2026 Volume 35 Issue 4
Article Contents

Cui-Xian Guo, Xiao-Ming Zhao, Su-Peng Kou. PT phase transition and non-Hermitian skin effect in vacancy-induced subbands of the Haldane model with gain and loss[J]. Chinese Physics B, 2026, 35(4): 047103. doi: 10.1088/1674-1056/ae44f4
Citation: Cui-Xian Guo, Xiao-Ming Zhao, Su-Peng Kou. PT phase transition and non-Hermitian skin effect in vacancy-induced subbands of the Haldane model with gain and loss[J]. Chinese Physics B, 2026, 35(4): 047103. doi: 10.1088/1674-1056/ae44f4

PT phase transition and non-Hermitian skin effect in vacancy-induced subbands of the Haldane model with gain and loss

  • Received Date: 13/11/2025
    Accepted Date: 11/02/2026
    Available Online: 01/04/2026
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(8)

Article Metrics

Article views(289) PDF downloads(1) Cited by(0)

Access History

PT phase transition and non-Hermitian skin effect in vacancy-induced subbands of the Haldane model with gain and loss

Abstract: Defects can profoundly alter the properties of quantum materials, inducing emergent phenomena beyond the scope of their parent counterparts. Here, we construct effective models to investigate the properties of vacancy-induced subbands in the Haldane model with gain and loss, revealing that their behaviors are critically dependent on the spatial configuration of the vacancy array. Specifically, a sequence of parity–time (PT) phase transitions occurs for the one-dimensional vacancy array across different sublattices, whereas the non-Hermitian skin effect (NHSE) — with its nontrivial spectral winding — manifests in a chain confined to a single sublattice. Notably, the NHSE arises in this subband even though the parent system lacks it. Our results demonstrate that vacancy defect engineering serves as a powerful approach to generate subsystems whose properties are decoupled from the parent Hamiltonian.

1.   Introduction
  • The past decade has witnessed a surge of interest in non-Hermitian physics,[1–6] extending its reach across optics,[7–13] acoustics,[14–17] photonic quantum walks[18,19] and single-spin systems.[20] This expansion has been driven by the discovery of novel phenomena without Hermitian counterparts. Two of the most remarkable are the parity–time (PT) phase transition[14–22] and the non-Hermitian skin effect (NHSE).[28–40] The former describes a symmetry-breaking transition, marked by the spectrum becoming complex and the appearance of exceptional points (EPs) signaling the transition and the simultaneous coalescence of both eigenvalues and eigenstates. The NHSE manifests as the exponential boundary localization of an extensive number of eigenstates and is rooted in the nontrivial spectral winding of complex energy bands, which fundamentally reshapes the bulk–boundary correspondence in topological systems. These novel effects have subsequently spurred the development of innovative applications, such as unidirectional transport,[23,24] single-mode lasers[25,26] and optical funnels.[27]

    Concurrently, defect engineering has long been established as a powerful tool for manipulating the properties of quantum materials.[41,42] In Hermitian settings, from dopants in semiconductors[43] to vacancies in graphene[44] and magnetic impurities in topological insulators,[45] defects can induce bound states, alter electronic structure and even generate entirely new phases of matter. The role of defects in non-Hermitian systems, however, presents a frontier rich with potential. A fundamental question persists: can a designed vacancy defect lattice in a non-Hermitian system yield a subsystem with emergent properties decoupled from, or absent in, the parent system?

    In this work, we address this question by theoretically investigating vacancy-induced subbands in a non-Hermitian Haldane model under the periodic boundary condition (PBC). We construct effective models to elucidate the properties of these defect-engineered one-dimensional chains. Our key finding is that their behavior is critically governed by the spatial configuration of the vacancy array. We demonstrate that a chain of vacancies spanning different sublattices undergoes a sequence of PT phase transitions, whereas a chain confined to a single sublattice exhibits the NHSE, characterized by its nontrivial spectral winding. Strikingly, this NHSE emerges in the vacancy subband even when it is entirely absent in the parent system.

2.   Model and single vacancy state
  • We investigate the non-Hermitian Haldane model[46] in the presence of vacancy defects. The total Hamiltonian is given by

    where H^0 represents the pristine non-Hermitian Haldane model on a honeycomb lattice with staggered imaginary on-site potentials, defined as

    and H^v is introduced to cancel the couplings and potentials at the vacancy sites

    Here t and t′ denote the nearest-neighbor and next-nearest-neighbor hopping strengths, respectively. The phase factor is defined as ϕij = ±π/2 for clockwise (counterclockwise) next-nearest-neighbor hopping. The gain and loss are implemented via on-site potentials iγ on sublattice A and −iγ on sublattice B, with γ quantifying the non-Hermitian strength. The set V denotes the indices of the vacancy defects.

    Considering H^0 under PBC as the parent system, the corresponding Bloch Hamiltonian is given by H^(k)=∑k(a^k†,b^k†)ℋ(k)(a^k,b^k)T, where

    with

    The lattice vectors are defined as a1 = a0(0,1), a2=a0(32,−12), a3=a0(−32,−12), and b1 = a2 – a3, b2 = a3 – a1, b3 = a1 – a2 with a0 = 1. The energy spectrum of the parent Hamiltonian reads

    The eigenstates of this parent model are extended Bloch states, and no skin effect is present.

    By adding H^v, we effectively create the vacancy defects. Therefore, our primary focus is to investigate the properties of the defect states induced by these vacancies in H^.

    A single vacancy is created by removing a site i0, as shown in Fig. 1(a), which eliminates both its on-site potential and all associated hopping terms. This defect gives rise to a bound state localized near the vacancy site (Fig. 1(b)), with the corresponding energy lying within the bulk band gap (Fig. 1(c)). Specifically, the defect state exhibits a purely imaginary energy: positive for a vacancy on sublattice B and negative for one on sublattice A (Fig. 1(d)).

3.   PT-phase transition in subbands from vacancies on different sublattices
  • We now examine the system described by Hamiltonian H^(j1,j2) containing two vacancies located at sites j1 and j2 on different sublattices (A and B), separated by a distance L along the zigzag direction (Fig. 2(a)). The coupling between the corresponding defect states leads to inter-state tunneling. As shown in Figs. 2(b) and 2(c), our numerical results indicate that the eigenvalues of the two defect states undergo a PT-phase transition as γ increases, accompanied by distinct changes in the eigenstates.

    To model this tunneling, we construct an effective Hamiltonian H^eff within the biorthogonal subspace spanned by the single-vacancy eigenstates. Let |ViR〉 and 〈ViL| denote the right and left eigenstates of the Hamiltonian H^(ji) containing a single vacancy at site ji, satisfying H^(ji)|ViR〉=E|ViR〉 and 〈ViL|H^(ji)=〈ViL|E. Assuming these states form a biorthogonal basis such that 〈ViL|VjR〉=δij, the effective Hamiltonian can then be obtained by projecting the full Hamiltonian H^(j1,j2) onto this basis as H^eff=(|V1R〉,|V2R〉)ℋeff(〈V1L|,〈V2L|)T, where

    with the matrix elements hij=〈ViL|H^(j1,j2)|VjR〉.

    Using the explicit forms of |ViR〉 and 〈ViL|, we derive the effective Hamiltonian

    Here σx,y,z are Pauli matrices and hx,hy,hz∈ℝ. Figure 3(a) illustrates the evolution of hx, hy and hz as a function of γ for L/|b1| = 3.5. Since hy ≈ 0, we neglect this term, allowing the effective Hamiltonian to be approximated as

    yielding the eigenvalues

    This effective Hamiltonian possesses PT symmetry: [H^eff,PT]=0, where P=σx and T denotes complex conjugation. The system resides in the PT-unbroken phase when |hx| > |hz|, characterized by two real eigenvalues E1 and E2. Conversely, when |hz| > |hx|, the system enters the PT-broken phase, and the eigenvalues become purely imaginary. The EP emerges at |hz| = |hx| (i.e., at γ = γ0), marked by the simultaneous coalescence of eigenvalues and eigenstates.

    To quantify the spatial distribution of the defect states over the two vacancy sites, we calculate the projection of the defect states |Ψi〉 onto the normalized right basis vectors |V∼jR〉, defined as |V∼jR〉=|VjR〉/〈VjR|VjR〉. Note that the defect states |Ψi〉 are self-normalized (i.e., 〈Ψi|Ψi〉 = 1) by default. This normalization ensures that the projection reflects the physical weight of the defect states on each vacancy site. Figure 3(b) illustrates the squared overlap |〈V∼jR|Ψi〉|2 as a function of γ. In the PT-unbroken phase, the defect states are distributed nearly uniformly across the two vacancy sites (with minor deviations from perfect uniformity attributed to the hy term). In contrast, in the PT-broken regime, these states rapidly localize onto separate vacancies as γ increases. Figure 3(c) plots the similarity 〈Ψ1|Ψ2〉 between the two defect states. At the EP, the similarity reaches unity, indicating state coalescence, as predicted by the effective model. The analytical eigenenergies from the effective model agree closely with numerical results (Figs. 2(b) and 2(c)), validating our approach. While the distance between the two vacancies is fixed L/|b1| = 3.5 in Figs. 3(a)–3(c), figure 3(d) subsequently demonstrates the influence of the vacancy separation L on both the transition point γ0 and the initial energy level splitting at γ = 0 (inset). Figure 3(d) illustrates the dependence of the transition point γ0 on the vacancy separation L, while the inset displays the initial energy level splitting δE0 = |E1(γ = 0) – E2(γ = 0)| between the two defect states, which represents the intrinsic tunneling strength in the Hermitian limit. The results indicate that as the vacancy distance increases, the tunnel coupling between the two defect states weakens at γ = 0, which consequently leads to an earlier onset of the PT-broken phase with increasing γ.

  • Here, we extend our investigation to a one-dimensional (1D) array of N alternating A–B vacancies located at sites j1 ⋯ jN along the zigzag direction (i.e., the x-direction), with adjacent spacing L. Generalizing the two-vacancy case, we construct an effective non-Hermitian tight-binding Hamiltonian to describe this subsystem

    where c^j† and c^j (j = j1 ⋯ jN) represent the particle creation and annihilation operators at the vacancy site j. Here hx(γ) is the coupling between the nearest-neighbor sites, and ±ihz(γ) are the on-site potentials on B or A sublattices, respectively. The γ-dependence of hx(γ) and hz(γ) follows from the two-defect analysis.

    For a periodic defect array, we apply the Fourier transform c^j∈A=1N∑kxeikxRja^kx, c^j∈B=1N∑kxe−ikxRjb^kx, yielding the momentum-space effective Hamiltonian H^eff(kx)=∑kx(a^kx†,b^kx†)ℋeff(kx)(a^kx,b^kx)T with

    Diagonalizing ℋeff(kx) gives the spectrum

    The effective Hamiltonian H^eff(kx) possesses PT symmetry [H^eff(kx),PT]=0, where P=σx and T performs complex conjugation.

    For a fixed momentum kx, the system is in the PT-unbroken phase when |hz(γ)| < |hx(γ) cos(kx)|, yielding real eigenvalues E±(kx). When |hz(γ)| > |hx(γ) cos(kx)|, the eigenvalues become purely imaginary, corresponding to the PT-broken phase. The EP occurs at |hz(γ)| = |hx(γ) cos(kx)|. The PT phase transition depends on kx, and symmetry breaking occurs gradually in pairs of PT-symmetric dimers as γ varies.

    Figures 4(A) and 4(B) present the real and imaginary parts of the analytical and numerical energies for the ten defect states as a function of γ. The defect states form PT-symmetric pairs that sequentially transition into the PT-broken phase as γ increases. The paired states in the PT-unbroken phase share identical spatial distributions spanning multiple defect sites [Figs. 4(a)–4(c) and 4(f)–4(h)], whereas the distributions of the remaining PT-broken pairs become distinct [Figs. 4(d), 4(e) and 4(i), 4(j)]. These results align with the predictions of our effective model. Furthermore, we observe that many states exhibit a spatial asymmetry (with higher probability weights on the right-side defects). This likely originates from the hy term — which is neglected in our effective model — analogous to the mechanism discussed in the two-defect case.

4.   NHSE in subbands from vacancies on the same sublattices
  • We next analyze two vacancies on the same sublattices (e.g., A-sublattice), separated by a distance L along the zigzag direction (Fig. 5(a)). The tunneling between the two defect states differs from that in the different-sublattice case. As shown in Figs. 5(b) and 5(c), the eigenvalues of the two defect states undergo a spectral transition from complex to purely imaginary as γ increases.

    Within the framework established in the preceding section, we model the tunneling via the effective Hamiltonian

    with eigenvalues

    Here σx,y are Pauli matrices, σ0 is the 2 × 2 identity matrix and hx, hy, and h0 are real γ-dependent parameters plotted in Fig. 6(a). For γ < 0.15, hy remains nearly constant while hx increases approximately linearly with γ. The effective Hamiltonian is non-Hermitian and respects sublattice symmetry: σzH^effσz=−H^eff.

    At the critical value γ0, where hx = hy, the effective Hamiltonian reduces to H^eff=2ihxσ−+ih0σ0 with σ−=12(σx−iσy), signaling an EP. Here, both eigenvalues and eigenvectors coalesce: E1 = E2 = ih0 and |ψ1〉 = |ψ2〉 = (0,1)T. As γ increases and surpasses the critical value γ0, the energy spectrum undergoes a transition from complex to purely imaginary, with the EP at γ = γ0 marking this critical point of transition.

    Figure 6(b) illustrates the projection of the two defect states onto the selected basis, and figure 6(c) displays their similarity |〈ψ1|ψ2〉|. For γ ≠ 0, both defect states tend to be more localized at the right vacancy defect. As the system approaches the EP, the disparity in distribution between the two vacancies becomes more pronounced. At the EP (i.e., γ = γ0), both defect states become completely localized at the right vacancy defect, with a similarity of 1, consistent with the predictions of our effective model. The analytical eigenenergies from the effective model agree closely with numerical results (Figs. 5(b) and 5(c)). With the distance between vacancies fixed at L/|b1| = 6 in Figs. 6(a)–6(c), figure 6(d) examines the dependence of both the transition point γ0 and the initial energy level splitting (inset) on the vacancy separation L. It is shown that the tunneling effect at γ = 0 is suppressed by increasing L, and the occurrence of the EP for γ ≠ 0 is explicitly shown to depend on L.

  • Having established the formation of a non-Hermitian dimer from two vacancies on the same sublattice, we extend our analysis to a 1D array of N such vacancies located at sites j1 ⋯ jN, all residing on the same A-sublattice with period L along the zigzag direction (i.e., the x-direction). This subsystem can be described by the tight-binding Hamiltonian with nonreciprocal hopping:

    where c^j† and c^j (j = j1 ⋯ jN) represent the particle creation and annihilation operators at the vacancy site j, i[hx(γ) ±hy(γ)] is the nonreciprocal hopping amplitudes between nearest-neighbor sites and ih0(γ) is the on-site potential. The parameters hx(γ), hz(γ) and h0(γ) evolve with γ, following the behavior established in the two-vacancy scenario.

    For a periodic vacancy array, we apply the Fourier transform c^j=1N∑kxeikxRja^kx, yielding the momentum-space effective Hamiltonian H^eff(kx)=∑kxa^kx†ℋeff(kx)a^kx with

    and energy spectrum

    Figure 7(a) shows the complex energy spectrum for the non-Hermitian Haldane model with 20 A-sublattice vacancies, where the vacancies form a periodic lattice. The defect state energies lie within the bulk gap and form a closed loop in the complex plane (Fig. 7(b)), indicating nontrivial spectral topology. Figure 7(c) plots the γ-dependence of the effective Hamiltonian matrix elements following the two A-sublattice vacancies scenario.

    Using the extracted values hx(0.05), hy(0.05) and h0(0.05) in Fig. 7(c), we compute Eeff from Eq. (16). The resulting closed energy loop in Fig. 7(d) matches the numerical results in Fig. 7(b), validating our effective model. The nonzero spectral winding around points such as d1 or d2 implies the presence of the NHSE for the vacancy array subsystem under open boundary condition.

    To verify the NHSE in this 1D vacancy–defect subsystem, we construct eight vacancies on the A-sublattice with open boundaries embedded in the parent model. Figures 8(A) and 8(B) display the real and imaginary parts of the energies for eight defect states as a function of γ. Their spatial profiles at γ = 0.05 (Figs. 8(a)–8(h)) reveal that all eigenmodes are localized toward the left vacancy boundary, demonstrating that the subsystem exhibits the NHSE as predicted by our effective model. Strikingly, this pronounced skin localization emerges in the 1D A-sublattice vacancy array subsystem, even though the parent model itself lacks the NHSE.

5.   Conclusion
  • In summary, we have systematically investigated the emergent phenomena in vacancy-induced subbands within a non-Hermitian Haldane model. By constructing effective models, we have demonstrated that the spatial configuration of the vacancy array dictates the fundamental character of the resulting subsystem. For a 1D chain of vacancies alternating between different sublattices, the subsystem exhibits a sequence of PT phase transitions, with eigenvalues transitioning from real to imaginary pairs. In stark contrast, a vacancy chain confined to a single sublattice gives rise to the NHSE, characterized by nontrivial spectral winding. The most profound implication of our work is the demonstration that vacancy defect engineering can create effective subsystems whose physical properties are entirely decoupled from those of the parent Hamiltonian. The emergence of the NHSE in a single-sublattice vacancy chain — despite its absence in the parent system — highlights the power of this approach.

Figure (8)  Reference (46)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return