theta}, while POj(α)P−1=−z2ON−1−j(−α)PO_j(\alpha)P^{-1}=-z^2O_{N-1-j}(-\alpha) and the factor −z2-z^2 cancels between OO and O†O^\dagger. The mirror is thus a linear map from Lα(θK,θB)\mathcal{L}_{\alpha}(\theta_K,\theta_B) to L−α(−θK,−θB)\mathcal{L}_{-\alpha}(-\theta_K,-\theta_B), so that every winding is odd in α\alpha and vanishes at α=0,π\alpha=0,\pi for both ranges, which is the triviality criterion of Ref. [15].

Mirror-chiral symmetry. The sublattice sign by itself reverses HH, keeps ODO^D, and multiplies OTO^T by (−1)l(-1)^l. Its antiunitary completion is A(ρ)=PΓρ∗ΓP\mathcal{A}(\rho)=P\Gamma\rho^*\Gamma P. Conjugation acts as (α,H,θ)→(−α,−H,−θ)(\alpha,H,\theta)\to(-\alpha,-H,-\theta), because it flips the factor ii in −i[H,⋅]-i[H,\cdot]; Γ\Gamma then restores HH and, for l=1l=1 only, shifts α\alpha by π\pi; finally PP reverses α\alpha and θ\theta once more. The result is

ALα(θK,θB)A−1={Lα(θK,θB),l=2,Lα+π(θK,θB),l=1,(50)\mathcal{A}\mathcal{L}_{\alpha}(\theta_K,\theta_B)\mathcal{A}^{-1} = \begin{cases} \mathcal{L}_{\alpha}(\theta_K,\theta_B), & l=2,\\ \mathcal{L}_{\alpha+\pi}(\theta_K,\theta_B), & l=1, \end{cases} \tag*{(50)}

which reduces to Eq. (11) at zero twist.

For l=2l=2, (R2) gives ν(Λ∗)=−ν(Λ)\nu(\Lambda^*)=-\nu(\Lambda) for every winding, and det⁡[L(θ)−Λ]\det[\mathcal{L}(\theta)-\Lambda] is real for real Λ\Lambda. The product JA\mathcal{J}\mathcal{A}, ρ↦PΓρTΓP\rho\mapsto P\Gamma\rho^{T}\Gamma P, is linear and sends (θK,θB)→(−θB,−θK)(\theta_K,\theta_B)\to(-\theta_B,-\theta_K). The ket loop turns into the bra loop run backward and the density loop into itself run backward, so (R1) and (R3) give νK=−νB\nu_K=-\nu_B and νC=0\nu_C=0, Eq. (21). For l=1l=1, A\mathcal{A} links α\alpha to α+π\alpha+\pi instead; then να+π(Λ∗)=−να(Λ)\nu_{\alpha+\pi}(\Lambda^*)=-\nu_\alpha(\Lambda), and on the real axis every winding reverses under α→α+π\alpha\to\alpha+\pi, in agreement with Eq. (16).

Equation (21) follows from A\mathcal{A} alone, and it is therefore exact for a uniform chain and in the presence of a staggered mass ∑i(−1)ini\sum_i(-1)^i n_i, which the mirror maps onto its negative. Weak breaking of A\mathcal{A}, by on-site disorder or by hoppings without mirror symmetry, leaves the integer windings practically unchanged, as expected for a symmetry-protected relation [5, 6]. The orientation of the bonds is also relevant. When every bond is ordered with its A site first, α\alpha is reversed on the intercell bonds and the l=1l=1 pile-up is nearly removed (the steady-state center of mass at α=π/4\alpha=\pi/4, N=50N=50, falls from +0.85+0.85 to +0.09+0.09 on the scale [−1,1][-1,1]).

Steady state, Hermitian points, and zero modes

For a unique steady state, the mirror and Eq. (11) imply

ρss(−α)=Pρss(α)P,ρss(α+π)=A[ρss(α)](l=1),ρss(α)=A[ρss(α)](l=2).(51)\rho_{\mathrm{ss}}(-\alpha)=P\rho_{\mathrm{ss}}(\alpha)P,\qquad \rho_{\mathrm{ss}}(\alpha+\pi)=\mathcal{A}[\rho_{\mathrm{ss}}(\alpha)]\quad(l=1),\qquad \rho_{\mathrm{ss}}(\alpha)=\mathcal{A}[\rho_{\mathrm{ss}}(\alpha)]\quad(l=2). \tag*{(51)}

Since A[ρ]ii=ρN−1−i,N−1−i\mathcal{A}[\rho]_{ii}=\rho_{N-1-i,N-1-i}, the occupations obey ρii(−α)=ρii(α+π)=ρN−1−i,N−1−i(α)\rho_{ii}(-\alpha)=\rho_{ii}(\alpha+\pi)=\rho_{N-1-i,N-1-i}(\alpha) for l=1l=1, whereas for l=2l=2 they are mirror symmetric at every α\alpha and admit no directional pile-up. For H=0H=0, conjugation alone maps Lα\mathcal{L}_{\alpha} onto L−α\mathcal{L}_{-\alpha}; combined with the mirror this gives ρss=PρssP\rho_{\mathrm{ss}}=P\rho_{\mathrm{ss}}P, a mirror-symmetric density at every α\alpha. In the energy basis, the eigenmodes dE=∑iϕE(i)cid_E=\sum_i\phi_E(i)c_i of the real HH have real amplitudes, and for a nondegenerate spectrum PΓP\Gamma maps dEd_E onto ±d−E\pm d_{-E}. The population of dEd_E in A[ρ]\mathcal{A}[\rho] is therefore that of d−Ed_{-E} in ρ\rho, which together with Eq. (51) is Eq. (13).

Hermitian points. At αH=π/2+nπ\alpha_H=\pi/2+n\pi the dissipator is self-adjoint and L†=i[H,⋅]+D\mathcal{L}^{\dagger}=i[H,\cdot]+\mathcal{D}. Conjugation takes LαH†\mathcal{L}^{\dagger}_{\alpha_H} to L−αH\mathcal{L}_{-\alpha_H}, and the mirror takes this back to LαH\mathcal{L}_{\alpha_H}, so that L†=S−1LS\mathcal{L}^{\dagger}=S^{-1}\mathcal{L}S with the antiunitary S(X)=PX∗PS(X)=PX^*P. This is Eq. (29). A right eigenmode ρR\rho_R with eigenvalue λ\lambda thus has the left partner ρL∝PρR∗P\rho_L\propto P\rho_R^*P, for both ranges. Detailed balance at infinite temperature, L†(X)=[L(X∗)]∗\mathcal{L}^{\dagger}(X)=[\mathcal{L}(X^*)]^*, would instead give ρL∝ρR∗\rho_L\propto\rho_R^*, with both at the same boundary [12], whereas the mirror in SS places them at opposite boundaries.

Zero modes. H\mathcal{H} is invariant under PP, Γ\Gamma, and conjugation. The pair consists of dLd_L, on sublattice A at the left edge, and dRd_R, on sublattice B at the right edge, which PΓP\Gamma exchanges up to sign. With pL,R=⟨dL,R†dL,R⟩p_{L,R}=\langle d_{L,R}^{\dagger}d_{L,R}\rangle,

pL(−α)=pR(α),pL(α+π)=pR(α)(l=1),pL(α)=pR(α)(l=2),(52)p_L(-\alpha)=p_R(\alpha),\qquad p_L(\alpha+\pi)=p_R(\alpha)\quad(l=1),\qquad p_L(\alpha)=p_R(\alpha)\quad(l=2), \tag*{(52)}

and PZM=2/NP_{\mathrm{ZM}}=2/N at α=π/2+nπ\alpha=\pi/2+n\pi. The splitting within the pair is exponentially small in NN, so the occupation of an individual member depends on the basis chosen inside the pair; we therefore use the pair weight PZM=⟨Π⟩P_{\mathrm{ZM}}=\langle\Pi\rangle and its edge-resolved parts pL,Rp_{L,R}.

For next-nearest-neighbor bonds the two ends are equivalent at every α\alpha and NN. The pair is favored over equipartition for π/2<α<3π/2\pi/2<\alpha<3\pi/2, by up to an order of magnitude near α≈0.9π\alpha\approx0.9\pi, yet its share of the particle shrinks as the chain grows. At α=π\alpha=\pi the jump on (j,j+2)(j,j+2) empties ψj+ψj+2\psi_j+\psi_{j+2} and so drains the sign-alternating zero mode at the size-independent rate 2γ(1−t12/t22)2\gamma(1-t_1^2/t_2^2), while bulk states with the same pattern become invisible to the bath at large NN. No fixed pattern therefore survives, and the kink is absent. This favoring is unrelated to the bipolar skin effect of coherences, since the zero-mode weight is mirror symmetric about α=π\alpha=\pi, where the bipolar imprint changes sign.

Vacuum sector, corner rule, and skin modes

We explain the vacuum-sector loops of Figure 4(c), Figure 4(d), Figure 5(c), and Figure 5(d), why they are weaker than Eq. (21), how the shading and the corner rule were obtained, and why the edge-localized eigenmodes behave as skin modes. Figure S1 and Figure S2 extend the main-text panels to five phases, for both ranges.

Vacuum block of the full Liouvillian on a 24-site twisted ring

Figure S1. Vacuum block of the full Liouvillian on a 24-site twisted ring, top row l=2l=2, bottom row l=1l=1. Red: counterclockwise loops (νK=+1\nu_K=+1, right edge); blue: clockwise (νK=−1\nu_K=-1, left edge); gray: arcs of zero area (νK=0\nu_K=0). Arrows give the sense of rotation as θ\theta increases, and xˉ∈[−1,1]\bar{x}\in[-1,1] is the mean position of the open-chain modes (N=60N=60) belonging to each loop. t1=1.5t_1=1.5, t2=2.5t_2=2.5, γ=1\gamma=1.

Density block on the open chain

Figure S2. Density block on the open chain (N=30N = 30), top row l=2l = 2, bottom row l=1l = 1. Points are the eigenvalues, colored by the corner in which the eigenmode localizes; shading is the sign of the ket winding νK(Λ)\nu_K(\Lambda) on a 12-site twisted ring. For l=2l = 2 every edge-localized mode sits in a crossed corner, in equal numbers; for l=1l = 1 every one sits in a same-edge corner. The few teal points near λ=0\lambda= 0 at l=1l = 1, α=π/2\alpha= \pi/2 lie in a small pocket with νK<0\nu_K < 0 and sit at the left edge. t1=1.5t_1 = 1.5, t2=2.5t_2 = 2.5, γ=1\gamma= 1.

Blocks and vacuum loops

On the space spanned by ∣vac⟩|\mathrm{vac}\rangle and ci†∣vac⟩c_i^\dagger|\mathrm{vac}\rangle, group the rows and columns of ρ\rho by particle number. Because HH and every OjO_j conserve particle number, each term of Eq. (3) leaves the particle numbers of the row and of the column unchanged, so the blocks (ket number | bra number) = (0∣0)(0|0), (1∣0)(1|0), (0∣1)(0|1), (1∣1)(1|1) evolve independently.

The (1|0) block. Its elements are ρi,vac=⟨vac∣ciρ∣vac⟩\rho_{i,\mathrm{vac}}=\langle\mathrm{vac}|c_i\rho|\mathrm{vac}\rangle. Going through Eq. (3), we find that HρH\rho acts on the ket and is kept, ρH\rho H vanishes because H∣vac⟩=0H|\mathrm{vac}\rangle=0, OjρOj†O_j\rho O_j^\dagger vanishes because Oj†∣vac⟩=0O_j^\dagger|\mathrm{vac}\rangle=0, and Oj†OjρO_j^\dagger O_j\rho is kept while ρOj†Oj\rho O_j^\dagger O_j vanishes. What remains is

ddtρ⋅,vac=Gρ⋅,vac,G=−iH−γ2∑jOj†Oj(53)\frac{d}{dt}\rho_{\cdot,\mathrm{vac}}=G\rho_{\cdot,\mathrm{vac}},\qquad G=-iH-\frac{\gamma}{2}\sum_j O_j^\dagger O_j \tag*{(53)}

a matrix equation for NN numbers. The jump term therefore does not contribute in this block.

The (1|1) block. Writing Oj=∑abca†(oj)abcbO_j=\sum_{ab}c_a^\dagger(o_j)_{ab}c_b and sorting terms by the index they act on,

L11=G⊗I+I⊗Gˉ+γ∑joj⊗oˉj.(54)\mathcal{L}_{11}=G\otimes I+I\otimes\bar{G}+\gamma\sum_j o_j\otimes\bar{o}_j. \tag*{(54)}

Without the last term every eigenvalue would be λrow+λcolumn∗\lambda_{\mathrm{row}}+\lambda_{\mathrm{column}}^*, i.e., NN shifted copies of each vacuum loop, so the density-block spectra are dense. The jump term modifies them only through entries with ∣i−j∣≤l|i-j|\le l.

Meaning. The (1∣1)(1|1) block describes the particle itself, and the (1∣0)(1|0) block describes its amplitude. For qubits, ρi,vac=⟨σi−⟩\rho_{i,\mathrm{vac}}=\langle\sigma_i^-\rangle, while for fermions, where parity forbids it as a coherence of the state, the same block governs the propagator “inject at jj, find at ii” through the quantum regression theorem [3].

Vacuum loops. Twisting the single index of the (1∣0)(1|0) block on a 24-site ring moves its NN eigenvalues along one closed loop per band (Figure S1). For l=2l=2 the two loops run in opposite senses for every 0<α<π0<\alpha<\pi. The loop above the real axis has νK=−1\nu_K=-1 and its open-chain modes sit at the left edge, whereas the loop below has νK=+1\nu_K=+1 and its modes sit at the right edge. The two loops are exact mirror images of each other, and ∣xˉ∣|\bar{x}| is largest at the Hermitian point. The mean position of the open-chain modes moves closer to the edges as NN grows. For l=1l=1 there is one orientation, νK=+1\nu_K=+1, with all modes at the right edge; the two parts of the loop merge into a single loop crossing the real axis for α\alpha between about π/4\pi/4 and 3π/43\pi/4. Both ranges collapse to arcs at α=0,π\alpha=0,\pi.

Why the vacuum loops are weaker than Eq. (21). A real bipartite HH with jumps that preserve chirality already gives Gˉ(θ)=ΓG(θ)TΓ\bar{G}(\theta)=\Gamma G(\theta)^T\Gamma. The spectrum of G(θ)G(\theta) then equals its complex conjugate, and νK(λ∗)=−νK(λ)\nu_K(\lambda^*)=-\nu_K(\lambda), i.e., the two loops have opposite windings. This requires neither the mirror part of A\mathcal{A} nor the jump term. For hoppings without mirror symmetry the vacuum loops stay exactly paired, while the full-Liouvillian relation (21) no longer holds exactly.

Ket-winding shading and the transient coherence

The shading in Figs. 4(b), 5(b), and S2 is sgn⁡νK(Λ)\operatorname{sgn}\nu_K(\Lambda) evaluated on a grid of reference points Λ\Lambda covering the density-block spectrum of a 12-site twisted ring. For an open-chain eigenmode ρij\rho_{ij} we define the ket and bra centers

xˉK=∑ijxi∣ρij∣2∑ij∣ρij∣2,xˉB=∑ijxj∣ρij∣2∑ij∣ρij∣2,(55)\bar{x}_K = \frac{\sum_{ij} x_i |\rho_{ij}|^2}{\sum_{ij} |\rho_{ij}|^2}, \qquad\bar{x}_B = \frac{\sum_{ij} x_j |\rho_{ij}|^2}{\sum_{ij} |\rho_{ij}|^2}, \tag*{(55)}

with xi=2i/(N−1)−1∈[−1,1]x_i = 2i/(N - 1) - 1 \in[-1, 1], and call the mode edge localized if ∣xˉK∣|\bar{x}_K| and ∣xˉB∣|\bar{x}_B| both exceed 0.2; its corner is [sgn⁡xˉK,sgn⁡xˉB][\operatorname{sgn}\bar{x}_K,\operatorname{sgn}\bar{x}_B]. The corner rule predicts the ket side sgn⁡νK(λ)\operatorname{sgn}\nu_K(\lambda) and, through Eqs. (19) and (21), the bra side.

For l=2l = 2 only crossed corners occur, in equal numbers (Figure S2, top row), while for l=1l = 1 every localized mode sits in a same-edge corner (bottom row). The rule holds equally in the trivial phase, and its rare failures lie on the boundaries between regions of opposite νK\nu_K. For l=1l = 1 the sign of νK\nu_K is uniform except at the Hermitian point, where it turns negative in a small pocket close to λ=0\lambda= 0; the few modes inside it sit at the left edge, as the rule requires, while the slowest decaying mode stays extended.

To check that the localized modes are skin modes, we add a weak hop between sites 0 and N−1N - 1. It shifts their eigenvalues by an amount that grows exponentially with NN, and faster still for l=1l = 1, whereas at α=0\alpha= 0, where no winding is present, the shift decreases with NN. The left eigenvector of each localized l=2l = 2 mode occupies the mirror corner of its right partner, as required for skin modes [12], and the anticorrelation between ket and bra positions strengthens with size; for l=1l = 1 the two positions are instead positively correlated.

The bipolar structure is nevertheless transient. A bulk coherence, isolated by subtracting the full evolutions of (cj0†±cj0+1†)∣vac⟩/2(c^\dagger_{j_0} \pm c^\dagger_{j_0+1})|\mathrm{vac}\rangle/\sqrt{2}, spreads to both ends equally for l=2l = 2, to the right end only for l=1l = 1, and to neither at α=0\alpha= 0, and in every case it eventually decays. The l=2l = 2 steady state retains only a negligible end-to-end coherence, and what survives is a mirror-symmetric enhancement of short-range coherence at both edges, without a directional skin effect of the density.

References

  1. [1]G. Lindblad, On the generators of quantum dynamical semigroups, Commun. Math. Phys. 48, 119 (1976).DOI
  2. [2]V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, Completely positive dynamical semigroups of N-level systems, J. Math. Phys. 17, 821 (1976).DOI
  3. [3]H.-P. Breuer and F. Petruccione, The Theory of Open Quantum Systems (Oxford University Press, Oxford, 2002).
  4. [4]Z. Gong, Y. Ashida, K. Kawabata, K. Takasan, S. Higashikawa, and M. Ueda, Topological phases of non-Hermitian systems, Phys. Rev. X 8, 031079 (2018).DOI
  5. [5]K. Kawabata, K. Shiozaki, M. Ueda, and M. Sato, Symmetry and topology in non-Hermitian physics, Phys. Rev. X 9, 041015 (2019).DOI
  6. [6]N. Okuma, K. Kawabata, K. Shiozaki, and M. Sato, Topological origin of non-Hermitian skin effects, Phys. Rev. Lett. 124, 086801 (2020).DOI
  7. [7]K. Zhang, Z. Yang, and C. Fang, Correspondence between winding numbers and skin modes in non-Hermitian systems, Phys. Rev. Lett. 125, 126402 (2020).DOI
  8. [8]D. S. Borgnia, A. J. Kruchkov, and R.-J. Slager, Non-Hermitian boundary modes and topology, Phys. Rev. Lett. 124, 056802 (2020).
  9. [9]A. Chaduteau, D. Lee, F. Schindler, and A. Prem, Topology from decoherence (2026), arXiv:2607.07801 [quant-ph].arxiv.org/abs/2607.07801
  10. [10]R. Jha, S. Haas, and A. Prem, Geometric decoherence time in lindbladian dynamics, Phys. Rev. B 114, 105132 (2026).DOI
  11. [11]F. Song, S. Yao, and Z. Wang, Non-Hermitian skin effect and chiral damping in open quantum systems, Phys. Rev. Lett. 123, 170401 (2019).DOI
  12. [12]T. Gaga, M. Nakagawa, R. Hamazaki, and M. Ueda, Liouvillian skin effect: Slowing down of relaxation processes without gap closing, Phys. Rev. Lett. 127, 070402 (2021).
  13. [13]F. Yang, Q.-D. Jiang, and E. J. Bergholtz, Liouvillian skin effect in an exactly solvable model, Phys. Rev. Res. 4, 023160 (2022).DOI
  14. [14]Z. Wang, L. Lu, Y. Peng, R. Qi, Y. Wang, and J. Jie, Accelerating relaxation dynamics in open quantum systems with Liouvillian skin effect, Phys. Rev. B 108, 054313 (2023).DOI
  15. [15]S. Hamanaka, K. Yamamoto, and T. Yoshida, Interaction-induced Liouvillian skin effect in a fermionic chain with a two-body loss, Phys. Rev. B 108, 155114 (2023).
  16. [16]L. Mao, X. Yang, M.-J. Tao, H. Hu, and L. Pan, Liouvillian skin effect in a one-dimensional open many-body quantum system with generalized boundary conditions, Phys. Rev. B 110, 045440 (2024).arxiv.org/abs/2401.15614
  17. [17]X. Feng and S. Chen, Boundary-sensitive Lindbladians and relaxation dynamics, Phys. Rev. B 109, 014313 (2024).DOI
  18. [18]C. Ekman and E. J. Bergholtz, Liouvillian skin effects and fragmented condensates in an integrable dissipative Bose-Hubbard model, Phys. Rev. Res. 6, L032067 (2024).DOI
  19. [19]S. Long, H.-S. Yin, C. Yang, S. Mu, J.-W. Zhang, and L. Li, Controlling Liouvillian topological phases via Hamiltonian band topology under chiral symmetry (2026), arXiv:2602.22323 [quant-ph].arxiv.org/abs/2602.22323
  20. [20]K. Roy, S. Shahab, and S. Basu, Controlling dissipative topology through Floquet driving: From transient diagnostics to boundary states isolation (2025), arXiv:2511.23229 [cond-mat.mes-hall].arxiv.org/abs/2511.23229
  21. [21]N. Okuma and M. Sato, Non-Hermitian topological phenomena: A review, Annu. Rev. Condens. Matter Phys. 14, 83 (2023).DOI
  22. [22]S. Diehl, A. Micheli, A. Kantian, B. Kraus, H. P. Büchler, and P. Zoller, Quantum states and phases in driven open quantum systems with cold atoms, Nat. Phys. 4, 878 (2008).DOI
  23. [23]B. Kraus, H. P. Büchler, S. Diehl, A. Kantian, A. Micheli, and P. Zoller, Preparation of entangled states by quantum Markov processes, Phys. Rev. A 78, 042307 (2008).DOI
  24. [24]S. Diehl, A. Tomadin, A. Micheli, R. Fazio, and P. Zoller, Dynamical phase transitions and instabilities in open atomic many-body systems, Phys. Rev. Lett. 105, 015702 (2010).DOI
  25. [25]S. Diehl, E. Rico, M. A. Baranov, and P. Zoller, Topology by dissipation in atomic quantum wires, Nat. Phys. 7, 971 (2011).DOI
  26. [26]C.-E. Bardyn, M. A. Baranov, C. V. Kraus, E. Rico, A. İmamoğlu, P. Zoller, and S. Diehl, Topology by dissipation, New J. Phys. 15, 085001 (2013).
  27. [27]L. Bonnes, D. Charrrier, and A. M. Läuchli, Dynamical and steady-state properties of a Bose-Hubbard chain with bond dissipation: A study based on matrix product operators, Phys. Rev. A 90, 033612 (2014).DOI
  28. [28]Y. Liu, Z. Wang, C. Yang, J. Jie, and Y. Wang, Dissipation-induced extended-localized transition, Phys. Rev. Lett. 132, 216301 (2024).DOI
  29. [29]Y. Peng, C. Yang, H. Hu, and Y. Wang, Dissipation-assisted preparation of topological boundary states (2024), arXiv:2412.04152 [cond-mat.mes-hall].arxiv.org/abs/2412.04152
  30. [30]Y. Peng, C. Yang, and Y. Wang, Manipulating the relaxation time of boundary-dissipative systems through bond dissipation, Phys. Rev. B 110, 104305 (2024).DOI
  31. [31]Y. Liu and Y. Wang, General strategy for realizing Mpemba effects in open quantum systems, Phys. Rev. A 114, L010802 (2026).DOI
  32. [32]D. J. Strachan, A. Purkayastha, and S. B. Clark, Non-markovian quantum Mpemba effect, Phys. Rev. Lett. 134, 220403 (2025).DOI
  33. [33]A. Chaduteau, D. K. K. Lee, and F. Schindler, Lindbladian versus postselected non-Hermitian topology, Phys. Rev. Lett. 136, 016603 (2026).DOI
  34. [34]W. P. Su, J. R. Schrieffer, and A. J. Heeger, Solitons in polyacetylene, Phys. Rev. Lett. 42, 1698 (1979).DOI
  35. [35]J. K. Asbóth, L. Oroszlány, and A. Pályi, A Short Course on Topological Insulators, Lecture Notes in Physics, Vol. 919 (Springer, Cham, 2016) arXiv:1509.02925.
  36. [36]K. Roy, K. Gogoi, and S. Basu, Topological characterization of a non-Hermitian ladder via Floquet non-Bloch theory, Phys. Rev. B 111, 115424 (2025).DOI
  37. [37]K. Roy, L. Kalita, B. Tanatar, and S. Basu, Floquet generation of hybrid-order topology and Z₂-like bipolar localization, Phys. Rev. Res. 8, 043094 (2026).DOI
  38. [38]K. Roy and S. Basu, Topological properties of a periodically driven Creutz ladder, Phys. Rev. B 108, 045415 (2023).DOI
  39. [39]A. J. Daley, Quantum trajectories and open many-body quantum systems, Adv. Phys. 63, 77 (2014).DOI
  40. [40]F. Reiter and A. S. Sørensen, Effective operator formalism for open quantum systems, Phys. Rev. A 85, 032111 (2012).DOI
  41. [41]A. Griessner, A. J. Daley, S. R. Clark, D. Jaksch, and P. Zoller, Dark-state cooling of atoms by superfluid immersion, Phys. Rev. Lett. 97, 220403 (2006).DOI
  42. [42]D. Marcos, A. Tomadin, S. Diehl, and P. Rabl, Photon condensation in circuit quantum electrodynamics by engineered dissipation, New J. Phys. 14, 055005 (2012).
  43. [43]M. Aidelsburger, M. Atala, M. Lohse, J. T. Barreiro, B. Paredes, and I. Bloch, Realization of the Hofstadter Hamiltonian with ultracold atoms in optical lattices, Phys. Rev. Lett. 111, 185301 (2013).DOI
  44. [44]H. Miyake, G. A. Silviaoglou, C. J. Kennedy, W. C. Burton, and W. Ketterle, Realizing the Harper Hamiltonian with laser-assisted tunneling in optical lattices, Phys. Rev. Lett. 111, 185302 (2013).DOI
  45. [45]M. Atala, M. Aidelsburger, J. T. Barreiro, D. Abanin, T. Kitagawa, E. Demler, and I. Bloch, Direct measurement of the Zak phase in topological Bloch bands, Nat. Phys. 9, 795 (2013).
  46. [46]W. S. Bakr, J. I. Gillen, A. Peng, S. Fölling, and M. Greiner, A quantum gas microscope for detecting single atoms in a Hubbard-regime optical lattice, Nature 462, 74 (2009).DOI
  47. [47]J. F. Sherson, C. Weitenberg, M. Endres, M. Cheneau, I. Bloch, and S. Kuhr, Single-atom-resolved fluorescence imaging of an atomic Mott insulator, Nature 467, 68 (2010).DOI
  48. [48]M. Atala, M. Aidelsburger, M. Lohse, J. T. Barreiro, B. Paredes, and I. Bloch, Observation of chiral currents with ultracold atoms in bosonic ladders, Nat. Phys. 10, 588 (2014).DOI
  49. [49]J. T. Barreiro, M. Müller, P. Schindler, D. Nigg, T. Monz, M. Chwalla, M. Hennrich, C. F. Roos, P. Zoller, and R. Blatt, An open-system quantum simulator with trapped ions, Nature 470, 486 (2011).DOI
  50. [50]P. M. Harrington, E. J. Mueller, and K. W. Murch, Engineered dissipation for quantum information science, Nat. Rev. Phys. 4, 660 (2022).DOI
  51. [51]X. Mi et al., Stable quantum-correlated many-body states through engineered dissipation, Science 383, 1332 (2024).DOI
  52. [52]K. Shimomura and M. Sato, General criterion for non-Hermitian skin effects and application: Fock space skin effects in many-body systems, Phys. Rev. Lett. 133, 136502 (2024).DOI
  53. [53]T. Prosen, Third quantization: a general method to solve master equations for quadratic open Fermi systems, New J. Phys. 10, 043026 (2008).DOI
  54. [54]S. Lieu, M. McGinley, and N. R. Cooper, Tenfold way for quadratic Lindbladians, Phys. Rev. Lett. 124, 040401 (2020).DOI

Paper details

vlnc:W_9rzruhx7
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 1

downloads · 0

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents