Boundary-Sensitive PT Symmetry Breaking

• The paper demonstrates that PT symmetry breaking can be induced by boundary effects—such as Robin conditions and edge-localized gain/loss—rather than by modifications of bulk parameters.
• Advanced models including tight-binding lattices, Floquet systems, and non-Bloch frameworks reveal that boundary-induced phenomena govern the transition between real and complex spectral regimes.
• The work differentiates PT symmetry breaking from lasing by showing that the spectral transition remains coupling-insensitive, emphasizing the crucial role of boundary conditions in non-Hermitian dynamics.

Boundary-sensitive PT symmetry breaking denotes PT-symmetry-breaking phenomena in which the transition between real and complex spectra, or between unimodular and non-unimodular scattering eigenvalues, is controlled by boundary conditions, boundary matching, or boundary-generated non-Hermiticity rather than by bulk parameters alone. In the current literature this appears in open scattering systems mapped to Robin boundary conditions, in tight-binding lattices where open and periodic boundaries produce different PT thresholds, in non-Bloch settings where the generalized Brillouin zone replaces the ordinary Brillouin zone under open boundary conditions, and in Floquet systems whose effective Hamiltonian is Hermitian under periodic boundary conditions but acquires non-Hermitian boundary terms under open boundary conditions (Ambichl et al., 2013, Longhi, 2019, Li et al., 2023, Li et al., 24 Mar 2026).

1. Conceptual structure of the transition

In Hamiltonian problems with $[\mathcal{PT},H]=0$, the unbroken phase is the regime in which every eigenfunction is also an eigenstate of $\mathcal{PT}$ and all eigenvalues are real; in the broken phase, some eigenfunctions are not eigenstates of $\mathcal{PT}$, and eigenvalues appear in complex-conjugate pairs. The transition occurs at exceptional points where two real eigenvalues coalesce and become a complex-conjugate pair (Zhu et al., 2014). In PT-symmetric scattering, the same distinction is expressed at the level of the non-unitary scattering matrix $S$: in the unbroken phase the eigenvalues $\sigma_i$ are unimodular, $|\sigma_i|=1$, while in the broken phase they form inverse-conjugate pairs with $\sigma_1\sigma_2^*=1$, and the transition occurs when two unimodular eigenvalues coalesce at an exceptional point and then split off the unit circle (Ambichl et al., 2013).

What makes the phenomenon boundary-sensitive is not a single universal mechanism. In some systems, the relevant spectrum itself changes when one passes from periodic to open boundaries, so that PT symmetry is absent at the Bloch level but present at the non-Bloch level under open boundary conditions (Longhi, 2019). In other systems, the bulk Hamiltonian remains unchanged but boundary-only gain/loss terms couple directly to edge sectors, causing the first PT-broken modes to be boundary localized (Zhu et al., 2014). There are also cases in which the PT threshold is encoded in effective boundary conditions of a bounded problem, even though the physical device is an unbounded scattering system (Ambichl et al., 2013).

2. Scattering systems and Robin boundary conditions

For PT-symmetric scattering systems with balanced gain and loss, a central result is the one-to-one map between scattering eigenchannels and Robin boundary conditions of an associated bounded problem. If $S\vec v_i=\sigma_i\vec v_i$, then the corresponding interior field $\psi_i(x)$ satisfies

$$-\partial_x\psi_i(0)=\lambda\,\psi_i(0),\qquad \partial_x\psi_i(L)=\lambda\,\psi_i(L),$$

with

$\mathcal{PT}$0

In the PT-unbroken scattering phase, $\mathcal{PT}$1 and $\mathcal{PT}$2; in the PT-broken phase, $\mathcal{PT}$3 is complex. The exceptional line of the scattering problem is the envelope of the real spectra of the family of Robin-bounded problems, so PT breaking in scattering is encoded in boundary conditions rather than in a resonance condition (Ambichl et al., 2013).

The same work proves a “mirror theorem” for symmetric external coupling elements. Writing the PT-transition condition in transfer-matrix language as

$\mathcal{PT}$4

one finds that adding arbitrary symmetric mirrors leaves $\mathcal{PT}$5 invariant: $\mathcal{PT}$6 Hence the exceptional line in $\mathcal{PT}$7 is unchanged by adding symmetric mirrors, even if those mirrors are lossy or amplifying and destroy the global PT symmetry of the full device. In the boundary-condition language, the mirrors shift the effective Robin parameter $\mathcal{PT}$8, but the PT-unbroken region is the union over all real $\mathcal{PT}$9, so the phase boundary is invariant. This is why the paper characterizes the transition as boundary-sensitive yet coupling-insensitive. It also distinguishes PT breaking from lasing and CPA-lasing: lasing thresholds are strongly sensitive to external coupling, whereas the PT-breaking threshold of the scattering matrix is not (Ambichl et al., 2013).

3. Boundary modes, topology, and local non-Hermiticity

Boundary-localized non-Hermiticity gives a particularly sharp form of boundary-sensitive PT breaking in lattice models. In the PT-symmetric non-Hermitian SSH chain with imaginary potentials $\mathcal{PT}$0 and $\mathcal{PT}$1 on the two end sites, the topologically nontrivial phase $\mathcal{PT}$2 is spontaneously PT-broken for arbitrarily small $\mathcal{PT}$3: the spectrum contains $\mathcal{PT}$4 real eigenvalues and one purely imaginary conjugate pair $\mathcal{PT}$5. By contrast, in the topologically trivial phase there is a finite $\mathcal{PT}$6 where PT first breaks and a second transition $\mathcal{PT}$7 where four off-axis complex eigenvalues become purely imaginary. The distinction is controlled by whether Hermitian zero-mode edge states already exist at the boundary sites where gain and loss act (Zhu et al., 2014).

A more recent two-dimensional realization shows that spontaneous PT breaking can convert fragile topology into stable boundary phenomenology. In a PT-symmetric three-band model with fragile Euler topology, non-Hermitian couplings drive the two lower bands from a real spectrum into a pair of complex-conjugate bands separated by an imaginary spectral gap. On a cylinder, robust in-gap boundary modes then traverse that imaginary gap, and the net number of such modes is protected by an operator version of anomaly cancellation. PT-symmetric boundary perturbations can change details and add or remove pairs of modes, but cannot change the net spectral flow without closing the gap. In this setting, the observable boundary consequence of PT breaking is not merely complexification of eigenvalues but a protected in-gap edge channel structure (Yang et al., 6 Jan 2026).

Local non-Hermiticity also produces boundary-sensitive localization even without topological bands. For a Hermitian 1D lattice plus a local non-Hermitian perturbation, a perturbation at the boundary generically induces scale-free localization of continuous-spectrum eigenstates, with decay length proportional to system size. When the same impurity is placed a finite distance $\mathcal{PT}$8 from a boundary, the scale-free states are promoted to exponentially localized modes, and the number of such modes is proportional to $\mathcal{PT}$9. When the perturbation is PT symmetric, PT breaking is always accompanied by the emergence of scale-free or exponential localization, and a concise band-structure condition determines both when continuous-spectrum PT breaking can occur and the precise PT-breaking energy window (Li et al., 2023).

4. Open chains, rings, and global boundary closure

Periodic closure can alter PT-breaking behavior even when the local gain/loss structure is unchanged. In the PT-symmetric ring with gain at one site, loss at another, and two uniform tunneling amplitudes $S$0 and $S$1 along the two paths connecting them, the critical impurity strength is

$S$2

In the large-$S$3 limit this threshold is insensitive to the impurity distance: it depends only on the difference of the two tunneling amplitudes. The same model exhibits a dynamical signature with no open-chain counterpart: the chirality $S$4 of a wave packet is zero at $S$5, grows linearly for small $S$6, reaches the universal maximum $S$7 at the PT threshold, and then decreases even though the total intensity increases exponentially beyond threshold (Scott et al., 2012).

A complementary study of a nonuniform PT-symmetric ring shows how changing the boundary condition by adding a single bond can globally reshape the PT phase diagram. The model interpolates between an open chain and a ring through a boundary coupling $S$8. For $S$9, where the added bond is the weakest link, closing the chain weakens the PT-symmetric phase for essentially all impurity positions; for $\sigma_i$0, where the added bond is of the order of the bandwidth, closing the chain strengthens the PT-symmetric phase. The same work shows that the chirality of wave-packet motion on the ring is maximal at the PT threshold, but unlike the two-path uniform ring it is not necessarily equal to 1 and depends on impurity position and hopping profile. The common feature is that a boundary modification that is local in the Hamiltonian can produce a global change in the PT threshold and in the dynamical signatures of the transition (Scott et al., 2012).

5. Non-Bloch PT breaking and the generalized Brillouin zone

In systems with non-Hermitian skin effect, boundary sensitivity becomes a bulk spectral effect. The photonic quantum-walk model with balanced gain and loss provides a canonical example: under periodic boundary conditions, the Bloch quasi-energy spectrum is complex for any nonzero $\sigma_i$1, so there is no Bloch PT-unbroken phase. Under open boundary conditions, however, the correct bulk description uses the generalized Brillouin zone

$\sigma_i$2

or equivalently $\sigma_i$3. The non-Bloch quasi-energies are then

$\sigma_i$4

which are real when

$\sigma_i$5

The transition can be detected in bulk dynamics through the Lyapunov exponent $\sigma_i$6, which is zero below threshold and positive above it (Longhi, 2019).

The geometric mechanism of this non-Bloch PT breaking was identified later in terms of the generalized Brillouin zone itself: in one dimension, non-Bloch PT symmetry breaking occurs by the formation of cusps in the generalized Brillouin zone. At the threshold, saddle-point energies of the non-Bloch Hamiltonian coalesce on the real axis; beyond threshold, the spectrum branches into complex-conjugate pairs. The same analysis predicts non-Bloch van Hove singularities in the density of states and in end-to-end Green’s functions, with exponents fixed by the saddle-point order rather than by ordinary Bloch-band extrema (Hu et al., 2022).

Exact quasiperiodic models show that the boundary sensitivity can be tied to localization transitions. In PT-symmetric non-Hermitian mosaic quasicrystals, Avila’s global theory yields exact mobility edges, and the mobility edge is identical to the boundary of PT-symmetry breaking. In the reciprocal PT-symmetric case, extended states correspond to the PT-symmetric sector and localized states to the PT-broken sector. When nonreciprocal hopping is added, localized states remain boundary-insensitive, but extended states become skin states under open boundary conditions; skin states and localized states coexist and are separated by the same analytically determined mobility edge (Liu et al., 2020).

A further dimensional result sharpens the boundary-sensitive character of non-Bloch PT breaking. In one dimension, the non-Bloch PT-breaking threshold generally remains nonzero in the large-size limit. In two and higher dimensions, by contrast, the threshold universally approaches zero as system size increases. The paper identifies the product $\sigma_i$7 as the effective measure of PT-breaking tendency: perturbation theory requires this product to be small, and its growth with system size causes the breakdown of perturbation theory and the onset of PT breaking. The result is a dimensional surprise: the interplay of PT symmetry, skin effect, and open boundaries is qualitatively different in $\sigma_i$8 (Song et al., 2021).

6. Floquet boundary generation, continuum regulators, and singular stability boundaries

A distinct mechanism appears in driven systems. In the one-dimensional Floquet model built from two noncommuting shift steps, the Floquet Hamiltonian is Hermitian under periodic boundary conditions,

$\sigma_i$9

but under open boundary conditions the Baker–Campbell–Hausdorff expansion generates non-Hermitian boundary terms localized near the two ends. The PT-breaking transition occurs when the bulk quasienergy bandwidth

$|\sigma_i|=1$0

reaches the full frequency Brillouin zone $|\sigma_i|=1$1, namely at

$|\sigma_i|=1$2

This criterion is qualitatively different from static non-Hermitian systems, where PT breaking is typically tied to band touching or exceptional points in a pre-existing non-Hermitian Hamiltonian. In the Floquet model, non-Hermiticity is boundary-generated, and the PT-broken phase exhibits scale-free localization with imaginary parts of quasienergies scaling as $|\sigma_i|=1$3 (Li et al., 24 Mar 2026).

Continuum models show equally strong dependence on how boundaries are posed. In the square box with hard-wall boundary conditions, the PT-symmetric potential $|\sigma_i|=1$4 yields eigenvalues that are complex for sufficiently small $|\sigma_i|=1$5, while $|\sigma_i|=1$6 exhibits real eigenvalues for sufficiently small $|\sigma_i|=1$7. The difference is traced to point-group symmetry and to whether first-order perturbation corrections vanish or are nonzero within the irreducible representations fixed by the box and the boundary conditions. Here the same non-Hermitian strategy behaves differently because the confining domain fixes the relevant parity structure and degeneracies (Fernández et al., 2013).

An even more explicit boundary formulation occurs in the attractive inverse-square problem. There the Hamiltonian

$|\sigma_i|=1$8

requires a boundary condition at the origin, encoded in point-particle effective field theory by a $|\sigma_i|=1$9 term with running coupling $\sigma_1\sigma_2^*=1$0. The renormalization-group equation for the boundary coupling has two real fixed points in the subcritical regime $\sigma_1\sigma_2^*=1$1, corresponding to a PT-symmetric phase, but above $\sigma_1\sigma_2^*=1$2 the fixed points move to imaginary values and represent perfect sink and perfect source boundary conditions around which the flow executes limit-cycle evolution. From this viewpoint, fall to the centre is a PT-symmetry-breaking transition of the boundary coupling rather than a bulk modification of the inverse-square potential (Sundaram et al., 2021).

A related classical picture appears in systems with indefinite damping and balanced gain and loss. Perfectly PT-symmetric, marginally stable configurations occupy singularities on the boundary of asymptotic stability, and PT-symmetry breaking occurs at exceptional points through a non-semisimple $\sigma_1\sigma_2^*=1$3 resonance. In parameter space, these PT-symmetric systems lie on singular surfaces with Whitney-umbrella or Plücker-conoid structure, so small mismatches of gain and loss produce dissipation-induced destabilization in a strongly direction-dependent way (Kirillov, 2011).

Boundary-sensitive PT symmetry breaking is therefore not a single theorem but a family of mechanisms linked by one common feature: the PT transition is organized by how states match, terminate, or localize at boundaries. In scattering it is encoded in Robin boundary conditions; in lattice and topological models it is often initiated by edge or boundary sectors; in non-Bloch systems it is defined only after open boundaries select a generalized Brillouin zone; and in Floquet or singular-continuum problems the boundary can generate the non-Hermiticity that the bulk Hamiltonian does not contain. This suggests that, in non-Hermitian PT physics, “bulk versus boundary” is frequently not a separation of roles but the central dynamical content of the symmetry-breaking transition itself.