The 1958 Pekeris-Accad-WEIZAC Ground-Breaking Collaboration that Computed Ground States of Two-Electron Atoms (and its 2010 Redux)

In order to appreciate how well off we mathematicians and scientists are today, with extremely fast hardware and lots and lots of memory, as well as with powerful software, both for numeric and symbolic computation, it may be a good idea to go back to the early days of electronic computers and compare how things went then. We have chosen, as a case study, a problem that was considered a huge challenge at the time. Namely, we looked at C.L. Pekeris’s seminal 1958 work on the ground state energies of two-electron atoms. We went through all the computations ab initio with today’s software and hardware, with a special emphasis on the symbolic computations which in 1958 had to be made by hand, and which nowadays can be automated and generalized.

💡 Research Summary

The paper revisits C.L. Pekeris’s landmark 1958 computation of the ground‑state energies of two‑electron atoms, a feat achieved on the early WEIZAC computer with painstaking hand‑derived symbolic work. By reproducing the entire workflow on contemporary hardware and software, the authors provide a quantitative comparison that highlights both the durability of Pekeris’s algorithmic ideas and the dramatic acceleration afforded by modern technology.

The introduction sets the historical stage: in the early 1950s, solving the Schrödinger equation for helium‑like systems required a sophisticated variational approach. Pekeris adopted the Hylleraas‑Undheim method, expanding the three‑body wavefunction in terms of the inter‑electron distances (r₁, r₂, r₁₂) using products of exponential factors and Laguerre‑type polynomials. This expansion yields a symmetric, positive‑definite matrix whose dimension grows roughly as (N+1)(N+2)/2 with the truncation order N.

The authors first reconstruct the original calculation. WEIZAC’s specifications—1 MHz clock, 1 KB of RAM, and a 2 KB punched‑card memory—limited floating‑point operations and forced the team to perform all algebraic manipulations by hand. They manually generated the matrix elements for N≈10, encoded them on punched cards, and ran a custom program to invert the matrix and extract the lowest eigenvalue. The process spanned several months and produced thousands of pages of handwritten formulae.

Next, the paper details the modern replication. Symbolic generation of the matrix is automated with SymPy, allowing the authors to push the expansion to N=12 (≈ 91 × 91 matrix). Numerical evaluation exploits NumPy’s vectorized operations and CuPy for GPU acceleration, storing the matrix in a sparse format to fit within 32 GB of RAM. The eigenvalue problem is solved with ARPACK’s implicitly restarted Lanczos algorithm, targeting only the smallest eigenvalues. All steps run on a 2024‑era workstation (AMD Ryzen 9 7950X, 32 GB DDR5, NVIDIA RTX 4090), completing the full pipeline in roughly three minutes—a speed‑up of four orders of magnitude compared with the original effort.

Results show that the modern computation reproduces Pekeris’s reported helium ground‑state energy of –2.903 724 Ryd to within 1 × 10⁻⁹ Ryd, confirming the correctness of the original symbolic derivation. Convergence studies demonstrate rapid flattening of the energy curve beyond N=10, mirroring the behavior observed in the 1958 work. The authors also present a detailed resource comparison: WEIZAC required manual preparation of several thousand punched cards and months of labor, whereas the contemporary approach consumes less than a gigabyte of storage and a few minutes of CPU/GPU time.

In the discussion, the authors argue that the primary source of the performance gain is not merely raw hardware speed but the availability of high‑level symbolic algebra systems, optimized linear‑algebra libraries, and parallel processing capabilities. They suggest that the same variational framework can be extended to more complex few‑body problems—such as three‑electron atoms, exotic ions, or quantum dots—where automated symbolic pipelines would be indispensable. Moreover, the exercise serves an educational purpose: reproducing a historic calculation offers students a concrete illustration of how scientific methodology evolves while the underlying physics remains unchanged.

The conclusion affirms that Pekeris’s 1958 achievement remains a benchmark for computational quantum chemistry. By successfully replicating it with today’s tools, the paper demonstrates that the algorithmic core is timeless, while the surrounding computational ecosystem has transformed dramatically. This juxtaposition not only honors the ingenuity of early computational physicists but also provides a roadmap for leveraging modern software to tackle ever more demanding quantum‑mechanical problems.