Journal of Chemical Theory and Computation最新文献

Preparing ground states of correlated molecules on near-termquantumcomputers requires algorithms that are both accurate and resource-efficient.We introduce the qubit-adaptive variational quantum imaginary timeevolution (QAITE) algorithm, which synthesizes the systematic convergenceof imaginary-time evolution with hardware-efficient, adaptive ansatzconstruction. By integrating the linear-scaling qubit operator poolinto the adaptive variational quantum imaginary time evolution framework,QAITE deterministically constructs a compact ansatz without costlyclassical optimization loops. Numerical simulations for molecularsystems (H₄, LiH, H₂O, BeH₂), benchmarkedagainst our own implementations of established adaptive algorithmsunder identical conditions, show that QAITE achieves chemical accuracywith a 15–45% reduction in CNOT gates compared to the standardadaptive variational quantum imaginary time evolution (AVQITE) method.For hydrogen chains (R_HH = 1.5 Å, N = 6–16 qubits), QAITE’s circuit complexitygrows approximately quadratically and remains the lowest of the fourmethods at every system size, reaching 31% fewer CNOT gates than AVQITEat 16 qubits─a saving of 1322 gates. This efficiency extendsto heterogeneous (LiH)_n and (HBe)_n chains, where QAITE maintains a CNOT gateratio below 1 for all sizes studied (e.g., 0.81 for (LiH)₂). We further show that the deterministic parameter update is numericallyrobust: the results are insensitive to the regularization of the quantumgeometric tensor over at least 4 orders of magnitude, and chemicalaccuracy is retained under realistic shot noise applied to both thequantum geometric tensor and the force vector. By outperforming establishedadaptive algorithms in both absolute gate count and scaling, QAITEestablishes a promising, scalable pathway for quantum molecular simulationson noisy intermediate-scale quantum devices.

When a chemical system of interest contains heavy elements,itis important to obtain a balanced description of relativistic effectsand electron correlation. The exact two-component (X2C) Hamiltonianincludes both scalar-relativistic and spin–orbit effects witha reasonable computational cost. By using the spin–orbit X2CHamiltonian in state-averaged complete active space self-consistentfield (SA-CASSCF) calculations, it is possible to obtain equivalentdescriptions of the states with different spin characters. For moleculargeometry optimizations within this framework, the nuclear gradientsshould be efficiently evaluated. In this work, we describe a state-specificanalytical gradient algorithm for X2C-SA-CASSCF theory with the density-fittingapproximation. To realize such calculations, we formulate and implementthe response (Z-vector) equation for X2C-SA-CASSCFtheory. Using the obtained state-specific nuclear gradients, we optimizethe equilibrium geometries of metal porphyrins and the singlet–singletand singlet–triplet minimum energy crossing points or minimumenergy conical intersections of 2-thiouracil.

Free energy perturbation (FEP) methods provide a robustcomputationalframework for predicting protein-ligand binding affinities and haveseen increased adoption in drug discovery workflows. Among these,absolute binding free energy (ABFE) calculations are especially valuableas they enable affinity prediction for novel compounds without priorexperimental data. However, their broader adoption is limited by setupcomplexity, sampling challenges, and variable performance across systems.These challenges are particularly pronounced in therapeutically relevantcomplex membrane protein systems such as G protein-coupled receptors(GPCRs). This underscores the need to evaluate ABFE performance inGPCR systems and to develop strategies to mitigate these errors. Here,we evaluate ABFE calculations across three GPCR systems: the adenosineA_2A receptor (A_2AR), the dopamine D3 receptor(D3R), and the metabotropic glutamate receptor 5 (mGlu5). We focuson three sources of systematic error in these systems: (i) inadequateresolvation of the binding pocket in apo-like alchemical states, (ii)inconsistent treatment of metal ions, and (iii) insufficient samplingof receptor conformational transitions between holo-like and apo-likeensembles. To address these issues, we introduce two methodologicalimprovements: a biased grand-canonical-like water transport schemeto improve binding pocket resolvation, and a replica-exchange umbrellasampling (REUS)-based framework to quantify and correct the free energyassociated with receptor relaxation to apo-like conformations. Together,these approaches reduce the systematic overestimation of binding affinities,improving agreement with experimental data while maintaining rankingperformance. Overall, our results highlight the importance of accuratelyrepresenting the apo state in ABFE calculations and provide a practicalframework for applying high-accuracy free energy methods to membraneprotein targets.

Entanglementis a key bottleneck limiting the efficiency of tensor-networkand quantum simulations of molecular electronic structure. Here, wesystematically assess and extend Clifford disentanglers as a structure-preservingapproach to entanglement reduction: they can modify the entanglementstructure of qubit wave functions while retaining the Pauli-stringform of qubit Hamiltonians. To enable a practical search over Cliffordtransformations, we classify Clifford operators by their action onthe Schmidt spectrum across a bipartition, reducing the two- and four-qubitsearch spaces to 20 and 91392 representatives, respectively. Embeddedin an iterative Clifford-augmented matrix product state framework,these transformations reduce the energy errors at fixed bond dimensionfor the molecular test cases studied and mitigate the dependence onorbital ordering and Fermion-to-qubit mappings. We further show thatClifford disentanglers can also benefit quantum simulations such asshallow-circuit variational quantum eigensolver calculations. Together,these results establish Clifford disentanglers as a useful structure-preservingentanglement-engineering tool for tensor-network and quantum simulationsof molecular electronic structure, while also clarifying their dependenceon correlation and motivating future developments.

Membrane-embedded signalosome assembly is co-gated byconformationalselection, oligomeric interfaces, and the lipid microenvironment,yet short-lived, tightly coupled intermediate states are hard to inferfrom static structures. SLC15A4 recruits TASL to support IRF5 signaling,but available structures largely capture distinct conformational endpoints,leaving assembly bottlenecks, interface/lipid coupling, and inhibitorinterception without a quantitative energetic rationale. Here, wedeveloped a computational energetic framework connecting the restingapo dimer to a TASL-permissive state. In the apo state, SLC15A4 maintaineda persistent dimeric arrangement organized around a TM9/TM12-centeredinterfacial scaffold, while the interfacial cholesterol moleculesmaintained persistent contacts with the interface; luminal hydrophobicpacking and a multivalent hydrogen-bond network around N428 were associatedwith interface persistence, while the dimer retained characteristiccoupled motions of MFS helical bundles. Along the prescribed insertionpathway, TASL exhibited a pronounced apparent free-energy rise nearthe pocket, accompanied by rapid contact formation and a single-protomerswitch from luminal-open to cytosolic-open, yielding a modeled asymmetricintermediate with altered cross-protomer interface geometry. TASLwas stabilized by directional anchoring at a few key cationic sites.Pathway-dependent free-energy profiles further indicated that TASLengagement relaxed luminal-side interfacial stabilization and loweredthe free-energy cost of inter-subunit separation without compromisingpocket binding, potentially contributing to the formation of a signaling-competentSLC15A4–TASL module. Finally, our results suggest that an inhibitorlimits access to the TASL-permissive state by locking the luminal-openconformation and occupying the central cavity, increasing the free-energycost along the prescribed TASL entry and reshaping electrostatic determinants.Together, these results connect experimentally resolved structuralstates through a computational energetic framework in which TASL-associatedsingle-protomer switching may couple to luminal interface weakeningand reduced interface persistence, motivating therapeutic strategiestargeting the SLC15A4–TASL–IRF5 axis via conformationallocking and competitive cavity occupation.

The potential energy landscape (PEL) formalism is a powerfultoolwithin statistical mechanics to study the thermodynamic behavior ofclassical and quantum liquids as well as non-annealing glasses. Inthe quantum case, this theoretical framework is based on the path-integralformulation of statistical mechanics where the liquid/non-annealingglass is mapped onto a classical system of ring-polymers. In previousstudies, and motivated by path-integral (PI) computer simulations,the PEL formalism was developed assuming that the ring-polymers werediscrete, i.e., each ring-polymer was composed of a finite numberof beads, n_b. In thiswork, we revisit the PEL formalism for the case where the ring-polymersof the system are continuous, i.e., n_b → ∞ (as required in the PI formulationof statistical mechanics). We find that in this limit, the PEL ofthe quantum liquid is unique and well-defined. The main PEL propertiesof the system, including the average inherent structure energy E_IS, configurational entropy S_IS, and vibrational Helmholtzfree energy F_vib, arealso well-defined (i.e., they converge). Notably, in the harmonicapproximation of the PEL, F_vib can be expressed analytically; it is identical to the freeenergy of a multidimensional quantum harmonic oscillator with thecorresponding frequencies given by the normal mode frequencies ofthe classical liquid counterpart (accessible from classical computersimulations). When applied to non-annealing glasses and crystals,this new version of the PEL formalism reduces to the so-called quasi-harmonic approximation, commonly used in the studyof solids. Overall, the PEL formalism for n_b → ∞ provides an alternative approachto its previous version (finite n_b); it is also more elegant and intuitive, and simpler to apply.For a Gaussian and harmonic PEL, this formalism (n_b → ∞) allows one to obtainthe Helmholtz free energy of the quantum liquid using classical computersimulations, avoiding the more expensive PI computer simulations.

Integral-features density functional theory (IF-DFT)replaces theconventional integration over local ingredients with a nonlinear mappingfrom global integral descriptors, enabling machine-learned densityfunctionals with broad transferability. Here we report ML26@MN15,a new IF functional that extends the earlier ML25@MN15 model by incorporating12 additional correlation features─including overlap-projectedrung-3.5 terms, CS1 dynamic-correlation ingredients, and VV10 nonlocalcorrelation─and by treating Hartree–Fock exchange asan integral feature rather than as a fixed hybrid percentage. ML26@MN15employs 79 integral features evaluated on MN15 ingredients and a single-hidden-layerneural network with 5000 nodes to produce a size-extensive exchange–correlationenergy. Trained on 185 databases comprising 7242 reference data, ML26@MN15achieves a data-averaged energetic mean unsigned error of 0.89 kcal/mol,improving upon ML25@MN15 by 15% and outperforming a group of previousleading functionals across all eight chemical partitions examined.For the widely used MDB2019S, MGCDB84, and GMTKN55 data sets, ML26@MN15yields the lowest averaged errors among all tested functionals, includingthe results for GMTKN55 by ML-based Skala-1.1 and DM21 density functionalapproximations. The new functional also maintains strong performancefor systems containing heavy elements and exhibits improved behavioron a challenging self-interaction benchmark. These results show thatexpanding the integral-feature space with physically motivated nonlocalingredients yields a density functional with unprecedented accuracyacross diverse chemical benchmarks. The broad and consistent accuracyof ML26@MN15 highlights the promise of integral-features density functionalapproximations as a scalable framework for incorporating additionalnonlocal physics into future functional development.

TDDFT for open-shell systems, whether spin-conservingor spin-flip,has long suffered from spin contamination. This problem arises becausethe single-excitation space built upon a single Kohn–Sham determinantis not spin-complete. Adopting spin tensor reference states thereforeoffers an elegant and promising route to resolving this issue. Inthis work, we revisit the tensor TDDFT equations within the Tamm–Dancoffapproximation (TDA) from a noncollinear perspective. We show that,for S = 1/2 reference states, the internal consistencyof the spin tensor formulation can, with the aid of the zero-excitation-energytheorem, be recast as a set of constraints that the exchange–correlationkernel must satisfy. Standard noncollinear functionals, however, generallyfail to meet these constraints. To address this, we propose a kernelreconstruction scheme that is independent of the specific functionalform and free of empirical parameters. This scheme enforces the requiredconstraints, restoring internal consistency in the full spin tensorstructure, with spin adaptation following as a natural consequence.Furthermore, when extended to tensor reference states with other valuesof S, such as S = 1 for the oxygen molecule, the scheme eliminatesthe so-called artifact states, namely solutions with severely underestimatedexcitation energies. In addition, the scheme allows the target statesthat ROKS reference states aim to describe to be expressed and computed,at the TDA level, within the same unified framework as other states,a capability that spin-adapted spin-conserving TDDFT has so far lacked.