Adding Temperature to the Adiabatic Connection: Hybrid Functionals for Static and Dynamic Correlation

arXiv:2610.22284 · physics.chem-ph, quant-ph · Submitted 2026-10-15 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "Adding Temperature to the Adiabatic Connection".

Mira: The gist The framework formally unifies information-density matrix functional theory (i-DMFT) and thermally-assisted-occupation (TAO-)DFT as special cases,

Kai: First, who's behind it and why it matters.

Paper summary: Mira: So to wrap up the paper "Adding Temperature to the Adiabatic Connection: Hybrid Functionals for Static and Dynamic Correlation", they are proposing a way to bridge the gap between capturing static correlation with temperature-dependent occupations and handling dynamic correlation via standard DFT approximations.

Kai: It's about extending the adiabatic connection mechanism into a thermal dimension, creating these thermally enhanced hybrid functionals that operate at mean-field cost while covering both types of correlation simultaneously.

Lev: The main implication for us in hardware is that if we can find a way to reliably estimate this correlation temperature theta, we might get better predictive power from less demanding methods than what i-DMFT offered before.

Mira: The paper clarifies the theoretical foundation of DFT as a thermodynamic theory and shows how the resulting approximation neatly separates static and dynamic correlation, which is pretty clean conceptually.

Kai: They show that this framework formally unifies i-DMFT and TAO-DFT as special cases, defining their respective prescriptions for the temperature parameter.

Lev: But they also admit a limitation: finding that correct temperature theta is still a hard job; it’s not some universal constant you just plug in.

Mira: The work sets up clear recipes for determining theta, which are diverse in cost but test how well they capture the physics across different chemical systems.

Kai: Overall, this paper shows that thermally enhanced hybrids are a systematic way to treat correlation types at a lower computational price than full methods, provided you can manage the uncertainty in selecting that temperature parameter.

Conclusion: Kai: So, we've been looking at this paper that's titled "Adding Temperature to the Adiabatic Connection: Hybrid Functionals for Static and Dynamic Correlation."

Mira: Yeah, basically, they figured out a way to mix static correlation—the stuff that’s hard to handle—with dynamic correlation using an effective temperature.

Lev: What does that actually mean for us when we're thinking about building real hardware?

Kai: It means we can use these thermally enhanced hybrids in methods that are already low-cost, like DFT approximations, instead of needing the super expensive information-density matrix theory.

Mira: The authors formally connect this to information-density matrix functional theory and thermally assisted occupation density functional theory as special cases.

Lev: So they're not inventing a whole new theoretical framework from scratch, but just adding this thermal dimension to what we already have.

Kai: Exactly. They show that you can capture static correlation through fractional occupations generated by this effective "correlation temperature."

Mira: And the dynamic part, the standard dynamic correlation, is handled by established density functional approximations.

Lev: But the authors also pointed out that choosing that correct correlation temperature theta is still a real hurdle for any practical application.

Kai: That’s the catch. They test three different ways to pick this theta, like fitting dissociation energies or fitting natural occupation numbers to see what works best.

Mira: The results they got from testing on biphenyl and other systems show that these hybrids can reproduce dissociation curves much better than just using mean-field Hartree-Fock.

Lev: It seems like a solid step forward for getting more accurate electronic structure without needing the full cost of correlated calculations everywhere.

Kai: Yeah, it shows a systematic route toward treating both correlation types with lower computational demands across different chemical systems.

Florian Matz, Paul G. Graf, Markus Penz, Christian Schilling

Arnold Sommerfeld Centre for Theoretical Physics · Oslo Metropolitan University · University of Geneva

physics.chem-ph, quant-ph

Submitted: 2026-10-15

Updated: 2026-10-15

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 60/100

The gist: The gist The framework formally unifies information-density matrix functional theory (i-DMFT) and thermally-assisted-occupation (TAO-)DFT as special cases, clarifying their theoretical basis and the

Key concepts

Adiabatic Connection Mechanism
This is a method used in DFT to connect the ground state to the fully interacting system by following a path in parameter space. The authors extend this mechanism by incorporating a temperature dimension, allowing the theory to account for thermal effects during this connection.
Thermally Enhanced Hybrid Functionals
These are new approximations for exchange-correlation energy that separate static and dynamic correlation. Static correlation is modeled using fractional occupations derived from an effective 'correlation temperature' within thermal Hartree-Fock, while dynamic correlation is treated by established density functional approximations.
Correlation Temperature ($ heta$)
This parameter represents an effective temperature used in the thermal Hartree-Fock calculation to capture static correlation. Determining the correct value of $\theta$ is crucial for accurately describing how fractional occupations change during molecular dissociation, which helps reproduce key properties like dissociation curves.

Terminology

Summary

The gist The framework formally unifies information-density matrix functional theory (i-DMFT) and thermally-assisted-occupation (TAO-)DFT as special cases, clarifying their theoretical basis and the respective prescriptions for the temperature parameter>

How it works

The paper extends the adiabatic connection mechanism from density functional theory to include a temperature dimension, giving rise to thermally enhanced hybrid functionals in which static correlation is captured via fractional occupations generated by an effective “correlation temperature” within thermal Hartree-Fock, while dynamic correlation is described by established density functional approximations>

  1. The framework derives DFT as a thermodynamic theory by minimizing the trial grand potential and then performing a Legendre transform to define a convex density functional as the free energy Fλ,τ (ρ) = sup v ρdr>

  2. This leads to the definition of exchange-correlation energy Eθxc = F1,0 − F0,θ − EH via an integral along a path γ in (λ, τ) parameter space between a thermal KS system and the fully interacting system>

  3. The resulting approximation for the exchange-correlation functional is given by Ehyb+θxc ≈ EHF+θx + (1 − a)EDFAx + E DFAc, where 'a' ranges between 0.20 [49] and 0.25 [50, 51]>

  4. This approximation neatly differentiates between static and dynamic correlation, ceding the dynamic part to a DFT functional while the first term handles static correlations with an adequate choice of effective temperature θ>

Analysis of Numerical Results

The paper demonstrates that this new hybrid method can reproduce electron density, dissociation curves, and natural occupation numbers much more faithfully than the previously proposed i-DMFT method based only on the Hartree-Fock mean-field interaction>

correlation-corrected thermally enhanced calculations almost exactly reproduce the FCI results at 1.4 and 2.41 Bohr, and reproduce the cusp at the nucleus significantly better at 8 Bohr

The inclusion of a mean-field correlation functional results in smoothly changing natural occupation numbers over the entire dissociation, even though they still deviate from the CASSCF results

The reproduction of the correct barrierless dissociation surface is obviously a sign for the qualitative superiority of the DFT+θ family of methods

Approaches for Obtaining Correlation Temperatures

A crucial difficulty remains in choosing the electronic temperature θ that enters as a product −θS in the free energy F0,θ>

The methodologies that emerged for the determination of θ so far are relatively diverse in cost and complexity, but most importantly in their proposed universality

A global θ is also not to be expected based on the results of studies of the correlation conjecture and i-DMFT

In this work, we compare some of these and newly devised approaches for the determination of θ along the dissociation of dinitrogen (14 electrons) in the 6-31G basis (18 orbitals)

The paper investigates three main recipes for determining θ:

  1. Fitting the dissociation energy to a reference calculation, where this can be a low-level DFT calculation used to interpolate the dissociation curve>

  2. Fitting natural occupation numbers to those of a high-level method, motivated by the fact that fractional occupations are meant to describe strong/static correlation>

  3. Finding θ such that the ground-state energy of the FCI reference is recovered at every step during a deformation of the molecule>

Benchmarking Larger Systems

A benchmark was conducted on nine different molecules, including single bonds between two sp2 hybridized carbon atoms and seven differently substituted biphenyls, to test transferability>

"The results of these calculations can be taken from Tab. V. Analogous to the approaches discussed in Sections IV and VI, we obtained the dissociation energy simply as the energy difference between starting point and end point in the FCI reference"

"Thermally enhanced Hartree-Fock and B3LYP calculations can reproduce the dissociation energies obtained on the CASSCF level of theory, but a different value of θ has to be applied for every molecule in order to reproduce the correct trend of dissociation energies"

The average values for θ are 0.0865 EH with HF and 0.0309 EH with B3LYP, and the standard deviations are 0.0023 EH (2.6%) and 0.019 EH (6.2%), respectively

**"In general, such a connection between bond strength and θ is apparent as illustrated in Fig.

Improvements for AI systems

  1. textbfDefinition of Thermally Enhanced Hybrid Functionals (Eq. 19): The improved system can capture both static and dynamic correlation simultaneously at mean-field cost by mixing thermal HF with a DFA functional, as described by the expression E hyb+θ xc ≈ aEHF+θ x + (1 − a)E DFA x + E DFA c.

  2. textbfSystematic Resolution of i-DMFT Failures: The improved system can systematically resolve known failures of i-DMFT in dissociation barriers and densities by testing different schemes for determining the correlation temperature and investigating their accuracy for a benchmark set derived from biphenyl, as detailed in Section IV.

  3. textbfThermodynamic Foundation for DFT (Section II): The improved system is grounded in a thermodynamic theory of DFT, allowing it to be formulated via the trial grand potential (Eq. 2) and subsequently mapped onto a convex density functional (Eq. 4), providing a rigorous basis for thermally enhanced approximations where F λ,τ (ρ) = sup v [omegaλ,τ (v) − Z v(r)ρ(r)dr.

  4. textbfAdaptive Correlation Temperature Determination: The improved system can determine the correlation temperature through various schemes, including fitting to reference occupation numbers or reference energies, as shown in Fig. 7 of Section V, enabling the selection of an optimal global θ or geometry-specific values for different molecules.

  5. textbfAccurate Dissociation Curve Prediction: The improved system can accurately predict dissociation curves by employing two main recipes: the fitting of the dissociation energy to a reference calculation (interpolation) or the fitting of natural occupation numbers to a high-level method, which is shown to reproduce FCI results for H2 and N2.

  6. textbfCharacterization of Correlation Types: The improved system can neatly differentiate between static and dynamic correlation because the approximation neatly differentiates between static and dynamic correlation, as modern DFAs handle dynamic effects while the first term in Eq. (19), aEHF+θ x, takes over the responsibility for static correlations.

  7. textbfBenchmarking Against Complex Systems: The improved system can be validated on larger systems like biphenyl by demonstrating that Thermally enhanced Hartree-Fock and B3LYP calculations can reproduce the dissociation energies obtained on the CASSCF level of theory, although it requires a different value of θ has to be applied for every molecule in order to reproduce the correct trend of dissociation energies.

Sources

Related papers