Wei Hu

Heterogeneous high-performance computing is increasingly important for large-scale Kohn–Sham density functional theory (DFT) calculations. However, although GPU device memory provides high bandwidth, its limited capacity makes peak memory usage a critical factor that determines the maximum accessible system size. In this work, we analyze and optimize the peak memory usage of PyPWDFT, a Python native plane-wave DFT code, for heterogeneous GPU computing. The main computational stages of plane-wave DFT calculations are systematically examined to identify the dominant contributors to peak memory usage and to derive theoretical estimates of their memory requirements. Guided by this analysis, several targeted optimization strategies are introduced, including the explicit release of the CuPy memory pool, decomposition of vectorized operations, preservation of array memory contiguity, broader use of in-place operations, and additional computations when they can substantially reduce memory usage. The measured peak memory usage agrees closely with the theoretical analysis, demonstrating that with targeted memory optimization, a native Python implementation can achieve memory efficiency comparable to that of compiled-language implementations and enable DFT calculations for a 1536-atom silicon system on a single NVIDIA H200 GPU. Additional memory and performance benchmarks on A100, A800, H100, and H200 GPUs further demonstrate the general applicability of the optimized implementation across different heterogeneous GPU computing platforms.

We benchmark a representative set of exchange–correlation functionals for the 2PA strengths of centrosymmetric quadrupolar dyes using CC2 as reference. Owing to the Ci symmetry, the dominant 2PA channel of these compounds proceeds through a symmetry-allowed Ag → Au → Ag pathway, significantly differing from the process in dipolar dyes. We analyze the response using a three-state framework. Many functionals reproduce the qualitative CC2 structure–property trends, indicating that TD-DFT can capture the main chemical effects. However, substantial deviations remain in absolute magnitudes and in the ordering of closely related systems. These errors primarily originate from the transition-dipole-moment contribution, especially the upper S1 → S2 transition, while energetic detuning plays a secondary role. Low-exact-exchange global hybrids, such as B3LYP and PBE0, tend to overestimate the 2PA strengths, whereas range-separated hybrids typically underestimate them. A reasonably balanced performance is obtained with higher-exact-exchange hybrids, particularly MN15, BMK, M06–2X, and BH&HLYP. Overall, MN15 yields the lowest global error for this benchmark set, while CAM-B3LYP remains effective for preserving relative trends despite its systematic underestimation of absolute responses.

KSSOLV (Kohn–Sham Solver) is a MATLAB package for plane-wave Kohn–Sham density functional theory (DFT) calculations. Here, we present the KSSOLV Toolbox, a MATLAB-based graphical and workflow-oriented environment built on KSSOLV to simplify the setup, execution, and analysis of plane-wave DFT calculations. The toolbox integrates structure import, symmetry analysis, self-consistent and non-self-consistent calculations, postprocessing, and visualization within a unified interface, while retaining access to the underlying KSSOLV computational backend. To improve reproducibility and project management, calculation settings, workflows, and results are organized in a single project file. We further assess the numerical consistency and practical computational behavior of the toolbox through benchmark calculations against established plane-wave codes. Comparisons with Quantum ESPRESSO show close agreement in total energies and atomic forces, and single-thread CPU timing tests against M-SPARC, DFTK.jl, PyPWDFT, and Quantum ESPRESSO clarify the computational performance of KSSOLV for PBE and HSE06 calculations on Si systems of different sizes. The KSSOLV Toolbox is intended as a transparent and accessible platform for plane-wave DFT workflow organization, method development, validation, and rapid prototyping.

The coordinate descent full configuration interaction (CDFCI) is an accurate and efficient method for calculating ground-state energies of molecular systems. In this work we introduce two perturbative extensions of original CDFCI. We first propose a direct CDFCI + PT to further correct variational energy by perturbation theory. We also propose an iterative perturbative CDFCI (IP-CDFCI) to avoid evaluating the overlarge perturbative space and catch the information encoded in variational stage. In IP-CDFCI, we store three sparse vectors in a Hash table: c, b and d, where c is the variational wave function, b is the compressed approximation of Hc and d is the perturbative wave function, respectively. By utilizing the redundant memory in the storage of c, we can calculate perturbation without any extra memory compared to original CDFCI. We also use a strategy to eliminate the expensive repetitive evaluation of diagonal elements of Hamiltonian by creating a vector h which stores the diagonal entries of Hamiltonian. This strategy trades memory usage for computational time and reduces the computing time significantly. Furthermore, we use quadruple-precision floating-point numbers to improve the numerical stability of perturbative energy. We demonstrate the effectiveness of the single-threaded IP-CDFCI and the multithreaded CDFCI + PT, respectively, by performing calculations on several small molecular benchmarks as well as challenging correlated system: the chromium dimer.

Excitonic insulator (EI) is a long-sought quantum phase with a many-body gap opening due to the Bose-Einstein condensate of electron-hole pairs, providing an intriguing platform for exploring many-body effects. However, due to the large dielectric screening, few materials are reported to be EIs without external fields. Using the Green's function-based many-body perturbation theory calculations and Bardeen-Cooper-Schrieffer (BCS)-like mean field theory, we present a heteroatom-centered triangulene monolayer with a small band gap and bandwidth that exhibits an intrinsic beyond-room-temperature EI ground state. The GW plus Bethe-Salpeter equation calculation results show that the exciton binding energy exceeds the band gap, and the BCS-like model shows an additional excitonic gap existing. We also present a possible fingerprint of this system in the transmittance spectrum for the experimental observation. Our findings show that such precisely synthesized two-dimensional polymers will promise to enhance the exploration of nontrivial excitonic states and many quantum phenomena.

ChatMat is an autonomous multiagent framework integrating LLMs with DFT and ML-PES to accelerate material property prediction and exploration. By coordinating specialized agents, it executes complex workflows with minimal human intervention.

Nonlinear optical properties of molecular materials are governed by aggregation and cooperative effects in the condensed phase, requiring first-principles simulations on large molecular assemblies to achieve a realistic description of experiment. While modern electronic-structure theory implementations mitigate key computational challenges through low-memory two-electron integral engines and multifrequency response solvers, practical calculations of third-order nonlinear optical response in extended systems remain prohibitively expensive. The dominant bottleneck arises because conventional approaches require evaluating many independent components of the rank-four hyperpolarizability tensor, each of which is costly, making orientational averaging prohibitively expensive. Here, we introduce an analytic tensor-averaged formulation of cubic response theory within time-dependent Kohn-Sham density functional theory that removes this spatial bottleneck. By performing the orientational average directly at the level of perturbed densities and transformed Fock matrices, the full Cartesian second-order hyperpolarizability tensor is never constructed. Instead, the isotropic second-order hyperpolarizability is obtained directly. Exploiting the linearity of the Fock-matrix construction, this approach eliminates redundant spatial components and reduces the number of required exchange-correlation kernel integrations in the electronic quartic- and cubic-Hessian contractions by 72 and 90%, respectively. The resulting methodology enables cubic-response simulations of third-harmonic generation in large molecular aggregates with exchange-correlation functionals spanning multiple rungs of Jacob's ladder. Applications to oligo(thiophene-benzothiadiazole) (OTBP) clusters reveal pronounced aggregation-induced dampening of the third-harmonic response and demonstrate scalability to systems with nearly 5800 contracted basis functions. The present framework provides a quantitative tool for disentangling intrinsic molecular effects from supramolecular and morphological contributions, and for assessing the potential for further performance gains through crystal-structure engineering. In this way, theory can directly guide experimental efforts by identifying packing motifs that maximize cooperative enhancement and those that are detrimental. By resolving the missing spatial-domain optimization in the cubic response theory, the present tensor-averaged formulation shifts the dominant computational cost back to the frequency-dependent response solver, enabling routine material-level modeling of cooperative third-order nonlinear optical phenomena.