OpenWorldLab

Loading interactive view…

Why the table has its shape

Watch it converge

Every atom in this lab is computed, not drawn. Here is how: the solver guesses the charge, finds each electron’s orbital in the field that charge makes, builds a new charge from them, and repeats until the two agree.

What you are looking at

Neon, solved from scratch and replayed one iteration at a time. The cloud on the stage is each iteration’s charge. The charts in the Atom tab follow the change from one iteration to the next falling towards zero, and each orbital’s energy settling.

The real solve takes a few tens of milliseconds; the replay stretches it to about seven seconds.

The physics

Each electron moves in the field of the nucleus and the average charge of all the electrons, V(r)=−Z/r+VH[ρ]+Vxc[ρ]V(r) = -Z/r + V_H[\rho] + V_{xc}[\rho]. Its orbital solves

[−12∇2+V(r)]ψi=εiψi,\left[-\tfrac12 \nabla^2 + V(r)\right] \psi_i = \varepsilon_i \psi_i,

and the orbitals make the charge, ρ=∑ifi∣ψi∣2\rho = \sum_i f_i |\psi_i|^2. The field depends on the charge and the charge on the field, so the solver iterates: a self-consistent field. It starts from the Thomas–Fermi screening of a neutral atom and mixes each new field with the last few (Pulay mixing), so it settles in a few dozen steps instead of oscillating.

When the field stops changing, to one part in a billion, the eigenvalues match NIST’s tables for the same model to within 10⁻⁶ hartree.

What to measure

Count the iterations and compare the final energies with NIST’s.

  1. Count the iterations neon needs, and watch where most of the change happens.
  2. Read the converged 2p energy and compare with NIST SRD 141: −0.498 Ha, −13.5 eV.
  3. Try argon and a heavy atom such as gold.

A common misconception

“Computers find atoms by solving the Schrödinger equation exactly.” No one can for more than two electrons. Practical methods replace the electrons’ interactions by an average field and then make that field consistent with itself. What converges here is that approximation, not the exact atom.

What the model leaves out

  • The atom is spherically averaged and spin-unpolarised: an open subshell’s electrons are spread evenly over its orbitals, with no Hund’s-rule spin alignment and no multiplets.
  • Exchange and correlation are approximated locally (LDA). Each electron feels a little of its own charge, so outer orbitals are too weakly bound.
  • No spin–orbit coupling: the scalar-relativistic option contracts s and p shells but does not split p into p½ and p³⁄₂.

Things to try

  • Turn on relativity and watch gold converge: more iterations, and the 6s orbital settles closer in.
  • Compare the first iteration’s energies with the last: the Thomas–Fermi start is already close, so the cloud changes little, but 1s moves by almost 60 eV and 2s and 2p by about 3 eV.

References

  • W. Kohn and L. J. Sham, Self-consistent equations including exchange and correlation effects, Physical Review 140, A1133 (1965).
  • S. Kotochigova, Z. H. Levine, E. L. Shirley, M. D. Stiles and C. W. Clark, Local-density-functional calculations of the energy of atoms, Physical Review A 55, 191 (1997); NIST Standard Reference Database 141, which this lab’s solver matches.

More experiments, and how the lab works, are in the overview of atoms and orbitals.