---
title: 'Stealthy Disorder: Hidden Constraints in Systems'
url: https://www.emergentmind.com/topics/stealthy-disorder
type: topic
---

# Stealthy Disorder: Hidden Constraints in Systems

Stealthy disorder denotes a class of systems in which apparent disorder is accompanied by hidden constraints that suppress the observable signature ordinarily used to diagnose randomness, scattering, or attack. In condensed-matter, statistical-mechanical, and wave-transport settings, the canonical form is stealthy hyperuniform disorder: configurations that are disordered in real space yet satisfy a reciprocal-space exclusion condition such as \(S(\mathbf{k})=0\) for \(0<|\mathbf{k}|\le K\), thereby eliminating single scattering over a finite low-\(k\) interval and suppressing long-wavelength fluctuations [1503.06436][1508.07669]. In cyber-physical control and sequential decision systems, the same adjective denotes covert perturbations that keep residuals, alarms, or homogeneity tests unchanged or bounded while materially altering states, estimates, or actions [2002.01557][1710.02597][2402.13487]. Across these literatures, the unifying theme is not randomness alone, but concealment relative to a specified detection observable.

## 1. Formal criteria and terminological scope

In the many-particle and wave literature, stealthiness is defined through the structure factor or spectral density. For a finite \(N\)-particle configuration,
\[
S(\mathbf{k})=\frac{|\tilde n(\mathbf{k})|^2}{N}, \qquad \tilde n(\mathbf{k})=\sum_{j=1}^N e^{-i\mathbf{k}\cdot\mathbf{r}_j},
\]
and a configuration is stealthy hyperuniform when
\[
S(\mathbf{k})=0 \qquad \text{for } 0<|\mathbf{k}|\le K.
\]
Hyperuniformity is the weaker requirement \(S(\mathbf{k})\to 0\) as \(|\mathbf{k}|\to 0\); stealthiness strengthens this to exact suppression on a finite reciprocal-space domain [2404.16819]. In one-dimensional spin chains, the analogous quantity is the spin structure factor
\[
S(k)=\frac{1}{N}\rho_\sigma(k)\rho_\sigma(-k), \qquad \rho_\sigma(k)=\sum_{j=1}^N \sigma_j e^{ikj},
\]
and the main focus is often the smallest positive wavevector \(K=2\pi/N\), so that \(S(2\pi/N)=0\) implies stealthy hyperuniformity [1508.07669].

In cyber-physical systems, stealth is defined relative to the detector residue. For a stochastic linear CPS, an attack is strictly stealthy if
\[
\|\triangle z_t\|=0,\quad \forall t\in\mathbb N,
\]
and stealthy if there exists \(\delta>0\) such that
\[
\|\triangle z_t\|\le \delta,\quad \forall t\in\mathbb N.
\]
A system is vulnerable if such an attack can drive the estimation error bias unboundedly large [2002.01557]. In chi-squared residual detection, a zero-alarm attack satisfies \(z_k\le \alpha\) at every attacked step, whereas a hidden attack preserves the nominal alarm rate \(\Pr(z_k>\alpha)=\mathcal A\) [1710.02597]. In stochastic bandits, a stealthy reward-poisoning attack is one whose probability of detection under a homogeneity test is at most the detector’s own false-positive rate \(\delta\) [2402.13487].

These definitions are not interchangeable, but they share a common architecture: a covert perturbation is engineered so that the monitoring statistic remains null, bounded, or statistically normal.

## 2. Stealthy hyperuniform disorder as a state of matter

The foundational statistical-mechanical construction starts from bounded, isotropic pair potentials whose Fourier transform is nonnegative and compactly supported,
\[
\tilde v(k)=V(k)\,\Theta(K-k), \qquad V(k)>0 \text{ for } 0\le k<K.
\]
For such “stealthy” potentials, the total energy is minimized when density fluctuations vanish on the support of \(\tilde v\), i.e. when \(S(k)=0\) for \(0<k\le K\). The resulting classical ground states are highly degenerate and can be simultaneously disordered and hyperuniform [1503.06436].

A central control parameter is
\[
\chi=\frac{M}{d(N-1)},
\]
where \(M\) is the number of independent constrained wavevectors. In the thermodynamic limit,
\[
\chi = \frac{v_d K^d}{2d\,\rho\,(2\pi)^d}.
\]
For \(d\ge 2\), disordered stealthy ground states occur for \(\chi<1/2\), whereas only ordered phases are allowed for \(\chi>1/2\); \(\chi=1/2\) is the boundary between generic disorder and forced order. Ordered states include crystals and stacked-slider phases, the latter being nonperiodic but still ordered and not representative of disordered stealthy states [2404.16819].

The reciprocal-space exclusion zone has a real-space geometric consequence: stealthy systems have bounded hole size. In nearest-neighbor statistics, this means that the void exclusion probability \(E_V(r)\) and the associated density \(H_V(r)\) have compact support, vanishing for \(r\ge r_c\). The probability of holes close to the critical-hole size is argued to decay asymptotically faster for disordered stealthy systems than for crystalline lattices, implying that near-critical holes are rarer in disordered states [2009.00123].

A recurrent misconception is that any system with strong low-\(k\) suppression is automatically equivalent to a crystal. The literature is more specific. Gyromorphs, for example, were proposed as disordered media with rings of Bragg-like peaks that seemed to reproduce stealthy-hyperuniform functionalities without stealthiness or hyperuniformity. A later analysis showed instead that gyromorphs are hyperuniform and, in the large-\(G\) nearly isotropic limit, belong to Class III hyperuniformity, whereas stealthy hyperuniform media belong to Class I. Correspondingly, gyromorphs display size-dependent pseudogaps populated by localized states rather than the smooth band gaps found for highly stealthy hyperuniform materials [2606.22300].

## 3. Spin, electronic, and magnetic realizations

The one-dimensional Ising-like spin-chain realization provides a discrete analog of stealthy many-particle systems. On a periodic integer lattice with \(\sigma_j=\pm 1\), a configuration is hyperuniform when \(\lim_{k\to 0}S(k)=0\) and stealthy when \(S(k)=0\) for \(0<k\le K\). Exhaustive enumeration over all spin configurations on periodic chains with \(N=2,3,\ldots,36\) identified all stealthy hyperuniform spin chains in that range. The number of stealthy configurations grows approximately exponentially, \(\sim a^N\) with \(a\approx 1.4\); for prime \(N\), only the all-up and all-down configurations are stealthy; and most stealthy patterns for \(N\le 36\) are reducible to smaller unit cells, so the genuinely disordered cases are the irreducible ones [1508.07669].

Inverse statistical-mechanical design was then used to ask which radial spin-spin interactions \(J(R)\) stabilize a chosen target chain. The Hamiltonian takes the form
\[
H=-\sum_{i=1}^N\sum_{R=1}^{R_C}J(R)\sigma_i\sigma_{i+R}
     =-N\sum_{R=1}^{R_C}J(R)S_2(R),
\]
with \(-1\le J(R)\le 1\). Solution classes were defined as follows: Class I, the target is the unique ground state up to translations, reflections, and global spin inversion; Class II, the target is a ground state but degenerate in \(S_2(R)\); and Class III, no stabilizing interaction is found. The stabilizing interactions are radial and long-ranged relative to the chain length; for \(N=36\), reported cutoff distances lie roughly in the range \(9\le R_C\le 23\). The resulting disordered stealthy hyperuniform ground states are explicitly distinguished from spin glasses: their couplings are deterministic and engineered, and their defining feature is the reciprocal-space constraint \(S(k)=0\) over a low-\(k\) interval [1508.07669].

A related lattice-electron realization appears in the Hubbard model on the honeycomb lattice with stealthy hyperuniform bond disorder. The bond configuration is optimized so that
\[
S({\bf k})=0 \quad \text{for } 0<|{\bf k}|<k_c.
\]
At \(U=0\), diagonalization shows that a linear density of states robustly emerges around \(E=0\), while the stealth property mainly changes higher-energy wave functions and the DOS near the band edge. Near the band edge, random disorder produces stronger localization, whereas in the stealthy hyperuniform case the high-energy states become more extended. Under a real-space Hartree approximation, the interacting system still undergoes a semimetal-to-antiferromagnet transition, but the critical interaction strength is larger than in the random case; the average magnetization is smaller over the same interaction window, and the stealthy pattern suppresses the fraction of strongly magnetized \(C_3\) sites [2604.19041].

These realizations show that stealthy disorder is not confined to continuum point processes. It can be encoded in spins, bonds, and local environments, provided the decisive suppression occurs in the relevant reciprocal-space or detector-space observable.

## 4. Transparency, scattering suppression, and localization

In wave transport, the defining physical interpretation of stealthiness is the suppression of single scattering. One-dimensional disordered stealthy hyperuniform layered media, consisting of up to \(10{,}000\) thin high-dielectric slabs separated by low-dielectric intervals, exhibit no apparent evidence of Anderson localization or deviations from transparency for a continuous frequency band \(0\le \omega<\omega_T\). Transfer-matrix calculations of the Lyapunov exponent
\[
\lambda(L)=\left\langle \frac{1}{L}\log \|\Pi(L)\| \right\rangle, \qquad 
\lambda=\limsup_{L\to\infty}\lambda(L), \qquad \xi=\frac{1}{\lambda},
\]
show numerically zero \(\lambda\) in the transparent band, in contrast to perturbed lattices, RSA, and equiluminous systems, which localize for all \(\omega>0\). The observed transparency edge is always close to but below the strong-contrast upper bound
\[
\omega_c(\chi;\varepsilon_2,\varepsilon_1,\phi)
=\frac{\pi \chi}{\sqrt{\phi(\varepsilon_2-1)+\varepsilon_1}}.
\]
The study is explicitly finite-size and does not claim a proof of absent localization in the thermodynamic limit [2507.22377].

A direct experimental validation was reported for two-dimensional water waves. In a \(150\times 60\) cm tank containing \(N=401\) PMMA cylinders at density \(\rho=0.2\ \mathrm{cm^{-2}}\), one medium was uncorrelated and the other stealthy hyperuniform with \(\chi=0.5\), yielding \(K\simeq 2.24\ \mathrm{cm^{-1}}\) and \(K/2\simeq 1.12\ \mathrm{cm^{-1}}\). Because the scattering mean free path obeys
\[
\frac{1}{\ell_s}=\rho \int_0^{2\pi} S(q)\,|f(q)|^2\,d\theta,
\]
stealthy hyperuniformity predicts \(\ell_s\to\infty\) for \(k\le K/2\). The measured extinction coefficient shows a sharp transition at the predicted threshold; below \(K/2\), the paper reports that the ratio of extinction coefficients is about \(1.5\), while above threshold it drops to about \(0.9\). The standard deviation of the effective scattering coefficient nearly vanishes in the stealthy regime, indicating self-averaging suppression of scattering at the level of individual samples [2602.07067].

In two-dimensional photonic-crystal slabs, stealthy-hyperuniform disorder has been probed through disorder-induced linewidth broadening. For a Hermitian quadratic band, the transition between stealthy and non-stealthy regimes occurs at
\[
k=\frac{K}{2},
\]
below which leading-order single scattering is forbidden. Reflection spectroscopy confirms a sharp linewidth increase at this boundary. The same study also shows that residual single scattering in the stealthy regime arises from an intrinsically non-Hermitian effect: the effective mass is complex because of radiative loss out of the slab, so residual broadening scales with the imaginary part of the effective mass. Multiple scattering is observed as disorder strength increases, implying diminishing transparency even within the stealthy regime [2507.05253].

The one-dimensional Anderson model with stealthy disorder provides the corresponding quantum-disordered lattice analogue. Here the disorder power spectrum is prescribed as
\[
S(q)=\Theta(|q|-k_0), \qquad \chi=\frac{k_0}{2\pi}.
\]
The standard weak-disorder formula
\[
\frac{1}{\xi(k)}=\frac{W^2}{8\sin^2 k}\,S(2k)
\]
shows that when \(2k<k_0\), first-order back-scattering is eliminated. Higher-order processes then control localization, producing regimes with
\[
\xi \sim W^{-2n}, \qquad n\ge 2.
\]
For fixed energy and small but finite disorder strength, any finite system has a range of \(\chi\) for which \(\xi\) exceeds the system size, a phenomenon termed effective delocalization. The paper is explicit that this is not a true metallic phase in the thermodynamic limit [2509.13502].

## 5. Generalizations: non-Hermitian, anisotropic, and deep-subwavelength stealthiness

The non-Hermitian generalization replaces a real material field by a complex potential
\[
V_a(\mathbf r)=V_r(\mathbf r)+iV_i(\mathbf r),
\]
with generalized structure factor
\[
S(\mathbf{k}) = S_A(\mathbf{k}) + S_c(\mathbf{k}),
\]
\[
S_A(\mathbf{k}) = \left\langle |V_r(\mathbf{k})|^2 \right\rangle + \left\langle |V_i(\mathbf{k})|^2 \right\rangle,
\]
\[
S_c(\mathbf{k}) = 2\,\mathrm{Im}\!\left[\left\langle V_r(\mathbf{k})V_i^*(\mathbf{k})\right\rangle\right].
\]
A notable result is that non-Hermitian hyperuniformity reduces to simultaneous hyperuniformity of the real and imaginary parts separately, because the odd cross term cannot change the long-wavelength requirement. Stealthiness is different: real–imaginary cross-correlations are irrelevant for hyperuniformity but essential for stealthiness, since \(S_c(-\mathbf{k})=-S_c(\mathbf{k})\) permits directional cancellation or enhancement of scattering. This yields unidirectional scattering phases inaccessible in Hermitian disordered materials and in non-Hermitian crystals [2602.09458].

The same framework broadens stealthiness from isotropic exclusion zones to sectorial or angle-selective conditions,
\[
S(|\mathbf{k}|\le K;\theta(\mathbf{k})\in \phi)\approx 0,
\]
which place stealthiness within a statistical crystallography of scattering symmetries rather than only a scalar low-\(k\) suppression criterion [2602.09458].

A complementary deep-subwavelength extension appears in one-dimensional multilayers. There, stealthy hyperuniformity is treated as an intermediate microstructural phase between a crystal and a Poisson-like uncorrelated stack, defined by
\[
S(|k|<K)\sim 0.
\]
Inverse design of the layer positions at fixed EMT-effective index enables controlled transitions that separately modify short-range and long-range order. The physical mechanism is not ordinary homogenization but the breakdown of effective medium theory under interface-dominated transport, especially in the Goos-Hänchen and EMT-evanescent regimes. Localization is quantified by
\[
\zeta = -\frac{L}{\langle \ln T(\theta)\rangle}.
\]
Increasing \(S(k)\) at a target \(k\) selectively annihilates the corresponding Fabry–Perot resonance, whereas decreasing \(S(k)\) preserves or creates it, allowing angle-selective manipulation of localization even at deep-subwavelength scales [2410.07875].

These generalizations show that stealthy disorder is not restricted to isotropic Hermitian transparency windows. It extends to gain–loss media, directional scattering phases, and subwavelength structures in which interface phase shifts and evanescent components dominate transport.

## 6. Stealthy attacks in cyber-physical and decision systems

In control-theoretic usage, stealthy disorder is a covert actuation or sensing perturbation that hides from a specified detector. For a stochastic linear CPS with state and output equations
\[
x_{t+1} = Ax_t + Bu_t + w_t,\qquad y_t = Cx_t + v_t,
\]
attacks can enter through actuators, sensors, or both. The central theorem states that vulnerability is equivalent to the existence of an unstable reachable zero-dynamics mode compatible with the attack channels: there must exist a vector \(v\) and matrix \(Q\) such that \(v\) is an unstable eigenvector of \(A+B^aQ\), \(Cv\in \mathrm{span}(\Gamma^a)\), and \(v\) is reachable for the attacked estimation-error dynamics. Strict vulnerability is characterized by non-invertibility of an associated attacked system. For invulnerable systems, the estimation error bias remains bounded under any stealthy attack, with universal bound
\[
\|\triangle e_k\|_2 \le \|R\|_{1,sp}\delta.
\]
Thus, stealthiness does not imply harmlessness; it determines whether destabilization can occur without residual exposure [2002.01557].

A closely related sensor-attack literature studies residual detectors of chi-squared type,
\[
z_k=r_k^T\Sigma^{-1}r_k.
\]
Zero-alarm attacks enforce \(z_k\le \alpha\) at all attacked times, whereas hidden attacks preserve the nominal alarm frequency \(\Pr(z_k>\alpha)=\mathcal A\). Because hidden attacks constrain only the probability of threshold crossing, not the magnitude of above-threshold events, they can produce much larger reachable sets than zero-alarm attacks. This result sharpens the distinction between pathwise invisibility and statistical indistinguishability [1710.02597].

For multi-agent average-consensus systems, a glocal architecture combines a centralized observer with cluster-level unknown-input observers. The global layer is made deliberately unobservable in a privacy-preserving way, while local observers monitor residuals and trigger communication-topology switching when covert attacks or zero-dynamics attacks are suspected. Under explicit rank and switching conditions, topology switching breaks the attacker’s model matching and makes the attack detectable globally [2109.14094].

In partially observed continuous-time linear systems, stealthiness has been reformulated through path-space likelihood. If the attacker injects state and observation perturbations \((\rho_t,\tau_t)\), the expected log-likelihood of the corrupted innovation process defines the stealthiness cost
\[
\mathcal S(\rho,\tau)=\frac12 E\int_0^T (H_t\Delta X_t+\tau_t)(\sigma_W\sigma_W)^{-1}(H_t\Delta X_t+\tau_t)\,dt.
\]
The optimal attack then solves a trade-off between performance degradation and detectability. Deterministic attacks reduce to an offline LQ design; adaptive attacks require a hierarchical optimization framework but still admit semi-explicit optimal controls through separation and Markovian reduction [2605.05545].

Active exposure rather than passive detection is the premise of the “Lure-and-Reveal” framework for multi-sensor uncertain systems using error-state Kalman filtering. Once the detector enters a suspect mode, the defender injects random exposure shakes into the control input,
\[
u_k=u_k^*+u_k^d,
\]
so that the defender’s state estimate diverges from the attacker’s manipulated estimate and the attack can no longer remain stealthy. The framework provides an explicit minimum shake magnitude for finite-time exposure and an upper bound ensuring compensability of the closed-loop performance [2605.10098].

In stochastic multi-armed bandits, stealthy reward poisoning is framed relative to a homogeneity test applied to each arm’s empirical means. Under this detector, many previously proposed attacks become easy to detect because they create non-homogeneous reward histories. For UCB1 and \(\epsilon\)-greedy, successful stealthy attacks depend on the environment and on the realized reward of the first pulled arm; when \(\Delta_{1K}^0\) exceeds the detector margin, effective attacks are detected with non-negligible probability, while when \(\Delta_{1K}^0<\beta(1)\) the paper constructs attacks that remain stealthy and force the learner to pull the target arm almost always. The same work also shows that more general algorithms can admit stealthy attacks that almost always succeed [2402.13487].

Stealthy disorder therefore names two technically distinct but structurally analogous research programs. In one, disorder is engineered so that its reciprocal-space signature vanishes over a designed domain, producing transparency, bounded holes, and unusual localization behavior. In the other, covert perturbations are engineered so that the detector’s residual, alarm process, or innovation law remains unchanged or acceptably bounded, even while the underlying dynamics are materially altered.

Source: https://www.emergentmind.com/topics/stealthy-disorder