Atomic-Density Field: Concepts & Applications

• Atomic-density field is a continuous representation that replaces discrete atomic details with spatially resolved density functions encoding occupancy, probability, and structure.
• It is applied in atomic physics, nuclear theory, materials modeling, biomolecular simulation, and quantum optics to extract key physical properties and predict phenomena.
• Field-based methods enable variational, statistical, and computational tractability by linking microscopic interactions to macroscopic observables such as shell structure, saturation, and grain-boundary characteristics.

The expression atomic-density field denotes a family of field-based descriptions in which explicitly discrete atomic or many-body degrees of freedom are replaced by a spatial density, probability density, or coarse-grained occupancy field. In the cited literature, the object can be the electron density of an atom, the set of local densities entering a nuclear energy density functional, a continuous atomic density function for crystalline matter, a one-dimensional projected density built from molecular dynamics trajectories, a smoothed atomic density used for machine-learning descriptors, or a mixed-state probability-density structure in cavity QED (Borgoo et al., 2011, Nakatsukasa, 2012, Kapikranian et al., 2013, Giorgino, 2013, Willatt et al., 2018, Fang et al., 2013). This suggests that there is no single canonical definition; rather, the common theme is the replacement of explicit particle-by-particle descriptions by fields that are variationally, statistically, or computationally tractable.

1. Terminology and scope

Across the literature, the same phrase refers to distinct but structurally related objects. In each case, the field is defined over space and used to encode either occupancy, probability, charge, or coarse-grained structural information.

Domain	Field object	Primary role
Atomic physics	Electron density $\rho(\mathbf r)$ or $D(r)=r^2\rho(r)$	Shell structure, correlation, relativistic effects, periodic trends
Nuclear theory	Local densities such as $\rho(\vec r)$, $\tau(\vec r)$, $\vec J(\vec r)$, $\kappa(\vec r)$	Self-consistent mean field, saturation, deformation, collective response
Materials modeling	Continuous atomic density function $\rho(\mathbf r)$	Grain-boundary structure and energy
Biomolecular simulation	One-dimensional slab-averaged density profile	Structural localization and comparison to diffraction data
Atomistic ML	Smoothed atomic density or learned electron-density field	Symmetry-adapted descriptors and DFT surrogates
Cavity QED and BECs	Quantum probability-density structure of atomic motion	Ordered spatial modes and density-dependent optical response

A recurring distinction is between a field and a count. In several materials and atomistic contexts, the density field carries both multiplicity and spacing information, whereas a coordination number or a raw atom list does not (Kamachali et al., 22 Aug 2025, Willatt et al., 2018). In nuclear and electronic structure theory, the field is not merely descriptive: it is the variational variable from which energies, forces, and response properties are obtained (Nakatsukasa, 2012, Das et al., 11 Nov 2025).

2. Atomic electron density as a field variable

In atomic structure theory, the atomic-density field is the electron density of an atom viewed as a spatial field. It is defined from the first-order reduced density matrix,

$$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$

with the spin-free density

$$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$

or equivalently

$$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$

Within the multi-configuration Hartree–Fock framework, the density inherits both configuration interaction and orbital relaxation through the one-particle density matrix $D(r)=r^2\rho(r)$0, giving

$D(r)=r^2\rho(r)$1

For open-shell atoms, this density is not automatically spherical when computed for a specific $D(r)=r^2\rho(r)$2 component; its angular content is constrained by angular momentum and parity and can contain spherical harmonics up to $D(r)=r^2\rho(r)$3 (Borgoo et al., 2011).

A central construction is the spherical density obtained by averaging over all magnetic sublevels,

$D(r)=r^2\rho(r)$4

which restores spherical symmetry. The resulting radial form is

$D(r)=r^2\rho(r)$5

Here $D(r)=r^2\rho(r)$6 is interpreted as the probability of finding an electron between $D(r)=r^2\rho(r)$7 and $D(r)=r^2\rho(r)$8, independent of direction. In this form, the density field cleanly reveals shell structure and supports comparison across correlation levels and between nonrelativistic and relativistic calculations (Borgoo et al., 2011).

The same work treats the density field as an information-bearing object rather than a mere visualization. Quantum similarity measures,

$D(r)=r^2\rho(r)$9

and Kullback–Leibler-based functionals are used to compare atoms. Plain density-based quantum similarity does not recover the periodicity of Mendeleev’s table very well, whereas KL-based measures with a noble-gas reference reveal clear periodic trends (Borgoo et al., 2011). This establishes the atomic-density field as a carrier of shell, correlation, relativistic, and periodic information.

3. Nuclear density fields and the energy-density-functional framework

In nuclear density functional theory, the atomic-density field is the local set of nuclear densities that self-consistently generate the mean field and determine observables. The motivation is empirical nuclear saturation: an almost constant density

$\rho(\vec r)$0

and an approximately constant binding energy per particle. A simple independent-particle picture with a state-independent attractive potential cannot reproduce both saturation and binding consistently. The inconsistency is exhibited by the simultaneous estimates

$\rho(\vec r)$1

and

$\rho(\vec r)$2

which cannot be made mutually consistent with a single state-independent potential (Nakatsukasa, 2012).

The paper further notes a momentum-dependent mean field,

$\rho(\vec r)$3

corresponding to an effective mass

$\rho(\vec r)$4

which is too small compared with empirical values. The conclusion is that the effective interaction must depend on the state of the nucleon, or equivalently on the nuclear medium. In Brueckner theory this appears through the $\rho(\vec r)$5-matrix,

$\rho(\vec r)$6

whose dependence on the Pauli operator $\rho(\vec r)$7 and the starting energy $\rho(\vec r)$8 is then replaced in practice by density dependence via the local density approximation and the density matrix expansion (Nakatsukasa, 2012).

The energy-density-functional statement is

$\rho(\vec r)$9

Because the effective interaction depends on density, the variational derivative generates rearrangement terms like $\tau(\vec r)$0, which are essential for getting the saturation mechanism right. In the modern formulation, the relevant variables include the normal density $\tau(\vec r)$1, kinetic density $\tau(\vec r)$2, spin-orbit density $\tau(\vec r)$3, pairing density $\tau(\vec r)$4, and, in time-dependent problems, spin density $\tau(\vec r)$5 and current density $\tau(\vec r)$6. The Hartree–Fock–Bogoliubov equation,

$\tau(\vec r)$7

yields the ground-state densities

$\tau(\vec r)$8

while the time-dependent extension

$\tau(\vec r)$9

leads, upon linearization, to the QRPA matrix equation (Nakatsukasa, 2012).

The physical significance is illustrated by photoabsorption in Nd isotopes. The ground-state deformation is obtained by minimizing the Skyrme energy functional; $\vec J(\vec r)$0 is spherical, while nuclei with $\vec J(\vec r)$1–92 gradually develop deformation. For spherical isotopes such as $\vec J(\vec r)$2 and $\vec J(\vec r)$3, the photoabsorption cross section shows a single giant dipole peak around $\vec J(\vec r)$4 MeV. As neutron number increases and the ground state becomes more deformed, the peak broadens, matching experiment (Nakatsukasa, 2012). The density field is therefore the bridge between microscopic interaction, saturation, deformation, and collective response.

4. Continuous atomic density in crystalline matter and grain boundaries

In materials modeling, the atomic-density field often appears as a continuous atomic density function (ADF). In this usage, atoms are not treated as sharp point particles but as a continuous field $\vec J(\vec r)$5, interpreted as the time-averaged probability density of finding an atom near position $\vec J(\vec r)$6. In a perfect crystal, the field forms a periodic array of sharp peaks at equilibrium atomic sites; in a liquid it becomes nearly uniform. The free energy combines nonlocal interaction and a logarithmic local entropic term, and the logarithmic form keeps $\vec J(\vec r)$7 in the interval $\vec J(\vec r)$8, unlike older polynomial PFC free energies that permit negative density values (Kapikranian et al., 2013).

For bcc iron, the interaction kernel is fitted to the structure factor $\vec J(\vec r)$9 of liquid iron at the melting point. The reported fit uses

$\kappa(\vec r)$0

leading to

$\kappa(\vec r)$1

Symmetrical [100] and [110] tilt grain boundaries in bcc iron are then generated by crystallizing a liquid layer between two grains. The method yields well-localized atomic peaks for a wide family of [110] grain boundaries, including the special cusps at $\kappa(\vec r)$2 $\kappa(\vec r)$3 and $\kappa(\vec r)$4 $\kappa(\vec r)$5, and the ADF energy curve reproduces both cusps very well. Atomic coordinates extracted from the density peaks and relaxed by molecular dynamics with an embedded-atom method potential for Fe are reported to be “quite close” to the equilibrium MD structures (Kapikranian et al., 2013).

A related but distinct usage treats the atomic-density field as a systematically coarse-grained descriptor of grain-boundary disorder. The field is defined as a Gaussian smoothing of the atomic number density,

$\kappa(\vec r)$6

with $\kappa(\vec r)$7 in bulk-like regions and $\kappa(\vec r)$8 near a grain boundary. For the BCC-Fe study, the coarse-graining radius is

$\kappa(\vec r)$9

The study analyzes 408 distinct BCC-Fe grain boundaries relaxed by molecular statics at $\rho(\mathbf r)$0 K and shows that many boundaries have $\rho(\mathbf r)$1 but still have nonzero density deficit $\rho(\mathbf r)$2. This is the central distinction between density and coordination: the density field carries both bond depletion and spacing variation, whereas coordination alone is insensitive to spacing (Kamachali et al., 22 Aug 2025).

The same work reports that the excess free volume per unit area satisfies

$\rho(\mathbf r)$3

with an almost perfect linear trend, while $\rho(\mathbf r)$4 versus coordination deficit is scattered and weakly correlated. Grain-boundary energy also increases systematically with the density deficit, and a density-based entropy $\rho(\mathbf r)$5 is approximately equal to a shell-descriptor entropy

$\rho(\mathbf r)$6

with a near-unity slope in direct comparison (Kamachali et al., 22 Aug 2025). The field thus acts as a compact surrogate for topology, spacing, excess volume, energy, and configurational disorder.

5. Projected and machine-learned density fields

In atomistic simulation analysis, an atomic-density field may be a projected histogram field. The Density Profile Tool for VMD computes one-dimensional density profiles by binning atoms into slabs of thickness $\rho(\mathbf r)$7 along a chosen axis. With slab indicator

$\rho(\mathbf r)$8

the density in bin $\rho(\mathbf r)$9 is

$$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$0

Choosing $$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$1, $$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$2, $$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$3, $$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$4, or $$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$5 yields number, mass, charge, electron, or neutral electron density. The method assumes point-like atoms and is aimed especially at planar systems such as lipid bilayers, where mass and electron densities provide membrane thickness, location of structural elements, and comparison to X-ray diffraction experiments (Giorgino, 2013).

In machine learning for chemistry and materials, descriptors are frequently built from a smoothed atomic density. In the bra-ket formalism,

$$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$6

each atom contributes a localized bump $$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$7 together with a chemical-species ket $$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$8. Symmetry invariance is then obtained by group averaging, and tensor products before symmetrization encode body-order correlations. In this construction, the $$\gamma_1(\mathbf x_1,\mathbf x_1') = N \int \Psi(\mathbf x_1,\mathbf x_2,\ldots,\mathbf x_N)\, \Psi^*(\mathbf x_1',\mathbf x_2,\ldots,\mathbf x_N)\, d\mathbf x_2\cdots d\mathbf x_N,$$9 rotationally invariant representation is explicitly equivalent to the standard SOAP power spectrum, while $$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$0 yields the bispectrum (Willatt et al., 2018). The essential object is therefore not a basis choice but the correlation structure of the smoothed density field.

A more recent development elevates the field variable from descriptor to variational output. In field theoretic atomistics (FTA), the primary inputs are the external potential field $$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$1 and a smeared nuclear field $$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$2, while the electron-density field is recovered variationally from the learned energy map through the Hohenberg–Kohn relation

$$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$3

The framework also introduces a thermodynamically well-behaved total-energy saddle-point formulation, and atomic forces are obtained as functional-derivative outputs rather than as separately learned labels. The model uses atom-centered probe functions to sample the fields and combines an atomic cluster expansion with neural networks. On the rMD17 aspirin dataset, the field-trained model achieved approximately $$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$4 monopole, $$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$5 dipole, and $$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$6 quadrupole errors for $$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$7, whereas omitting field losses led to errors two orders of magnitude larger (Das et al., 11 Nov 2025). In this setting, the atomic-density field is a learned, atom-centered representation of the electron-density field tied by exact variational identities to external potential and total energy.

6. Quantum optical probability densities, dense vapors, and terminological limits

In cavity QED, the atomic-density field is not a classical density profile but a quantum probability-density structure. For a pumped atom in a dissipative strong-coupling cavity, the atomic probability density organizes into two ordered spatial modes—an odd mode with peaks only on odd lattice sites and an even mode with peaks only on even lattice sites—together with a residual mode coupled to vacuum. The steady state is described by the mixed-state density operator

$$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$8

This goes beyond the transient pure-state picture

$$\rho(\mathbf r_1)=\int \gamma_1(\mathbf x_1,\mathbf x_1)\,d\sigma_1,$$9

which is reasonable only at very short times before dissipation becomes important. The ordered-mode weights increase with pumping, while the residual mode decreases, and the density-operator model reproduces the reduced cavity state with fitting errors ranging from $$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$0 to $$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$1 in the reported examples (Fang et al., 2013).

A related usage appears in superradiant Rayleigh scattering from a Bose–Einstein condensate with a slightly distorted density envelope. There, a very small longitudinal density distortion produces strongly asymmetric scattering-mode populations and abnormal asymmetric amplification: the optical field in the diluter region of the atomic cloud is more greatly amplified. The condensate density is modeled as a linearly distorted Thomas–Fermi profile, and the asymmetry is traced to density-dependent gain and nonlocal end-fire-mode buildup rather than to simple local-density arguments (Wang et al., 2013).

In dense rubidium vapor, the relevant field is the atomic number density of the medium. Selective-reflection experiments on the $$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$2 D2 line at densities

$$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$3

show that increasing density transforms self-broadening from inhomogeneous to homogeneous. At the maximal density, a strong pump field splits the spectrum into two symmetrical resonances, interpreted in the dressed-state picture, with

$$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$4

Here the density field is not an atom-resolved spatial profile but a medium parameter governing optical broadening and nonlinear spectral structure (Sautenkov et al., 16 Oct 2025).

The term also has an unrelated mathematical usage that should not be conflated with any spatial field. In arithmetical congruence monoids, the atomic density is the limiting proportion of atoms in

$$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$5

with exact value

$$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$6

while in numerical semigroup algebras the atomic density is the asymptotic proportion of irreducible elements by degree, and for every $$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$7 it is $$\rho(\mathbf r)= \left\langle \Psi \left| \sum_{i=1}^N \delta(\mathbf r-\mathbf r_i)\right| \Psi \right\rangle, \qquad \int \rho(\mathbf r)\,d\mathbf r = N.$$8 (Olsson et al., 2023, Antoniou et al., 2020). These works concern asymptotic factorization theory, not spatial density fields.

The broad literature therefore supports a precise but plural conclusion: an atomic-density field is any field variable that replaces explicit atomic constituents by a density-like object whose spatial, variational, or statistical structure carries the physically relevant information. Depending on context, that information may encode shell structure, nuclear saturation, grain-boundary geometry, projected molecular organization, symmetry-adapted atomistic descriptors, cavity-induced ordering, or density-controlled optical response.