Isomorph: Invariance in Physics, Logic & Supply Chains

• Isomorph is a multifaceted concept denoting invariance under structured transformations across condensed-matter physics, formal semantics, and supply-chain simulations.
• It captures constant excess entropy and reduced-unit dynamics, enabling simplified universal scaling in liquids, crystals, and high-pressure materials.
• In computer science and logistics, it facilitates data-type refinements and state-space conservation in mechanized program transformation and digital twin modeling.

In contemporary research, isomorph denotes several distinct but technically related ideas. In condensed-matter physics, it names a curve in a thermodynamic phase diagram along which structure and dynamics are invariant when expressed in reduced units, a central concept of isomorph theory for Roskilde-simple systems (Schrøder et al., 2014). In logic and theoretical computer science, it appears in formal treatments of isomorphism, including forcing-based notions of “probably isomorphic” structures, set-theoretic semantics for dependent type theory, and mechanized data-type transformations in ACL2 (Farah et al., 2 Jul 2025, McAllester, 2014, Coglio et al., 2020). More recently, ISOMORPH has also been adopted as the name of a supply-chain digital twin for simulation, dataset generation, and forecasting benchmarks (Zhang et al., 12 May 2026).

1. Isomorphs in condensed-matter theory

In isomorph theory, an isomorph is a curve in the phase diagram $(\rho,T)$ consisting of state points with constant excess entropy $S_{\rm ex}\equiv S-S_{\rm ideal}$ (Veldhorst et al., 2013). Two state points $(\rho_1,T_1)$ and $(\rho_2,T_2)$ are defined as isomorphic if for physically relevant pairs of microconfigurations with coordinates that match when scaled by density,

$$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$

the Boltzmann factors are proportional,

$$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$

A Roskilde-simple system is defined by the property that the order of the potential energies of configurations at one density is maintained when these are scaled uniformly to a different density (Schrøder et al., 2014).

The underlying physical statement is hidden scale invariance. A convenient representation is

$$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$

where $h(\rho)$ sets the density-dependent energy scale and $g(\rho)$ is an additive term that does not affect forces, structure, or dynamics (Friedeheim et al., 2018). This implies that, along curves satisfying $h(\rho)/T=\mathrm{const}$, the canonical configurational distributions are the same in reduced units. Standard reduced units are

$S_{\rm ex}\equiv S-S_{\rm ideal}$0

with corresponding reduced transport quantities such as $S_{\rm ex}\equiv S-S_{\rm ideal}$1 and $S_{\rm ex}\equiv S-S_{\rm ideal}$2 (Veldhorst et al., 2013).

A practical characterization uses equilibrium virial–potential-energy correlations. The Pearson correlation coefficient is

$S_{\rm ex}\equiv S-S_{\rm ideal}$3

and the density-scaling exponent is

$S_{\rm ex}\equiv S-S_{\rm ideal}$4

Roskilde-simple systems typically satisfy $S_{\rm ex}\equiv S-S_{\rm ideal}$5, and along an isomorph one has $S_{\rm ex}\equiv S-S_{\rm ideal}$6 (Schrøder et al., 2010). A recurrent misconception is that isomorphs require exact inverse-power-law interactions. The Buckingham-liquid study showed instead that strong correlations and isomorphs do not depend critically on the mathematical form of the repulsion being an inverse power law (Veldhorst et al., 2011).

2. Invariants, response functions, and experimental tests

A central strength of isomorph theory is that it yields explicit invariants for measurable response functions. For dielectric spectroscopy of isotropic molecular liquids, one can combine linear response, the fluctuation-dissipation theorem, and the isomorph condition $S_{\rm ex}\equiv S-S_{\rm ideal}$7 to obtain four equivalent isomorph-invariant quantities:

$S_{\rm ex}\equiv S-S_{\rm ideal}$8

where $S_{\rm ex}\equiv S-S_{\rm ideal}$9 is the electric susceptibility and $(\rho_1,T_1)$0 is the Kirkwood–Frölich factor product (Xiao et al., 2014).

The explicit experimental prediction tested for the van der Waals liquid 5-phenyl-4-ether (5PPE) was

$(\rho_1,T_1)$1

using the experimentally justified approximation $(\rho_1,T_1)$2 (Xiao et al., 2014). The liquid was studied at isochronal states in the temperature range $(\rho_1,T_1)$3 K and pressure range $(\rho_1,T_1)$4 MPa, with relaxation times around $(\rho_1,T_1)$5 s and $(\rho_1,T_1)$6 s. The dielectric setup achieved amplitude reproducibility $(\rho_1,T_1)$7, the empty capacitance was measured as $(\rho_1,T_1)$8 pF at ambient pressure, and 42 pairs of isochronal states were identified from 155 measured state points. Using $(\rho_1,T_1)$9 from prior work and densities from a Tait equation of state, the predicted and measured loss spectra showed good match, with a relative deviation at the loss peak within $(\rho_2,T_2)$0 to $(\rho_2,T_2)$1 for the $(\rho_2,T_2)$2 prediction (Xiao et al., 2014).

Dynamic mechanical analysis yields an analogous invariance statement for viscoelastic response. In simulations of the Kob–Andersen binary Lennard-Jones system, the standard reduced modulus is

$(\rho_2,T_2)$3

with reduced frequency

$(\rho_2,T_2)$4

Using SLLOD with time-dependent strain rates to impose sinusoidal shear, reduced loss-modulus curves $(\rho_2,T_2)$5 were found to collapse along isomorphic state points when plotted against the unscaled temperatures, provided the isomorphic temperatures were chosen so that reduced forces remain invariant (Moch et al., 2022). The study considered densities $(\rho_2,T_2)$6, $(\rho_2,T_2)$7, $(\rho_2,T_2)$8, and $(\rho_2,T_2)$9, a reference angular frequency $$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$0, and strain amplitude $$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$1. A force-based construction using temperature-matched configurations remained applicable at $$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$2, whereas a quenched-configuration variant broke down for the largest density rescalings because of force-vector decorrelation (Moch et al., 2022).

3. Scope across liquids, polymers, crystals, plasmas, and confinement

The theory was first developed for strongly correlating liquids and generalized Lennard-Jones systems, for which isomorphs were shown to be curves along which structure, dynamics, and some thermodynamic properties are invariant in reduced units (Schrøder et al., 2010). For generalized Lennard-Jones pair potentials with exponents $$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$3 and $$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$4, the density-scaling function has the form

$$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$5

and in the standard $$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$6 case all isomorphs can be scaled onto a single curve in the $$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$7 plane (Schrøder et al., 2010). This already implied that melting and freezing lines are closely tied to isomorphs on the liquid side.

Subsequent work broadened the class of model systems. A binary Buckingham liquid, despite its exponential repulsion, was shown to be strongly correlating and to possess isomorphs; reduced radial distribution functions and both incoherent and coherent intermediate scattering functions were approximately invariant, and the viscous-state dynamics were closely mimicked by a purely repulsive inverse-power-law reference system with exponent $$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$8 (Veldhorst et al., 2011). Flexible Lennard-Jones chains likewise exhibited isomorphs: segmental and chain-center-of-mass incoherent intermediate scattering functions, end-to-end vector autocorrelation functions, most Rouse-mode correlators, and mean-square displacements collapsed across density changes up to about $$\rho_1^{1/3} \mathbf{R}_1 = \rho_2^{1/3} \mathbf{R}_2,$$9 in reduced units, while jumps between isomorphic state points produced instantaneous equilibration without slow relaxation (Veldhorst et al., 2013).

The concept also extends to crystals. Simulations of face-centered-cubic Lennard-Jones crystals showed that reduced radial distribution functions, velocity autocorrelation functions, phonon dynamics, and even slow vacancy-jump dynamics are approximately invariant along isomorphs (Albrechtsen et al., 2014). Other crystalline systems with isomorphs included the Wahnström binary Lennard-Jones crystal with the $$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$0 Laves structure, monatomic FCC Buckingham crystals, a purely repulsive finite-separation model, and an ortho-terphenyl molecular crystal. In contrast, a NaCl crystal model and SPC/E hexagonal ice did not exhibit isomorph invariances, supporting the broader conjecture that crystalline solids with isomorphs include most or all formed by atoms or molecules interacting via metallic or van der Waals forces, whereas covalently- or hydrogen-bonded crystals are not expected to have isomorphs (Albrechtsen et al., 2014).

High-pressure metallic crystals provided a further extension. Molecular-dynamics simulations using effective medium theory for Au, Ni, Cu, Pd, Ag, and Pt found strong hidden scale invariance at condensed-state densities: reduced-unit radial distribution functions collapsed almost perfectly along isomorphs, reduced velocity autocorrelation functions and vibrational density-of-states proxies showed good collapse, and jumps between isomorphic points led to instantaneous equilibration (Friedeheim et al., 2018). A notable difference from simple Lennard-Jones behavior is that $$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$1 varies substantially with density. For the Au crystal isomorph, as density increased from $$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$2 to $$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$3 g/cm$$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$4 and pressure from $$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$5 to $$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$6 GPa, $$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$7 decreased from $$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$8 to $$\exp(-U(\mathbf{R}_1)/k_B T_1)=C_{1,2}\exp(-U(\mathbf{R}_2)/k_B T_2).$$9, while $$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$0 increased from $$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$1 to $$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$2 (Friedeheim et al., 2018).

Soft-matter and plasma systems also fall within the scope of isomorph theory when virial–potential-energy correlations are strong. For the Yukawa fluid, $$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$3 at all simulated state points, and isomorphs identified by both direct isomorph check and an analytical construction displayed invariance of the reduced radial distribution function, static structure factor, mean-square displacement, and incoherent intermediate scattering function (Veldhorst et al., 2015). The analytically derived form

$$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$4

reproduces the known melting-line shape, which the theory interprets as an isomorph (Veldhorst et al., 2015).

Confinement does not automatically destroy isomorphs. In slit-pore simulations with crystalline walls, both the single-component Lennard-Jones liquid and the Kob–Andersen binary Lennard-Jones mixture retained good isomorph behavior for pore widths $$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$5: reduced density profiles parallel and perpendicular to the walls, reduced mean-square displacements, and higher-order structures from topological cluster classification were nearly invariant along confined isomorphs (Carter et al., 2021). The breakdown at $$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$6, where $$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$7, is a useful counterexample to the assumption that reduced-unit invariance is automatic in strongly inhomogeneous environments (Carter et al., 2021).

4. Nonequilibrium, shear, and topological extensions

Isomorph theory is not restricted to equilibrium. For Couette shear flows generated by the SLLOD equations of motion, the reduced equations are identical along an isomorph provided the reduced strain rate

$$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$8

is held fixed (Separdar et al., 2012). Under this condition, simulations of both the single-component Lennard-Jones liquid and the Kob–Andersen binary Lennard-Jones mixture showed collapse of the reduced radial distribution function, the transverse self-intermediate scattering function, and the reduced viscosity

$$U(\mathbf{R}) \approx h(\rho)\,\Phi(\rho^{1/3}\mathbf{R}) + g(\rho),$$9

as a function of reduced strain rate, in both linear and shear-thinning regimes (Separdar et al., 2012).

A more general nonequilibrium reformulation introduces the systemic temperature $h(\rho)$0, defined as the temperature of the equilibrium state point with average potential energy equal to $h(\rho)$1 (Dyre, 2020). Systemic isomorphs are lines of constant excess entropy in the phase diagram defined by density and systemic temperature, and the reduced dynamics is invariant along a systemic isomorph if there is a constant ratio between the systemic and the bath temperature. In thermal equilibrium, $h(\rho)$2 and the original formalism is recovered (Dyre, 2020). This framework rationalizes earlier observations of isomorph invariance in nonlinear steady-state shear flows, zero-temperature plastic flows, and glass-state isomorphs.

A different extension uses topological information rather than thermodynamic excess entropy as the scaling variable. For soft-sphere fluids with repulsive $h(\rho)$3 Mie potentials, the Shannon entropy of the Voronoi-cell-topology distribution,

$h(\rho)$4

was shown to provide a scaling law for reduced transport properties comparable to conventional excess-entropy scaling (Yoon et al., 2019). Across $h(\rho)$5, the Voronoi excess entropy and thermodynamic excess entropy were almost linearly related, with $h(\rho)$6 between $h(\rho)$7 and $h(\rho)$8. The work further suggested that the Frenkel line is a topological isomorphic line, marked by $h(\rho)$9 and $g(\rho)$0, where the functional form of the scaling relation changes qualitatively (Yoon et al., 2019).

At the same time, higher-order structure can expose the limits of approximate isomorph invariance. In the Kob–Andersen Lennard-Jones glassformer and its mapped inverse-power-law reference system, two-point structure and dynamics were nearly identical, but topological cluster classification showed that bicapped square antiprisms, the locally favored 11A structures, had populations up to $g(\rho)$1 higher in the Lennard-Jones system and lifetimes up to $g(\rho)$2 higher than in the inverse-power-law reference system (Malins et al., 2013). The structural relaxation times were almost identical, while the four-point dynamical susceptibility was marginally higher in the inverse-power-law system. This indicates that higher-order structural observables need not be as tightly constrained by isomorph theory as two-point reduced-unit observables (Malins et al., 2013).

5. Isomorphism in logic, type theory, and program transformation

Outside condensed matter, the term appears in formal semantics and theorem proving. In a set-theoretic formulation of dependent type theory, types are divided into small and large types—sets and proper classes respectively—and each proper class, such as “group” or “topological space,” has an associated notion of isomorphism (McAllester, 2014). Isomorphism is handled by defining a groupoid structure on the space of all definable values. The values are simultaneously objects and morphisms—“morphoids”—which supports sound inference rules for deriving isomorphisms and for substitution of isomorphics (McAllester, 2014).

A forcing-theoretic generalization was introduced under the name probably isomorphic. Two structures $g(\rho)$3 in the same language are probably isomorphic if they, or in the metric case their completions, are isomorphic after forcing with the Lebesgue measure algebra (Farah et al., 2 Jul 2025). For discrete structures, or extremal models of a non-degenerate simplicial theory, the paper proved the equivalence

$g(\rho)$4

thereby linking forcing-based isomorphism to randomization structures in continuous logic (Farah et al., 2 Jul 2025).

In mechanized program derivation, isomorphism is treated operationally. In ACL2, “types” are represented by predicates old and new, and an isomorphism is a pair of total ACL2 functions iso and osi that are inverse bijections when restricted to those domains (Coglio et al., 2020). The APT tools implement this through defiso, isodata, and propagate-iso. Once versions of the interface functions of a data type have been derived on the isomorphic representation, higher-level functions can be generated by substitution, and the tools automatically produce proofs of equivalence (Coglio et al., 2020). The paper gives examples ranging from refinement of finite sets to duplicate-free ordered lists or bit vectors to record extensions that cache derived fields.

6. ISOMORPH as a supply-chain digital twin

In a distinct and acronymic usage, ISOMORPH names a public, open-source digital twin of a multi-echelon supply-chain logistics network (Zhang et al., 12 May 2026). The simulator advances a directed routing graph in discrete time: demand arrives at the destination, is served from stock or recorded as backlog, and triggers replenishment through the network. The state vector

$g(\rho)$5

tracks on-hand inventory, destination backlog, outstanding orders, in-transit shipments, and a smoothed demand estimate, closing the dynamics as a Markov chain on a hybrid state space (Zhang et al., 12 May 2026).

The released benchmark includes catalogue sizes $g(\rho)$6 and $g(\rho)$7, a horizon $g(\rho)$8, six one-at-a-time scenario sweeps producing 30 additional rollouts, and 20 Latin-hypercube perturbations over demand-side parameters (Zhang et al., 12 May 2026). The demand process is

$g(\rho)$9

with $h(\rho)/T=\mathrm{const}$0 composed of yearly and weekly seasonality, clipped AR(1) drift, per-item bursts, and shared macro-shocks. Replenishment follows $h(\rho)/T=\mathrm{const}$1 rules, routing uses Dijkstra weights $h(\rho)/T=\mathrm{const}$2, and dispatch to the destination is governed by

$h(\rho)/T=\mathrm{const}$3

with pipeline multiplier $h(\rho)/T=\mathrm{const}$4 (Zhang et al., 12 May 2026).

The system was designed to encode three pathwise conservation laws, including per-node mass conservation

$h(\rho)/T=\mathrm{const}$5

and global internal-network mass conservation

$h(\rho)/T=\mathrm{const}$6

which serve as verification tools for simulator extensions (Zhang et al., 12 May 2026). The released data reproduces the bullwhip effect at empirically consistent magnitudes: tier-level monthly means on $h(\rho)/T=\mathrm{const}$7 fall in $h(\rho)/T=\mathrm{const}$8, within the $h(\rho)/T=\mathrm{const}$9 interval spanning the $S_{\rm ex}\equiv S-S_{\rm ideal}$00th–$S_{\rm ex}\equiv S-S_{\rm ideal}$01th percentiles reported in the literature (Zhang et al., 12 May 2026).

The benchmark was also used for zero-shot forecasting with Chronos-T5, Moirai-1.1-R, TimesFM-2.0, and Lag-Llama. Using context length $S_{\rm ex}\equiv S-S_{\rm ideal}$02, horizons $S_{\rm ex}\equiv S-S_{\rm ideal}$03, and MASE relative to a seasonal-naive baseline, TimesFM achieved $S_{\rm ex}\equiv S-S_{\rm ideal}$04 and $S_{\rm ex}\equiv S-S_{\rm ideal}$05 on the $S_{\rm ex}\equiv S-S_{\rm ideal}$06 baseline, while Lag-Llama yielded $S_{\rm ex}\equiv S-S_{\rm ideal}$07 and $S_{\rm ex}\equiv S-S_{\rm ideal}$08 (Zhang et al., 12 May 2026). The same digital twin and Latin-hypercube perturbations were used to generate forward uncertainty-quantification bands, illustrating a role for foundation models as fast surrogates for forward UQ under parameter uncertainty (Zhang et al., 12 May 2026).

7. Conceptual unity and points of divergence

Across these domains, the common theme is not a single formalism but a recurring principle: a complicated object or process is organized by an invariance under a structured transformation. In condensed matter, the relevant transformation is uniform scaling in density and temperature, and the main content of the theory is reduced-unit invariance along configurational adiabats (Schrøder et al., 2014). In logic and computer science, the transformation is relabeling, transport, or representation change under a bijection, with correctness expressed by commuting diagrams or equivalence theorems (McAllester, 2014, Coglio et al., 2020). In the supply-chain digital twin, the name functions as an acronym rather than a direct statement of mathematical invariance, although the simulator itself is organized around an explicit state-space and conservation structure (Zhang et al., 12 May 2026).

Several boundary conditions recur. In physics, isomorph invariance is strongest for Roskilde-simple liquids, many van der Waals systems, and many metallic crystals, but may fail for hydrogen-bonded, ionic, network-forming, or ultra-confined systems (Xiao et al., 2014, Albrechtsen et al., 2014, Carter et al., 2021). Power-law density scaling is only an approximation to the more general $S_{\rm ex}\equiv S-S_{\rm ideal}$09 scaling and breaks down across sufficiently large density ranges (Veldhorst et al., 2013). Higher-order structural observables may vary even when two-point reduced-unit observables collapse well (Malins et al., 2013). In formal settings, by contrast, isomorphism is exact once the relevant domains and transport maps are specified, but the scope of admissible constructions is determined by the logic or theorem-proving environment (Farah et al., 2 Jul 2025, Coglio et al., 2020).

The term isomorph therefore names a family of ideas rather than a single object. Its most developed physical usage is the one codified by isomorph theory: curves of constant excess entropy along which reduced structure and dynamics are invariant. Its broader technical use retains the same structural intuition—equivalence under a transformation that preserves the relations of interest—even when the transformation is a forcing extension, a data-type refinement, or a configurable logistics simulator.