Density Scaling Laws: Fundamentals & Applications

• Density scaling laws are relations that express observables as functions of density or density-weighted fields, reducing multi-parameter dependencies into master curves or segmented power laws.
• These laws span a wide range of applications from glass-forming liquids and polymers to granular packings, turbulence, and network data, providing a unified framework for diverse phenomena.
• Methodologies employ single and segmented power laws, isomorph functions, and dimensional analyses to capture scaling behaviors and transitions in both experimental and simulated systems.

Density scaling laws are relations in which observables collapse when expressed as functions of density, inverse density, or density-weighted fields rather than as independent functions of multiple control parameters. In the cited literature, “density” denotes specific volume inverse in glass-forming liquids, packing fraction in dense particulate systems, mass density in fluids and self-gravitating matter, population density in settlement systems, and dataset density in machine learning corpora. The common mathematical theme is reduction to a master curve or a piecewise power law, but the physical content ranges from hidden scale invariance and isomorph theory to jamming constraints, inverse cascades, hydrodynamic self-similarity, and breakpoint structure in heterogeneous spatial systems (Fragiadakis et al., 2010, Xia et al., 2017, Sutton et al., 12 Sep 2025).

1. Canonical forms of density scaling

A basic form is the single power law

$D_i(\rho)=a\,\rho^b,$

where $D_i(\rho)$ is an indicator density, $a$ is a pre-exponential constant, and $b$ is the scaling exponent. In the rural–urban setting, $b<1$ is sublinear, $b=1$ linear, and $b>1$ superlinear. The same literature also uses a segmented law with a breakpoint $\rho_c$,

$$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$

which distinguishes rural and urban regimes (Sutton et al., 12 Sep 2025).

In viscous liquids and polymers, the canonical variable is not $\rho$ itself but the combined density–temperature scaling variable

$D_i(\rho)$0

with $D_i(\rho)$1 the specific volume. Relaxation time, viscosity, and diffusion are then written as

$D_i(\rho)$2

For strongly correlating liquids, a more general form replaces the power law by an isomorph function,

$D_i(\rho)$3

so that density enters through a nontrivial function $D_i(\rho)$4 rather than a constant exponent $D_i(\rho)$5 (Fragiadakis et al., 2010, Bøhling et al., 2011).

In disordered granular packing, density scaling is local and geometric. The average distance of the first $D_i(\rho)$6 neighbors is

$D_i(\rho)$7

and the packing fraction obeys

$D_i(\rho)$8

Here the exponent is shell-dependent and reaches a minimum at the first-shell boundary (Xia et al., 2017).

In compressible MHD turbulence and compressible turbulence more broadly, density appears as a weighting factor. For solar-wind Elsässer fields,

$D_i(\rho)$9

and the third-order law is written in terms of $a$0 rather than $a$1. In compressible turbulence spectra, the density-weighted velocity $a$2 plays an analogous role (Carbone et al., 2010, Sun, 2015).

Stationary driven fluids admit yet another master-curve representation. For uniaxial compressible Navier–Stokes–Fourier flows, the density field can be written as

$a$3

and in the separable case the exponents are $a$4, $a$5, $a$6, with $a$7 the kinetic field (Hurtado et al., 15 Dec 2025).

2. Liquids, supercooled states, and glass-transition scaling

Density scaling in glass-forming liquids and polymers is the empirical observation that structural relaxation times, viscosities, and diffusion constants measured over a range of temperatures and pressures superpose when plotted against $a$8. This has been demonstrated for roughly a hundred materials. The theoretical rationale is the inverse power-law approximation

$a$9

for which the scaling exponent is

$b$0

However, the strict derivation applies to reduced quantities, not raw experimental observables. The reduced dynamical variables are

$b$1

Only the exponent obtained from reduced quantities, $b$2, can be sensibly related to the intermolecular potential. In the deeply supercooled regime the difference between reduced and unreduced scaling is small, but above the melting point it can be substantial; in dodecane, for example, unreduced viscosity gives $b$3, whereas reduced viscosity gives $b$4 (Fragiadakis et al., 2010).

Power-law density scaling is not exact over arbitrarily wide density ranges. In the Kob–Andersen binary Lennard–Jones mixture, a single exponent collapses any two isochores well but fails to collapse the three isochores $b$5, $b$6, and $b$7 simultaneously. Isomorph theory replaces

$b$8

by

$b$9

For a standard 12–6 Lennard–Jones potential,

$b<1$0

This generalization preserves single-variable collapse while allowing the local effective exponent $b<1$1 to vary with density (Bøhling et al., 2011).

Volumetric data can be incorporated into the same framework. In Lennard–Jones-based model systems, the same exponent that scales dynamics also scales $b<1$2–$b<1$3 data through an equation of state derived from the short-range effective potential $b<1$4 weak background, with $b<1$5. In the KABLJ model, $b<1$6 from raw $b<1$7, $b<1$8, and $b<1$9; in the Lewis–Wahnström OTP model, $b=1$0 and $b=1$1 (Grzybowski et al., 2011).

Generalized equations of state make the density dependence explicit through a one-parameter function $b=1$2. For KABLJ,

$b=1$3

Fitting volumetric data gives $b=1$4 from the configurational EOS and $b=1$5 from the total-pressure EOS, and the same $b=1$6 collapses all eleven isotherms of reduced structural relaxation times when plotted against $b=1$7 (Grzybowski et al., 2013).

Packing fraction enters analytically in generalized mode-coupling theory through the reduced control parameter

$b=1$8

Near the ideal glass transition, GMCT yields two divergent time scales,

$b=1$9

with

$b>1$0

For the Percus–Yevick hard-sphere system under MF closures, the critical packing fractions are $b>1$1, $b>1$2, and $b>1$3 for $b>1$4, $b>1$5, and $b>1$6, respectively, while the exponents shift systematically as higher-order correlations are included (Luo et al., 2020).

3. Granular packing, jamming, and non-cubic density–distance scaling

In disordered granular packing of spherical particles, X-ray tomography identifies a non-cubic scaling law between packing fraction and characteristic nearest-neighbor distance,

$b>1$7

More precisely, for the average distance $b>1$8 of the first $b>1$9 neighbors,

$\rho_c$0

and the minimum exponent occurs at $\rho_c$1, corresponding to the first-shell boundary. Over packings with global packing fraction $\rho_c$2,

$\rho_c$3

The first-shell boundary is located at $\rho_c$4 (Xia et al., 2017).

The experimental system consists of $\rho_c$5 glass beads with $\rho_c$6 polydispersity, packed in a cylindrical cell by tapping, hopper deposition, and flow-pulse protocols. Synchrotron X-ray CT at $\rho_c$7 voxel resolution yields particle centroids and radii with $\rho_c$8 diameter precision, and the analysis is restricted to beads at least $\rho_c$9 diameters from walls, amounting to $$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$0 beads per scan. Packing fraction is defined as

$$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$1

These definitions matter because the observed exponent is extracted from a strictly local neighbor statistic rather than from an assumed continuum field (Xia et al., 2017).

The non-cubic exponent is traced to the internal structure of the first neighbor shell. Within $$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$2, the six closest neighbors change distance very little with $$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$3, corresponding to local exponents $$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$4, whereas neighbors $$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$5 shrink strongly with $$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$6, corresponding to $$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$7. The measured $$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$8 is the weighted average of these two groups. Conditional on the quasi-contact number $$D_i(\rho)= \begin{cases} a_1\,\rho^{b_1}, & \rho<\rho_c,\ a_2\,\rho^{b_2}, & \rho\ge \rho_c, \end{cases}$$9, defined as neighbors whose surface separation is $\rho$0, the local law is

$\rho$1

with $\rho$2 increasing from $\rho$3 at $\rho$4 toward $\rho$5 for smaller $\rho$6. The decomposition

$\rho$7

shows that more than $\rho$8 of the deviation from $\rho$9 arises from the $D_i(\rho)$00-dependence alone (Xia et al., 2017).

The physical interpretation is therefore jamming-related. The exponent $D_i(\rho)$01 emerges close to the isostatic, marginally jammed state $D_i(\rho)$02, while as $D_i(\rho)$03 decreases and $D_i(\rho)$04 drops below $D_i(\rho)$05, the local exponent rises toward $D_i(\rho)$06 and the global exponent rises from $D_i(\rho)$07 to $D_i(\rho)$08. Metallic glasses report $D_i(\rho)$09–$D_i(\rho)$10 under density changes, and these values had been attributed to “fractal” medium-range order. The granular results show that no true fractal is required: the non-cubic law is fully explained by local non-affine displacements of contact versus non-contact neighbors and their isostatic count. Friction, bead stiffness, thermal effects, and departures from the hard-sphere jamming point shift $D_i(\rho)$11 and therefore modify $D_i(\rho)$12 (Xia et al., 2017).

4. Density weighting in turbulence and driven fluids

In fast polar solar wind, the incompressible MHD analogue of the Kolmogorov–Yaglom law is

$D_i(\rho)$13

where $D_i(\rho)$14. To incorporate compressible fluctuations, a phenomenological density-weighted form introduces

$D_i(\rho)$15

and the corresponding third-order law becomes

$D_i(\rho)$16

The Ulysses analysis uses 8-minute averages in the first half of 1996, 11-day sliding windows, and the Taylor hypothesis $D_i(\rho)$17. The compressible law is observed in $D_i(\rho)$18 of the time for $D_i(\rho)$19 and $D_i(\rho)$20 for $D_i(\rho)$21, whereas the incompressible law appears in only $D_i(\rho)$22 of the time. Despite density fluctuations of only $D_i(\rho)$23, the compressible cascade rate is an order of magnitude larger than the incompressible one and is comparable to the heating rate required to explain the non-adiabatic solar wind; the compressible derivation, however, remains purely phenomenological and lacks an exact proof in compressible MHD (Carbone et al., 2010).

Dimensional analysis gives analogous density-weighted scaling in compressible turbulence. The ordinary kinetic-energy spectrum takes the form

$D_i(\rho)$24

while the density-weighted spectrum satisfies

$D_i(\rho)$25

with $D_i(\rho)$26. Using Barenblatt incomplete similarity,

$D_i(\rho)$27

As $D_i(\rho)$28, $D_i(\rho)$29, and $D_i(\rho)$30, the effective slope tends to $D_i(\rho)$31. The density-weighted velocity $D_i(\rho)$32 is introduced because it more robustly follows the $D_i(\rho)$33 law across a wide range of Mach numbers (Sun, 2015).

Stationary compressible Navier–Stokes–Fourier flows admit a different form of density scaling. In uniaxial steady states, pressure $D_i(\rho)$34, shear stress $D_i(\rho)$35, and heat current $D_i(\rho)$36 are constants, and elimination of temperature yields

$D_i(\rho)$37

Inverting $D_i(\rho)$38 gives

$D_i(\rho)$39

while spatial scaling of the kinetic field yields

$D_i(\rho)$40

For density–temperature separable constitutive laws, this simplifies to

$D_i(\rho)$41

with $D_i(\rho)$42, $D_i(\rho)$43, $D_i(\rho)$44. Large-scale molecular dynamics simulations of 2D hard disks and 3D Lennard–Jones fluids show excellent data collapse in the bulk after discarding boundary layers of order one cell wide (Hurtado et al., 15 Dec 2025).

Hydrodynamic self-similarity also governs ATI-formed plasma channels for laser wakefield accelerators. After the ionization pulse, the over-pressured plasma drives a cylindrical shock, and the late-time density profile depends only on

$D_i(\rho)$45

For parabolic channels matched to a Gaussian driver, the on-axis density scales linearly with initial gas density,

$D_i(\rho)$46

with fitted relations $D_i(\rho)$47 for $D_i(\rho)$48 and $D_i(\rho)$49 for He. The matched spot radius obeys

$D_i(\rho)$50

with fit constants $D_i(\rho)$51 for nitrogen and $D_i(\rho)$52 for helium. Simulation profiles at $D_i(\rho)$53, $D_i(\rho)$54, and $D_i(\rho)$55 collapse onto a single curve over $D_i(\rho)$56–$D_i(\rho)$57 and $D_i(\rho)$58–$D_i(\rho)$59, with deviations from ideal exponents $D_i(\rho)$60 (Zhang et al., 15 Aug 2025).

5. Gravitational, network, and settlement-scale density laws

A cascade-based theory of dark matter proposes an inverse kinetic-energy cascade with constant rate

$D_i(\rho)$61

Dimensional analysis then gives the two-thirds law

$D_i(\rho)$62

which, combined with the virial theorem, yields the five-thirds law for enclosed mass and the four-thirds law for mean density,

$D_i(\rho)$63

At the scale radius $D_i(\rho)$64,

$D_i(\rho)$65

For fully virialized haloes with vanishing radial flow, the asymptotic inner slope is

$D_i(\rho)$66

Rotation-curve fits from SPARC, DMS, and SOFUE yield a best-fit slope $D_i(\rho)$67, but the same framework states explicitly that nonzero radial flow or non-steady accretion can steepen or flatten $D_i(\rho)$68 relative to $D_i(\rho)$69 (Xu, 2022).

Finite-density scaling laws also govern condensation in zero-range processes on scale-free networks. With $D_i(\rho)$70 particles and hopping rate $D_i(\rho)$71, condensation occurs for

$D_i(\rho)$72

In the condensed regime,

$D_i(\rho)$73

and the average occupation takes the unified form

$D_i(\rho)$74

where $D_i(\rho)$75 for $D_i(\rho)$76 and $D_i(\rho)$77 for $D_i(\rho)$78. Relaxation is hierarchical, with

$D_i(\rho)$79

for the inverse participation ratio. Monte Carlo simulations on Barabási–Albert networks with $D_i(\rho)$80, $D_i(\rho)$81, and $D_i(\rho)$82 validate both the steady-state collapse and the transient scaling (Su et al., 2017).

Population density introduces a segmented scaling paradigm in rural–urban systems. For Middle Layer Super Output Areas in England and Wales, $D_i(\rho)$83, 117 indicators are converted to per-hectare densities and fitted in log-space by OLS. Segmented models are compared to single power laws using Davies’ test, AIC, and BIC, and 92 of the 117 indicators exhibit a significant breakpoint at

$D_i(\rho)$84

Typical exponents have median rural $D_i(\rho)$85 and urban $D_i(\rho)$86 for accelerating phenomena, while some mortality indicators show urban exponents $D_i(\rho)$87–$D_i(\rho)$88. Crime, property transactions, road accidents, and mortality display distinct rural-to-urban transitions, and finer MSOA resolution reveals segmented behaviors not visible in coarser units. For dementia and ischaemic heart disease stratified by older age groups, $D_i(\rho)$89 remains near unity whereas $D_i(\rho)$90, which is interpreted as an urban protective effect (Sutton et al., 12 Sep 2025).

6. Universality, invariance, and breakdown

In biological continuum mechanics, density scaling appears as an invariance statement rather than as a fitted exponent. For a coupled system of incompressible Navier–Stokes fluid dynamics, nonlinear elasticity, and rigid-body mechanics, the scaling group is

$D_i(\rho)$91

with

$D_i(\rho)$92

Demanding invariance of the full PDE system forces all densities—fluid, rigid, stiff tissue, and soft tissue—to scale with the same factor $D_i(\rho)$93. In a fixed terrestrial gravitational field, $D_i(\rho)$94, so $D_i(\rho)$95, and the natural choice is $D_i(\rho)$96. The resulting terrestrial density law is therefore size-invariant: $D_i(\rho)$97 (Liu et al., 17 Feb 2025).

In LLMs, “data density” denotes redundancy in embedding space rather than a physical density. Cluster density is defined by

$D_i(\rho)$98

and the overall dataset density is

$D_i(\rho)$99

with a weighted centroid distance $a$00. High $a$01 indicates tightly packed, less diverse data. Classical scaling laws,

$a$02

are extended by a sub-optimal law,

$a$03

with $a$04 and $a$05 logistic in the over-training ratio $a$06. Over more than 400 runs spanning 20 M to 7.03 B parameters, high-density regimes $a$07 and high OTR $a$08 show sub-scaling, the loss–compute exponent $a$09 falls from $a$10 at OTR $a$11 to $a$12 by OTR $a$13, and the sub-optimal law reduces MAPE by 40–90% in dense-data regimes (Chen et al., 13 Jul 2025).

Across these literatures, observed exponents are frequently regime-dependent rather than universal. The granular $a$14 law is tied to the isostatic contact number and varies with coordination, friction, and related parameters rather than indicating a true fractal structure (Xia et al., 2017). Power-law density scaling in supercooled liquids is accurate over modest density ranges but fails over larger density variations, where $a$15 replaces $a$16 (Bøhling et al., 2011). The compressible Yaglom law in the solar wind is phenomenological rather than exact (Carbone et al., 2010). The dark-matter inner slope $a$17 requires vanishing radial flow and is modified by accretion history (Xu, 2022). Rural–urban exponents depend on spatial granularity and demographic stratification, with a consistent breakpoint emerging only after fine-grained analysis (Sutton et al., 12 Sep 2025). This suggests that density scaling laws are best understood as symmetry-based, cascade-based, or empirically stabilized effective descriptions whose validity is set by coordinate choice, constitutive assumptions, and the dynamical regime under study.