Verdict: no acceleration — 0 resolutions in 2026; 1 dated resolution (2003) over 2000–2025
Per-problem resolution timeline
Open problems sit on the right edge. Hover a point for resolver and notes.
Definition
The Clay Mathematics Institute named seven problems in 2000 and attached a
US$1 million prize to each [@clay2000millennium]. A "discovery" in this
series is a row moving to resolved, dated by the year of the resolving
work. The list needs no subproblem splitting and has no contested rows:
seven problems, one resolved, six open.
The one resolved row is the Poincaré conjecture. The ledger dates it 2003,
the year of the last of Perelman's arXiv preprints; the row's notes column
records that the preprints ran 2002 to 2003 and that the prize was announced
in 2010.
Facts
rows: 7 scored; 1 resolved with a dated year; 6 open
by-year: 2003: 1
ai-attributed: 0 of 1 dated resolutions
open rows: bsd, hodge, navier_stokes, p_vs_np, yang_mills, riemann
The collection-wide cumulative index redraws the
ledger as rows remaining:
poincare — Poincaré conjecture
status: resolved
resolved: 2003
resolver: Perelman
notes: arXiv 2002–2003; Clay prize announced 2010
"Nearly a century passed between its formulation in 1904 by Henri Poincaré
and its solution by Grigoriy Perelman, announced in preprints posted on
ArXiv.org in 2002 and 2003."
— Clay Mathematics Institute, Poincaré Conjecture page, read 2026-08-14 [@clay2000millennium]
Method
The seven rows are hand-scored from the Clay Mathematics Institute's own
pages, which the source column names for every row; there is no
fetch.py. The Institute's site lists the Poincaré conjecture under solved
problems and the other six as open, and the ledger mirrors that split. Of
the three candidate dates for the one event — preprints 2002–2003, prize
announced 2010 — the resolved_year takes 2003, the end of the preprint
span.
figure.py calls the shared problem_list_chart() in
../../lib/families.py, which keeps the rows whose
status is resolved with a non-empty resolved_year and counts
resolution events by year from the 2000 list_year to the present. No
ai_problem argument is passed, because no row carries an AI credit. The
cumulative view is the shared ledger_remaining_chart().
check.py recomputes the fact lines and the register entry from
the CSV.
Limitations
sample size. One dated resolution in 26 years; no rate or trend is
estimable from this series.
dating. Preprints 2002–2003, prize announced 2010; the ledger dates
the event 2003, and a different defensible choice moves the only step by
up to seven years.
acknowledgement lag. Seven years passed between the 2003 preprints and
the 2010 prize announcement, so a recent resolution of another row could
predate its appearance here by years.
overlap. riemann is Hilbert row 8a and Smale row 1; p_vs_np is Smale
row 3. The prestige ledgers are not independent samples.
effort. Resolution landmarks are not effort-adjusted discovery rates.
AI attribution
No row in millennium-problems.csv names an AI
system in its resolver or notes columns; the one dated resolution is
Perelman's, dated 2003. No AI credit appears on the Clay Mathematics
Institute's problem pages as of the 2026-08-14 read.
Sources
[@clay2000millennium] — the Institute's own list and status pages, the
ledger source for every row and the register quote.
[@arxiv2026horizonmath] — a 2026 benchmark of over 100 predominantly
unsolved problems chosen so that "verification is computationally
efficient and simple"; frontier models score near 0% on it.
[@sherry2021fast] — measured improvement rates across algorithm families,
including multi-decade stationary stretches, with no AI involved.
