| CIPSI | |
|---|---|
| Name | CIPSI |
| Developer | Christiane Paoli and collaborators (originators) / community development |
| Introduced | 1973 |
| Genre | Selected configuration interaction method |
| Operating system | Cross-platform |
CIPSI
CIPSI is a selected configuration interaction algorithm originally developed to build compact variational wavefunctions for correlated electrons by iterative selection of important Slater determinants. In quantum mechanics and quantum chemistry, CIPSI matters because it provides a systematically improvable route to near‑full configuration interaction accuracy with far fewer determinants, bridging traditional quantum many‑body theory and modern high‑performance computing approaches used in studying correlated electronic structure and emergent quantum materials.
CIPSI (Configuration Interaction using a Perturbative Selection made Iteratively) is a deterministic selected configuration interaction method that grows a variational subspace by adding determinants chosen according to their estimated second‑order perturbative contribution to the energy. The method dates to work by Hervé Malrieu's group and later formalized by Christian Seigneur? and others in the 1970s and 1980s; it is closely associated in practice with implementations by researchers such as Roland Guillemin, Jean‑Pierre Malrieu, and later contributors in the quantum chemistry community. CIPSI aims to approximate the Full configuration interaction (FCI) solution of the nonrelativistic electronic Schrödinger equation while controlling computational cost via energy‑based selection metrics and perturbative corrections (often denoted E^(2)).
CIPSI is grounded in the variational principle and multi‑reference many‑body perturbation theory. The method constructs a sequence of variational spaces spanned by selected Slater determinants (or configuration state functions) and uses Rayleigh–Schrödinger second‑order perturbation theory to estimate the residual correlation energy. Its theoretical basis connects to Configuration interaction theory, Møller–Plesset perturbation theory, and the concept of sparse representation in the space of determinants; it also relates to modern ideas in reduced density matrices and tensor network approaches such as Density matrix renormalization group (DMRG) when both aim to capture strong static correlation. CIPSI selection criteria often employ Epstein–Nesbet or other partitioning schemes to rank external determinants by their coupling to the current variational state.
Several algorithmic variants of CIPSI exist, differing in selection thresholds, perturbative estimators, and treatment of spin and symmetry. Practical techniques include semi‑stochastic evaluation of the perturbative correction to reduce memory, importance sampling inspired by quantum Monte Carlo strategies, and use of natural orbitals or localized orbitals to accelerate convergence. Modern hybrid algorithms combine CIPSI with selected configuration interaction perturbation theory (sCI+PT), and integrate with stochastic methods like Full configuration interaction quantum Monte Carlo (FCIQMC) and Heat‑bath configuration interaction (HCI). Implementations exploit sparse linear algebra, bitstring-based determinant hashing, and parallelism strategies derived from HPC practices used at centers such as Argonne National Laboratory, Lawrence Berkeley National Laboratory, and university supercomputing facilities.
CIPSI has been applied to compute highly accurate potential energy surfaces, excitation energies, and spectroscopic constants for small to medium molecules where multireference correlation is critical, for example in benchmark studies on dissociation, electron affinity, and ionization potential calculations. In materials science, CIPSI‑based wavefunctions have informed cluster and embedded calculations for correlated solids and molecular magnets, and provided reference data for developing and validating density functional theory (DFT) functionals and emerging quantum algorithms for chemistry on quantum computers (e.g., variational quantum eigensolver benchmarks). CIPSI results often serve as targets in benchmark sets produced by groups at institutions like CNRS, Max Planck Society, University of California, Berkeley, and national computational chemistry collaborations.
CIPSI achieves systematic improvability: by lowering selection thresholds and extending the variational space, energies converge toward the FCI limit, with the perturbative E^(2) providing a practical estimate of the remaining correlation. Benchmarks comparing CIPSI to FCI, DMRG, HCI, and coupled cluster methods (e.g., CCSD(T)) demonstrate competitive accuracy for multireference problems and excited states. Convergence behavior depends on orbital basis choice (e.g., canonical vs natural orbitals), active space selection akin to CASSCF, and system entanglement. Community benchmarking campaigns and datasets from groups such as the Quantum Chemistry Benchmarking Forum and published works in journals like The Journal of Chemical Physics and Chemical Physics Letters document typical errors, scaling trends, and cases where CIPSI remains more efficient than brute‑force methods.
Practical CIPSI codes emphasize memory efficiency, determinant screening, and distributed computing. Public and academic implementations appear in packages and libraries interfacing with Molpro, GAMESS, and standalone projects maintained by academic groups. Performance engineering draws on bitwise operations for determinant representation, pipelined communication patterns, and GPU acceleration in select components. Scalability studies report substantial parallel efficiency on thousands of cores for large determinant spaces, though bottlenecks remain in perturbative accumulation and I/O. Efforts to integrate CIPSI into workflow systems for high‑throughput quantum chemistry and into quantum‑classical co‑design for near‑term quantum devices are active at research centers including CEA, Université Paris‑Saclay, and national labs.
CIPSI's development and dissemination intersect with values of equitable access to scientific tools and reproducible research. Open‑source implementations, shared benchmark datasets, and educational resources lower barriers for researchers globally, including underfunded institutions and scholars in the Global South. Community efforts to document algorithms, provide permissive licensing, and run cloud‑accessible tutorials help counter inequities in access to HPC resources. Responsible deployment includes clear citation practices acknowledging originators and contributors, and collaboration across publicly funded labs (e.g., European Research Council projects, national science agencies) to prioritize broad training, transparent benchmarks, and inclusive participation in advancing high‑accuracy quantum many‑body methods.
Category:Quantum chemistry Category:Computational chemistry methods