Search

Effects of finite-temperature quasiparticle self-energy corrections to x-ray absorption spectra are investigated within the finite-temperature quasiparticle local density GW approximation up to temperatures $T$ of order the Fermi temperature. To facilitate the calculations, we parametrize the quasiparticle self-energy using low-order polynomial fits. We show that temperature-driven decrease in the electron lifetime substantially broadens the spectra in the near-edge region with increasing $T$. However, the quasiparticle shift is most strongly modified near the onset of plasmon excitations.

Though calculations based on density functional theory (DFT) are used remarkably widely in chemistry, physics, materials science, and biomolecular research and though the modern form of DFT has been studied for almost 60 years, some mathematical problems remain. For context, we provide an outline of the basic structure of DFT, then pose several questions regarding both its time-independent and time-dependent forms. Progress on any of these would aid in development of better approximate functionals and in interpretation.

We report APW+${lo}$ (augmented plane wave plus local orbital) density functional theory (DFT) calculations of molecule systems using the domain specific SIRIUS multi-functional DFT package. Compared to other packages the additional APW and FLAPW task and data parallelism and the additional eigensystem solver provided by the SIRIUS package can be exploited for performance gains in in the ground state Kohn-Sham calculation. This is in contrast with the use of SIRIUS as a library backend to some other APW+${lo}$ or FLAPW (full-potential linearized AWP) code. We benchmark the code and demonstrate performance on several magnetic molecule and metal organic framework systems. We show that the SIRIUS package in itself is capable of handling systems as large as a few hundreds of atoms in the unit cell without losing the accuracy needed for magnetic systems., Comment: 41 pages, 6 figures

The liquid-liquid phase transition in high-pressure Hydrogen is a problem of longstanding and controversy. The recent Nature paper by Cheng et al. [vol. 585, p. 217] makes a set of strong claims to the effect that all the previous density functional theory molecular dynamics (MD-DFT) and quantum Monte Carlo calculations of that transition are incorrect because of finite size effects and, in the MD-DFT case, short run times. The basis of those claims is their use of large systems and long durations for classical MD driven by a machine-learnt potential (MLP) which they developed. The straightforward test of their claims is to do MD-DFT on systems as large or larger than Cheng et al. used and for significantly longer durations than in the previous MD-DFT simulations. We have done so and find that neither diagnosis of theirs (size effects, duration limits) is correct. Instead, we find that the MLP does not drive MD in fidelity with the underlying DFT electronic structure that it is supposed to replicate. The result is that the MLP-driven MD results are artifactual, not systematically connected to the theoretical underpinning on which the MLP was trained.

The recent major modification, r$^2$SCAN, of the SCAN (strongly constrained and appropriately normed) meta-GGA exchange-correlation functional is shown to give substantially better spin-crossover electronic energies (high spin minus low spin) on a benchmark data set than the original SCAN. The deorbitalized counterpart r$^2$SCAN-L is almost as good as SCAN and much faster in periodically bounded systems. A combination strategy for balanced treatment of molecular and periodic spin-crossover therefore is recommended., Comment: 18 pages, 3 figures

A recent modification, r$^2$SCAN, of the SCAN (strongly constrained and appropriately normed) meta-GGA exchange-correlation functional mostly eliminates numerical instabilities and attendant integration grid sensitivities exhibited by SCAN. Here we show that the successful deorbitalization of SCAN to SCAN-L (SCAN with density Laplacian dependence) carries over directly to yield r$^2$SCAN-L. A major benefit is that the high iteration counts that hindered use of SCAN-L are eliminated in r$^2$SCAN-L. It therefore is a computationally much faster meta-GGA than its orbital-dependent antecedent. Validation data for molecular heats of formation, bond lengths, and vibration frequencies (G3/99X, T96-R, T82-F test sets respectively) and on lattice constants, and cohesive energies (for 55 solids) and bulk moduli (for 40 solids) are provided. In addition, we show that the over-magnetization of bcc Fe from SCAN persists in r$^2$SCAN but does not appear in r$^2$SCAN-L, just as with SCAN-L., Comment: 4 pages. For associated Supplemental Material file see http://www.qtp.ufl.edu/ofdft/research/publications.shtml

Using conceptually and procedurally consistent density functional theory (DFT) calculations with an advanced meta-GGA exchange-correlation functional in ab initio molecular dynamics simulations, we determine the insulator-metal transition (IMT) of warm dense fluid hydrogen over the pressure range 50 to 300 GPa. Inclusion of nuclear quantum effects via path-integral molecular dynamics (PIMD) sharpens the metallic transition and lowers the transition temperature relative to results from Born-Oppenheimer (BO) MD. BOMD itself gives improved agreement with experimental results compared to previous DFT predictions. Examination of the ionic pair correlation function in the context of the abrupt conductivity increase at the transition confirms a metallic transition due to the dissociation of molecular hydrogen that coincides with an abrupt band gap closure. Direct comparison of the PIMD and BOMD results clearly demonstrates an isotope effect on the IMT. Distinct from stochastic simulations, these results do not depend upon any ad hoc combination of ground-state and finite-T methodologies.

For orbital-free {\it ab initio} molecular dynamics, especially on systems in extreme thermodynamic conditions, we provide the first pseudo-potential-adapted generalized gradient approximation (GGA) functional for the non-interacting free energy. This is achieved by systematic finite-temperature extension of our recent LKT ground state non-interacting kinetic energy GGA functional (Phys. Rev. B \textbf{98}, 041111(R) (2018)). We test the performance of the new functional first via static lattice calculations on crystalline aluminum and silicon. Then we compare deuterium equation of state results against both path-integral Monte Carlo and conventional (orbital-dependent) Kohn-Sham results. The new functional, denoted LKTF, outperforms the previous best semi-local free energy functional, VT84F (Phys.\ Rev.\ B \textbf{88}, 161108(R) (2013)), and provides modestly faster simulations. We also discuss subtleties of identification of kinetic and entropic contributions to non-interacting free-energy functionals obtained by extension from ground state orbital-free kinetic energy functionals.

Recent investigations have found that the strongly constrained and appropriately normed (SCAN) meta-GGA exchange-correlation functional significantly over-magnetizes elemental Fe, Co, and Ni solids. For the paradigmatic case, bcc Fe, the error relative to experiment is $\gtrsim 20 \%$. Comparative analysis of magnetization results from SCAN and its \textit{deorbitalized} counterpart, SCAN-L, leads to identification of the source of the discrepancy. It is not from the difference between Kohn-Sham (SCAN-L) and generalized Kohn-Sham (SCAN) procedures. The key is the iso-orbital indicator $\alpha$ (the ratio of the local Pauli and Thomas-Fermi kinetic energy densities). Its \textit{deorbitalized} counterpart, $\alpha_L$, has more dispersion in both spin channels with respect to magnetization in an approximate region between 0.6 Bohr and 1.2 Bohr around an Fe nucleus. The overall effect is that the SCAN switching function evaluated with $\alpha_L$ reduces the energetic disadvantage of the down channel with respect to up compared to the original $\alpha$, which in turn reduces the magnetization. This identifies the cause of the SCAN magnetization error as insensitivity of the SCAN switching function to $\alpha$ values in the approximate range $0.5 \lesssim \alpha \lesssim 0.8$ and oversensitivity for $\alpha \gtrsim 0.8$., Comment: 5 pages, 5 figures, revised version

A procedure for removing explicit orbital dependence from meta-generalized-gradient approximation (mGGA) exchange-correlation functionals by converting them into Laplacian-dependent functionals recently was developed by us and shown to be successful in molecules. It uses an approximate kinetic energy density functional (KEDF) parametrized to Kohn-Sham results (not experimental data) on a small training set. Here we present extensive validation calculations on periodic solids that demonstrate that the same deorbitalization with the same parametrization also is successful for those extended systems. Because of the number of stringent constraints used in its construction and its recent prominence, our focus is on the SCAN meta-GGA. Coded in \textsc{vasp}, the deorbitalized version, SCAN-L, can be as much as a factor of three faster than original SCAN, a potentially significant gain for large-scale ab initio molecular dynamics.

A simple, novel, non-empirical, constraint-based orbital-free generalized gradient approximation (GGA) non-interacting kinetic energy density functional is presented along with illustrative applications. The innovation is adaptation of constraint-based construction to the essential properties of pseudo-densities from the pseudo-potentials that are essential in plane-wave-basis {\it ab initio} molecular dynamics. This contrasts with constraining to the qualitatively different Kato-cusp-condition densities. The single parameter in the new functional is calibrated by satisfying Pauli potential positivity constraints for pseudo-atom densities. In static lattice tests on simple metals and semiconductors, the new LKT functional outperforms the previous best constraint-based GGA functional, VT84F (Phys.\ Rev.\ B \textbf{88}, 161108(R) (2013)), is generally superior to a recently proposed meta-GGA, is reasonably competitive with parametrized two-point functionals, and is substantially faster., Comment: 3 figures

The density linear response function for an inhomogeneous system of electrons in equilibrium with an array of fixed ions is considered. Two routes to its evaluation for extreme conditions (e.g., warm dense matter) are considered. The first is from a recently developed short-time kinetic equation; the second is from time-dependent density functional theory (tdDFT). The result from the latter approach agrees with that from kinetic theory in the "adiabatic approximation", providing support and context for each. Both provide a connection to the phenomenological Kubo-Greenwood method for calculating transport properties. A brief proof of the van Leeuwen theorem (an essential underpinning of tdDFT) extended to the mixed states of equilibrium ensembles is given., Comment: 21 pages

Kinetic energy density functionals (KEDFs) are central to orbital-free density functional theory. Limitations on the spatial derivative dependencies of KEDFs have been claimed from differential virial theorems. We point out a central defect in the argument: the relationships are not true for an arbitrary density but hold only for the minimizing density and corresponding chemical potential. Contrary to the claims therefore, the relationships are not constraints and provide no independent information about the spatial derivative dependencies of approximate KEDFs. A simple argument also shows that validity for arbitrary $v$-representable densities is not restored by appeal to the density-potential bijection., Comment: 5 pages

We explore the simplification of widely used meta-generalized-gradient approximation (mGGA) exchange-correlation functionals to the Laplacian level of refinement by use of approximate kinetic energy density functionals (KEDFs). Such deorbitalization is motivated by the prospect of reducing computational cost while recovering a strictly Kohn-Sham local potential framework (rather than the usual generalized Kohn-Sham treatment of mGGAs). A KEDF that has been rather successful in solid simulations proves to be inadequate for deorbitalization but we produce other forms which, with parametrization to Kohn-Sham results (not experimental data) on a small training set, yield rather good results on standard molecular test sets when used to deorbitalize the meta-GGA made very simple, TPSS, and SCAN functionals. We also study the difference between high-fidelity and best-performing deorbitalizations and discuss possible implications for use in ab initio molecular dynamics simulations of complicated condensed phase systems.

In high magnetic field calculations, anisotropic Gaussian type orbital (AGTO) basis functions are capable of reconciling the competing demands of the spherically symmetric Coulombic interaction and cylindrical magnetic ($B$ field) confinement. However, the best available {\it a priori} procedure for composing highly accurate AGTO sets for atoms in a strong $B$ field [Phys.\ Rev. A {\bf 90}, 022504 (2014)] yields very large basis sets. Their size is problematical for use in any calculation with unfavorable computational cost scaling. Here we provide an alternative constructive procedure. It is based upon analysis of the underlying physics of atoms in $B$ fields that allows identification of several principles for the construction of AGTO basis sets. Aided by numerical optimization and parameter fitting, followed by fine tuning of fitting parameters, we devise formulae for generating accurate AGTO basis sets in an arbitrary $B$ field. For the hydrogen iso-electronic sequence, a set depends on $B$ field strength, nuclear charge, and upon orbital quantum numbers. For multi-electron systems, the basis set formulae also include adjustment to account for orbital occupations. Tests of the new basis sets for atoms H through C ($1 \le Z \le 6$), and ions Li$^+$, Be$^+$, and B$^+$, in a wide $B$ field range ($0 \le B \le 2000$ a.u.), show an accuracy better than a few $\mu$H for single-electron systems, and a few hundredths to a few mHs for multi-electron atoms. The relative errors are similar for different atoms and ions in a large $B$ field range, from a few to a couple of tens of millionths, thereby confirming rather uniform accuracy across the nuclear charge $Z$ and $B$ field strength values. Residual basis set errors are two to three orders of magnitude smaller than the electronic correlation energies in muti-electron atoms ...

Currently, the most common method to calculate transport properties for materials under extreme conditions is based on the phenomenological Kubo-Greenwood method. The results of an inquiry into the justification and context of that model are summarized here. Specifically, the basis for its connection to equilibrium DFT and the assumption of static ions are discussed briefly., Comment: Submitted to Contributions to Plasma Physics for the 2017 Strongly Coupled Coulombs Systems conference proceedings

As the foundation for a new computational implementation, we survey the calculation of the complex electrical conductivity tensor based on the Kubo-Greenwood (KG) formalism (J.\ Phys.\ Soc.\ Jpn. \textbf{12}, 570 (1957); Proc.\ Phys.\ Soc.\ \textbf{71}, 585 (1958)), with emphasis on derivations and technical aspects pertinent to use of projector augmented wave datasets with plane wave basis sets (Phys.\ Rev.\ B \textbf{50}, 17953 (1994)). New analytical results and a full implementation of the KG approach in an open-source Fortran 90 post-processing code for use with Quantum Espresso (J.\ Phys.\ Cond.\ Matt.\ \textbf{21}, 395502 (2009)) are presented.Named KGEC ([K]ubo [G]reenwood [E]lectronic [C]onductivity), the code calculates the full complex conductivity tensor (not just the average trace). It supports use of either the original KG formula or the popular one approximated in terms of a Dirac delta function. It provides both Gaussian and Lorentzian representations of the Dirac delta function (thoughthe Lorentzian is preferable on basic grounds). KGEC provides decomposition of the conductivity into intra- and inter-band contributions as well as degenerate state contributions. It calculates the dc conductivity tensor directly. It is MPI parallelized over k-points, bands, and plane waves, with an option to recover the plane wave processes for their use in band parallelization as well. It is designed to provide rapid convergence with respect to $\mathbf k$-point density. Examples of its use are given.

Realizing the potential for predictive density functional calculations of matter under extreme conditions depends crucially upon having an exchange-correlation (XC) free energy functional accurate over a wide range of state conditions. Unlike the ground-state case, no such functional exists. We remedy that with systematic construction of a generalized gradient approximation XC free-energy functional based on rigorous constraints, including the free energy gradient expansion. The new functional provides the correct temperature dependence in the slowly varying regime and the correct zero-T, high-T, and homogeneous electron gas limits. Its accuracy in the warm dense matter regime is attested by excellent agreement of the calculated deuterium equation of state with reference path integral Monte Carlo results at intermediate and elevated T. Pressure shifts for hot electrons in compressed static fcc Al and for low density Al demonstrate the combined magnitude of thermal and gradient effects handled well by this functional over a wide T range.

A previous analysis of scaling, bounds, and inequalities for the non-interacting functionals of thermal density functional theory is extended to the full interacting functionals. The results are obtained from analysis of the related functionals from the equilibrium statistical mechanics of thermodynamics for an inhomogeneous system. Their extension to the functionals of density functional theory is described., Comment: invited paper in honor of Andreas Savin

Approximations to the many-fermion free energy density functional that include the Thomas-Fermi (TF) form for the non-interacting part lead to singular densities for singular external potentials (e.g. attractive Coulomb). This limitation of the TF approximation is addressed here by a formal map of the exact Euler equation for the density onto an equivalent TF form characterized by a modified Kohn-Sham potential. It is shown to be a "regularized" version of the Kohn-Sham potential, tempered by convolution with a finite-temperature response function. The resulting density is non-singular, with the equilibrium properties obtained from the total free energy functional evaluated at this density. This new representation is formally exact. Approximate expressions for the regularized potential are given to leading order in a non-locality parameter and the limiting behavior at high and low temperatures is described. The non-interacting part of the free energy in this approximation is the usual Thomas-Fermi functional. These results generalize and extend to finite temperatures the ground-state regularization by Parr and Ghosh (Proc. Nat. Acad. Sci. 83, 3577 (1986)) and by Pratt, Hoffman, and Harris (J. Chem. Phys. 92, 1818 (1988)) and formally systematize the finite-temperature regularization given by the latter authors., Comment: Version 2 clarifies notation issue identified by reviewer. Arguments and conclusions are unchanged from version 1

Effects of explicit temperature dependence in the exchange-correlation (XC) free-energy functional upon calculated properties of matter in the warm dense regime are investigated. The comparison is between the KSDT finite-temperature local density approximation (TLDA) XC functional [Phys.\ Rev.\ Lett.\ \textbf{112}, 076403 (2014)] parametrized from restricted path integral Monte Carlo data on the homogeneous electron gas (HEG) and the conventional Monte Carlo parametrization ground-state LDA XC functional (Perdew-Zunger, "PZ") evaluated with $T$-dependent densities. Both Kohn-Sham (KS) and orbital-free density functional theory (OFDFT) are used, depending upon computational resource demands. Compared to the PZ functional, the KSDT functional generally lowers the direct-current (DC) electrical conductivity of low density Al, yielding improved agreement with experiment. The greatest lowering is about 15\% for T= 15 kK. Correspondingly, the KS band structure of low-density fcc Al from KSDT exhibits a clear increase in inter-band separation above the Fermi level compared to the PZ bands. In some density-temperature regimes, the Deuterium equations of state obtained from the two XC functionals exhibit pressure differences as large as 4\% and a 6\% range of differences. However, the Hydrogen principal Hugoniot is insensitive to explicit XC $T$-dependence because of cancellation between the energy and pressure-volume work difference terms in the Rankine-Hugoniot equation. Finally, the temperature at which the HEG becomes unstable is $T\geq$ 7200 K for $T$-dependent XC, a result that the ground-state XC underestimates by about 1000 K.

F.E. Harris has been a significant partner in our work on orbital-free density functional approximations for use in ab initio molecular dynamics. Here we mention briefly the essential progress on single-point functionals since our original paper (2006). Then we focus on the advantages and limitations of generalized gradient approximation (GGA) non-interacting kinetic-energy functionals. We reconsider the constraints provided by near-origin conditions in atomic-like systems and their relationship to regularized versus physical external potentials. Then we seek the best empirical GGA for the non-interacting KE for a modest-sized set of molecules with a well-defined near-origin behavior of their densities. The search is motivated by a desire for insight into GGA limitations and for a target for constraint-based development.

Smooth, highly accurate analytical representations of Fermi-Dirac (FD) integral combinations important in free-energy density functional calculations are presented. Specific forms include those that occur in the local density approximation (LDA), generalized gradient approximation (GGA), and fourth-order gradient expansion of the non-interacting free energy as well as in the LDA and second-order gradient expansion for exchange. By construction, all the representations and their derivatives of any order are continuous on the full domains of their independent variables. The same type of technique provides an analytical representation of the function inverse to the FD integral of order $1/2$. It plays an important role in physical problems related to the electron gas at finite temperature. From direct evaluation, the quality of these improved representations is shown to be substantially superior to existing ones, many of which were developed before the era of large-scale computation or early in the era.

Implementation of orbital-free free-energy functionals in the Profess code and the coupling of Profess with the Quantum Espresso code are described. The combination enables orbital-free DFT to drive ab initio molecular dynamics simulations on the same footing (algorithms, thermostats, convergence parameters, etc.) as for Kohn-Sham (KS) DFT. All the non-interacting free-energy functionals implemented are single-point: the local density approximation (LDA; also known as finite-T Thomas-Fermi, ftTF), the second-order gradient approximation (SGA or finite-T gradient-corrected TF), and our recently introduced finite-T generalized gradient approximations (ftGGA). Elimination of the KS orbital bottleneck via orbital-free methodology enables high-T simulations on ordinary computers, whereas those simulations would be costly or even prohibitively time-consuming for KS molecular dynamics (MD) on very high-performance computer systems. Example MD simulations on H over a temperature range 2,000 K <= T <=4,000,000 K are reported, with timings on small clusters (16-128 cores) and even laptops. With respect to KS-driven calculations, the orbital-free calculations are between a few times through a few hundreds of times faster.

An accurate analytical parametrization for the exchange-correlation free energy of the homogeneous electron gas, including interpolation for partial spin-polarization, is derived via thermodynamic analysis of recent restricted path integral Monte-Carlo (RPIMC) data. This parametrization constitutes the local spin density approximation (LSDA) for the exchange-correlation functional in density functional theory. The new finite-temperature LSDA reproduces the RPIMC data well, satisfies the correct high-density and low- and high-$T$ asymptotic limits, and is well-behaved beyond the range of the RPIMC data, suggestive of broad utility.

We report the first wholly non-empirical generalized gradient approximation, non-interacting free energy functional for orbital-free density functional theory and use that new functional to provide forces for finite-temperature molecular dynamics simulations in the warm dense matter (WDM) regime The new functional provides good-to-excellent agreement with reference Kohn-Sham calculations under WDM conditions at a minuscule fraction of the computational cost of corresponding orbital-based simulations.

We compare the behavior of the finite-temperature Hartree-Fock model with that of thermal density functional theory using both ground-state and temperature-dependent approximate exchange functionals. The test system is bcc Li in the temperature-density regime of warm dense matter (WDM). In this exchange-only case, there are significant qualitative differences in results from the three approaches. Those differences may be important for Born-Oppenheimer molecular dynamics studies of WDM with ground-state approximate density functionals and thermal occupancies. Such calculations require reliable regularized potentials over a demanding range of temperatures and densities. By comparison of pseudopotential and all-electron results at ${\mathrm T} = 0$K for small Li clusters of local bcc symmetry and bond-lengths equivalent to high density bulk Li, we determine the density ranges for which standard projector augmented wave (PAW) and norm-conserving pseudopotentials are reliable. Then we construct and use all-electron PAW data sets with a small cutoff radius which are valid for lithium densities up to at least 80 g/cm$^3$.

Solving the Euler equation which corresponds to the energy minimum of a density functional expressed in orbital-free form involves related but distinct computational challenges. One is the choice between all-electron and pseudo-potential calculations and, if the latter, construction of the pseudo-potential. Another is the stability, speed, and accuracy of solution algorithms. Underlying both is the fundamental issue of satisfactory quality of the approximate functionals (kinetic energy and exchange-correlation). We address both computational issues and illustrate them by some comparative performance testing of our recently developed modified-conjoint generalized gradient approximation kinetic energy functionals. Comparisons are given for atoms, diatomic molecules, and some simple solids., Comment: submitted to Computer Physics Communications

Many-electron systems at substantial finite temperatures and densities present a major challenge to density functional theory. Very little is known about the free-energy behavior over the temperature range of interest, for example, in the study of warm dense matter. As a result, it is difficult to assess the validity of proposed approximate free-energy density functionals. Here we address, at least in part, this need for detailed results on well-characterized systems for purposes of testing and calibration of proposed approximate functionals. We present results on a comparatively simple, well-defined, but computationally feasible model, namely thermally occupied Hartree-Fock states for eight one-electron atoms at arbitrary positions in a hard-walled box. We discuss the main technical tasks (defining a suitable basis and evaluation of the required matrix elements) and discuss the physics which emerges from the calculations.

Finite temperature density functional theory requires representations for the internal energy, entropy, and free energy as functionals of the local density field. A central formal difficulty for an orbital-free representation is construction of the corresponding functionals for non-interacting particles in an arbitrary external potential. That problem is posed here in the context of the equilibrium statistical mechanics of an inhomogeneous system. The density functionals are defined and shown to be equal to the extremal state for a functional of the reduced one-particle statistical operators. Convexity of the latter functionals implies a class of general inequalities. First, it is shown that the familiar von Weizs\"acker lower bound for zero temperature functionals applies at finite temperature as well. An upper bound is obtained in terms of a single-particle statistical operator corresponding to the Thomas-Fermi approximation. Next, the behavior of the density functionals under coordinate scaling is obtained. The inequalities are exploited to obtain a class of upper and lower bounds at constant temperature, and a complementary class at constant density. The utility of such constraints and their relationship to corresponding results at zero temperature are discussed., Comment: submitted for publication in Phys. Rev. B

The Lieb-Oxford bound is a constraint upon approximate exchange-correlation functionals. We explore a non-empirical tightening of that bound in both universal and electron-number-dependent form. The test functional is PBE. Regarding both atomization energies (slightly worsened) and bond lengths (slightly bettered), we find the PBE functional to be remarkably insensitive to the value of the Lieb-Oxford bound. This both rationalizes the use of the original Lieb-Oxford constant in PBE and suggests that enhancement factors more sensitive to sharpened constraints await discovery., Comment: six figures (color)

We present a substantial extension of our constraint-based approach for development of orbital-free (OF) kinetic-energy (KE) density functionals intended for the calculation of quantum-mechanical forces in multi-scale molecular dynamics simulations. Suitability for realistic system simulations requires that the OF-KE functional yield accurate forces on the nuclei yet be relatively simple. We therefore require that the functionals be based on DFT constraints, local, dependent upon a small number of parameters fitted to a training set of limited size, and applicable beyond the scope of the training set. Our previous "modified conjoint" generalized-gradient-type functionals were constrained to producing a positive-definite Pauli potential. Though distinctly better than several published GGA-type functionals in that they gave semi-quantitative agreement with Born-Oppenheimer forces from full Kohn-Sham results, those modified conjoint functionals suffer from unphysical singularities at the nuclei. Here we show how to remove such singularities by introducing higher-order density derivatives. We give a simple illustration of such a functional used for the dissociation energy as a function of bond length for selected molecules., Comment: 16 pages, 9 figures, 2 tables, submitted to Phys. Rev. B