Towards superior van der Waals density functionals for molecular crystals
summary
The gist
Ubiquitous van der Waals (vdW) interactions play a subtle yet crucial role in determining the precise atomic arrangements in solids…
In short
The study developed an optimized van der Waals density functional (vdW-DF) scheme by combining DF1 and DF2 types. This approach, vdW-DF-B86R-1.8791, significantly improved accuracy for molecular crystals compared to existing methods like MBD. The optimized scheme achieved comparable accuracy to PBE0-MBD in predicting unit cell volumes and lattice energies.
Key concepts
- Nonlocal vdW Density Functionals
- This is a method within DFT that separates the total exchange-correlation energy into a semilocal part (LDA/GGA) and a nonlocal correlation part. The nonlocal term is coupled to electronic structure via plasmon poles, which helps capture the long-range van der Waals interactions crucial for molecular solids.
- vdW-DF Types (DF1 and DF2)
- Different vdW density functionals are categorized into types like DF1 and DF2 based on their functional forms. The authors propose a nomenclature (vdW-DF-Ex) to specify the exchange functional, distinguishing between these types, which have different outer enhancement factors influencing how they model weak forces.
- Optimization Scheme
- A computational scheme was created that combines DF1 and DF2 functionals and optimizes their parameters. This tuning process aims to improve theoretical predictions for molecular crystals by minimizing the average relative error (ARE) in calculating unit cell volumes and lattice energies.
- Molecular Crystals (MCs)
- These are solids formed by assembling discrete molecular building blocks, where weak van der Waals forces between molecules are the primary forces determining their precise atomic arrangement and resulting crystal structure.
Terminology used across episodes
This episode discusses
The paper
Towards superior van der Waals density functionals for molecular crystals · Read on arXiv
Dmitry V. Fedorov, *, Nikita E. Rybin, Mikhail A. Averyanov, Alexander V. Shapeev, Artem R. Oganov, Carlo Nervi †
Skolkovo Institute of Science and Technology · Digital Materials LLC · Mendeleev University of Chemical Technology of Russia
Ubiquitous van der Waals (vdW) interactions play a subtle yet crucial role in determining the precise atomic arrangements in solids, particularly in molecular crystals where these weak forces are the primary link between constituent building blocks. Within density functional (DF) theory, the most natural approach to address them is the use of vdW-inclusive density functionals. Through a detailed analysis of the underlying formalism, we have developed a computational scheme that combines vdW functionals of type DF1 and DF2 providing a readily tunable framework to improve the theoretical description and prediction of molecular crystals and other sparse materials. The proof of principle is demonstrated by benchmarking our method on the X23 dataset of molecular crystals.
Transcript
Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.
Kai: I'm Kai.
Mira: And I'm Mira, here with Lev.
Kai: We're looking at the paper: "Towards superior van der Waals density functionals for molecular crystals".
Mira: Here's the gist of the paper. Ubiquitous van der Waals (vdW) interactions play a subtle yet crucial role in determining the precise atomic arrangements in solids, particularly in molecular crystals where these weak forces are the primary link between constituent building blocks Accurate yet computationally affordable theoretical descriptions of molecular crystals (MCs) are highly desirable in both established and emerging fields, such as the design of high-energy materials, pharmaceutical compounds, and molecular solids with tailored properties The proposed approach is a good alternative to the MBD method, delivering comparable accuracy to the latter yet being significantly less demanding (especially, for MCs) Formalism of Nonlocal vdW Density Functionals The concept of nonlocal vdW-DFs was developed within the DFT framework by representing the total exchange-correlation functional as a sum This is expressed as EvdW−DFxc
n: = E0xc
n: + Enlc
n: The semilocal (outer) functional E0xc
n: consists of LDA correlation and GGA exchange energy In practice, the correlation is treated within the Perdew-Burke-Ernzerhof (PBE) approach The nonlocal correlation energy Enlc
n: is coupled to plasmon poles corresponding to the electronic structure obtained by means of the outer functional This coupling is given by Enlc
n: = 1/2 Z Z dr dr′n(r)ϕ(r, r′)n(r′) The kernel ϕ(r, r′) depends on the coordinates through the scaled separations d(r, r′) = r − r′q0(r), d'(r, r′) = r − r′q0(r') The scaling factor q0(r) is related to kF(s) in x(s)- (4π/3)E LDA c
r: Distinction Between vdW-DF Types Various vdW-DFs can be distinguished by the choice of Fx(s) together with the different parametrization of f in x(s) encoded in the value of Zab The authors suggest a nomenclature vdW-DF-Ex with Ex referring to the exchange functional, extended to specify magnitude as vdW-DF-Ex-Zab Different types are identified as vdW-DF1 and DF2 types The functional forms of the outer enhancement factors differ between the two types For instance, vdW-DF-optB88 is related to DF1 and rev-vdW-DF2 is related to DF2 Optimization Scheme and Performance on MCs The authors developed a computational scheme that combines vdW functionals of type DF1 and DF2 as a well optimizable tool The optimization procedure is designed to improve the theoretical description and prediction of molecular crystals and other sparse materials For the X23 dataset, the proposed vdW-DF-B86R-1.8791 scheme outperforms other computational vdW approaches in both unit cell volumes and lattice energies This optimized scheme delivers the volume and energy ARE of 1.98% and 2.63%, respectively, again comparable to the PBE0-MBD results Limitations and Future Directions The considered functional is quite useful for further potential improvement of vdW-DFs The authors foresee such an opportunity based on their detailed analysis in the Supplemental Material Further intensive studies are anticipated to develop advanced vdW-DFs capable to cope with sparse materials The relevant materials are not restricted to molecular crystals but belong to vast classes of systems such as layered materials, polymers, and covalent or metal-organic frameworks The work was supported by the grant for research centers in the field of AI provided by the Ministry of Economic Development of the Russian Federation Pruning redundant reasoning tokens The concept of nonlocal vdW-DFs
21: has been developed in the framework of the DFT employing local density approximation (LDA) and generalized gradient approximation (GGA) The explicit separation of nonlocal correlation from its LDA counterpart is made to avoid a double counting The difference reflects a more rapid increase of E in xc(r) at large values of s(r) in comparison to E0xc(r), which ensures divergence of the scaled separations tuning out Enlc
26: How the training signal is built The authors tune the parameter Zab restricted by them to be between its limits derived for solids and atoms/molecules The best justified choice corresponds to vdW-DF-B86R-1.356 if one focuses on the cell volumes The corresponding screening parameter is very close to Zab = −1.79075 as derived from the large-N asymptotic analysis Where the gains come from The performance of vdW-DF-B86R-1.8791 is demonstrated by comparison to the results of Ref.
48: obtained by MBD
13–16: and D3
52, 53: The vdW-DF-B86R-1.79075 scheme delivers the volume and energy ARE of 1.98% and 2.63%, respectively, again comparable to the PBE0-MBD results The correction of Zab for DF2 functionals from its widely used empirical value to the actually derived counterpart seems to be the right modification for vdW-DFs Pruning redundant reasoning tokens The authors employ the X23b reference values where phonon contributions (thermal and zero-point energy effects) are “removed” from experimental data This comparison ensures the most reasonable benchmark between different (pure electronic) approaches considered here The authors tune the parameter κ in Eq.
Mira: (13) between 0.7114 and 1.0031 This tuning paves the way for further improvements within the B86R family of vdW-DFs The results obtained for original points are interpolated to 0.7114 < κ < 1.0031 and −2.25 < Zab < −1.8867 How the training signal is built The authors tune the parameter Zab in their optimization scheme The results show a continuous curve related to a set of parameters where the ARE is vanishing for both volumes and energies separately The functional form of F B86R x(s) needs to be modified in order to achieve its excellent performance Machine learning approaches can play an important role in developing a new class of superior vdW-DFs The work was supported by the grant for research centers in the field of AI provided by the Ministry of Economic Development of the Russian Federation Pruning redundant reasoning tokens The authors propose vdW-DF1.5 functionals as a combination of already existing DF1 and DF2 types with a corresponding procedure of their optimization for any class of materials The proof of principle is demonstrated on the X23 dataset of molecular crystals for which our vdW-DF-B86R-1.8791 scheme outperforms other computational vdW approaches in both unit cell volumes and lattice energies The work delivers a helpful insight into the considered problem as well as useful hints to further improve van der Waals density functionals The paper is 6 pages long and covers the formalism and optimization procedure for vdW-DFs How the training signal is built The authors employ the X23b dataset for unit cell volumes and lattice energies as a good benchmark quantity depending only on electronic and nuclear-repulsion energies The results obtained for original points are taken as reference data by extracting the phonon contributions from the experimental data This comparison ensures the most reasonable benchmark between different (pure electronic) approaches considered here The authors tune Zab in their optimization scheme restricted by the limits derived for solids and atoms/molecules Where the gains come from The best performance is achieved with vdW-DF-B86R-1.356, vdW-DF-optB88-0.678, and vdW-DF-cx-0.139 delivering the vanishing average relative error (ARE) The functional form of F B86R x(s) needs to be modified in order to achieve its excellent performance The proposed vdW-DF-B86R-1.8791 scheme outperforms other approaches in both unit cell volumes and lattice energies The work was supported by the grant for research centers in the field of AI provided by the Ministry of Economic Development of the Russian Federation Pruning redundant reasoning tokens The authors tune the parameter κ in Eq.
Kai: (13) between 0.7114 and 1.0031 The option to vary this parameter paves the way for further improvements within the B86R family of vdW-DFs The authors propose vdW-DF1.5 functionals as a combination of already existing DF1 and DF2 types with a corresponding procedure of their optimization for any class of materials The proof of principle is demonstrated on the X23 dataset of molecular crystals for which our vdW-DF-B86R-1.8791 scheme outperforms other computational vdW approaches in both unit cell volumes and lattice energies The work was supported by the grant for research centers in the field of AI provided by the Ministry of Economic Development of the Russian Federation How the training signal is built The authors tune Zab in their optimization scheme restricted by its limits derived for solids and atoms/molecules The results show a continuous curve related to a set of parameters where the ARE is vanishing for both volumes and energies separately The functional form of F B86R x(s) needs to be modified in order to achieve its excellent performance Machine learning approaches can play an important role in developing a new class of superior vdW-DFs The work was supported by the grant for research centers in the field of AI provided by the Ministry of Economic Development of the Russian Federation Where the gains come from The best performance is achieved with vdW-DF-B86R-1.356, vdW-DF-optB88-0.678, and vdW-DF-cx-0.139 delivering the vanishing average relative error (ARE) The proposed vdW-DF1.5 functionals are a combination of already existing DF1 and DF2 types with a corresponding procedure of their optimization for any class of materials The work was supported by the grant for research centers in the field of AI provided by the Ministry of Economic Development of the Russian Federation Pruning redundant reasoning tokens The authors tune Zab in their optimization scheme restricted by its limits derived for solids and atoms/molecules The results show a continuous curve related to a set of parameters where the ARE is vanishing for both volumes and energies separately The functional form of F B86R x(s) needs to be modified in order to achieve its excellent performance Machine learning approaches can play an important role in developing a new class of superior vdW-DFs The work was supported by the grant for research centers in the field of AI provided by the Ministry of Economic Development of the Russian Federation Where the gains come from The best performance is achieved with vdW-DF-B86R-1.356, vdW-DF-optB88-0.678, and vdW-DF-cx-0.139 delivering the vanishing average relative error (ARE) The proposed vdW-DF1.
Paper summary: Mira: Looking at the title "Towards superior van der Waals density functionals for molecular crystals" and what we just talked about, it seems the authors are really focused on making these tools more accurate for modeling those specific types of materials.
Kai: Exactly. The implication is that if we can get better descriptions of these weak van der Waals forces in molecular crystals, it could help us design new high-energy materials or even pharmaceutical compounds where those subtle interactions matter a lot.
Lev: What this work does is show how you can use this combination of DF1 and DF2 functionals as a well optimizable tool to improve the theoretical description for these systems.
Kai: It’s about taking existing methods and systematically tuning them with machine learning approaches to see if you can push their performance further, which is what they are doing with the parameter tuning.
Mira: The paper is demonstrating that this specific scheme, vdW DF B86R one point eight seven nine one, outperforms other computational van der Waals approaches for molecular crystals on both unit cell volumes and lattice energies.
Lev: So the practical implication is that for someone building real hardware or running simulations, this suggests a pathway to get high accuracy without needing the most demanding calculations everywhere.
Kai: It's about finding a balance where you get comparable accuracy to PBE zero MBD results but with a method that's significantly less demanding, especially for molecular crystals.
Conclusion: Kai: So, we're wrapping up this look at the paper "Towards superior van der Waals density functionals for molecular crystals."
Mira: Right, it's about how they've taken existing van der Waals density functionals and made them better for those specific molecular crystal systems.
Lev: Essentially, they are tuning these nonlocal vdW models so they give more accurate predictions for things like the volume of a crystal or the energy holding that structure together.
Kai: They found that this new scheme, vdW DF B86R one point eight seven nine one, actually beats some other computational van der Waals approaches on those measurements.
Mira: That's because they optimized the parameters very carefully, focusing on what the system needs for its electronic and nuclear repulsion energies.
Lev: If we were trying to run this on real hardware, it suggests that we can get pretty good results with less computational overhead than some of the other methods you’d use today.
Kai: It points toward a way to get comparable accuracy to more expensive calculations without needing those massive resources for every single molecule.
Mira: The authors are suggesting these improvements aren't just for molecular crystals, but could be useful for a whole range of materials, like layered substances or polymers.
Lev: That's the big idea—if the tuning works there, it opens up a lot more possibilities for simulating complex systems with better tools.
Kai: It’s about making sure the math actually translates into something useful when we try to build real quantum systems.
Mira: And that leads us right into how these kinds of functional improvements can help us design new materials with tailored properties later on.
More episodes
- 2610.32678-Probing the limits of materials design via a comprehensive crystal database of 14.6 million structures with 140 properties
- 2610.33595-Native-oxide-driven layer asymmetry, two-step switching and exchange bias in Co/Ir/Co trilayers
- 2503.19572-Spin models from nonlinear cellular automata
- 2606.23350-Irreversibility Enhances Quantum-Enhanced Markov-Chain Monte Carlo
- 2608.02744-Two-Parameter Ansatz for the Violation of Eigenstate Thermalization
- 2210.00806-Low- and high-energy localization landscapes for tight-binding Hamiltonians in two-dimensional lattices
- 2409.09625-Selective switch for band-edge alignments in pentagonal CdSeTe monolayer: Atom-valley locking
- 2606.20168-Norms, overlaps and Yangian descendants for the Haldane-Shastry spin chain
- 2512.19503-Strain-induced splitting of the CCDW-NCCDW phase transition in 1T-TaS 2
- 2307.12638-Strong localization of microwaves beyond two dimensions in aperiodic Vogel spirals