Polar Clustering: Concepts & Applications

• Polar clustering is the aggregation of polar entities, manifesting as finite, coherently moving clusters in active matter, dipolar droplets in quantum gases, and structured groups in data analysis.
• It involves varied methodologies from particle alignment and dipolar interactions to clustering in polar coordinates and singularity theory, each yielding unique phase behaviors and distribution metrics.
• This phenomenon bridges disciplines by linking orientational order and geometric invariants, offering practical insights into phase transitions, confined dynamics, and non-reciprocal interactions.

Searching arXiv for recent and related papers on polar clustering. arxiv_search(query="^^^^^^^^2all:polar clustering OR ti:\2^^^^^^^^"polar clustering\"^^^^^^^^2 OR abs:\2^^^^^^^^"polar clustering\"", max_results=^^^^2 OR abs:\2all:polar clustering OR ti:\2^^^^)

arxiv_search(query="polar clustering", max_results=^^^^2 OR abs:\2all:polar clustering OR ti:\2^^^^)

“Polar clustering” is not a single standardized term across the research literature. In active-matter theory it denotes the formation of finite, coherently moving, locally polar clusters, including micro-clustered phases of polar active particles; in one-dimensional polar gases it denotes spatially clustered, quasi-localized dimers and droplets generated by long-range dipolar interactions; in data analysis it denotes clustering procedures formulated in polar or circular coordinates; and in singularity theory it denotes the grouping of the polar curves of 2-variable function germs into polar clusters (&&&^^^^2all:polar clustering OR ti:\2^^^^&&&, &&&^^^^2 OR abs:\2^^^^&&&, Sun et al., 2023, Migus et al., 2021).

^^^^2 OR abs:\2^^^^. Scope of the term

In the cited literature, the term is attached to several distinct objects rather than to a single universal formalism. The common element is the role of polarity: polarity may mean local heading order in active matter, dipolar interactions in lattice gases, angle–radius structure in circular data, or polar curves in singularity theory.

Domain	Meaning of polar clustering	Representative papers
Polar active matter	Finite, coherently moving, locally polar clusters; polarized clumps; micro-clustered regimes	(&&&^^^^2all:polar clustering OR ti:\2^^^^&&&, Martín-Gómez et al., 2018, Kreienkamp et al., 2024)
Confined active matter	Smaller polarized clusters orbiting a trap center as rigid bodies	(Canavello et al., 2023)
Polar flocks and active polar fluids	Clustering, percolation, and phase separation coupled to Toner–Tu correlations or meso-scale turbulence	(Kyriakopoulos et al., 2019, Worlitzer et al., 2020)
Dipolar quantum gases	Dimer clustering, disorderless quasi-localization, and self-bound droplets	(&&&^^^^2 OR abs:\2^^^^&&&)
Polar/circular data analysis	Clustering in polar coordinates, probabilistic polar-coordinate mixture models, or polar-region functional clustering	(Sun et al., 2023, &&&^^^^2 OR abs:\22^^^^&&&, &&&^^^^2 OR abs:\2^^^^3&&&)
Mathematical geometry	Grouping polar curves; higher order polar and reciprocal polar loci	(Migus et al., 2021, &&&^^^^2 OR abs:\2^^^^5&&&)

This distribution of meanings also explains why the phrase can refer either to a physical state of matter, a statistical procedure, or a geometric invariant. A plausible implication is that any encyclopedia treatment of the topic must be domain-sensitive rather than definitionally monolithic.

2. Polar clustering in polar active matter

In the most direct contemporary usage, polar clustering arises in Vicsek-like or aligning active-particle systems in which the clusters are themselves locally polar. A representative formulation introduces point-like self-propelled units in 2D with positions PRESERVED_PLACEHOLDER_^^^^2all:polar clustering OR ti:\2^^^^ and heading angles PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2^^^^, with dimensionless dynamics

$$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$

where $\mathrm{Pe}=v_0/(D_r r_c)$ is the Péclet number and $g=J/D_r$ is the dimensionless alignment rate (&&&^^^^2all:polar clustering OR ti:\2^^^^&&&). In that model, increasing $g$ at fixed density produces the sequence homogeneous isotropic (HI) $\to$ polar bands (PB) $\to$ cross-sea (CS) $\to$ homogeneous polar (HP) $\to$ micro-clustered (mC), with simplified sequences at small or very large activity.

The micro-clustered regime is defined as a polar-ordered state at large PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2all:polar clustering OR ti:\2^^^^ in which the system breaks into many small, self-propelled, locally polar flocks of finite typical size; there is no single macroscopic band or percolating structure. Its structural signature is a scale-free cluster-size distribution without a separate macroscopic peak. In the disordered phase the cluster size distribution is

PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2 OR abs:\2^^^^

whereas in the ordered phases it is fitted by a double power law,

PRESERVED_PLACEHOLDER_^^^^2 OR abs:\22^^^^

In the mC regime at PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2^^^^3 and PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2^^^^4, the large-PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2^^^^5 exponent approaches PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2^^^^6, all particles are accounted for within the scale-free distribution, and no infinite-cluster peak remains (&&&^^^^2all:polar clustering OR ti:\2^^^^&&&).

The same work establishes that the isotropic-to-polar transition is strongly first order. The evidence is a negative Binder-cumulant minimum that sharpens with system size, together with bimodal distributions of local polarization and density near onset, indicating coexistence of gas-like and liquid-like regions. Mean-field Fokker–Planck theory without density fluctuations yields a continuous transition at PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2^^^^7, but the ^^^^2 OR abs:\2^^^^D Toner–Tu reduction generates an effective cubic term PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2^^^^8 in the free-energy density, and that advective coupling drives the transition first order. In this setting, finite-time orientational relaxation is not a perturbative detail but a control parameter that reorganizes clustering morphologies, band widths, and the ordered-state topology of the phase diagram (&&&^^^^2all:polar clustering OR ti:\2^^^^&&&).

3. Other active-matter realizations

A second line of work studies repulsive Active Brownian Particles with explicit polar alignment. In that model the translational dynamics is

PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2^^^^9

the orientation dynamics is

$$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$^^^^2all:polar clustering OR ti:\2^^^^

and the control parameters are $$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$^^^^2 OR abs:\2^^^^, $$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$2, and $$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$3 (Martín-Gómez et al., 2018). At $$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$4, the flocking threshold satisfies $$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$5 for $$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$6. Above threshold, the system displays a regime of finite polar clusters (mC) and, at larger $$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$7, a macroscopic-cluster phase (MC) with band-like, lane-like, or hybrid traveling structures. A central conclusion is that global polar order and large-scale density segregation decouple: there is a broad region $$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$8 in which the system is globally polar but composed of finite polar clusters.

Under harmonic confinement, sterically interacting active Brownian particles with an alignment torque proportional to the total mechanical force generate a distinct family of polar clusters. The rotational dynamics is

$$d \mathbf{r}_i = \mathrm{Pe}\,\mathbf{u}_i\,dt, \qquad d\theta_i = - g \sin(\theta_i - \psi_i)\, dt + \sqrt{2}\, dB_i,$$9

and the dominant polar state is the orbiting polar or “ferromagnetic” state, in which all particles’ orientations are nearly parallel and the cluster revolves around the trap center without rotating about its own centroid (Canavello et al., 2023). In dimensionless form, the centroid obeys

$\mathrm{Pe}=v_0/(D_r r_c)$^^^^2all:polar clustering OR ti:\2^^^^

and for $\mathrm{Pe}=v_0/(D_r r_c)$^^^^2 OR abs:\2^^^^ the orbiting polar cluster has

$\mathrm{Pe}=v_0/(D_r r_c)$2

At lower filling fraction $\mathrm{Pe}=v_0/(D_r r_c)$3, the single trapped cluster breaks into smaller polarized clusters, each one orbiting the potential center as a rigid body. Here polar clustering is therefore inseparable from confinement-induced rigid-body kinematics.

Non-reciprocal binary mixtures provide a third mechanism. With torque

$\mathrm{Pe}=v_0/(D_r r_c)$4

non-reciprocity means $\mathrm{Pe}=v_0/(D_r r_c)$5 (Kreienkamp et al., 2024). For antagonistic couplings $\mathrm{Pe}=v_0/(D_r r_c)$6, one species forms compact, polarized clusters while the other remains more dilute, yielding asymmetric clustering and a chase-and-run dynamics in which a dense polarized clump of one species persistently pursues a diffuse cloud of the other. Even without repulsion, non-reciprocal alignment alone produces single-species polarized clumps. This distinguishes alignment-driven asymmetric polar clustering from reciprocal flocking and from scalar phase separation.

4. Percolation, turbulence, and misconceptions about flocking

A common misconception in the Vicsek literature is that flocking is a percolation transition. A detailed study of the 2D Vicsek model shows that the order–disorder transition is not related in any way to a percolation transition (Kyriakopoulos et al., 2019). Clusters are defined as connected components of the metric-neighbor graph, with largest-cluster mass fraction

$\mathrm{Pe}=v_0/(D_r r_c)$7

and largest-cluster extension fraction

$\mathrm{Pe}=v_0/(D_r r_c)$8

In the ordered Toner–Tu phase at $\mathrm{Pe}=v_0/(D_r r_c)$9, the percolation threshold extrapolates to $g=J/D_r$^^^^2all:polar clustering OR ti:\2^^^^, well inside the ordered region. The transition is anisotropic: transverse spanning appears at lower density than full two-dimensional spanning in finite systems, although both thresholds converge in the thermodynamic limit. Longitudinally, the measured ratios $g=J/D_r$^^^^2 OR abs:\2^^^^ and $g=J/D_r$2 are compatible with standard 2D percolation, but the correlation-length exponent is anomalous, $g=J/D_r$3. The cluster-size distribution at threshold obeys $g=J/D_r$4 with $g=J/D_r$5, while away from threshold broad pseudo-power laws always terminate in an exponential cutoff. The paper attributes the modified $g=J/D_r$6 to long-range density correlations in the Toner–Tu phase via the Harris–Weinrib criterion.

A related continuum route to polar clustering appears in active polar fluids with density-dependent motility. The extended model

$g=J/D_r$7

$g=J/D_r$8

combines a short-wavelength instability of the Dunkel model with a MIPS-like long-wavelength instability generated by $g=J/D_r$9 (Worlitzer et al., 2020). The long-wave mode is controlled by an effective diffusivity

$g$^^^^2all:polar clustering OR ti:\2^^^^

and a critical damping

$g$^^^^2 OR abs:\2^^^^

The resulting phase diagram contains D, IT, MIC/MIPS, and ITC regions. In the ITC regimes, dense nearly immobile clusters coexist with vortical dilute regions, turbulence can trigger nucleation in metastable parameter ranges, and altered Ostwald ripening produces vortices during droplet dissolution. Anomalous velocity statistics, with a strong peak at $g$2 and kurtosis $g$3, appear in all phases where the system segregates into regions of high and low densities (Worlitzer et al., 2020).

5. Dipolar quantum gases and disorderless quasi-localization

In a different physical meaning, polar clustering refers to the clustering of dipolar excitations in a one-dimensional gas of polar bosons in a deep optical lattice. The system is described by the hard-core extended Bose–Hubbard Hamiltonian

$g$4

with nearest-neighbor tunneling $g$5 and dipolar interaction scale $g$6 (&&&^^^^2 OR abs:\2^^^^&&&). For $g$7, the tightest two-body bound eigenstate is a nearest-neighbor dimer with probability $g$8, and the effective dimer hopping amplitude is

$g$9

Dimers are therefore heavy, yet they continue to interact via a $\to$^^^^2all:polar clustering OR ti:\2^^^^ tail. The critical separation for two-dimer clustering obeys

$\to$^^^^2 OR abs:\2^^^^

which expresses the balance between dimer–dimer interaction energy and the second-order kinetic scale $\to$2.

This yields disorderless quasi-localization: the Hamiltonian is translation invariant, but dimers can form clusters with fixed or very slowly fluctuating relative distances. The same mechanism generates self-bound lattice droplets in expansion protocols. Using a contrast criterion $\to$3, the paper finds

$\to$4

for droplet formation from an initially full box. Within such droplets the average density satisfies $\to$5, so mobile but confined holons persist and the droplet remains superfluid rather than crystalline. In this literature, “polar clustering” therefore designates interaction-driven clustering of dimers and droplets generated purely by long-range dipolar interactions in a clean lattice (&&&^^^^2 OR abs:\2^^^^&&&).

6. Polar-coordinate and polar-region data clustering

In statistics and machine learning, the phrase refers to clustering procedures that respect polar or circular structure. For circular data $\to$6, one proposal maps each point to the lateral surface of a cylinder,

$\to$7

unwraps it to reconstructed coordinates

$\to$8

and then periodically replicates the data along the angular direction (Sun et al., 2023). The method is designed to overcome the wrap-around and phase-ambiguity problems of Cartesian clustering and can be combined with k-means, DBSCAN, and hierarchical clustering. The central guarantee is that the correct clustering result appears within the reconstructed dataset given sufficient periodic repetitions of the data.

A second polar-coordinate formulation arises in Cryo-EM 2D classification. In a Fourier–Bessel steerable PCA basis, each coefficient is written as

$\to$9

and a planar rotation acts as

$\to$^^^^2all:polar clustering OR ti:\2^^^^

componentwise (&&&^^^^2 OR abs:\22^^^^&&&). Probabilistic PolarGMM then defines a Gaussian mixture model in the amplitude–phase variables $\to$^^^^2 OR abs:\2^^^^, couples it to discrete searches over rotations and translations, and updates cluster responsibilities by expectation maximization. On simulated 7^^^^2all:polar clustering OR ti:\2^^^^S, Bgal, and T2^^^^2all:polar clustering OR ti:\2^^^^ Cryo-EM datasets, it improves clustering metrics and alignment errors relative to EMAN2 and RELION.

A third use concerns polar-region functional data. Scalable model-based Gaussian process clustering embeds the Vecchia approximation into a GP-mixture EM algorithm, reducing dense GP costs from $\to$2 time and $\to$3 memory per cluster to $\to$4 and $\to$5 (&&&^^^^2 OR abs:\2^^^^3&&&). Applied to North Pole monthly temperature anomalies from ^^^^2 OR abs:\2^^^^9^^^^2all:polar clustering OR ti:\2 OR abs:\2^^^^–2^^^^2all:polar clustering OR ti:\222^^^^, smoothed by a 5-year moving average, the method groups the ^^^^2 OR abs:\22^^^^ monthly curves into summer months (June–August), winter months (October–February), and transition months (March, April, September). Here “polar clustering” concerns clustering data from the polar region rather than clustering in polar coordinates.

Polar-coordinate decoupling also appears in vector quantization. For a weight vector $\to$6, Polar Coordinate Decoupled Vector Quantization writes

$\to$7

and quantizes directions and magnitudes separately (&&&3^^^^2all:polar clustering OR ti:\2^^^^&&&). On LLaMA-2-7B, separately clustering the directions and magnitudes of weight vectors yields zero-shot accuracy drops of $\to$8 and $\to$9, respectively, which motivates a direction codebook and a magnitude codebook with distinct metrics. At the 2-bit level the method outperforms baseline methods by at least $\to$^^^^2all:polar clustering OR ti:\2^^^^ zero-shot accuracy.

7. Polar clusters in singularity theory and algebraic geometry

In singularity theory, polar clustering has a precise intrinsic definition. For a mini-regular holomorphic germ $\to$^^^^2 OR abs:\2^^^^, let $\to$2 be the Puiseux roots of $\to$3 and $\to$4 the Puiseux roots of $\to$5. For a polar arc $\to$6, define

$\to$7

With $\to$8 a tangent line in the tangent cone and $\to$9 the corresponding bar of the Kuo–Lu tree, the polar cluster is

$\to$^^^^2all:polar clustering OR ti:\2^^^^

Topological equivalence $\to$^^^^2 OR abs:\2^^^^ induces bijections

$\to$2

and therefore preserves coarser partitions $\to$3 and $\to$4, partial Łojasiewicz exponents, tangential Łojasiewicz exponents, and tangential Milnor numbers (Migus et al., 2021). In the Lipschitz category these clusters are refined by gradient canyons, whose degrees and contact patterns become bi-Lipschitz invariants and recover the Henry–Parusiński moduli.

A neighboring algebraic-geometric usage replaces tangent spaces by higher order osculating spaces. For a projective variety $\to$5, the higher order polar locus is

$\to$6

where $\to$7 has codimension $\to$8 (&&&^^^^2 OR abs:\2^^^^5&&&). The corresponding polar class satisfies

$\to$9

Higher order reciprocal polar loci are defined through higher order Euclidean normal bundles PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2all:polar clustering OR ti:\2all:polar clustering OR ti:\2^^^^, with

PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2all:polar clustering OR ti:\2 OR abs:\2^^^^

and for generic PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2all:polar clustering OR ti:\22^^^^ and PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2all:polar clustering OR ti:\2^^^^3,

PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2all:polar clustering OR ti:\2^^^^4

For PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2all:polar clustering OR ti:\2^^^^5-reflexive varieties, the degree of the top polar class of order PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2all:polar clustering OR ti:\2^^^^6 equals the degree of the PRESERVED_PLACEHOLDER_^^^^2 OR abs:\2all:polar clustering OR ti:\2^^^^7-th dual variety. This is a different, but structurally related, sense in which “polar” objects are grouped and organized.

Across these literatures, polar clustering consistently links a notion of polarity to a notion of aggregation. In active matter the polarity is orientational and the aggregates are moving flocks or clumps; in dipolar gases it is interaction-mediated and the aggregates are dimers or droplets; in data analysis it is coordinate geometry or polar-region structure; and in singularity theory it is the geometry of polar curves and higher order polar loci. This suggests that the phrase is best treated as a family resemblance term rather than as the name of a single theory.