---
title: Covariant Energy Density Functionals
url: https://www.emergentmind.com/topics/covariant-energy-density-functionals-cedfs
type: topic
---

# Covariant Energy Density Functionals

Searching arXiv for the cited CEDF papers to ground the article in the specified literature.
Covariant energy density functionals (CEDFs) are the functional realizations of covariant density functional theory (CDFT), a relativistically consistent framework for the microscopic description of bulk and spectral nuclear properties. In this approach, nuclei are described by Dirac nucleons moving in self-consistent scalar and vector mean fields generated either by effective meson exchange or by point-coupling interactions. CEDFs are used for the global description of binding energies, charge radii, deformations, neutron skins, drip lines, fission barriers, and dense-matter equations of state, and they have been benchmarked across large portions of the nuclear chart with explicit assessments of systematic and statistical uncertainties [1508.05526] [2111.04782].

## 1. Covariant formulation and mean-field structure

The defining feature of a CEDF is Lorentz covariance. In the meson-exchange formulation, the basic Lagrangian density couples the nucleon Dirac field $\psi$ to the isoscalar-scalar $\sigma$, isoscalar-vector $\omega_\mu$, isovector-vector $\rho_\mu$, and electromagnetic $A_\mu$ fields. A representative form is
\[
\mathcal{L}
=
\bar\psi
\Bigl[
\gamma^\mu
\bigl(
i\partial_\mu
-
g_\omega \omega_\mu
-
g_\rho \vec\tau\!\cdot\!\vec\rho_\mu
-
e A_\mu (1-\tau_3)/2
\bigr)
-
(m+g_\sigma \sigma)
\Bigr]\psi
+\cdots
\]
with the corresponding meson and photon kinetic and mass terms. In nonlinear meson models such as NL3*, the density dependence is encoded through the $\sigma$ self-interaction
\[
U(\sigma)=\frac12 m_\sigma^2\sigma^2+\frac13 g_2\sigma^3+\frac14 g_3\sigma^4,
\]
whereas in density-dependent meson-exchange models and in density-dependent point-coupling models the couplings are explicit functions of the baryon or vector density, so that no explicit $U(\sigma)$ is required [1508.05526] [1604.07296].

Variation of the functional yields the Dirac equation for single-particle spinors,
\[
\Bigl[
-\,i\boldsymbol\alpha\!\cdot\!\nabla
+
\beta\,(m+S(\mathbf r))
+
V(\mathbf r)
\Bigr]\psi_k(\mathbf r)
=
\varepsilon_k\psi_k(\mathbf r),
\]
with scalar and vector self-energies generated by the covariant fields. In the Hartree approximation, the scalar field is typically $S(\mathbf r)=g_\sigma \sigma(\mathbf r)$, while the time-like vector field contains $\omega$, $\rho$, and Coulomb contributions. The total energy functional in the static mean-field approximation is built from the occupied Dirac orbitals plus meson-field, Coulomb, and, when the couplings are density-dependent, rearrangement contributions [1508.05526] [2111.04782].

Point-coupling CEDFs eliminate explicit mesons in favor of local contact interactions and gradient terms. A representative interaction part is
\[
\mathcal{L}_{\rm int}
=
-\frac12\alpha_S(\rho_v)(\bar\psi\psi)^2
-\frac12\alpha_V(\rho_v)(\bar\psi\gamma_\mu\psi)(\bar\psi\gamma^\mu\psi)
-\frac12\alpha_{TV}(\rho_v)(\bar\psi\gamma_\mu\vec\tau\psi)\!\cdot\!(\bar\psi\gamma^\mu\vec\tau\psi)
-\delta_S\,\nabla_\mu(\bar\psi\psi)\nabla^\mu(\bar\psi\psi)+\cdots .
\]
This representation is central to DD-PC1, PC-PK1, and later neural-network or reduced-basis developments built on point-coupling RMF structure [2111.04782] [2606.30326] [2406.01747].

Pairing correlations are usually treated within the relativistic Hartree-Bogoliubov (RHB) framework. In the global surveys and many later developments, the particle-hole channel is defined by the CEDF and the particle-particle channel by a finite-range Gogny-type force or a finite-range separable pairing force tuned to reproduce pairing gaps [2111.04782] [1404.4901].

## 2. Major families and representative parametrizations

Three major families of CEDFs are in widespread use: nonlinear meson models (NL), density-dependent meson-exchange models (DD-ME), and point-coupling models (PC). These families differ mainly in how they encode density dependence and finite-range effects, but all retain the covariant scalar-vector structure of the mean field [2111.04782].

| Family | Representative functionals | Characteristic features |
|---|---|---|
| NL | NL3*, NL5(C,D,E) | Nonlinear $\sigma^3+\sigma^4$ self-interactions |
| DD-ME | DD-ME2, DD-ME$\delta$ | Density-dependent meson-nucleon couplings |
| PC | DD-PC1, PC-PK1 | Zero-range contact terms plus gradients |

NL3* is a nonlinear meson-coupling functional fitted to bulk properties of 12 spherical nuclei and selected nuclear-matter properties; in another formulation it is described as fitted to binding energies, charge radii, neutron skins of selected nuclei, and nuclear incompressibility. DD-ME2 and DD-ME$\delta$ are density-dependent meson-exchange functionals calibrated to binding energies and charge radii, with DD-ME$\delta$ also incorporating pairing information and an isovector-scalar $\delta$ channel. DD-PC1 and PC-PK1 are point-coupling functionals fitted, respectively, to deformed and spherical nuclear data sets, with DD-PC1 emphasizing binding energies and deformations of 64 nuclei and PC-PK1 including charge radii and spin-orbit splittings [1508.05526] [2111.04782] [1604.07296].

The density dependence of DD-ME2 is commonly written as
\[
g_i(\rho)=g_i(\rho_{\rm sat})\,a_i\,\frac{1+b_i(x+d_i)^2}{1+c_i(x+d_i)^2},
\qquad x=\rho/\rho_{\rm sat},
\]
with typical values $g_\sigma(\rho_{\rm sat})=10.5396$, $g_\omega(\rho_{\rm sat})=13.0189$, and $g_\rho(\rho_{\rm sat})=3.6836$ in one summary and $3.8052$ in another parameter listing tied to a different source presentation. DD-PC1 is characterized by density-dependent four-fermion couplings in the scalar, vector, and isovector-vector channels and associated gradient terms [2111.04782] [1604.07296] [2011.13368].

A standard point of comparison among major CEDFs is the set of nuclear-matter saturation properties. One frequently quoted compilation gives, for NL3*, DD-ME2, DD-ME$\delta$, DD-PC1, and PC-PK1, respectively:
- $\rho_0 = 0.150, 0.152, 0.152, 0.152, 0.154\ {\rm fm}^{-3}$,
- $E/A = -16.31, -16.14, -16.12, -16.06, -16.12\ {\rm MeV}$,
- $K_\infty = 258, 251, 219, 230, 238\ {\rm MeV}$,
- $J = 38.68, 32.40, 32.35, 33.00, 35.6\ {\rm MeV}$,
- $L = 122.6, 49.4, 52.9, 68.4, 113\ {\rm MeV}$,
- $m^*/m = 0.67, 0.66, 0.61, 0.66, 0.65$.
In that assessment, NL3* and PC-PK1 violate the empirical “SET2b” constraints in $J$ and $L$ [1508.05526].

## 3. Calibration targets, observables, and uncertainty measures

CEDFs are calibrated to selected combinations of finite-nucleus observables and nuclear-matter properties. The data entering the fit may include binding energies, charge radii, neutron skins, deformations, spin-orbit splittings, pairing information, and pseudo-data for symmetric matter, pure neutron matter, or other infinite-matter constraints. This heterogeneous calibration practice is central to later discussions of why similarly performing functionals can have different nuclear-matter characteristics, and conversely why functionals close to empirical nuclear-matter windows need not perform best for finite nuclei [1508.05526] [1604.07296].

For ground-state studies, the standard observables are the binding energy $E_B(Z,N)$, the charge radius
\[
R_{\rm ch}=\sqrt{\langle r^2\rangle_{\rm charge}},
\]
and the quadrupole deformation
\[
\beta_2=\frac{4\pi}{5}\frac{Q_2}{A R_0^2},
\qquad
R_0=1.2\,A^{1/3}\ {\rm fm}.
\]
For charge-radii systematics one often starts from the point-proton radius and includes finite-size and Darwin-Foldy corrections,
\[
r_{\rm ch}(Z,N)=\sqrt{\langle r^2\rangle_p + R_p^2 + (N/Z)R_n^2 + \frac{3\hbar^2}{4m^2}},
\]
with the compact approximation
\[
r_{\rm ch}=\sqrt{\langle r^2\rangle_p + 0.64\ {\rm fm}^2}.
\]
Differential radii along isotopic chains are defined by
\[
\delta\langle r^2\rangle_{\rm ch}^{N,N'}
=
r_{\rm ch}^2(N)-r_{\rm ch}^2(N'),
\]
and odd-even staggering in charge radii by the three-point indicator
\[
\Delta r^{(3)}(Z,N)
=
(-1)^N
\bigl[
r_{\rm ch}(Z,N+1)-2r_{\rm ch}(Z,N)+r_{\rm ch}(Z,N-1)
\bigr].
\]
These definitions are central in the global charge-radius literature within CDFT [2111.04782].

For mass systematics, the global rms deviation for a set of $N$ even-even nuclei is
\[
\Delta E_{\rm rms}
=
\sqrt{
\frac{1}{N}
\sum_{i=1}^{N}
\bigl(E_{\rm calc}(i)-E_{\rm exp}(i)\bigr)^2
}.
\]
Theoretical systematic uncertainty in the “global-spread” approach is quantified at each $(Z,N)$ by
\[
\Delta E(Z,N)=|E_{\max}(Z,N)-E_{\min}(Z,N)|,
\]
that is, the spread among representative functionals. Statistical uncertainty within a single parametrization is propagated from the covariance matrix,
\[
(\delta O)^2
=
\sum_{i,j}
\frac{\partial O}{\partial p_i}
\,C_{ij}\,
\frac{\partial O}{\partial p_j}.
\]
The same distinction between systematic spreads and covariance-based statistical errors is used in charge-radius assessments, where the spread over NL3*, DD-ME2, DD-ME$\delta$, and DD-PC1 is taken as a systematic error estimate [1508.05526] [2111.04782].

## 4. Global performance across the nuclear chart

Large-scale axial RHB calculations with modern CEDFs have established a consistent but not uniform level of accuracy across the nuclear landscape. In one broad survey of even-even nuclei with $Z\le 104$ between the two-proton and two-neutron drip lines, the rms mass deviations across 835 nuclei were reported as $3.00$ MeV for NL3*, $2.45$ MeV for DD-ME2, $2.40$ MeV for DD-ME$\delta$, and $2.15$ MeV for DD-PC1 [1404.4901]. In a closely related later assessment over approximately 640 even-even nuclei, the ordering was DD-PC1 ($2.01$ MeV), DD-ME$\delta$ ($2.29$ MeV), DD-ME2 ($2.39$ MeV), NL3* ($2.96$ MeV), with PC-PK1 quoted at $2.58$ MeV in the saturation-property comparison [1508.05526] [1604.07296].

Systematic spreads in binding energies are modest near stability and large at the neutron-rich edge. In the 2015 assessment, $\Delta E(Z,N)$ is about $3$ MeV in the valley of stability and can grow up to about $15$ MeV toward the neutron drip line [1508.05526]. In the 2014 global survey, mass spreads increase from roughly $3$–$6$ MeV near stability to more than $15$ MeV near the two-neutron drip line [1404.4901]. These results establish that extrapolation uncertainty is not a peripheral issue but a structural feature of present-day CEDFs.

Charge radii are described more accurately and with smaller functional spreads than masses. Across the nuclear chart, NL3*, DD-ME2, DD-ME$\delta$, DD-PC1, PC-PK1, and NL5(C–E) achieve
\[
\Delta r_{\rm ch}^{\rm rms}\simeq 0.023\text{--}0.033\ {\rm fm}
\]
with a mean of about $0.028$ fm, corresponding to about $0.6\%$ on a typical $4.8$ fm charge radius. Systematic spreads in predicted absolute radii are $\lesssim 0.01$ fm for medium-heavy and heavy nuclei and grow to about $0.02$ fm for $Z\le 40$ [2111.04782]. In earlier global surveys, rms deviations for 351 measured even-even charge radii were $0.04$ fm for NL3*, $0.037$ fm for DD-ME2, $0.041$ fm for DD-ME$\delta$, and $0.040$ fm for DD-PC1, with improved values of $0.025$–$0.033$ fm if light He and anomalous Cm isotopes were excluded [1404.4901].

Two-particle separation energies, drip lines, and deformation maps reveal where the functional dependence is most consequential. Reported rms deviations are about $1.05$–$1.23$ MeV for $S_{2n}$ and about $0.95$–$1.29$ MeV for $S_{2p}$, with the proton side generally better reproduced than the neutron side [1404.4901]. The two-proton drip line predicted by DD-ME2 and DD-ME$\delta$ agrees with experiment within two neutrons for $Z\le 86$, whereas the two-neutron drip line exhibits much larger model spreads, often by tens of neutrons between major shell closures. Shell gaps at $N=126$, $184$, and $258$ remain robust in all EDFs examined in that survey [1404.4901].

For deformations, the spreads $\Delta\beta_2(Z,N)$ are $\lesssim 0.01$ in spherical and well-deformed regions but can exceed $0.3$ at shape-coexistence boundaries [1404.4901]. In actinides and superheavy nuclei, systematic studies with DD-PC1, DD-ME2, NL3*, and PC-PK1 identify spherical shell closures at $Z=120$, $N=184$, and $N=258$ as major sources of uncertainty in ground-state deformations and fission barriers. Theoretical uncertainties in barrier heights are moderate in well-deformed actinides but can peak around $4$ MeV at $Z\approx 110$, $N\approx 164$, and along $N\approx 184$; near $N\approx 240$, NL3* and PC-PK1 give barriers below $2$ MeV while DD-ME2 and DD-PC1 stay above $4$ MeV [2011.13368].

## 5. Correlations, shell structure, and beyond-mean-field physics

A central result of the CEDF literature is that there is no strong one-to-one correlation between global finite-nucleus performance and any single nuclear-matter parameter such as $J$, $L$, or $K_\infty$. A representative example compares DD-ME2 and DD-PC1: despite similar $J$ values ($32.40$ versus $33.00$ MeV), their mass predictions differ by up to about $5$ MeV over large portions of the chart. Conversely, DD-ME2 and DD-ME$\delta$ have nearly identical $(J,L)$ values and display $\Delta E(Z,N)<1.5$ MeV for about half of all nuclei up to $Z=104$. The conclusion drawn in the global assessments is that differences in shell structure and in the finite-nucleus fit play a role as large as bulk nuclear-matter parameters [1508.05526] [1604.07296].

This conclusion directly addresses a recurrent misconception: strict enforcement of empirical nuclear-matter windows is neither necessary nor sufficient for optimal finite-nucleus performance. The 2016 analysis states that functionals that come close to satisfying current nuclear-matter constraints can have problems in the description of existing nuclear data, whereas functionals carefully fitted to finite nuclei but violating some nuclear-matter constraints may perform better for binding energies and other ground- and excited-state properties. The stated reason is that finite nuclei depend not only on bulk nuclear matter but also on shell effects and surface properties [1604.07296].

Charge-radius systematics sharpen this point further. CEDF calculations reproduce absolute and differential charge radii globally, but outliers appear in regions of shape coexistence and in light nuclei with soft surfaces, where static mean field neglects beyond-mean-field correlations [2111.04782]. In deformed actinides and light superheavy nuclei, inclusion of octupole deformation increases $\delta\langle r^2\rangle$ by about $0.1\ {\rm fm}^2$ in low-$N$ Th and U isotopes, improving agreement with laser-spectroscopy data [2111.04782]. For odd-even staggering, recent CEDF studies identify two additional mechanisms beyond the traditional DFT picture: level-ordering changes induced by particle-vibration coupling (PVC), and fragmentation of the blocked state. In the cited schematic model, replacing
\[
r_{\rm ch}(N+1)\to r_{\rm ch}(N+1)-(1-S)\,[r_{\rm ch}(N+1)-r_{\rm ch}(N)]
\]
boosts $\Delta r^{(3)}$ by roughly $1/S$, bringing the predicted staggering close to experiment for $S\approx 0.7$–$0.9$ [2111.04782].

Self-consistency is also essential for differential radii. A specific example from the Pb isotopes shows that filling the $\nu 1i_{11/2}$ subshell from $^{208}$Pb to $^{220}$Pb increases the rms radius of the $n=1$ proton orbitals by about $0.21$ fm, but only by about $0.03$ fm for $n=2$ orbitals. This orbital-dependent polarization cannot be captured with a frozen proton core or a uniform radius formula $R=1.2A^{1/3}$, and it is identified as crucial for reproducing the magnitude of radius kinks at shell closures [2111.04782].

Beyond-mean-field quadrupole dynamics substantially improves mass systematics. A global 5DCH study with PC-PK1 for 575 even-even nuclei found dynamic correlation energies associated with rotational and quadrupole vibrational motion ranging from about $0$ to $4.4$ MeV, mostly between $2.0$ and $3.5$ MeV. After including these correlations, the rms mass deviation was reduced to $1.14$ MeV, smaller than the mean-field values quoted for NL3* ($2.96$ MeV), DD-ME2 ($2.39$ MeV), DD-ME$\delta$ ($2.29$ MeV), and DD-PC1 ($2.01$ MeV); the rms deviation for two-nucleon separation energies was reduced by about $34\%$ relative to a cranking prescription [1502.06908].

Functional extensions at the Hartree-Bogoliubov level have followed the same logic. A 2022 study including tensor terms in the vector-isoscalar channel reported improvements in RMS binding energies, spin-orbit splittings, and shell gaps across the chart, including deformed nuclei. In infinite matter, the Dirac mass increased from $0.57$ for DD-ME2 to $0.63$ with tensor terms, while the binding-energy RMS over 828 nuclei improved from about $2.9$ MeV for DD-ME2 to $2.1$ MeV without tensor and $1.9$ MeV with tensor; the spin-orbit RMS changed from $0.70$ MeV to $0.59$ MeV and then $0.52$ MeV, and the shell-gap RMS from $1.47$ MeV to $1.35$ MeV and then $1.12$ MeV [2210.11142].

## 6. Numerical optimization, emulation, astrophysical refinement, and emerging directions

The post-2020 CEDF literature has focused heavily on calibration methodology and numerical control. A major advance is the anchor-based optimization approach for global fits to approximately 2000 even-even masses, in which the objective function augments the usual least-squares term with quadratic anchor penalties on selected parameters:
\[
\chi^2(\mathbf p)
=
\sum_i w_i\bigl(O_i^{\rm th}(\mathbf p)-O_i^{\rm exp}\bigr)^2
+
\sum_j \lambda_j (p_j-p_j^0)^2.
\]
With a derivative-based minimizer and anchors taken from established functionals or microscopic expectations, the number of full mass-table evaluations is reduced from about $10^3$ to about $10$, so the total cost becomes roughly $\sim 10\times$ a single global RHB run [2511.00327].

These optimization protocols are closely tied to the control of basis-truncation and atomic-to-nuclear mass-conversion errors. An empirical infinite-basis extrapolation,
\[
E(N_{\rm cut})=E_\infty + A\,e^{-\alpha N_{\rm cut}},
\]
using energies computed at three values of $N_{\rm cut}$, reduces the fermionic basis-truncation error from about $1$ MeV at $N_{\rm cut}=20$ to $\lesssim 10$ keV across the known chart [2511.00327]. A subsequent study incorporated both fermionic and bosonic infinite-basis corrections and total electron binding energies in the fit protocol. It defined pseudodata as
\[
B^{\rm pseudo}(Z,N)
=
B^{\rm AME}(Z,N)-\bigl[B_{\rm el}(Z)+\Delta B^{F\infty}(Z,N)\bigr],
\]
and found that neglect of these corrections had induced a global calculation error of order $0.8$ MeV or higher for the three major classes of CEDFs. For the new Z-type functionals, the reported rms deviations for 882 even-even nuclei are $2.357$ MeV for NL5(Z), $1.901$ MeV for PC-Z, and $1.601$ MeV for DD-MEZ, with negligible numerical errors of order $20$–$50$ keV; DD-MEZ reaches $1.557\pm 0.028$ MeV in a complementary summary [2507.17082] [2511.00327].

Reduced-order and optimized-basis methods address the same bottleneck from a different angle. A universal reduced basis for the single-particle Dirac equation, built by proper orthogonal decomposition of high-fidelity snapshots, reproduces bound levels with energy errors $\delta E\lesssim 10^{-4}$ in dimensionless units using reduced bases of size about $10$–$12$ per $\kappa$ channel. Direct reduced-basis diagonalization yields a speed-up of order $\sim 150\times$ relative to full Runge-Kutta integration, and linearization of Woods-Saxon-like potentials pushes the speed-up to order $10^4$ [2406.01747]. In point-coupling calculations with harmonic-oscillator bases, a globally optimized frequency scaling
\[
f_{\rm opt}(A)\approx 1.394 + 0.00161\,A,
\qquad
\hbar\omega_0 = f_{\rm opt}(A)\,41\,A^{-1/3}\ {\rm MeV},
\]
combined with empirical formulas for the minimal shell cutoff, reduces the necessary fermionic space by about $15$–$20$ shells for $\Delta B\approx 30$ keV accuracy and cuts CPU times by about $50$–$70\%$ [2605.30669].

Machine-learning extensions have begun to treat the energy density itself as the learning target. A 2026 point-coupling study used a physics-guaranteed neural network whose inputs at each radial point are
\[
\{\rho_S,\rho_V,\rho_{TV},\Delta\rho_S,\Delta\rho_V,\Delta\rho_{TV}\},
\]
and whose output is the local energy density. When trained on a correction to an existing CEDF, with
\[
\delta\varepsilon(r)=\mathcal E_{\rm NN}(r)-\mathcal E_{\rm CEDF}(r),
\qquad
E_{\rm total}=E_{\rm CEDF}+\int d^3r\,\delta\varepsilon(r),
\]
the approach improves the binding-energy accuracy from $644$ keV to $86$ keV in the known region and retains extrapolation accuracy of about $5$ MeV up to $30$ steps beyond known nuclei. The study reports that shell effects in two-neutron separation energies are effectively captured [2606.30326].

CEDFs also continue to be refined from the dense-matter side. Bayesian refinement of FSUGold2 and FSUGarnet with pulsar masses, NICER radii, tidal deformabilities, and $\chi$EFT neutron-matter constraints leads to posterior values such as $L=57.20(7)$ MeV for refined FSUGold2 and $L=55.79(6)$ MeV for refined FSUGarnet, with corresponding maximum neutron-star masses $2.30\pm 0.06\,M_\odot$ and $2.09\pm 0.05\,M_\odot$ [2301.09692]. At the same time, the PREX/CREX tension remains unresolved within standard isovector sectors; the cited work states that CEDFs with only two isovector couplings lie on a nearly linear $R_{\rm skin}^{48}$–$R_{\rm skin}^{208}$ correlation and cannot simultaneously reproduce the reported skins of both nuclei [2301.09692]. A later Bayesian study of density-dependent coupling parametrizations finds that current modeling requires freedom in the isoscalar channel at least up to the skewness coefficient $Q_{\rm sat}$ and in the isovector channel at least up to the curvature coefficient $K_{\rm sym}$ in order to capture supra-saturation variations in the equation of state and symmetry energy [2512.01503].

Taken together, these developments suggest a convergent picture rather than a settled one. Present CEDFs provide global mean-field mass accuracies at the $1.5$–$2$ MeV level, charge-radius accuracies at the few-$10^{-2}$ fm level, and controlled uncertainty estimates across the chart, but the isovector channel, shell-structure deficiencies, and the consistent inclusion of beyond-mean-field correlations remain the principal open problems. The explicit research directions identified in recent work are broader calibration datasets, quantified covariance analyses, explicit beyond-mean-field treatments such as 5DCH or GCM during the fit, improved spectroscopic quality through tensor or $\delta$-meson channels, and more flexible density dependencies or higher-order terms that better decouple isoscalar, isovector, and shell-structure aspects of the functional [1508.05526] [2511.00327]

Source: https://www.emergentmind.com/topics/covariant-energy-density-functionals-cedfs