Hyperuniform Scalar Fields

• Hyperuniform scalar fields are statistically homogeneous random fields defined by vanishing low-frequency spectral density, uncovering hidden large-scale order in continuous media.
• They can be constructed via spectral methods for Gaussian fields or through dynamic PDEs like the Cahn–Hilliard and Swift–Hohenberg equations, with classifications based on the small-k exponent α.
• Applications in imaging, optics, and heterogeneous materials leverage these properties using thresholding and topological analysis to design systems with enhanced performance.

Hyperuniform scalar fields are statistically homogeneous random fields in which long-wavelength fluctuations are anomalously suppressed, so that the spectral density vanishes at the origin, $\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$. In this sense they generalize hyperuniformity from point configurations and two-phase media to continuous fields, providing a continuum description of hidden large-scale order in disordered patterns. The concept now encompasses Gaussian random fields, phase-separating and pattern-forming PDEs, active field theories, optical intensity fields, and continuously varying material-property fields, with both isotropic and directional variants (Torquato, 2016, Ma et al., 2017, Torquato, 2018).

1. Formal definition and scaling theory

For a real-valued, statistically homogeneous scalar field $F(\mathbf{x})$ on $\mathbb{R}^d$, the mean is $\langle F\rangle$ and the autocovariance is

$$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$

Its spectral density is the Fourier transform

$$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$

Hyperuniformity is the condition $\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$, with the direct-space sum rule $\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$. The vanishing of the spectral density at the origin forces positive and negative parts of the autocovariance to cancel in aggregate, which distinguishes hyperuniform fields from ordinary disordered fields with finite $\tilde{\psi}(0)$ (Torquato, 2016, Torquato, 2018).

A standard classification uses the small-$k$ asymptotic form $F(\mathbf{x})$0 as $F(\mathbf{x})$1. Class I corresponds to $F(\mathbf{x})$2, Class II to $F(\mathbf{x})$3, and Class III to $F(\mathbf{x})$4; anti-hyperuniform systems have $F(\mathbf{x})$5. The same classification governs the decay of fluctuations of window-averaged fields. For a window average

$F(\mathbf{x})$6

the variance can be written as

$F(\mathbf{x})$7

and decays as $F(\mathbf{x})$8 for Class I, $F(\mathbf{x})$9 for Class II, and $\mathbb{R}^d$0 for Class III. Nonhyperuniform fields retain the central-limit scaling $\mathbb{R}^d$1 (Torquato, 2016, Torquato, 2018, Milor et al., 1 Sep 2025).

The isotropic criterion is not exhaustive. In statistically anisotropic settings, one must consider directional hyperuniformity, defined by $\mathbb{R}^d$2 along selected directions $\mathbb{R}^d$3. This directional generalization is necessary when the spectral function is nonanalytic or angle-dependent near the origin, and it anticipates the importance of angular diagnostics in applications where isotropy is itself a design variable (Torquato, 2016).

2. Construction routes and governing equations

One explicit route is spectral construction of Gaussian random fields. A field can be represented as a superposition of plane waves with random orientations and random phases, with amplitudes chosen so that the target spectral density satisfies $\mathbb{R}^d$4 as $\mathbb{R}^d$5. In two dimensions, constructions with $\mathbb{R}^d$6 and $\mathbb{R}^d$7 were exhibited explicitly, and a stealthy variant was obtained by translating the spectrum away from the origin so that $\mathbb{R}^d$8 for $\mathbb{R}^d$9. These constructions show that hyperuniformity is not restricted to particle systems or sharp interfaces; it can be built directly into continuous Gaussian fields (Ma et al., 2017).

A second route is dynamics. The Cahn–Hilliard equation for spinodal decomposition,

$\langle F\rangle$0

produces hyperuniform scalar fields in the scaling regime, with measured small-$\langle F\rangle$1 behavior $\langle F\rangle$2 in two dimensions. The Swift–Hohenberg equation,

$\langle F\rangle$3

produces labyrinth-like patterns that are effectively hyperuniform and often stealthy-like, with $\langle F\rangle$4 in the range $\langle F\rangle$5 to $\langle F\rangle$6. In both cases the field morphology is disordered locally but constrained spectrally at large scales (Ma et al., 2017).

Reaction–diffusion and pattern-forming models provide a broader nonlinear PDE framework. Random-field solutions of Cahn–Hilliard, Swift–Hohenberg, and Gray–Scott equations, computed with uniformly random initial conditions on periodic domains, generate isotropic two-phase fields whose spectral densities are rotationally symmetric and vanish at $\langle F\rangle$7. In these constructions, the ensemble-averaged amplitude statistics are bimodal, especially for Cahn–Hilliard, which makes thresholding stable and supports later conversion into binary imaging masks (Chen et al., 2021).

Active field theories add a nonequilibrium route. Dry scalar-order active field theories, including Effective Cahn–Hilliard, Active Model B, and Active Model B+, were shown to be universally Class I hyperuniform in two dimensions, with $\langle F\rangle$8 and $\langle F\rangle$9 in the long-time limit. Their two-point spectral measures are virtually indistinguishable from passive phase-separated hyperuniform fields, even though higher moments retain activity-dependent information (Zheng et al., 2023).

A distinct dynamic mechanism couples a conserved density field to a critical scalar friction field through a mobility of the form $$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$0. Near the critical point of a liquid-to-solid transition in a vibrated granular layer, the resulting density structure factor obeys $$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$1, with $$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$2 in the two-dimensional model and $$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$3 at criticality. This identifies a transport-blocking mechanism for dynamic hyperuniformity that is mediated by a correlated scalar field rather than by equilibrium phase separation (Castillo et al., 2018).

3. Thresholding, two-phase media, and topology

Thresholding converts a continuous scalar field into a two-phase medium. If $$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$4 is the indicator of one phase and $$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$5 its volume fraction, then the two-phase autocovariance is

$$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$6

with spectral density $$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$7. Hyperuniformity of the two-phase medium is the condition $$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$8, directly analogous to the scalar-field condition (Torquato, 2016, Torquato, 2018).

A central distinction is that thresholding does not generally preserve hyperuniformity. For hyperuniform Gaussian random fields, thresholding tends to destroy the hyperuniformity of the progenitor scalar field, and a two-phase random medium derived from a hyperuniform disordered Gaussian random field cannot be hyperuniform. By contrast, thresholding non-Gaussian fields that are already nearly two-phase, such as spinodal decomposition patterns in the Cahn–Hilliard scaling regime, can yield effectively hyperuniform two-phase media. This is one of the main reasons interface-dominated, bimodal fields are operationally important (Ma et al., 2017).

The lensless-imaging construction based on reaction–diffusion fields uses precisely this interface-rich regime. The scalar field is first binarized and then skeletonized with a medial-axis thinning algorithm that preserves the Euler number. Because the progenitor fields are already two-phase and isotropic, the resulting binary contour-line PSFs retain the isotropy and low-frequency suppression of the parent field to a useful degree (Chen et al., 2021).

Topological analysis has been introduced as a complementary characterization. A signed-distance filtration

$$\psi(\mathbf{r})=\left\langle\big(F(\mathbf{x})-\langle F\rangle\big)\big(F(\mathbf{x}+\mathbf{r})-\langle F\rangle\big)\right\rangle.$$9

was proposed for hyperuniform scalar fields with interfaces, especially Cahn–Hilliard fields. Persistence diagrams $$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$0 and $$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$1 derived from this filtration, together with Wasserstein distances between diagrams, recover known features of Cahn–Hilliard dynamics: $$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$2 at small $$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$3, approximate first-order convergence in $$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$4 toward the sharp-interface limit, and self-similarity under coarsening. The same framework generalized to Gaussian random fields with tunable $$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$5, where the closest match to Cahn–Hilliard was found at $$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$6 and $$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$7, and a neural-network proof of concept using binned persistence diagrams classified GRFs into HU vs. non-HU with accuracy $$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$8 overall (Milor et al., 1 Sep 2025).

4. Diagnostics, isotropy, and finite-size effects

The primary diagnostic remains the small-$$\tilde{\psi}(\mathbf{k})=\int_{\mathbb{R}^d}\psi(\mathbf{r})e^{-i\mathbf{k}\cdot\mathbf{r}}\,d\mathbf{r}.$$9 spectral density, but finite systems require operational proxies. One widely used hyperuniformity metric is

$\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$0

with finite-size proxy

$\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$1

Ideal hyperuniformity has $\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$2; the literature summarized here distinguishes effectively HU with $\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$3 and nearly HU with $\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$4. In practice, the smallest accessible wavenumber, the fitting range near $\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$5, and the image or simulation window strongly affect the estimate (Milor et al., 1 Sep 2025).

Window-variance scaling provides an independent check, especially when extrapolating $\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$6 is unreliable. Large observation windows act as low-pass filters, so the asymptotics of $\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$7 follow directly from the small-$\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$8 exponent. However, finite-size effects can mask the asymptotic regime. In hyperuniform vortex matter, decreasing the sample thickness depletes hyperuniform order through two crossovers: a thickness-controlled large-scale crossover that can drive the system toward asymptotically non-hyperuniform behavior in very thin samples, and a smaller-scale crossover associated with the dispersivity of elastic constants. The thickest sample studied, $\lim_{|\mathbf{k}|\to 0}\tilde{\psi}(\mathbf{k})=0$9, exhibited $\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$0, while thinner samples showed an upward bend and near-saturation of $\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$1 at the lowest accessible $\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$2 (Besana et al., 2024).

Angular diagnostics matter when isotropy is a target rather than a by-product. In the imaging application, isotropy was quantified through

$\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$3

where $\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$4 is the angular standard deviation of the radially averaged MTF. Hyperuniform-field PSFs showed smaller azimuthal variations than Perlin-noise contour PSFs, and the CH-based PSF showed a $\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$5 $\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$6 enhancement over $\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$7 realizations (Chen et al., 2021).

Experimental nearly hyperuniform fields illustrate the same diagnostic constraints. In ion-beam-patterned amorphous Ge on SiO$\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$8, the spectral density of the thresholded indicator field decayed as $\int_{\mathbb{R}^d}\psi(\mathbf{r})\,d\mathbf{r}=0$9 with inferred exponent $\tilde{\psi}(0)$0, and the hyperuniformity metric was $\tilde{\psi}(0)$1 for the averaged spectrum, with a range across images of $\tilde{\psi}(0)$2. Because the field of view was finite, the authors reported the structures as nearly hyperuniform rather than ideally hyperuniform (Claude et al., 2023).

5. Applications in imaging, optics, and heterogeneous materials

A technically mature application is lensless multispectral imaging. Hyperuniform scalar fields generated by reaction–diffusion equations were converted into sparse binary PSFs and then into diffractive Hyperuniform Phase Plates through a Gerchberg–Saxton phase-retrieval loop based on the non-paraxial Rayleigh–Sommerfeld integral. With a square aperture of side $\tilde{\psi}(0)$3, pixel pitch $\tilde{\psi}(0)$4, separation $\tilde{\psi}(0)$5, and wavelength $\tilde{\psi}(0)$6, the CH-based PSF produced a signal-to-background ratio of $\tilde{\psi}(0)$7 versus $\tilde{\psi}(0)$8 for a Perlin-noise PSF, a $\tilde{\psi}(0)$9 enhancement, and multispectral reconstructions gave SSIM $k$0 for CH versus $k$1 for Perlin across $k$2, $k$3, and $k$4 (Chen et al., 2021).

Optical speckle fields provide a different photonic realization. Amorphous speckle patterns formed by annular filtering of laser light produce intensity maxima whose point pattern is hyperuniform, with $k$5. After dividing the raw intensity by a Fourier-domain low-pass envelope, the continuous intensity field itself satisfies the scalar-field criterion $k$6. This is an experimentally simple route to hyperuniform scalar optical fields that does not rely on inverse design (Battista et al., 2018).

Nearly hyperuniform nanoarchitectures have also been fabricated directly. Ga$k$7 focused-ion-beam irradiation of thin amorphous Ge layers produced isotropic disordered porous networks with a characteristic wavevector $k$8, corresponding to $k$9 $F(\mathbf{x})$00. These networks exhibited $F(\mathbf{x})$01, nearly hyperuniform spectral suppression, and optical consequences including colorization and enhanced light absorption with respect to the flat Ge layer counterpart, up to one order of magnitude at some wavelength (Claude et al., 2023).

In heterogeneous materials, hyperuniform scalar fields can represent continuously varying local material properties $F(\mathbf{x})$02. For steady conduction, a first-order perturbation analysis yields

$F(\mathbf{x})$03

when $F(\mathbf{x})$04 at small $F(\mathbf{x})$05. Thus, in the weak-contrast limit, Class-I hyperuniform materials in the sense $F(\mathbf{x})$06 induce hyperuniform physical fields, albeit with exponent reduced by $F(\mathbf{x})$07. Away from the weak-contrast limit, second-order spectral convolution reintroduces low-$F(\mathbf{x})$08 power and the physical field develops a diverging spectral density at the origin. Numerical homogenization further showed that the mean effective conductivity increases with $F(\mathbf{x})$09 and saturates for $F(\mathbf{x})$10, while hyperuniform media are nearly isotropic and exhibit sharply reduced variance across realizations (Zhong et al., 10 Apr 2025).

6. Conceptual distinctions, misconceptions, and open problems

One common misconception is that hyperuniformity survives arbitrary nonlinear transformations. The established counterexample is thresholding: a hyperuniform Gaussian random field need not yield a hyperuniform two-phase medium after level cutting, and in the Gaussian case it generically does not. A related misconception is that hyperuniformity is fully captured by two-point functions. In dry scalar-order active field theories, the spectral density and window variance are virtually indistinguishable from passive Cahn–Hilliard behavior, yet skewness and kurtosis of coarse-grained fields retain clear activity-dependent signatures (Ma et al., 2017, Zheng et al., 2023).

Another distinction concerns weighted scalar attributes. Hyperuniformity of particle positions does not necessarily translate to hyperuniformity of the associated scalar-weight field, and the converse can also occur. In weighted particle systems, scalar weights such as Voronoi-cell volumes or excess side numbers can induce Class I hyperuniform weighted fields even when the underlying unweighted particle arrangement is nonhyperuniform or antihyperuniform, while uncorrelated zero-mean weights yield nonhyperuniform behavior. This separates scalar-field hyperuniformity from positional hyperuniformity and makes correlation between positions and weights essential (Torquato et al., 3 Mar 2026).

Current open problems concern robustness, anisotropy, and finite-size observability. The finite-thickness depletion seen in vortex matter shows that apparent large-scale suppression can be scale-dependent rather than asymptotic. For continuously varying material fields, the transfer of hyperuniformity from a property field to an induced physical field is controlled by contrast, and away from the weak-contrast limit the physical field can lose hyperuniformity even when the property field remains hyperuniform. Directional hyperuniformity in anisotropic scalar fields, preservation of hyperuniformity under nonlinear transformations, and standardized diagnostics for finite experimental windows remain active issues (Besana et al., 2024, Zhong et al., 10 Apr 2025, Torquato, 2016).

Taken together, these developments place hyperuniform scalar fields at the intersection of stochastic geometry, nonlinear PDEs, topological data analysis, photonics, and heterogeneous materials theory. The core unifying idea is simple but restrictive: a disordered field can possess hidden order if its low-frequency spectral content is forced to vanish. The difficulty, and the scientific richness, lie in how that condition is generated, diagnosed, preserved under transformations, and exploited in applications (Torquato, 2018).