Journal of Chemical Theory and Computation最新文献

Chemical reactivitymodeling in large systems over extendedtimescales is computationally and theoretically challenging; ab initio methods are too expensive, while classical force fieldscannot handle bond breaking/formation. Existing solutions like reactiveforce fields require extensive expertise to develop, and machine-learnedapproaches often need prohibitively large, high-quality data setsfor training. These barriers have hindered the modeling of reactivechemical processes. Multistate empirical valence bond (MSEVB) theoryoffers an alternative by enabling conventional unreactive force fieldsto be mixed in a controlled manner to selectively describe chemicalreactions. However, existing MSEVB implementations are often field-specific,have license-restricted distribution, or rely on complex parametrizationschemes. This work presents a new open-source, general MSEVB implementation,designed to make reactive modeling more accessible to nonexperts.Specifically, the implementation here focused on simple energy-basedcoupling, thereby eliminating the need for complex geometry-basedcoupling approaches, significantly lowering the barrier to entry forreactive force field development. This generalized approach is demonstratedto reproduce proton transfer behavior in Y-doped and undoped BaZrO₃, bulk water and liquid imidazole, as well as other formsof reactivity, such as electron migration in LiCoO₂. Wedemonstrate that only two fitted parameters are needed to accuratelymodel symmetric transfer across the three chemically diverse proton-conductingsystems studied.

Triplet–triplet annihilation upconversion dependson a near-resonantenergetic balance between the relevant singlet and triplet excitedstates. Here we introduce a computational framework for mechanicallytuning this balance by applying an external force that permits tolocate the molecular geometry reaching the singlet–tripletresonance condition, i.e., E_S1 – E_S0 = 2(E_T1 – E_S0). Our approach formulates this problem as a constrainedoptimization in which the force magnitude is minimized while enforcingthe target energetic resonance condition. Using a local quadraticmodel for the triplet-state potential energy surface together witha linearized description of the relevant energy gaps, we derive aminimum-force update procedure to obtain the optimal displacementand the associated external force. The resulting algorithm providesthe mechanically controlled geometry and the optimal force vectorrequired to reach it. The method offers a practical computationalroute to identify mechanically accessible structures with improvedupconversion characteristics, paving the way for the design of mechanoresponsivechromophores and materials with stress-tunable optical behavior.

The complicated physicochemical properties of metal complexesthatexhibit thermal spin crossover make it difficult for routine electronicstructure calculations to yield an accurate transition temperatureprediction, T_1/2. The difficulty liesin the intricate connection between the spin-crossover energy, whichis a molecular spectroscopic property, and T_1/2, a condensed-phase property. Here we show how to obtainspin-crossover energies systematically by reverse engineering of experimental T_1/2 data. The protocol is based upon fittingthe range separation parameter, ω, in the hybrid LC-ωPBEdensity functional to reproduce the experimental T_1/2 values for a series of metal complexes. We provideinsights into the sources of variations of at least ±15 kJ mol^–1 found from common exchange and correlation functionalsby comparing their performance against our reference data. By analysisof the sensitivity of transition temperatures to ±1% shifts inthe range separation parameter, we determined a typical uncertaintyof ±50 K for them, and a ±2 kJ mol^–1 uncertaintyin the extracted spin-crossover energies due to ±1% variationsof T_1/2. Lastly, we present results fromthe high-level, all-electron coupled cluster method for eight of thesmaller molecules in the reference data set, and discuss the influenceof the truncation of the excitation series upon the spin state energies.

In this article, low-rank separable analytical representationsof Coulomb-like potentials are presented. Cartesian coordinates areemployed for the sake of generality. The original potential is regularizedto avoid the numerical divergence associated with the coalescenceregion. Furthermore, a range-separation approach is adopted. The long-rangepart, being smooth and well-behaved, can be directly expressed inour functional sparse Tucker Finite Basis Representation (sparse TuckerFBR). On the other hand, the short-range part, which contains thetamed singularity, is expressed using a sparse spectral representation.The quality and accuracy of the full sparse Tucker expression is demonstratedby comparison to the two-interacting-electron-in-a-box model.

A covariant local matrix density approximation (LMDA)is introducedwithin multistate density functional theory based on subspace invarianceand the spectral decomposition of the matrix containing state andtransition densities. The exchange-correlation matrix functional isconstructed through spectral reconstruction of this matrix density,whereby local exchange-correlation functionals are incorporated asspectral-channel functionals within a covariant matrix-functionalformalism. The resulting formulation preserves exact normalizationof the exchange-correlation matrix hole, recovers ground-state densityfunctional theory in the single-state limit, and preserves spin-multipletdegeneracy through covariance of the spin matrix functional underspin rotations. Combined with multistate self-consistent-field (MSSCF)optimization, the present framework enables fully variational calculationsof interacting ground and excited states. Applications to atomic excitations,H₂ dissociation, ethylene torsion, and cyclobutadiene automerizationreveal that the resulting MSDFT/LMDA approach captures essential multistateand strong-correlation physics, including static correlation and spinsymmetry. The key insight is that the central obstacle in extendingdensity functional theory to coupled electronic states is not necessarilythe lack of appropriate scalar exchange-correlation approximations,but the lack of a covariant matrix-functional formalism in which suchfunctionals can operate. The present MSDFT/LMDA provides a directrealization of this matrix-functional formalism.

Open-shell nanographenes (NGs) possessunique electronicpropertiesthat hold promise for next-generation carbon-based materials, yettheir rational design is hampered by the lack of quantitative toolsto predict their radical character. Empirical rules such as Clar’saromatic sextet rule and the Fries rule provide useful qualitativeguides for closed-shell systems but inherently fail for open-shellNGs, where the ground state is better described as a resonance hybridof closed-shell and open-shell structures. Here, we present a furtherdevelopment of the generalized valence-bond-based block-correlatedcoupled cluster (GVB-BCCC) theory and demonstrate its applicationin qualitative and quantitative predictions of the radical characterin open-shell NGs. Four methodological advances are reported: (1)a systematic search procedure of multiple GVB solutions that enablesthe computational identification of distinct resonance structures(closed-shell, diradical, and tetraradical forms) through tailoredinitial orbital guesses; (2) the first implementation of GVB-BCCC4for triplet ground states, extending the framework to systems withtriplet ground states; (3) the first implementation of the unrelaxedone-particle density matrix for singlet GVB-BCCC4, enabling the extractionof natural orbital occupation numbers (NOONs) as a rigorous, quantitativemeasure of radical character directly from the GVB-BCCC4 wavefunction;and (4) the introduction of a relative local correlation energy (RLCE)index for evaluating local aromaticity in individual benzene-likerings. By exploiting the fact that distinct GVB solutions map to chemicallyintuitive resonance forms, our approach is able to identify the dominantresonance structure through accurate GVB-BCCC4 calculations on topof the respective GVB references. Critically, by comparing the uncorrelatedGVB solutions with the fully correlated GVB-BCCC4 results, we canunambiguously determine whether the radical character of a given systemoriginates from the π-skeleton topology (a resonance-driveneffect) or from interpair electron correlation─a distinctionthat remains inaccessible to empirical rules or conventional methods.Our approach is then applied to 11 representative NGs, including systemswith up to 66 π orbitals. The radical character indices fromGVB-BCCC4 show excellent agreement with DMRG-CASCI reference data,while broken-symmetry DFT is found to be unreliable for systems withpronounced polyradical character. The singlet–triplet energygaps calculated at the GVB-BCCC4 level also agree well with availableexperimental values. This work establishes GVB-BCCC as a predictivetool for understanding the structure–property relationshipsof open-shell systems and for guiding the rational design of carbon-basedfunctional materials.

Densitymatrix embedding theory (DMET), which is usuallybasedon a Schmidt decomposition of Slater determinants by partitioningthe full system into impurity and environment in terms of local orthogonalorbitals (LOs), has demonstrated considerable promise in electronicstructure studies because it enables the extraction of local propertiesusing a high-level solver within an embedded impurity subsystem withgreatly reduced degrees of freedom, thereby achieving a balance betweenaccuracy and computational cost. However, its application to excitedstates of strongly correlated systems, such as lanthanide complexes,remains challenging because the errors relative to all-electron resultscan still be significant. Motivated by the success of the previouslydeveloped atomic orbitals (AOs) based DMET framework (Ai, Li, andJiang, Phys. Rev. Lett. 2025, 135, 026502.), termed AO-DMET, whichattains improved accuracy by constructing the embedded subspace basedon a non-orthogonal decomposition of the Slater determinant in termsof AOs, we propose a new LO-based partitioning scheme that fully preservesthe impurity space spanned by corresponding AOs and can achieve accuracyclosely matching that of AO-DMET while retaining the orthogonal partitionand its associated computational efficiency. The performance of theproposed method is demonstrated through excitation energy calculationsfor several representative lanthanide complexes. These results establishan efficient and accurate partitioning scheme for describing excitedstates in strongly correlated systems within the DMET framework.

Predicting how proteinmutations affect drug bindingremains amajor challenge, particularly when the mutations are distal from thebinding site. In this study, we introduce a coupled simulation workflowthat combines classical molecular dynamics (MD) with high-throughputquantum mechanical (QM) analysis to reveal the electronic structuresignatures of mutation-induced drug resistance in the HIV-1 protease.Our workflow leverages GPU-accelerated MD to generate conformationalensembles, and performs in operando linear-scaling density functionaltheory (DFT) calculations on selected frames parallelized on a coupledpartition of CPU nodes. This design enables efficient, massively parallelquantum analysis of protein–ligand complexes at atomic resolution.Using this approach, we investigate resistance to the antiviral Darunavirin a multimutant HIV-1 protease variant. By mapping the network ofelectronic interactions across the binding interface, our resultshighlight the critical role of conformational sampling and quantuminsight in understanding distal mutation effects, and demonstratea scalable computational strategy for studying complex biophysicalmechanisms of drug resistance. We argue that this type of analysismay pave the way for designing inhibitors that maintain binding stabilityagainst systemic, mutation-induced destabilization.

Fluidnanofilms play a fundamental role in nanoscale interfacialthermodynamics, yet their interfacial behavior and treatment withinclassical density functional theory (cDFT) warrant further investigation.We investigate nanofilm interfacial properties using a cDFT frameworkbuilt upon the perturbed-chain statistical associating fluid theory(PC-SAFT) for both fluid–fluid and fluid–solid interfacialsystems. Comparison with molecular simulations shows that the long-standingnumerical discrepancies regarding free-standing nanofilms persistin the PC-SAFT functional predictions, supporting the view that theystem from thermal capillary-wave fluctuations rather than deficienciesof the density functional itself, as such fluctuations are inherentlyneglected in the mean-field approximation adopted by standard cDFT.Importantly, we establish a thermodynamically consistent frameworkfor interfacial free energy (IFE) calculation, which explicitly incorporatesnanofilm thermodynamic contributions and redefines the effective fluidvolume by excluding the solid-phase region. The proposed method exhibitsexcellent consistency with another method based on the relationshipbetween IFE and disjoining pressure for fluid systems, while bothfluid–fluid and fluid–solid IFEs differ substantiallyfrom those predicted by the conventional method. We find that neglectingnanofilm contributions induces prominent size-dependent variationsin the IFE and contact angle of hemicylindrical droplets, which isinconsistent with the extensive literature. Different methods alsolead to opposite signs of the line tension for a hemispherical argonnanodroplet on a strongly lyophobic surface. The proposed frameworkprovides a unified molecular-level basis for understanding interfacialprocesses involving nanofilms, including wetting, nucleation, adsorption,and other phenomena that rely on the accurate evaluation of IFEs.

Infrared spectroscopyserves as an indispensable analytical toolacross materials science, pharmaceutical development, and the lifesciences. Computational insights derived from first-principles methodsenhance the interpretation of vibrational spectra, yet their applicationremains hindered by high computational cost. Machine-learned interatomicpotentials offer a route to accelerate these predictions by multipleorders of magnitude without sacrificing ab initio accuracy. Here,we present an on-the-fly active learning framework for the automatedtraining of such potentials specifically tailored to infrared spectralprediction. Our approach employs a multitask acquisition strategy,in which model performance is evaluated iteratively on designatedobjectives, avoiding the dependence on fixed training sets. We demonstratethat this multitask scheme creates interatomic potentials that exhibitrobustness and fidelity in complex workflows for calculating infraredspectra.