Covariance Energy Functional Overview

• Covariance energy functional is a quantitative framework in nuclear DFT that uses covariance matrices to propagate parameter uncertainties into observable predictions.
• It employs χ² minimization, Hessian inversion, and error propagation to evaluate correlations and diagnose potential model instabilities in both non-relativistic and covariant EDFs.
• The method informs EDF optimization by identifying redundant parameters and guiding experimental constraints to enhance predictive reliability across nuclear and astrophysical models.

A covariance energy functional is a quantitative framework, principally used in nuclear density functional theory (DFT), to encode and propagate the uncertainties and correlations of model parameters and predicted observables. Within the context of nuclear energy density functionals (EDFs), both non-relativistic (e.g., Skyrme-type) and covariant (relativistic) implementations utilize covariance analysis to rigorously assess statistical errors, mutually constrain physical observables, and diagnose model instabilities. The covariance matrix, derived from the inverse Hessian of the fit's quality function, governs the propagation of parameter uncertainties into physical predictions—enabling systematic evaluation of both precision and limitations. The methodology finds applications in the optimization and error diagnostics of nuclear models, spanning finite nuclei, neutron-rich systems, and neutron-star equations of state (Roca-Maza et al., 2014, Erler et al., 2012, Afanasjev, 2015).

1. Mathematical Foundations: χ² Minimization, Hessian, and Covariance

Covariance energy functionals originate from statistical optimization of EDF parameters. The central objective function is the least-squares fit:

$$\chi^2(\vec{p}) = \sum_{k=1}^{m} \frac{[O_k^{\rm theo}(\vec{p}) - O_k^{\rm ref}]^2}{[\Delta O_k^{\rm ref}]^2}$$

Minimization yields the best-fit parameter vector $\vec{p}_0$. Near the minimum, $\chi^2$ is well approximated quadratically:

$$\chi^2(\vec{p}) \simeq \chi^2(\vec{p}_0) + \frac{1}{2} \sum_{i,j} (p_i-p_{0,i}) H_{ij} (p_j-p_{0,j})$$

where $$H_{ij} = [\partial^2 \chi^2/ \partial p_i \partial p_j]_{\vec{p}_0}$$ is the Hessian (curvature) matrix. Under the Gaussian approximation, the covariance matrix of the parameters is simply the inverse of $H$:

$$\mathrm{Cov}_{ij} = \langle\Delta p_i\,\Delta p_j\rangle = (H^{-1})_{ij}$$

This structure underpins error propagation and quantification in all EDF-based modeling (Roca-Maza et al., 2014).

2. Error Propagation Formalism for Derived Observables

For any observable $B$ derived from the fitted functional, the first-order expansion leads to the error propagation rule:

$$\sigma_B^2 = \sum_{i,j} \frac{\partial B}{\partial p_i} \, \mathrm{Cov}_{ij} \, \frac{\partial B}{\partial p_j}$$

Similarly, the covariance between two observables $A$ and $B$ reads

$$\mathrm{Cov}(A,B) = \sum_{i,j} \frac{\partial A}{\partial p_i} \, \mathrm{Cov}_{ij} \, \frac{\partial B}{\partial p_j}$$

with Pearson correlation coefficient

$$\rho_{AB} = \frac{\mathrm{Cov}(A,B)}{\sigma_A\,\sigma_B}, \quad \rho_{AB} \in [-1, +1]$$

This machinery allows propagation of uncertainties in model parameters to all predicted observables—binding energies, radii, symmetry energy, neutron skin thickness, neutron star radii, etc.—and quantifies their mutual correlations (Roca-Maza et al., 2014, Erler et al., 2012, Rios et al., 2014).

3. Applications to Non-Relativistic and Covariant EDFs: Protocols and Findings

Covariance energy functional analysis is implemented in both non-relativistic Skyrme EDFs (e.g., SLy5-min) and covariant EDFs (e.g., DDME-min1). In practice:

Non-Relativistic (Skyrme):

• Parameters: $\{t_0, t_1, t_2, t_3, x_0, ..., x_3\}$
• Fit observables: binding energies, charge radii of doubly-magic nuclei, neutron-matter EOS pseudo-data, saturation properties.
• Extracted quantities: e.g., symmetry energy $J = 32.60 \pm 0.71$ MeV for SLy5-min (Roca-Maza et al., 2014).

Covariant (Relativistic):

• Parameters: $$\{m_\sigma, g_\sigma(\rho_{\rm sat}), a_\rho, ...\}$$
• Fit observables: ground-state energies, charge radii, surface thicknesses of spherical nuclei.
• Extracted quantities: e.g., $J = 33.0 \pm 1.7$ MeV, $L=55\pm16$ MeV, $K_0=261\pm23$ MeV for DDME-min1 (Roca-Maza et al., 2014).

Both frameworks employ similar error propagation and correlation diagnostics, with correlation maps revealing clustering among isoscalar, isovector, and collective observables, and cross-channel couplings specific to the chosen functional.

4. Physical and Statistical Impact of Constraint Variations

The degree to which a property is constrained in the fit strongly influences inter-observable correlations:

• Relaxing a constraint (e.g., increasing the weight of neutron-matter EOS) increases correlations between the unconstrained observable and others (e.g., correlation of neutron skin thickness $\Delta r_{np}(^{208}\mathrm{Pb})$ with $L$, $J$, neutron resonance energies).
• Imposing a tight constraint (e.g., fixing $\Delta r_{np}(^{208}\mathrm{Pb})$ with a small uncertainty) largely decouples that property from all others (correlations $\rho_{AB}\rightarrow 0$).

This leads to the general rule: relaxing a constraint opens parameter space and allows stronger linear correlations, while tightening a constraint freezes out that direction, suppressing correlations (Roca-Maza et al., 2014).

5. Diagnosis and Implications of Functional Instabilities

Covariance energy functional analysis alone does not guarantee physical viability—unphysical instabilities may remain undetected. Instabilities manifest as:

Uniform-Matter (Linear-Response) Instabilities:

• Landau parameters ($F_l, F'_l, G_l, G'_l$) must satisfy $F_l > -(2l + 1)$ for stability.
• RPA response functions $\chi^{(\alpha)}(q,\omega)$: poles at $\omega\to 0$ signal critical densities $\rho_c(q)$ for phase separation or spin/isospin polarization.
• Skyrme+tensor models may develop low-$\rho_c$ instabilities in $S=1$ channels; finite-range forces less susceptible.

Finite-Nucleus Pathologies:

• Hartree-Fock iterations in some EDFs can yield "finite-size instabilities" (e.g., separation of neutron/proton densities) or spurious spin polarization in cranked solutions.

Such instabilities correspond to nearly zero or negative curvature eigenmodes in $H$, leading to unreasonably large covariance matrix elements ($\mathrm{Cov}_{ij} \gg 1$), signaling huge parameter or observable uncertainties. Effective EDF optimization must therefore combine covariance energy functional analysis with explicit stability (linear-response) constraints, excluding unphysical response poles in all channels up to a critical density threshold (Roca-Maza et al., 2014).

6. Guidance for EDF Optimization and Predictive Confidence

Covariance analysis enables:

• Quantitative estimation of statistical errors for model parameters and all derived observables.
• Identification of redundant parameters (via large parameter correlations).
• Mapping of physical correlations among observables.
• Strategic selection of experimental or astrophysical constraints (e.g., adding neutron star radii or maximum mass dramatically tightens error bars on the isovector sector (Erler et al., 2012)).

Its proper application thus informs the balance between nuclear-matter constraints and finite-nucleus phenomena in model optimization, sets rigorous error bars for extrapolations (e.g., neutron-rich nuclei, neutron star equations of state), and shapes confidence in predictive extrapolations across the nuclear landscape (Roca-Maza et al., 2014, Erler et al., 2012, Afanasjev, 2015).

7. Summary Table: Covariance Energy Functional Workflow

Stage	Mathematical Formulation	Physical Role
χ² minimization	$\chi^2(\mathbf{p})$	Parameter optimization
Hessian calculation	$H_{ij}=\partial^2\chi^2/\partial p_i \partial p_j$	Curvature, local error structure
Covariance matrix	$\mathrm{Cov}_{ij}=(H^{-1})_{ij}$	Statistical errors, correlations
Observable propagation	$$\sigma_B^2=\sum_{i,j}(\partial B/\partial p_i)\mathrm{Cov}_{ij}(\partial B/\partial p_j)$$	Uncertainty for any $B$
Correlation coefficient	$\rho_{AB} = \mathrm{Cov}(A,B)/(\sigma_A \sigma_B)$	Mutual dependence diagnostics
Instability analysis	Linear-response, Landau/RPA	Exclusion of unphysical solutions

The covariance energy functional thus underpins contemporary nuclear DFT methodology, anchoring statistical error quantification, QOI correlations, and robust model selection for the predictive description of nuclear and astrophysical observables (Roca-Maza et al., 2014, Erler et al., 2012, Rios et al., 2014, Afanasjev, 2015).