VibeMathedMath problems solved with AI

Sharp ionization bounds for helium and lithium: the exact values of Nc(2)N_c(2) and Nc(3)N_c(3)

For NN nonrelativistic spin-1/21/2 electrons around an infinitely heavy point nucleus of charge ZZ, let EN(Z)E_N(Z) be the ground energy of the Coulomb Hamiltonian HN(Z)H_N(Z) on the fermionic space, and let Nc(Z)N_c(Z) be the largest NN with EN(Z)<EN−1(Z)E_N(Z)<E_{N-1}(Z): the maximum number of electrons the nucleus can bind. Lieb (1984) proved Nc(Z)<2Z+1N_c(Z)<2Z+1. This is sharp for hydrogen but allows up to four electrons for helium and six for lithium, and later general bounds did not lower these two limits. What are the exact values of Nc(2)N_c(2) and Nc(3)N_c(3)?

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI co-developed
Method
Computation
Field
Spectral theory; many-body Schrödinger operators
Posed by
Left open by Lieb's bound (Phys. Rev. A 29, 1984); part of the ionization problem as surveyed by Nam (EMS Newsl. 118, 2020)
Year posed
1984
Years open
42y
Solved
2026-11-12
Model
GPT-6 Astra; Claude Fable 5.2 (Pharos harness)
Vendor
OpenAI; Anthropic
Collaborators
Alec Shelley, Charles F. Yang, Daniel M. Tartakovsky
Verification
Unreviewed
Publication
Preprint
Significance
30 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Theorem 1.1: Nc(2)=2N_c(2)=2 and Nc(3)=4N_c(3)=4. For Z=2Z=2, N≥3N\ge3 and for Z=3Z=3, N≥5N\ge5 the ground energy EN(Z)E_N(Z) is not an eigenvalue of HN(Z)H_N(Z), so the counts also hold when binding means existence of a ground state. Helium binds no extra electron in its ground state; lithium binds exactly one, with electron affinity above 0.0030648600.003064860 hartree. Nonbinding of the dianions uses a weighted form of Lieb's removal identity with a quadratic weight, reduced to finitely many inequalities verified by computer in interval and exact arithmetic. The computer-assisted statements assume the recorded executions and the stated floating-point conditions (Set-up 2.2). Not shown: any Z≥4Z\ge4, the uniform bound Nc(Z)≤Z+CN_c(Z)\le Z+C of the ionization conjecture, or anything about metastable or excited anion states.

What the AI did

From the Hexagon provenance statement: "The ideas were primarily human generated, while the proofs in this version were primarily AI generated. Fable, Astra, and the Pharos harness were all used." The paper's section Use of generative AI adds that the helium and lithium arguments and numerical witnesses were drafted with gpt-6-astra and fable-5.2, then audited, simplified and assembled with Pharos, an automated mathematical reasoning system built on Rethlas, and that this "was automated mathematical review, not proof-assistant verification". The helium-dianion proof and its certificate generators were prepared with fable-5.2. Hexagon labels the text a mix of human-written and AI-generated, with human understanding of some parts.

Verification

No independent mathematician has checked this yet. Read by this site on 19 November 2026: the Hexagon record and the 88-page PDF (hexagon:2611.00325v1) by Alec Shelley, Charles F. Yang and Daniel M. Tartakovsky (Stanford), including Theorem 1.1, the set-up of the computer-assisted statements and the section on generative AI. The statement was compared with Lieb's bound: Nc(Z)<2Z+1N_c(Z)<2Z+1 gives at most four electrons for Z=2Z=2 and six for Z=3Z=3, as the entry says. The results agree with what is known physically: helium has no stable ground-state anion and lithium has one. The proof is computer-assisted with no proof-assistant formalization, and the authors describe its review as automated. The Zenodo supplement (one 1.1 GB archive of certificates, readers and execution records) exists and names the same results, but was not downloaded or re-run here. The proof was not refereed, Hexagon is a moderated repository without peer review, and no expert reaction was found. The OpenAI release claims the uniform bound Nc(Z)≤Z+CN_c(Z)\le Z+C without a value of CC, so it does not give these two numbers and this is not a duplicate of that entry.

Sources

Submitted by MellowGander562 on

Changelog2 changes

Discussion