Chern insulator boundary criticality

Abstract: We investigate signatures of chirality at Chern insulator transitions in the presence of a boundary. The transition between a trivial insulator and a Chern insulator with Chern number $C=1$ is described by a massless Dirac fermion whose parity anomaly gives a critical Hall conductivity $σ_{xy}=\frac{1}{2}\frac{e<sup>2}{h}$. Using a Dirac mass domain wall construction, we show that the chiral edge mode delocalizes into the bulk at criticality, but the boundary fermion correlation function retains a chiral structure and acquires the scaling dimension of the bulk fermion. We compute current correlation functions and demonstrate that the anomaly of the bulk Hall response is matched by delocalized chiral modes near the boundary. Using only the residual conformal symmetry of a (2+1)d conformal field theory (CFT) in a half-space, we identify parity-odd terms in current and energy-momentum tensor correlation functions that encode these modes and determine their electromagnetic and gravitational anomaly coefficients. Our analysis therefore applies to general time-reversal breaking (2+1)d CFTs, beyond the free Dirac transition. We also extend our results to higher Chern number transitions and to the transition between a (3+1)d topological insulator and a trivial insulator.

• The paper explicitly determines previously unknown boundary fingerprints of critical bulk fermion modes in Chern insulator transitions, solving Dirac fermion dynamics numerically using the method of images.
• Through boundary correlation function analysis, Moy and Fradkin demonstrate that critical modes traverse anomaly inflow and delocalize from the bulk, while also distinguishing these modes from localized boundary modes.
• Conventional bulk-edge correspondence was extended into critical points where it links bulk-level topological invariants to parity-odd terms in the current or electromagnetic responses,

Overview

Moy and Fradkin study the boundary physics of continuous Chern insulator transitions, using a Dirac mass domain wall construction to model a physical edge. The central object of study is the transition between a trivial insulator ($C=0$) and a $C=1$ Chern insulator, described at long distances by a single massless (2+1)d Dirac fermion. Because of the parity anomaly, this critical point carries a half-integer Hall response $\sigma_{xy}=\tfrac{1}{2}e^2/h$ — an effective Chern number $C=1/2$ — so even though no localized edge mode survives at criticality, the bulk critical point is not time-reversal invariant and must leave chiral fingerprints at any boundary. The paper identifies these fingerprints in boundary correlation functions and demonstrates anomaly inflow carried by "delocalized" chiral modes that live partly in the critical bulk rather than strictly at the edge.

Domain wall construction and boundary fermion correlator

The lattice starting point is the Qi-Hughes-Zhang two-band Hamiltonian on a square lattice, whose Chern number passes through $0 \to 1 \to -1 \to 0$ as the mass parameter is varied; the authors focus on the $m=0$ transition between $C=1$ and $C=0$. A Pauli–Villars ("heavy fermion doubler") supplies the required regularization, with its contribution expressed precisely through the Atiyah–Patodi–Singer eta invariant of the Dirac operator rather than a globally gauge-invariant level-$\tfrac12$ Chern-Simons term.

To impose a boundary, the material occupies the half-space $y<0$ while the region $C=1$0 carries a large fixed negative mass $C=1$1, taken to infinity after solving for eigenmodes of the self-adjoint Dirac operator. In this limit the problem reduces to a half-space massless Dirac fermion obeying the conformally compatible chiral boundary condition $C=1$2. The propagator follows by the method of images:

$C=1$3

where $C=1$4 is the free bulk propagator. Restricting to $C=1$5 yields a boundary correlation function with the same $C=1$6 chiral matrix structure as a gapped-phase edge mode but dressed by an extra factor of $C=1$7:

$C=1$8

The boundary fermion therefore transforms chirally under rotations about the normal axis, yet acquires the scaling dimension of the bulk Dirac field rather than the $C=1$9d value. The asymmetric dependence on light-cone coordinates distinguishes this from an ordinary chiral CFT correlator and constitutes what the authors identify as a distinct chiral boundary universality class.

Approaching criticality from the gapped phase ($\sigma_{xy}=\tfrac{1}{2}e^2/h$0), the method of images fails because a massive Dirac fermion is not reflection-invariant; the propagator must be obtained from the full spectral decomposition. The result contains the localized bound-state chiral mode decaying as $\sigma_{xy}=\tfrac{1}{2}e^2/h$1 plus corrections. As $\sigma_{xy}=\tfrac{1}{2}e^2/h$2 the localization length diverges and the exponential profile crosses over to the power-law critical form, tracing explicitly how the edge mode delocalizes into the bulk.

Electromagnetic anomaly inflow

For a general (2+1)d CFT in a half-space without time-reversal symmetry, residual conformal symmetry permits a parity-odd structure in the current two-point function proportional to $\sigma_{xy}=\tfrac{1}{2}e^2/h$3, in addition to the known parity-even McAvity–Osborn terms. Current conservation constrains $\sigma_{xy}=\tfrac{1}{2}e^2/h$4 and $\sigma_{xy}=\tfrac{1}{2}e^2/h$5 via $\sigma_{xy}=\tfrac{1}{2}e^2/h$6, while full bulk conformal invariance forces $\sigma_{xy}=\tfrac{1}{2}e^2/h$7. For the free Dirac fermion the functions evaluate to $\sigma_{xy}=\tfrac{1}{2}e^2/h$8, $\sigma_{xy}=\tfrac{1}{2}e^2/h$9, $C=1/2$0, and the distributional limit of $C=1/2$1 as $C=1/2$2 produces a universal parity-odd contact term with coefficient $C=1/2$3, computed via a nontrivial integral transform of $C=1/2$4.

This contact term supplies exactly the gauge variation needed to cancel the anomaly of the effective level-$C=1/2$5 Chern-Simons response of the critical bulk (combined with the heavy doubler's response, which renders $C=1/2$6 a true vacuum). The matter current and Chern-Simons current are each separately non-conserved, with equal and opposite anomalies $C=1/2$7, so the total current is conserved. The implication is that when the bulk is gapless, anomaly inflow need not be saturated solely by boundary degrees of freedom: delocalized critical bulk modes near the edge can carry the anomaly instead. The authors note that their quadratic-order analysis captures only perturbative aspects of the eta invariant; global anomalies would require the full nonlocal object.

Chirality is also visible directly in chiral current correlators: at the boundary only $C=1/2$8 is nonzero,

$C=1/2$9

while the opposite-chirality correlator vanishes at the boundary and grows comparable deep in the bulk — confirming that the mode exists because of the edge even though it is not sharply localized there.

Gravitational anomaly and thermal response

Coupling to curved spacetime, the critical point's effective Chern number $0 \to 1 \to -1 \to 0$0 implies a thermal Hall conductivity $0 \to 1 \to -1 \to 0$1. The energy-momentum tensor two-point function of a time-reversal-breaking half-space CFT admits two parity-odd structures $0 \to 1 \to -1 \to 0$2, $0 \to 1 \to -1 \to 0$3 expressible through a single function $0 \to 1 \to -1 \to 0$4, constrained by conservation. For the Dirac critical point the explicit calculation gives $0 \to 1 \to -1 \to 0$5 with vanishing local gravitational Chern-Simons coefficient $0 \to 1 \to -1 \to 0$6, and the distributional boundary limit of $0 \to 1 \to -1 \to 0$7 carries anomaly coefficient $0 \to 1 \to -1 \to 0$8.

Under linearized diffeomorphisms, the heavy fermion doubler generates a level-$0 \to 1 \to -1 \to 0$9 gravitational Chern-Simons term whose boundary variation is exactly canceled by the $m=0$0 contribution from the gapless matter. Thus gravitational anomaly inflow at the transition is also matched by the delocalized chiral mode, consistent with the half-quantized thermal Hall conductivity. The analysis extends immediately to the transition into a $m=0$1 superconductor (a single Majorana fermion), where the coefficients are halved and $m=0$2.

Higher Chern number transitions

For a transition $m=0$3, the domain-wall model contains $m=0$4 step-function-mass fermions (delocalized at criticality) and $m=0$5 sign-changing-mass fermions (localized edge modes). Gauge invariance of the combined electromagnetic response constrains the allowed combinations through a critical version of bulk-edge correspondence:

$m=0$6

Anomaly matching restricts but does not uniquely determine the boundary state; which combination a given microscopic realization supports depends on non-universal information. The authors also note a robustness distinction: localized modes are protected, whereas delocalized modes from opposite edges of a finite-width sample can hybridize, requiring the thermodynamic limit $m=0$7 before tuning to criticality — an order-of-limits subtlety familiar from general boundary criticality.

Extension to the (3+1)d topological insulator transition

Applying the same machinery to the transition between a strong (3+1)d time-reversal-invariant topological insulator and a trivial insulator, the surface fermion obeys $m=0$8, and the method of images gives a boundary propagator resembling a (2+1)d Dirac cone dressed so that its scaling dimension matches the (3+1)d bulk fermion. Approaching from the gapped TI side, the ordinary surface Dirac cone dominates at low energies for small positive $m=0$9. Crucially, the current two-point function at this critical point contains only parity-even structures and remains time-reversal invariant, as expected since there is no parity anomaly here.

Limitations and open questions

Several caveats qualify the results. The step-function mass profile is chosen for exact solvability rather than microscopic fidelity, though it isolates universal long-distance physics. The anomaly analysis is perturbative (quadratic order in background fields); global aspects tied to the APS eta invariant are not resolved. For direct $C=1$0 transitions, an additional protecting symmetry must be assumed. Delocalized modes require the semi-infinite geometry, and their observability in finite samples demands separation of scales between system width and correlation length. Interacting surface states of the (3+1)d TI — such as the T-Pfaffian — and their evolution toward the bulk critical point remain unaddressed, as do disorder effects relevant to experiment.

Conclusion

This work establishes that the half-integer Hall response of a Chern insulator critical point has sharp, computable consequences at boundaries even when no localized edge mode exists: the boundary fermion acquires a chiral matrix structure with anomalous scaling dimension, and the electromagnetic and gravitational anomalies are inflow-matched by delocalized chiral modes encoded in previously unidentified parity-odd structures of half-space CFT correlators. These symmetry-determined structures apply to general time-reversal-breaking (2+1)d CFTs, yielding a critical analogue of bulk-edge correspondence that constrains — though does not uniquely fix — the possible boundary physics of topological quantum critical points.