Critical Edges as Surface-Event Sensors

Rare-event detectors such as germanium crystals searching for dark matter or neutrinoless double-beta decay share one weak spot: their surfaces. Alpha and beta particles from contamination on the outer layer deposit energy where electric fields are weak and charge gets trapped. The resulting events must be recognized afterwards from the shape of the electrical pulse. That works well at high energy. Below about 1.5 keV, electronic noise wipes out the pulse shape, and surface events become indistinguishable from the bulk recoils one is looking for (Performances of a prototype point-contact germanium detector immersed in liquid nitrogen for light dark matter search). This note asks a different question. Instead of inferring "surface or bulk" from a passive pulse after the fact, could the boundary itself be a sensor that switches state only when something happens at it?

Condensed-matter theory offers an unusual candidate for such a boundary. Some topological phases (states of matter whose bulk is insulating and featureless but whose boundaries carry protected structure) admit pure edge phase transitions. In these, the boundary changes phase while the bulk stays gapped and inert. At the transition point the edge becomes gapless and highly responsive, while the bulk, protected by its energy gap, stays silent. We propose a simulation study that tests whether an edge held at this critical point gives a better low-energy surface-versus-bulk discriminator than the same edge detuned into an ordinary gapped phase, and better than a non-topological critical surface layer. The answer is meant to be decisive in either direction before anyone tries to build hardware.

Core Claim

The claim is narrow and falsifiable. Take the two-dimensional $\mathbb{Z}_2$ topological order of Topological phase transition on the edge of two-dimensional $\mathbb{Z}_2$ topological order. That paper shows the edge of the toric-code lattice model can be tuned between two gapped phases, one in which the electric ($e$) excitations condense at the edge and one in which the magnetic ($m$) excitations do. At the tuning point $\lambda = 0$ the edge is gapless and described by the Ising conformal field theory. Schematically, the Hamiltonian is

$$H(\lambda) = H_{\text{bulk}} - \sum_{i \in \text{edge}} \left[(1+\lambda)\,A_i + (1-\lambda)\,B_i\right],$$

where $H_{\text{bulk}}$ is the gapped toric-code bulk, and $A_i$ and $B_i$ are local edge terms favoring $e$- and $m$-condensation respectively.

We claim three things. First, a local energy deposit at or near the critical edge produces an edge response that stays well above a fixed noise floor down to deposits far smaller than the bulk gap. Second, deposits deep in the bulk produce essentially no edge response until the excitations they create travel to the boundary. Third, both properties are lost when $\lambda$ is moved away from zero, and the bulk-quietness property is lost when the topological bulk is replaced by a trivial one.

The quantitative handle comes from conformal field theory. At criticality, the lowest edge excitation of an edge of length $L$ costs roughly $2\pi v x / L$, with $v$ the edge velocity and $x$ a scaling dimension. The edge's sensitivity threshold therefore falls as $1/L$. The same formula sets a hard temperature requirement: $k_B T$ must stay below $v/L$. We will report this tradeoff explicitly, because it is the most likely reason the idea fails.

What is new here is not the edge transition, which Chen et al. characterize algebraically. The new part is treating its critical point as a measuring device and asking for a detection-theoretic answer: a receiver-operating curve for surface-versus-bulk events as a function of deposited energy. Neither direction in the literature has asked this. The track "topology" cuts of Topological background discrimination in the PandaX-III neutrinoless double beta decay experiment use the word in a purely geometric sense and are unrelated.

Research Program

The program has three stages, each with a stopping rule.

The first stage is model construction. We will build the toric-code model with the tunable edge of Chen et al. on cylinders and strips, together with two controls. One control is the same model detuned to $\lambda = \pm 0.3$, where the edge is gapped. The other pairs a transverse-field Ising chain at its critical point with a trivial paramagnetic bulk, so the bulk has a matching gap but no topology. Because the $\mathbb{Z}_2$ edge transition is in the Ising universality class, this second control isolates topology from criticality exactly.

The second stage simulates energy deposits. An "event" is a short local pulse of energy $\epsilon$ applied at a chosen site, at depth $d$ from the edge. The surface is $d = 0$; bulk events range up to the system width. We evolve the system in time with matrix-product-state methods on strips up to a few hundred sites long, and with exact diagonalization on small systems to check the tensor-network results. The readout is the time trace of an edge observable, either the local edge energy density or the edge string operator that distinguishes the two condensed phases. We add white noise at a fixed fraction of the detuned edge's response to mimic electronics. Thermal effects come from finite-temperature purification at several values of $k_B T / (v/L)$.

The third stage translates the results to a detector setting. We map the dimensionless results onto the regime that defines the open problem: the CDEX threshold of 160 eVee and the region below 1.5 keVee where bulk/surface separation currently fails. We then ask what edge velocity, length, and temperature a material would need. This is where the constraints from materials science bite. Unhinging the Surfaces of Higher-Order Topological Insulators and Superconductors and Anomalous gapped boundaries between surface topological orders in higher-order topological insulators and superconductors with inversion symmetry construct interacting surface orders that gap out boundary modes while preserving symmetry. They supply the theoretical templates for an engineered boundary, but they explicitly note that known higher-order materials are weakly interacting, so their constructions are not currently realizable.

The closest weakly interacting stand-in is the time-reversal-protected step-edge channel in α-arsenic from A hybrid topological quantum state in an elemental solid, whose gap opens under a magnetic field. This is a single-particle edge gap closing, not a pure topological-order edge transition. We will model it as a free-fermion analogue to see how much of the critical-edge advantage survives without strong interactions. The unexplained non-topological "soft gap" reported there is a further constraint: any material route must show that such a depression in the density of states does not mimic or mask the edge response.

Baselines

The naive reference point is a detector with no edge sensor at all. Only the bulk channel is read out, and surface-versus-bulk is decided by deposit amplitude alone. Within our simulation this gives chance-level discrimination at low energy. Any proposal that does not clearly beat it is uninteresting.

The strongest existing method on real detectors is delayed charge recovery combined with the current-to-energy ratio, from α-event characterization and rejection in point-contact HPGe detectors. It reports 100% surface-alpha rejection (to within statistical uncertainty) for radii above 3 mm, with 0.2% bulk signal loss and an order-of-magnitude reduction in background in the double-beta region of interest. The learned alternative is the convolutional-autoencoder classifier of Deep learning based pulse shape discrimination for germanium detectors, which reports 90% background rejection at 90% signal acceptance. Neither is reported below 1.5 keVee. For that regime the relevant figure is CDEX's background of 2.5 cpkkd in 1.5–2.0 keVee, alongside the statement that separation fails below it. We cannot run these methods inside a lattice model. We therefore treat their reported operating points as targets: a critical edge is only worth pursuing if, after mapping to physical units, it projects separation at sacrifice levels comparable to 0.2% in an energy range where these methods have no reported result.

The two ablations do the real work, and they are run under identical conditions: same lattice sizes, same deposit energies and depths, same noise model, same time-evolution truncation, and the same readout classifier. That classifier is either a simple threshold on integrated edge response or a small one-dimensional convolutional network, trained with equal hyperparameter budgets for every arm. The detuned-edge ablation isolates criticality. The trivial-bulk critical-chain ablation isolates topology. The proposal beats both only if the critical topological edge achieves a strictly higher area under the receiver-operating curve, with non-overlapping bootstrap intervals, at deposits below the detuned edge gap. It must also show bulk-event leakage that stays flat with depth, rather than decaying slowly as it would in the trivial bulk.

Why not just…?

Why not just push machine-learning pulse-shape discrimination further on ordinary germanium? This is the default path and should remain so above about 1.5 keVee. The autoencoder classifier and the simulation-to-data translation network of CycleGAN-driven transfer learning for electronics response emulation in high-purity germanium detectors both work on waveform shape. CDEX reports that below 1.5 keVee the shape information is buried in electronic noise. No classifier can recover information the waveform does not contain. A boundary sensor attacks the problem by creating a second, physically separate channel, not by reprocessing the first one.

Why not just use classical discriminators such as delayed charge recovery? At the energies and radii where they apply, they already achieve 100% rejection at 0.2% signal sacrifice, and we do not claim to improve on that. They rely on clean pulse tails, which degrade under pileup (Model-Based Deep Learning Algorithm for Pulse Shape Discrimination in High Event Rates). More importantly, they rely on resolvable tail slopes, which disappear at sub-keV energies. The proposal targets only that leftover window.

Why not just use any critical surface layer, topological or not? This is the real threat. Ordinary critical sensors exist, and a critical Ising chain on a trivial substrate shares the same universality class and the same $1/L$ gain. If the trivial-bulk ablation matches the topological edge on both sensitivity and bulk leakage, then topology adds nothing and the idea should be dropped in favor of conventional critical films. The case for topology is that bulk excitations are protected and can reach the edge only by traveling to it, which predicts a sharper depth cutoff for bulk leakage. The second experiment measures exactly this.

Why not just shield harder? Passive immersion and optimized moderators, such as the 66.5% neutron-background reduction found with RESuM: A Rare Event Surrogate Model for Physics Detector Design, suppress external backgrounds. They do nothing about contamination already sitting on the detector surface, which is the background this proposal targets.

Proposed Experiments

The first decisive check is threshold scaling. For the critical edge, the minimum $\epsilon$ at which surface events are detected with 90% efficiency should fall roughly as $1/L$ at temperatures well below $v/L$, and it should saturate once $k_B T$ approaches $v/L$. The detuned edge should show a flat threshold set by its gap. If the critical threshold does not scale, or saturates at temperatures no material could reach after mapping to physical units, the hardware direction is closed.

The second check is depth selectivity. We plot false-positive rate against deposit depth for all three arms at fixed $\epsilon$ below the bulk gap. Topology earns its place only if the topological bulk shows a sharper cutoff than the trivial-bulk critical chain.

The third check is anyon-type filtering, which is specific to this system. Deposits create $e$ or $m$ excitation pairs. We test whether an edge in the $e$-condensed phase registers only $e$ arrivals, and an $m$-condensed edge only $m$ arrivals. If so, a detector that alternates the edge phase could tag event classes differentially, which no passive layer can do.

The fourth check repeats the first two in the free-fermion step-edge analogue modeled on α-arsenic, including a phenomenological soft gap. This tells us how much of any advantage survives in weakly interacting materials that exist today. A null result there, with a positive result in the interacting model, would make the strong-interaction gap identified by Tiwari et al. the concrete bottleneck to report.