Top 100 most famous open problems

Open questions ranked by model-estimated familiarity to English-reading audiences. This is not a survey or a ranking by mathematical importance.

Edition How the ranking works

Current Top 100 entries

  1. Rank 001Riemann HypothesisDo all zeros of the Riemann zeta function inside the strip with real part between zero and one lie exactly on its middle line?

    Exact statement

    For the meromorphic continuation of the Riemann zeta function zeta(s)=sum_{n>=1} n^(-s), initially defined for Re(s)>1, every zero s in the open critical strip 0<Re(s)<1 has Re(s)=1/2. This is the Riemann Hypothesis for zeta alone; it does not assert simplicity of its zeros.

    Number theory & arithmetic geometry

    Status

    Open

    Earlier note (2026-09-14): Riemann remains open. September 2026 progress on the proportion of simple critical-line zeros is partial, not proof that every nontrivial zero lies on the line. Uncorroborated or inaccessible proof claims were not accepted. October 7, 2026: OpenAI’s October 6 release reports partial progress. A zero-free half-plane Re(s)>7/8 for zeta excludes only the right-hand part of the critical strip. The target requires every nontrivial zero to have real part exactly 1/2; no conclusion excluding zeros between 1/2 and 7/8 is supplied. This is partial zero-location progress, not the Riemann Hypothesis. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A proof would sharply control fluctuations in the distribution of primes: the prime-counting function would differ from its logarithmic-integral approximation by at most O(√x log x). A disproof would invalidate this equivalent error bound.

    Sources

    Research on ProofAtlas

    Also ranked #2 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #1 of 286
    Math-interested nonspecialists
    #1 of 286
    90% bootstrap rank range
    1–3
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #1 of 286
    OpenAI GPT comparisons only
    #2 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  2. Rank 002P versus NPWhen a yes-or-no problem has short evidence for each yes answer that can be checked in polynomial time, must a deterministic computer also be able to solve the problem in polynomial time? Here polynomial time measures how the work grows with the input size.

    Exact statement

    Determine whether P = NP, that is, whether every decision problem whose solutions can be verified in polynomial time by a deterministic Turing machine can also be solved in polynomial time by a deterministic Turing machine.

    Theoretical computer science

    Status

    Open

    P versus NP remains open on the maintained Clay statement and the bounded claim sweep; no accepted exact-target resolution was established.

    Status reviewed .

    Why it matters

    Equality would give polynomial-time algorithms for every NP decision problem; separation would establish that some efficiently checkable problems resist all such algorithms. This addresses computational efficiency as input size grows.

    Sources

    Research on ProofAtlas

    Also ranked #1 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #2 of 286
    Math-interested nonspecialists
    #3 of 286
    90% bootstrap rank range
    1–3
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #2 of 286
    OpenAI GPT comparisons only
    #1 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  3. Rank 003Collatz conjectureStarting from any positive integer, must repeatedly halving even numbers and replacing odd numbers by three times the number plus one eventually reach one?

    Exact statement

    Prove or disprove that, for every positive integer N, repeated application of N -> N/2 when N is even and N -> 3N+1 when N is odd eventually reaches 1.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would control every trajectory generated by this single arithmetic rule. A disproof would reveal an orbit that never reaches one, through either a different cycle or an unbounded trajectory.

    Sources

    Research on ProofAtlas

    Try the Collatz Explorer

    Also ranked #175 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #3 of 286
    Math-interested nonspecialists
    #2 of 286
    90% bootstrap rank range
    1–3
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #3 of 286
    OpenAI GPT comparisons only
    #3 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  4. Rank 004Binary Goldbach Problem for Even IntegersCan every even integer greater than two be written as the sum of two prime numbers, allowing the same prime to be used twice?

    Exact statement

    Prove or disprove that every even integer n greater than 2 is the sum of two primes.

    Number theory & arithmetic geometry

    Status

    Open

    Goldbach remains open. July/August 2026 sources distinguish the conjecture from almost-prime results; the revised uniformity preprint explicitly does not yield pointwise Goldbach.

    Status reviewed .

    Why it matters

    A proof would guarantee that two prime summands suffice to represent every even integer above two. It would give a universal additive representation across all those integers, however large.

    Sources

    Research on ProofAtlas

    Also ranked #24 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #4 of 286
    Math-interested nonspecialists
    #4 of 286
    90% bootstrap rank range
    4–4
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #4 of 286
    OpenAI GPT comparisons only
    #4 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  5. Rank 005Twin Prime ConjectureAre there infinitely many pairs of prime numbers differing by exactly two, so that such pairs continue to occur however far along the number line we look?

    Exact statement

    Prove that there are infinitely many primes p such that p+2 is also prime.

    Number theory & arithmetic geometry

    Status

    Open

    Twin primes remains open with uncorroborated 2026 proof-claim warnings. Repository hosting and proof titles are not independent mathematical acceptance; bounded gaps, finite computations and prime-pair analogues are not exact gap-two infinitude.

    Status reviewed .

    Why it matters

    A proof would show that the exact gap of two recurs at arbitrarily large primes. It would turn this smallest positive gap between odd primes into an endlessly recurring feature of prime spacing.

    Sources

    Research on ProofAtlas

    Also ranked #20 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #5 of 286
    Math-interested nonspecialists
    #5 of 286
    90% bootstrap rank range
    5–5
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #5 of 286
    OpenAI GPT comparisons only
    #5 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  6. Rank 006Polynomial-time classical integer factorizationIs there a deterministic classical algorithm that finds a nontrivial factor of every composite integer in time polynomial in the number of its binary digits?

    Exact statement

    Determine whether there is a deterministic classical algorithm that, on input a composite integer N in binary, outputs a nontrivial factor of N in time polynomial in log N.

    Theoretical computer science

    Status

    Open

    Deterministic classical polynomial-bit-time factorization remains open. Conditional improvements to exponents in the integer being factored are not polynomial time in its bit length, and quantum factoring is outside this target.

    Status reviewed .

    Why it matters

    An algorithm would guarantee efficient factor finding for every composite input in this model. A negative result would establish that no deterministic classical algorithm can satisfy that polynomial-time bound.

    Sources

    Also ranked #17 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #6 of 286
    Math-interested nonspecialists
    #7 of 286
    90% bootstrap rank range
    6–7
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #8 of 286
    OpenAI GPT comparisons only
    #6 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  7. Rank 007Navier-Stokes Existence and Smoothness: Fefferman AFor an incompressible fluid filling three-dimensional space, with any positive viscosity and no external force, does every smooth, rapidly decaying initial velocity give a flow that stays smooth forever, with uniformly bounded total kinetic energy?

    Exact statement

    Prove Fefferman statement (A): for every viscosity nu > 0 and every real smooth divergence-free Schwartz vector field u0 on R^3, with force f identically zero, there are real velocity u and pressure p in C-infinity(R^3 x [0,infinity)) satisfying partial_t u + (u dot grad)u = nu Delta u - grad p, div u = 0 and u(x,0) = u0(x). The velocity energy integral over R^3 of |u(x,t)|^2 is bounded above by one finite constant for all t >= 0; that constant may depend on u0 and nu. Schwartz means every spatial derivative decays faster than every inverse polynomial, as in official condition (4). Smoothness includes time zero. The selected domain is whole space, not the torus.

    Analysis & PDE

    Status

    Open

    The entry concerns Fefferman A: unforced whole-space Navier–Stokes regularity for smooth rapidly decaying divergence-free data. OpenAI's September claim addresses forced alternatives C/D, not A. This exact target remains open; the broader prize announcement does not establish an accepted proof.

    Status reviewed .

    Why it matters

    This asks whether the specified three-dimensional viscous-flow equations can evolve every allowed starting flow for unlimited time without developing a singularity, while keeping total kinetic energy bounded.

    Sources

    Research on ProofAtlas

    Also ranked #5 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #7 of 286
    Math-interested nonspecialists
    #6 of 286
    90% bootstrap rank range
    6–7
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #6 of 286
    OpenAI GPT comparisons only
    #11 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  8. Rank 008The abc Conjecture of Masser and OesterleSuppose coprime positive integers satisfy a + b = c, and R is the product of the distinct primes dividing abc. For every positive ε, is c less than C(ε) times R^(1 + ε), with one constant C(ε) working for every such triple?Open · disputed claim

    Exact statement

    Prove that for every epsilon > 0, there exists a constant C(epsilon) such that for all coprime positive integers a + b = c, c < C(epsilon) rad(abc)^{1+epsilon}.

    Number theory & arithmetic geometry

    Status

    Open · disputed claim

    abc retains open_disputed_claim. Joshi's February 2025 revision is not adjudicated by the older Scholze–Stix objection to Mochizuki; the July 2026 abc-type proposal is another conjecture. Current acceptance of a full proof was not established.

    Status reviewed .

    Why it matters

    A proof would uniformly restrict how large an additive triple can be relative to its distinct prime factors. A disproof would find a positive epsilon for which no constant controls all such triples.

    Sources

    Research on ProofAtlas

    Also ranked #9 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #9 of 286
    Math-interested nonspecialists
    #8 of 286
    90% bootstrap rank range
    8–11
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #11 of 286
    OpenAI GPT comparisons only
    #7 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  9. Rank 009Yang-Mills Existence and Mass GapCan a nontrivial Yang–Mills quantum field theory in four-dimensional spacetime be constructed rigorously for every compact simple gauge group—the choice of internal symmetry—with a positive, finite gap between vacuum energy and the remaining energy spectrum?

    Exact statement

    For every compact simple gauge group G, prove that a nontrivial four-dimensional quantum Yang-Mills theory on R^4 exists with all existence requirements of Jaffe-Witten, Quantum Yang-Mills Theory, sections 3-4, including axiomatic properties at least as strong as their cited references [45,35], the required gauge-invariant local fields and short-distance compatibility specified there. Its vacuum Hamiltonian H has no spectrum in (0,Delta) for some Delta > 0, and the supremum of such Delta is finite. The official reconstruction and existence requirements are retained by reference, not replaced by an unrestricted equivalence of arbitrary Euclidean and Minkowski theories. An isolated first excited eigenstate, confinement and asymptotic completeness are not additional requirements.

    Mathematical physics

    Status

    Open

    The universal four-dimensional Yang–Mills existence/mass-gap target remains open. Three-dimensional outlines, individual gauge-group claims and withdrawn postings do not establish the full target.

    Status reviewed .

    Why it matters

    The construction would turn the specified four-dimensional quantum gauge-field model into a theory satisfying the required mathematical axioms. Its spectrum would have a positive, finite gap above vacuum energy, without requiring an isolated first excited eigenstate.

    Sources

    Also ranked #3 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #8 of 286
    Math-interested nonspecialists
    #10 of 286
    90% bootstrap rank range
    8–10
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #10 of 286
    OpenAI GPT comparisons only
    #8 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  10. Rank 010Birch and Swinnerton-Dyer ConjectureFor an elliptic curve over the rational numbers, its rank counts the independent infinite-order generators of its group of rational points. Does this rank equal the order of vanishing of its L-function at s = 1—the multiplicity of the zero there, taken as zero when the function is nonzero?

    Exact statement

    Prove or disprove that, for every elliptic curve E over the rational numbers, the rank of the finitely generated abelian group E(Q) of rational points equals the order of vanishing at s = 1 of the associated Hasse–Weil L-function L(E, s).

    Number theory & arithmetic geometry

    Status

    Open

    Earlier note (2026-09-14): BSD rank equality remains open. The located all-rank working-draft claim lacks independent acceptance; recent uses and special cases do not supply a universal proof. October 7, 2026: OpenAI’s October 6 release reports partial progress. The target is equality of analytic and Mordell-Weil ranks for every E/Q, rather than the full leading-term formula. This family gives that equality and the stronger leading-term conclusion when the full q-power Selmer corank is zero or one for some prime q. It leaves curves outside that hypothesis untreated, so it is a low-corank subcase only. The companion proves rank equality for E/Q with nonzero rational two-torsion and full 2-power Selmer corank zero or one, allowing arbitrary reduction at 2. The BSD rank target includes every E/Q without either restriction, so this is a limited subcase and does not follow as a full closure from the H10 result. The low-corank 2-converse explicitly gives equality of analytic and Mordell-Weil ranks for E/Q whose full 2-power Selmer corank is zero or one. The family also supplies analytic rank densities and their mean over twists, but an average or density-one assertion does not prove rank equality for every curve or every twist. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A proof would identify an algebraic measure of the curve’s rational points with an analytic measurement of its L-function. A counterexample would locate a curve where those two measurements differ.

    Sources

    Research on ProofAtlas

    Also ranked #6 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #12 of 286
    Math-interested nonspecialists
    #11 of 286
    90% bootstrap rank range
    9–13
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #15 of 286
    OpenAI GPT comparisons only
    #9 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  11. Rank 011Existence of an odd perfect numberIs there an odd positive integer whose positive divisors, including the integer itself, sum to exactly twice that integer, or must every perfect number be even?

    Exact statement

    Determine whether there exists an odd positive integer N satisfying sigma(N)=2N, where sigma is the sum-of-positive-divisors function; equivalently, prove that every perfect number is even or construct an odd perfect number.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A positive answer would establish a perfect number of odd parity. A negative answer would impose an absolute parity restriction on every integer satisfying this divisor-sum equation.

    Sources

    Research on ProofAtlas

    Also ranked #221 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #13 of 286
    Math-interested nonspecialists
    #9 of 286
    90% bootstrap rank range
    9–13
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #9 of 286
    OpenAI GPT comparisons only
    #12 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  12. Rank 012Normality of PiDo π’s digits give every pattern of the same length the same long-run frequency, no matter which integer base, from base 2 onward, is used?

    Exact statement

    Prove that pi is a normal number: in every integer base b >= 2, every block of k digits occurs in the base-b expansion of pi with asymptotic frequency b^{-k}.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would establish uniform limiting frequencies for every finite digit pattern across every base. A counterexample would identify a specific base and block whose frequency differs or fails to exist.

    Sources

    Also ranked #408 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #10 of 286
    Math-interested nonspecialists
    #12 of 286
    90% bootstrap rank range
    9–13
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #7 of 286
    OpenAI GPT comparisons only
    #17 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  13. Rank 013Hodge ConjectureHodge theory singles out certain rational cohomology classes, which encode topological features of a smooth complex projective variety. Must every such class be a rational combination of classes represented by algebraic subvarieties?

    Exact statement

    Determine whether every rational Hodge class on a smooth complex projective algebraic variety is a rational linear combination of classes of algebraic subvarieties (algebraic cycles).

    Geometry & topology

    Status

    Open

    Earlier note (2026-09-14): The rational Hodge conjecture remains open. Conditional witness axioms, selected Fermat varieties and inaccessible full-proof claims do not establish the all-varieties target. October 7, 2026: OpenAI’s October 6 release reports partial progress. The target quantifies over every smooth complex projective variety and every rational Hodge class. The family proves that assertion for CM abelian varieties, arbitrary products of projective K3 surfaces, and the specified additional abelian/cover families, in the stated codimensions. These are substantial subclasses, not all smooth projective varieties; none of the abstracts gives a reduction of the universal target to them. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    Cohomology records features of a variety, while algebraic cycles are built from its algebraic subvarieties. A proof would show that every rational class of the specified Hodge type can be accounted for by those algebraic pieces.

    Sources

    Research on ProofAtlas

    Also ranked #4 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #11 of 286
    Math-interested nonspecialists
    #14 of 286
    90% bootstrap rank range
    10–13
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #13 of 286
    OpenAI GPT comparisons only
    #10 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  14. Rank 014Infinitude of Mersenne PrimesAre there infinitely many prime exponents p for which the number 2^p−1 is itself prime, and hence infinitely many even perfect numbers?

    Exact statement

    Prove or disprove that there are infinitely many primes p for which the Mersenne number 2^p - 1 is prime. Equivalently (by the Euclid-Euler theorem), prove or disprove that there are infinitely many even perfect numbers.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A resolution would decide whether these two linked families continue indefinitely or eventually stop. It would settle infinitude without requiring a formula for locating every member.

    Sources

    Also ranked #234 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #14 of 286
    Math-interested nonspecialists
    #13 of 286
    90% bootstrap rank range
    14–16
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #12 of 286
    OpenAI GPT comparisons only
    #14 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  15. Rank 015Sums of Three Cubes Representation ConjectureApart from integers leaving remainder 4 or 5 when divided by 9, can every integer be written as a sum of three integer cubes? More strongly, does each allowed integer have infinitely many such representations?Open · subcases solved

    Exact statement

    Prove that every integer k not congruent to 4 or 5 modulo 9 admits a representation k = x^3 + y^3 + z^3 in integers, with Heath-Brown's refinement asking for infinitely many such representations for each eligible k.

    Number theory & arithmetic geometry

    Status

    Open · subcases solved

    Status review date not recorded in this edition.

    Why it matters

    Existence would show that these congruence exclusions account for every failure of representation. The stronger infinitude statement would ensure that each eligible integer has endlessly many representations, a separate requirement from finding one.

    Sources

    Also ranked #379 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #16 of 286
    Math-interested nonspecialists
    #15 of 286
    90% bootstrap rank range
    14–16
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #14 of 286
    OpenAI GPT comparisons only
    #15 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  16. Rank 016Beal's Conjecture (Tijdeman-Zagier Conjecture)Whenever positive integers satisfy Aˣ+Bʸ=Cᶻ with all three exponents greater than two, must a single prime divide all three bases A, B, and C?

    Exact statement

    Prove that if A^x + B^y = C^z with A, B, C, x, y, z positive integers and x, y, z all greater than 2, then A, B, and C have a common prime factor.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would rule out every such equation whose three bases have no common prime factor. A counterexample would exhibit an admissible power-sum identity without that shared divisibility.

    Sources

    Research on ProofAtlas

    Also ranked #328 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #15 of 286
    Math-interested nonspecialists
    #20 of 286
    90% bootstrap rank range
    15–16
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #16 of 286
    OpenAI GPT comparisons only
    #16 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  17. Rank 017Shortest superpermutationsFor any number of symbols, what is the shortest string that contains every ordering of those symbols as consecutive characters?Open · subcases solved

    Exact statement

    For each positive integer n, determine the minimum length of a word over n symbols that contains every permutation of the n symbols as a contiguous substring.

    Combinatorics & discrete geometry

    Also known as

    Minimal Superpermutation Length Problem

    Status

    Open · subcases solved

    The full question remains open; some special cases have been solved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #17 of 286
    Math-interested nonspecialists
    #21 of 286
    90% bootstrap rank range
    17–18
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #24 of 286
    OpenAI GPT comparisons only
    #13 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  18. Rank 018Magic square of squaresCan nine distinct positive square numbers fill a 3 × 3 grid so that every row, column and main diagonal has the same sum?Open · disputed claim

    Exact statement

    Determine whether there exists a 3 by 3 magic square whose nine entries are distinct positive perfect squares and whose three rows, three columns, and two main diagonals all have the same sum.

    Number theory & arithmetic geometry

    Also known as

    Magic Square of Squares

    Status

    Open · disputed claim

    A manuscript claims that no such square exists. The claim has not established a resolution in the evidence reviewed.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #18 of 286
    Math-interested nonspecialists
    #16 of 286
    90% bootstrap rank range
    17–22
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #17 of 286
    OpenAI GPT comparisons only
    #20 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  19. Rank 019Multiplicative persistenceIf you repeatedly multiply a number’s decimal digits together, must you reach a single digit within 11 steps?

    Exact statement

    Given n in N, let Pi(n) be the product of all its digits, and set P_m(n) the first positive integer such that Pi^{P_m(n)}(n) = Pi^{P_m(n)+1}(n). Determine whether for all n in N, P_m(n) <= 11.

    Probability, ergodic theory & dynamics

    Also known as

    Conjecture of multiplicative persistence: P_m(n) <= 11

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #19 of 286
    Math-interested nonspecialists
    #26 of 286
    90% bootstrap rank range
    18–22
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #29 of 286
    OpenAI GPT comparisons only
    #18 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  20. Rank 020Square Peg Problem (Toeplitz's Inscribed Square Conjecture)Must every continuous closed loop in the plane with no self-intersections contain four points forming a square, even when the loop is jagged or otherwise not smooth?Open · subcases solved

    Exact statement

    Prove or disprove that every continuous simple closed (Jordan) curve in the plane contains four points forming the vertices of a square.

    Geometry & topology

    Status

    Open · subcases solved

    Status review date not recorded in this edition.

    Why it matters

    A proof would guarantee the same inscribed shape on every Jordan curve, even when its geometry is irregular. A counterexample would identify a continuous closed curve that avoids every square.

    Sources

    Also ranked #431 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #20 of 286
    Math-interested nonspecialists
    #18 of 286
    90% bootstrap rank range
    18–22
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #18 of 286
    OpenAI GPT comparisons only
    #21 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  21. Rank 021Generalized Riemann Hypothesis for Dirichlet L-functionsDirichlet L-functions extend the zeta-function setting by using periodic arithmetic weights. For every Dirichlet character, do all zeros of its L-function with real part between zero and one lie on the middle line, with real part one-half?

    Exact statement

    For every integer q>=1 and every Dirichlet character chi modulo q, including principal and imprimitive characters, every zero s of the meromorphic continuation of L(s,chi)=sum_{n>=1}chi(n)n^(-s) in 0<Re(s)<1 satisfies Re(s)=1/2. The continuation is from Re(s)>1. Only open-strip zeros are quantified: principal-character poles and boundary zeros of imprimitive Euler factors are not counterexamples. No simplicity or central nonvanishing requirement is added.

    Number theory & arithmetic geometry

    Status

    Open

    Earlier note (2026-09-14): Dirichlet GRH remains open. No independently established full-target resolution was obtained; neighboring Riemann-family claims and source-access limitations remain qualified. October 7, 2026: OpenAI’s October 6 release reports partial progress. The theorem includes every Dirichlet L-function and allows the principal pole, but its zero-free boundary is 7/8 rather than 1/2. Thus its character quantification agrees broadly with the target while its zero-location conclusion is strictly weaker. The GRH target remains unresolved. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A proof would bound how far one must search for a prime in an arithmetic progression. For q > 3 and a coprime to q, the least prime congruent to a modulo q would be at most (φ(q) log q)², where φ(q) counts the residue classes coprime to q.

    Sources

    Also ranked #10 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #21 of 286
    Math-interested nonspecialists
    #24 of 286
    90% bootstrap rank range
    18–25
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #21 of 286
    OpenAI GPT comparisons only
    #23 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  22. Rank 022Chromatic Number of the Plane (Hadwiger--Nelson Problem)Allowing arbitrary colorings, with no requirement that color classes be measurable, what is the fewest colors needed for every point of the Euclidean plane so that points exactly one unit apart have different colors?

    Exact statement

    Determine the chromatic number of the Euclidean plane: the minimum number of colours needed to colour every point so that points at distance exactly one have distinct colours, with ordinary arbitrary colourings in ZFC and no measurability requirement.

    Combinatorics & discrete geometry

    Status

    Open

    Earlier note (null): Open; primary sources give 5 <= chi(R^2) <= 7. No exact-value resolution was located in the review. October 7, 2026: OpenAI’s October 6 release reports partial progress. The source rules out arbitrary five-colorings of the plane and explicitly requires no regularity of color classes, matching the full ordinary-coloring model. Together with the seven-color upper bound it narrows the answer to six or seven. The exact chromatic number is not determined, so this is partial progress and must not remove the full problem. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A resolution would pair a coloring of the entire plane with a proof that fewer colors cannot suffice. It would determine the exact coloring cost of this single geometric distance constraint.

    Sources

    Also ranked #213 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #22 of 286
    Math-interested nonspecialists
    #19 of 286
    90% bootstrap rank range
    19–27
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #26 of 286
    OpenAI GPT comparisons only
    #19 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  23. Rank 023Ramsey number R(5,5)What is the smallest number of vertices such that, however the edges between them are colored red or blue, some five vertices have all ten edges between them in one color?Open · subcases solved

    Exact statement

    Determine the smallest integer N such that every red/blue coloring of the edges of the complete graph K_N contains a monochromatic K_5; equivalently determine the diagonal Ramsey number R(5,5).

    Combinatorics & discrete geometry

    Also known as

    Classical Ramsey Number R(5,5)

    Status

    Open · subcases solved

    Bounds are known, but the exact minimum is unresolved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #27 of 286
    Math-interested nonspecialists
    #17 of 286
    90% bootstrap rank range
    21–28
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #20 of 286
    OpenAI GPT comparisons only
    #25 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  24. Rank 024BQP Versus NP: Complete Language-Class RelationHow do efficient quantum solving and efficient classical verification compare for yes-or-no problems? Does quantum computation with bounded error solve every problem with short, efficiently checkable evidence for yes answers—and do all problems it solves have such evidence?

    Exact statement

    Determine the truth values of BOTH NP subseteq BQP and BQP subseteq NP for unrelativized, advice-free classes of total languages over {0,1}. NP consists of languages with polynomial-length binary witnesses and a deterministic polynomial-time verifier. BQP consists of languages decided by polynomial-time uniformly generated polynomial-size quantum circuits (a deterministic Turing machine outputs Q_n on 1^n), using Watrous's fixed Toffoli, Hadamard and phase-i gates with |0> ancillas and discarding, followed by a standard-basis output measurement. On every yes input acceptance is at least 2/3, and on every no input it is at most 1/3. No arbitrary real gate constants, advice or oracle are available. Establish equality, either proper containment, or incomparability.

    Theoretical computer science

    Status

    Open

    Both unrelativized BQP/NP inclusions remain unresolved on the checked evidence. Random-permutation oracle separations do not settle the stated uniform total-language relationship.

    Status reviewed .

    Why it matters

    Resolving both inclusions would determine whether these computational classes coincide, one strictly contains the other, or neither contains the other. Each outcome gives a different relationship between efficient verification and bounded-error quantum computation.

    Sources

    Also ranked #16 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #23 of 286
    Math-interested nonspecialists
    #36 of 286
    90% bootstrap rank range
    21–30
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #23 of 286
    OpenAI GPT comparisons only
    #37 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  25. Rank 025Irrationality of Euler's ConstantIs Euler's constant, the limiting difference between the harmonic sum 1+1/2+⋯+1/n and the natural logarithm of n as n grows, an irrational number?

    Exact statement

    Prove or disprove that Euler's constant gamma = lim_{n->infty} (1 + 1/2 + ... + 1/n - log n) = 0.5772156649... is irrational.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A resolution would decide whether this limit can equal a ratio of integers. It would settle that exact arithmetic distinction, independently of how accurately its decimal expansion is computed.

    Sources

    Also ranked #258 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #26 of 286
    Math-interested nonspecialists
    #22 of 286
    90% bootstrap rank range
    22–27
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #22 of 286
    OpenAI GPT comparisons only
    #27 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  26. Rank 026Graph Isomorphism in Polynomial TimeCan a deterministic classical algorithm decide in polynomial time whether any two finite graphs describe the same connections after their vertices are relabeled?

    Exact statement

    Determine whether the graph isomorphism problem for general finite graphs is decidable by a deterministic classical Turing machine in polynomial time.

    Theoretical computer science

    Status

    Open

    General graph isomorphism in deterministic polynomial time remains open. Heated-space methods explicitly lack all-input guarantees; the older Xue claim was withdrawn after an author-reported counterexample. A separate inaccessible 2026 claim was not accepted.

    Status reviewed .

    Why it matters

    A positive answer would give an efficient general procedure for recognizing identical graph structure despite different labels. A negative answer would establish a polynomial-time barrier even for this precisely defined comparison task.

    Sources

    Also ranked #79 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #29 of 286
    Math-interested nonspecialists
    #25 of 286
    90% bootstrap rank range
    23–33
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #32 of 286
    OpenAI GPT comparisons only
    #22 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  27. Rank 027Erdos--Szekeres Convex Polygon ConjectureFor every n≥3, is 2^(n−2)+1 the smallest number of points in the plane that guarantees n vertices of a convex polygon when no three points lie on a line?Open · subcases solved

    Exact statement

    Prove or disprove that ES(n)=2^(n-2)+1 for every integer n>=3, where ES(n) is the least N such that every set of N points in the Euclidean plane in general position (no three collinear) contains n points in convex position.

    Combinatorics & discrete geometry

    Status

    Open · subcases solved

    Open with solved small cases. The SoCG 2025 paper explicitly keeps the planar conjecture open; its counterexample concerns an ordered-hypergraph generalization.

    Status review date not recorded in this edition.

    Why it matters

    A proof would establish the proposed exact threshold for every polygon size. A counterexample would refute that formula for at least one size, without determining the whole threshold function.

    Sources

    Also ranked #263 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #24 of 286
    Math-interested nonspecialists
    #43 of 286
    90% bootstrap rank range
    23–34
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #19 of 286
    OpenAI GPT comparisons only
    #63 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  28. Rank 028Infinitude of Fermat PrimesAre there infinitely many indices n≥0 for which the Fermat number 2^(2^n)+1 is prime, or do prime values eventually stop appearing?

    Exact statement

    Determine whether infinitely many Fermat numbers F_n=2^(2^n)+1 are prime; equivalently, prove that the set of n≥0 with F_n prime is infinite or prove that it is finite.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    Infinitude would guarantee prime Fermat numbers beyond every index bound. Finiteness would establish a bound after which every number in this particular sequence is composite.

    Sources

    Also ranked #410 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #30 of 286
    Math-interested nonspecialists
    #30 of 286
    90% bootstrap rank range
    26–33
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #25 of 286
    OpenAI GPT comparisons only
    #35 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  29. Rank 029Kelvin’s least-area partition problemHow can space be divided into cells of equal volume using the least average area of shared walls?Open · subcases solved

    Exact statement

    Determine a partition of three-dimensional Euclidean space into cells of equal unit volume that minimizes the average surface area of the interfaces (perimeter density), and in particular decide whether the Weaire-Phelan structure is optimal.

    Geometry & topology

    Status

    Open · subcases solved

    Status review date not recorded in this edition.

    Sources

    Also ranked #498 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #25 of 286
    Math-interested nonspecialists
    #44 of 286
    90% bootstrap rank range
    26–35
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #33 of 286
    OpenAI GPT comparisons only
    #28 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  30. Rank 030Unconditional Separation of Quantum and Classical Bounded-Error Polynomial Time (BQP != BPP)Can quantum computers solve some decision problem in polynomial time with bounded error that every classical randomized polynomial-time algorithm fails to solve with bounded error?

    Exact statement

    Prove unconditionally that bounded-error quantum polynomial time (BQP) strictly contains bounded-error probabilistic polynomial time (BPP).

    Quantum information & computation

    Status

    Open

    Unconditional unrelativized BQP versus BPP remains open. The September Fourier-hierarchy result is oracle-relative and does not establish an ordinary language-class separation.

    Status reviewed .

    Why it matters

    A proof would establish an unconditional separation between these quantum and classical computational resources. A negative resolution would equate their polynomial-time decision power when both are allowed bounded error.

    Sources

    Also ranked #19 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #28 of 286
    Math-interested nonspecialists
    #34 of 286
    90% bootstrap rank range
    24–38
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #44 of 286
    OpenAI GPT comparisons only
    #24 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  31. Rank 031Matrix-Multiplication Exponent Equals TwoCan two n-by-n rational matrices be multiplied exactly, without division, with a number of arithmetic operations arbitrarily close to quadratic: O(n^(2 + ε)) for every positive ε? Each rational arithmetic operation counts as one step, regardless of the number of bits in its operands.

    Exact statement

    Let R_Q(n) be the tensor rank over Q of the bilinear map multiplying two n by n rational matrices: the least r for which AB is a sum of r products alpha_j(A) beta_j(B) C_j, with rational linear forms alpha_j, beta_j and rational output matrices C_j. Set omega_Q = inf{tau in R : R_Q(n) = O(n^tau)}. Prove omega_Q = 2. Equivalently, for every epsilon > 0, exact multiplication admits division-free arithmetic circuits over Q of size O(n^(2+epsilon)), counting rational additions, subtractions, multiplications and constant scalar operations at unit cost. Circuits need not be uniformly generated and rational coefficient bit lengths are not charged. Constants and thresholds may depend on epsilon. Literal O(n^2) attainment is not required.

    Theoretical computer science

    Status

    Open

    Earlier note (2026-09-14): The exact rational-field matrix multiplication exponent target remains open. The August 2026 claimed exponent below 2.371177 is not exponent 2; complex-field results require field/scope review before transfer to Q. October 7, 2026: OpenAI’s October 6 release reports partial progress. The primary manuscript gives omega<=9/4 over C; the secondary manuscript explicitly gives omega<2.258 over every characteristic-zero field, including the full field Q. The all-field companion supplies a weaker square bound and the rectangular exponent improvement concerns a different shape. These are arithmetic upper-bound improvements but none gives omega_Q=2. In particular the strongest complex-only headline is not silently restated as an explicit rational-field theorem. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A proof would allow exact matrix multiplication with arithmetic exponents arbitrarily close to two and carry through applicable algebraic reductions to other computations. The statement measures arithmetic operations; it leaves coefficient bit costs and literal quadratic attainment outside its guarantee.

    Sources

    Also ranked #27 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #34 of 286
    Math-interested nonspecialists
    #27 of 286
    90% bootstrap rank range
    24–37
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #27 of 286
    OpenAI GPT comparisons only
    #32 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  32. Rank 032Perfect cuboidCan a rectangular box have positive whole-number lengths for all its edges, face diagonals and space diagonal?

    Exact statement

    Determine whether there exist positive integers a,b,c,d,e,f,g satisfying a^2+b^2=d^2, a^2+c^2=e^2, b^2+c^2=f^2, and a^2+b^2+c^2=g^2.

    Number theory & arithmetic geometry

    Also known as

    Perfect Cuboid Problem

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #36 of 286
    Math-interested nonspecialists
    #23 of 286
    90% bootstrap rank range
    26–34
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #34 of 286
    OpenAI GPT comparisons only
    #26 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  33. Rank 033Existence of One-Way FunctionsDo one-way functions exist: functions that a deterministic computer can evaluate in polynomial time, but that every randomized polynomial-time algorithm reverses on a random input’s output with success eventually smaller than every inverse power of the input length? Reversing means finding any input with that output.

    Exact statement

    Prove or disprove that there exists a deterministic polynomial-time computable function family f_n:{0,1}^n->{0,1}^{poly(n)} such that every probabilistic polynomial-time algorithm inverts f_n on a uniformly random input with only negligible probability.

    Cryptography, coding, information & optimization

    Status

    Open

    Unconditional existence of one-way functions remains open. Conditional constructions and hardness characterizations do not constitute an unconditional solved subcase of this existential target.

    Status reviewed .

    Why it matters

    A construction would unconditionally separate efficient forward computation from hard inversion on random inputs, even when the inverter uses randomness. Every randomized polynomial-time inverter would have only negligible success probability as input length grows.

    Sources

    Research on ProofAtlas

    Also ranked #8 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #35 of 286
    Math-interested nonspecialists
    #31 of 286
    90% bootstrap rank range
    26–38
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #38 of 286
    OpenAI GPT comparisons only
    #31 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  34. Rank 034Lonely runner conjectureMust each of n runners on a circular track, starting together at different constant speeds, at some moment be at least 1/n of a lap from every other runner?

    Exact statement

    Prove that for every integer n>=1 and every set of n distinct positive integer speeds s_1,...,s_n, there exists a real time t such that ||s_i t|| >= 1/(n+1) for every i, where ||x|| is distance to the nearest integer. Equivalently, among n+1 runners with pairwise distinct constant real speeds starting together on the unit circle, each runner is at some time at circular distance at least 1/(n+1) from every other runner.

    Combinatorics & discrete geometry

    Status

    Open

    Status review date not recorded in this edition.

    Sources

    Also ranked #485 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #38 of 286
    Math-interested nonspecialists
    #32 of 286
    90% bootstrap rank range
    30–40
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #41 of 286
    OpenAI GPT comparisons only
    #29 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  35. Rank 035Legendre's Conjecture on Primes Between Consecutive SquaresFor every positive integer n, is there always at least one prime number strictly between the consecutive squares n² and (n+1)²?

    Exact statement

    Prove that for every positive integer n there exists a prime p with n^2 < p < (n+1)^2.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would guarantee that none of these increasingly long intervals is empty of primes. A counterexample would identify a specific gap covering an entire interval between consecutive squares.

    Sources

    Research on ProofAtlas

    Also ranked #282 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #43 of 286
    Math-interested nonspecialists
    #28 of 286
    90% bootstrap rank range
    30–39
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #31 of 286
    OpenAI GPT comparisons only
    #34 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  36. Rank 036Graham’s number problemWhat is the smallest cube dimension in which every red/blue coloring of all vertex-to-vertex connections forces four coplanar vertices with all six connections the same color?

    Exact statement

    For n>=1, let K(Q_n) be the complete graph on Q_n={-1,+1}^n embedded in R^n. Let N* be the least n such that every red-blue colouring of all edges of K(Q_n) contains a monochromatic K4 whose four vertices lie in one affine two-dimensional plane. Determine the exact integer N*. Finite existence is known; by restricting a colouring to a cube facet, the defining property then holds for every dimension at least N*.

    Combinatorics & discrete geometry

    Also known as

    Graham-Rothschild Hypercube Ramsey Problem (Exact Planar K4 Threshold)

    Status

    Open

    Graham’s number is an upper bound associated with this problem, not its exact answer. The smallest dimension remains unresolved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #32 of 286
    Math-interested nonspecialists
    #38 of 286
    90% bootstrap rank range
    28–42
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #28 of 286
    OpenAI GPT comparisons only
    #61 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  37. Rank 037Exact Kissing Numbers in Higher DimensionsIn every dimension greater than four, what is the maximum number of equal unit spheres that can touch one central unit sphere without overlapping one another?Open · subcases solved

    Exact statement

    For every integer n > 4, determine the exact kissing number tau_n: the maximum number of pairwise nonoverlapping unit balls in Euclidean R^n that can touch a central unit ball of the same radius.

    Cryptography, coding, information & optimization

    Status

    Open · subcases solved

    Open with solved instances n=8 and n=24 within the target range; the March 2026 primary source still records a gap between 40 and 44 in dimension five. No full-sequence resolution was located. October 6, 2026: A four-dimensional packing-density/24-cell claim does not determine exact kissing numbers in every dimension n>4. The complete proof was not audited in this review.

    Status reviewed .

    Why it matters

    Exact kissing numbers would determine the largest possible local contact arrangement in each dimension. A complete answer needs both an arrangement attaining the number and a proof that more contacts are impossible.

    Sources

    Also ranked #103 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #44 of 286
    Math-interested nonspecialists
    #29 of 286
    90% bootstrap rank range
    32–39
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #37 of 286
    OpenAI GPT comparisons only
    #33 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  38. Rank 038Infinitely many amicable pairsAre there infinitely many pairs of different positive integers where each equals the sum of the other’s positive divisors smaller than itself?

    Exact statement

    Prove or disprove that there are infinitely many amicable pairs, i.e., pairs m != n with s(m) = n and s(n) = m, where s is the sum-of-proper-divisors function.

    Number theory & arithmetic geometry

    Also known as

    Infinitude of Amicable Pairs

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #31 of 286
    Math-interested nonspecialists
    #48 of 286
    90% bootstrap rank range
    32–44
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #30 of 286
    OpenAI GPT comparisons only
    #56 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  39. Rank 039No three in a lineOn a square grid of any size, what is the largest number of grid points you can choose with no three lying on a straight line?Open · subcases solved

    Exact statement

    Determine the maximum number of points that can be placed in an n x n grid with no three points collinear; in particular, determine whether the maximum is 2n for every n.

    Scope. A 1 × 1 grid has just one point. The 2n question concerns grids with n ≥ 2.

    Combinatorics & discrete geometry

    Also known as

    No-Three-in-Line Problem on the n by n Grid

    Status

    Open · subcases solved

    The full question remains open; some special cases have been solved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #37 of 286
    Math-interested nonspecialists
    #39 of 286
    90% bootstrap rank range
    33–44
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #45 of 286
    OpenAI GPT comparisons only
    #30 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  40. Rank 040Infinitely many Fibonacci primesAre infinitely many numbers in the Fibonacci sequence prime?

    Exact statement

    Prove or disprove that infinitely many Fibonacci numbers F_n are prime.

    Number theory & arithmetic geometry

    Also known as

    Infinitude of Fibonacci Primes

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #33 of 286
    Math-interested nonspecialists
    #45 of 286
    90% bootstrap rank range
    35–44
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #35 of 286
    OpenAI GPT comparisons only
    #51 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  41. Rank 041Landau's Fourth Problem on Primes of the Form n^2+1Are there infinitely many positive integers n for which n² + 1 is prime, so that prime numbers occur indefinitely immediately after perfect squares?

    Exact statement

    Prove or disprove that there are infinitely many primes of the form n^2+1 with n a positive integer.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would guarantee endlessly many prime values of this particular quadratic polynomial. A negative answer would mean that every sufficiently large value of n² + 1 is composite.

    Sources

    Also ranked #184 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #49 of 286
    Math-interested nonspecialists
    #33 of 286
    90% bootstrap rank range
    38–45
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #42 of 286
    OpenAI GPT comparisons only
    #36 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  42. Rank 042Infinitude of Sophie Germain PrimesAre there infinitely many primes p for which doubling p and adding one also gives a prime, forming a pair of primes p and 2p+1?

    Exact statement

    Determine whether there are infinitely many Sophie Germain primes, that is, primes p such that 2p + 1 is also prime.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A resolution would decide whether this linked prime pattern occurs at arbitrarily large scales. A positive answer would establish infinitude without requiring a count or density formula for the pairs.

    Sources

    Also ranked #481 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #48 of 286
    Math-interested nonspecialists
    #35 of 286
    90% bootstrap rank range
    35–46
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #39 of 286
    OpenAI GPT comparisons only
    #40 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  43. Rank 043Erdős–Straus conjectureFor every integer n ≥ 2, can 4/n be written as 1/x + 1/y + 1/z with positive whole numbers x, y and z?Open · disputed claim

    Exact statement

    For every integer n >= 2, do there exist positive integers x, y, z such that 4/n = 1/x + 1/y + 1/z? Specific residue classes modulo 840 and computational ranges (verified to n <= 10^18 as of a 2025 preprint) are settled, but no general proof or counterexample for all n is known.

    Number theory & arithmetic geometry

    Also known as

    Erdős–Straus Conjecture

    Status

    Open · disputed claim

    A 2026 manuscript claims a proof, while a later paper still describes the general question as open. Checks through 10¹⁸ establish a finite range, not the full conjecture.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #50 of 286
    Math-interested nonspecialists
    #37 of 286
    90% bootstrap rank range
    40–46
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #43 of 286
    OpenAI GPT comparisons only
    #38 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  44. Rank 044NP versus coNP ProblemWhenever yes answers to a decision problem have polynomial-length certificates checkable in polynomial time, do its no answers always have such certificates as well?

    Exact statement

    Determine whether NP = coNP, that is, whether the class NP is closed under complementation. Equivalently, by the Cook-Reckhow criterion, decide whether there exists a polynomially bounded propositional proof system.

    Theoretical computer science

    Status

    Open

    NP versus coNP remains open with an explicit claim warning. Lin's August 2026 v28 and Gordeev–Haeusler's May 2026 revision make opposing claims; older independent critiques do not automatically adjudicate those revisions. Restricted proof-system lower bounds are not the full separation.

    Status reviewed .

    Why it matters

    Equality would make NP closed under exchanging yes and no answers and provide a polynomially bounded propositional proof system. Separation would rule out every such proof system.

    Sources

    Also ranked #11 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #42 of 286
    Math-interested nonspecialists
    #42 of 286
    90% bootstrap rank range
    39–50
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #40 of 286
    OpenAI GPT comparisons only
    #50 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  45. Rank 045Higher-dimensional Euclidean Kakeya conjectureIn four or more dimensions, must a compact set containing a unit line segment in every direction have full Hausdorff dimension? This asks for full fractal dimension, even when the set need not have positive volume.Open · subcases solved

    Exact statement

    Prove or disprove that for every integer n >= 4, every compact set E in R^n that contains a unit line segment in every direction has Hausdorff dimension n.

    Analysis & PDE

    Status

    Open · subcases solved

    Earlier note (2026-09-14): The stated higher-dimensional Euclidean Kakeya target (n >= 4) remains open, with solved subcases. The solved three-dimensional theorem is outside that range. A February positive-volume preprint is unaccepted and does not overcome the arbitrary zero-measure Kakeya-set issue; September lifting results remain conditional. October 7, 2026: OpenAI’s October 6 release reports partial progress. The full Kakeya target requires full Hausdorff dimension in every dimension n at least four. The second manuscript proves precisely the n=4 subcase, even for noncompact and nonregular sets; dimensions n at least five remain outside the family. The separate three-dimensional maximal estimate does not turn this into an all-dimensional theorem. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A proof would show that accommodating all directions forces full Hausdorff dimension in these spaces. A counterexample would contain every required segment within a compact set of smaller dimension.

    Sources

    Also ranked #38 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #55 of 286
    Math-interested nonspecialists
    #40 of 286
    90% bootstrap rank range
    42–49
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #46 of 286
    OpenAI GPT comparisons only
    #39 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  46. Rank 046Unfolding convex polyhedraCan every convex polyhedron be cut only along its edges and flattened into one connected polygon without overlap?

    Exact statement

    Can every convex polyhedron be cut along its edges and unfolded flat to a single, non-overlapping, simple polygon?

    Geometry & topology

    Also known as

    Edge-Unfolding Convex Polyhedra

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #40 of 286
    Math-interested nonspecialists
    #53 of 286
    90% bootstrap rank range
    42–51
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #50 of 286
    OpenAI GPT comparisons only
    #43 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  47. Rank 047Frankl Union-Closed Sets ConjectureIn every finite family containing a nonempty set and closed under taking unions, must some element belong to at least half of the sets?

    Exact statement

    Prove that in every finite union-closed family of sets other than the empty set, there exists an element that belongs to at least half of the members of the family.

    Combinatorics & discrete geometry

    Status

    Open

    October 6, 2026: The claimed union-closed frequency lower bound 0.3828852549667978 improves prior bounds but remains below the conjectured 1/2. The complete proof was not audited in this review.

    Status reviewed .

    Why it matters

    A proof would force one element to occur frequently solely from closure under unions. A counterexample would show that every element can remain below half despite that closure property.

    Sources

    Research on ProofAtlas

    Also ranked #370 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #51 of 286
    Math-interested nonspecialists
    #41 of 286
    90% bootstrap rank range
    41–50
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #47 of 286
    OpenAI GPT comparisons only
    #41 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  48. Rank 048Thomson problemFor every positive integer N, which configurations of N identical point charges constrained to the unit sphere minimize their total Coulomb repulsion energy?Open · subcases solved

    Exact statement

    For each positive integer N, determine the configuration of N equal point charges on the unit sphere that minimizes total Coulomb energy.

    Mathematical physics

    Status

    Open · subcases solved

    Status review date not recorded in this edition.

    Why it matters

    A complete answer would identify globally optimal charge arrangements for every N and prove their optimality. Finding stable or locally minimizing arrangements alone would not establish the required energy minimum.

    Sources

    Research on ProofAtlas

    Also ranked #347 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #53 of 286
    Math-interested nonspecialists
    #46 of 286
    90% bootstrap rank range
    45–52
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #48 of 286
    OpenAI GPT comparisons only
    #44 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  49. Rank 049Polignac's conjectureDoes every positive even number occur infinitely often as the gap between two consecutive primes?

    Exact statement

    For every positive even integer 2k, there are infinitely many pairs of consecutive primes whose difference is 2k.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would show that every positive even gap recurs indefinitely among neighboring primes. A disproof would identify an even gap that eventually disappears or never occurs in consecutive-prime spacings.

    Sources

    Also ranked #102 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #39 of 286
    Math-interested nonspecialists
    #56 of 286
    90% bootstrap rank range
    39–58
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #36 of 286
    OpenAI GPT comparisons only
    #68 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  50. Rank 050Gauss Circle ProblemCount the integer grid points inside a disk centered at the origin and compare that count with the disk's area. What is the sharp size of the error as its radius r grows: is it bounded by a constant times r^(1/2+ε) for every ε>0?

    Exact statement

    Determine the sharp order of the error term E(r) in N(r) = pi*r^2 + E(r), where N(r) counts the integer lattice points in the disk of radius r centered at the origin; in particular, prove or disprove that E(r) = O(r^(1/2+eps)) for every eps > 0.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A resolution would identify the precision with which area predicts this lattice-point count. The specified bound would constrain the fluctuations to barely above a square-root scale in the radius.

    Sources

    Research on ProofAtlas

    Also ranked #237 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #74 of 286
    Math-interested nonspecialists
    #47 of 286
    90% bootstrap rank range
    47–54
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #51 of 286
    OpenAI GPT comparisons only
    #46 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  51. Rank 051Brocard’s problemAre 4, 5 and 7 the only positive integers n for which n! + 1 is a perfect square, where n! means 1 × 2 × ⋯ × n?

    Exact statement

    Determine whether (4,5), (5,11), and (7,71) are the only pairs of positive integers (n,m) satisfying n! + 1 = m^2; prove this assertion or exhibit another solution.

    Number theory & arithmetic geometry

    Also known as

    Brocard's Problem

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #61 of 286
    Math-interested nonspecialists
    #51 of 286
    90% bootstrap rank range
    47–54
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #53 of 286
    OpenAI GPT comparisons only
    #45 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  52. Rank 052Montgomerys Pair Correlation Conjecture for Zeros of the Riemann Zeta FunctionIf vertical separations between pairs of nontrivial zeta zeros are rescaled by the local mean spacing of zeros, does their limiting pair density at every fixed separation u > 0 equal 1 − (sin(πu)/(πu))²?

    Exact statement

    Prove that normalized vertical gaps between pairs of nontrivial zeros of the Riemann zeta function have limiting pair density 1 - (sin(pi u)/(pi u))^2 at fixed positive normalized separation.

    Number theory & arithmetic geometry

    Status

    Open

    Montgomery pair correlation remains open. Jerby's December 2025 proof claim assumes RH, which the stated target does not assume, and lacks established independent acceptance. It is recorded as an unverified conditional claim, not an accepted conditional theorem or an unconditional resolution.

    Status reviewed .

    Why it matters

    A proof would give the specified quantitative law for two-point spacing statistics of zeta zeros. A counterexample would show a departure from that density at a positive normalized separation.

    Sources

    Research on ProofAtlas

    Also ranked #75 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #78 of 286
    Math-interested nonspecialists
    #49 of 286
    90% bootstrap rank range
    47–56
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #59 of 286
    OpenAI GPT comparisons only
    #42 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  53. Rank 053Congruent Number ProblemExactly which positive integers occur as the area of a right triangle whose three side lengths are rational numbers?Open · subcases solved

    Exact statement

    Determine exactly which positive integers are congruent numbers, i.e. areas of right triangles with all three sides rational, equivalently for which squarefree n the elliptic curve y^2 = x^3 - n^2 x has a rational point with y != 0.

    Number theory & arithmetic geometry

    Status

    Open · subcases solved

    October 7, 2026: OpenAI’s October 6 release reports partial progress. The rank-equality theorem applies in particular to congruent-number curves y^2=x^3-n^2 x when a full q-power Selmer group has corank at most one; it then identifies whether their Mordell-Weil rank is zero or one. There is no unconditional classification of every squarefree n or a guarantee that every congruent-number curve meets the low-corank hypothesis. Congruent-number curves have rational two-torsion, so the companion identifies their actual rank when their full 2-power Selmer corank is zero or one. This does not decide congruence for every squarefree n: higher-corank twists and an exhaustive criterion are not covered. The low-corank 2-converse applies to those quadratic twists of y^2=x^3-x whose full 2-power Selmer corank is zero or one, giving the corresponding actual rank. Analytic densities and a mean over signed squarefree twists do not by themselves classify every individual positive congruent number. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A complete answer would classify these rational triangles by their integer areas. For squarefree n, it would also determine exactly when y² = x³ − n²x has a rational point with nonzero y.

    Sources

    Also ranked #145 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #83 of 286
    Math-interested nonspecialists
    #52 of 286
    90% bootstrap rank range
    49–57
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #60 of 286
    OpenAI GPT comparisons only
    #47 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  54. Rank 054Local Connectivity of the Mandelbrot Set (MLC Conjecture)At every point of the Mandelbrot set—the parameters c for which iterating z ↦ z² + c from zero stays bounded—are there arbitrarily small connected neighborhoods within the set?Open · subcases solved

    Exact statement

    Determine whether the Mandelbrot set M = {c in C : the orbit of 0 under z -> z^2 + c is bounded} is locally connected at every point.

    Probability, ergodic theory & dynamics

    Status

    Open · subcases solved

    Status review date not recorded in this edition.

    Why it matters

    A proof would control the set's local topology even at its most intricate parameters. A counterexample would identify a point where connected neighborhoods cannot be made arbitrarily small.

    Sources

    Also ranked #124 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #114 of 286
    Math-interested nonspecialists
    #50 of 286
    90% bootstrap rank range
    49–56
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #57 of 286
    OpenAI GPT comparisons only
    #53 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  55. Rank 055Smooth four-dimensional Poincare conjectureA homotopy equivalence identifies spaces up to continuous deformation, without requiring a smooth match point by point. If a smooth, closed four-dimensional manifold has the homotopy type of the four-dimensional sphere, must there be a smooth bijection between them with a smooth inverse?

    Exact statement

    Determine whether every smooth closed 4-manifold homotopy equivalent to the standard 4-sphere S^4 is diffeomorphic to S^4.

    Geometry & topology

    Status

    Open

    Smooth four-dimensional Poincare remains open. The reviewed X(41,189,73) project explicitly disclaims a peer-accepted counterexample and leaves geometric obligations outside its formal work. Restricted knot-route evidence does not decide the whole conjecture.

    Status reviewed .

    Why it matters

    A proof would show that the stated homotopy condition determines the sphere’s smooth structure. A counterexample would exhibit an exotic smooth four-sphere with a different diffeomorphism type.

    Sources

    Research on ProofAtlas

    Also ranked #21 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #62 of 286
    Math-interested nonspecialists
    #54 of 286
    90% bootstrap rank range
    50–57
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #49 of 286
    OpenAI GPT comparisons only
    #57 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  56. Rank 056Schanuel's conjectureIf n complex numbers have no nonzero rational linear relation, must at least n of those numbers and their exponentials be algebraically independent, meaning that no nonzero polynomial with rational coefficients relates them?

    Exact statement

    Prove or disprove that if z_1,...,z_n in C are linearly independent over Q, then the field Q(z_1,...,z_n,e^{z_1},...,e^{z_n}) has transcendence degree at least n over Q.

    Number theory & arithmetic geometry

    Status

    Open

    Schanuel remains open. Reformulations and almost-everywhere Schanuel properties do not establish its universal algebraic-independence conclusion.

    Status reviewed .

    Why it matters

    A proof would force at least n algebraically independent quantities into every field generated this way. A counterexample would reveal an exponential tuple with more algebraic dependence than the bound allows.

    Sources

    Research on ProofAtlas

    Also ranked #26 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #152 of 286
    Math-interested nonspecialists
    #55 of 286
    90% bootstrap rank range
    54–61
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #70 of 286
    OpenAI GPT comparisons only
    #48 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  57. Rank 057Hardy-Littlewood Prime k-Tuple ConjectureTake a fixed finite pattern of integer offsets. If no prime divides the product of the shifted entries for every integer shift, do shifts making every entry prime occur with exactly the asymptotic frequency predicted by the Hardy–Littlewood formula, including its correction for divisibility?

    Exact statement

    For every finite admissible set H={h_1,...,h_k} of distinct integers, prove that the number of n <= x for which every n+h_i is prime is asymptotic to S(H)x/(log x)^k, where S(H) is the Hardy-Littlewood singular series, or disprove this formula.

    Number theory & arithmetic geometry

    Status

    Open

    The all-fixed-admissible-tuples Hardy–Littlewood asymptotic remains open. Published short-average progress does not establish every fixed tuple. The located bitstring-sieve full-proof claim lacks independent acceptance.

    Status reviewed .

    Why it matters

    A proof would give quantitative frequencies for every admissible finite prime pattern. A counterexample would identify a pattern whose simultaneous-prime count departs from the proposed asymptotic law.

    Sources

    Also ranked #18 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #57 of 286
    Math-interested nonspecialists
    #58 of 286
    90% bootstrap rank range
    51–64
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #64 of 286
    OpenAI GPT comparisons only
    #59 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  58. Rank 058Periodic paths in triangular billiardsDoes every triangular billiard table allow a ball to follow a repeating path, with ideal reflections off its sides?Open · subcases solved

    Exact statement

    Determine whether all triangular billiards (convex polygonal billiards with a triangular boundary) have some periodic trajectory.

    Probability, ergodic theory & dynamics

    Also known as

    Do all triangular billiards have some periodic trajectory?

    Status

    Open · subcases solved

    The full question remains open; some special cases have been solved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #76 of 286
    Math-interested nonspecialists
    #57 of 286
    90% bootstrap rank range
    56–65
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #81 of 286
    OpenAI GPT comparisons only
    #49 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  59. Rank 059Elliott-Halberstam Conjecture on the Level of Distribution of PrimesWhen counting primes in arithmetic progressions, can their distribution errors be controlled for step sizes nearly as large as the counting limit x? Elliott–Halberstam predicts arbitrarily strong logarithmic savings in the sum of the largest errors over residue classes and shorter counting ranges, for step sizes up to x^θ with any 0 < θ < 1.

    Exact statement

    Prove that for every A > 0 and every 0 < theta < 1, the sum over q <= x^theta of the maximal Chebyshev error |psi(y; q, a) - y/phi(q)| over y <= x and (a,q)=1 is O(x / (log x)^A), extending Bombieri-Vinogradov from theta < 1/2 to every theta < 1.

    Number theory & arithmetic geometry

    Status

    Open

    Elliott–Halberstam remains open unconditionally. A recent implication assumes both GRH and Dirichlet-L pair correlation, while restricted-modulus results do not prove the stated all-moduli maximal-error estimate.

    Status reviewed .

    Why it matters

    A proof would control the combined error in prime distribution across arithmetic progressions for moduli extending to every fixed power below x. The estimate controls the sum of worst errors for each modulus, retaining uniformity over shorter cutoffs.

    Sources

    Research on ProofAtlas

    Also ranked #87 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #81 of 286
    Math-interested nonspecialists
    #59 of 286
    90% bootstrap rank range
    57–67
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #75 of 286
    OpenAI GPT comparisons only
    #55 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  60. Rank 060Bounded aliquot sequencesIf each positive integer is repeatedly replaced by the sum of its positive divisors smaller than itself, must the sequence stay bounded?

    Exact statement

    Prove or disprove that every aliquot sequence n, s(n), s(s(n)), ... obtained by iterating the sum-of-proper-divisors function s is bounded, equivalently that every such sequence terminates at 1, reaches a perfect number, or enters a sociable cycle.

    Number theory & arithmetic geometry

    Also known as

    Catalan-Dickson Conjecture on Boundedness of Aliquot Sequences

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #54 of 286
    Math-interested nonspecialists
    #68 of 286
    90% bootstrap rank range
    56–71
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #56 of 286
    OpenAI GPT comparisons only
    #81 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  61. Rank 061P versus PSPACECan every yes-or-no problem solvable by a deterministic computer using polynomially bounded memory also be solved deterministically in time polynomial in the input length?

    Exact statement

    Determine whether P = PSPACE, that is, whether every language decidable by a deterministic Turing machine in polynomial space is decidable by a deterministic Turing machine in polynomial time.

    Theoretical computer science

    Status

    Open

    P versus PSPACE remains open. Recent space simulations and fixed-exponent time lower bounds do not separate polynomial time from polynomial space; conditional equality implications are not unconditional resolutions.

    Status reviewed .

    Why it matters

    Equality would turn every polynomial-space decision procedure into a polynomial-time one. Separation would establish that some problems admit modest memory requirements while resisting every deterministic polynomial-time algorithm.

    Sources

    Also ranked #14 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #88 of 286
    Math-interested nonspecialists
    #62 of 286
    90% bootstrap rank range
    57–70
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #52 of 286
    OpenAI GPT comparisons only
    #92 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  62. Rank 062Erdos-Rado Sunflower ConjectureFor every integer r≥3, is there a constant C depending only on r such that any family of more than C^k sets of size at most k contains r sets whose pairwise intersections are all the same?

    Exact statement

    For every fixed integer r >= 3, prove or disprove that there is a constant C_r such that every family of more than C_r^k sets of size at most k contains an r-petal sunflower.

    Combinatorics & discrete geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would give a purely exponential threshold forcing a sunflower: sets sharing one common core and otherwise disjoint. It would quantify when sufficiently large set families must contain this reusable combinatorial pattern.

    Sources

    Research on ProofAtlas

    Also ranked #148 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #179 of 286
    Math-interested nonspecialists
    #60 of 286
    90% bootstrap rank range
    59–69
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #84 of 286
    OpenAI GPT comparisons only
    #52 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  63. Rank 063Inverse Galois problem over the rational numbersCan every finite group occur as the Galois group—the group of field symmetries—of some finite Galois extension of the rational numbers?Open · subcases solved

    Exact statement

    Prove or disprove that every finite group G is isomorphic to the Galois group Gal(L/Q) of some finite Galois extension L of the rational numbers Q.

    Algebra, representation & category theory

    Status

    Open · subcases solved

    The inverse Galois problem over Q remains open, with solved subcases. The August 2026 M23 preprint reports one finite group, not realization of every finite group; favorable informal commentary is not journal-proof certification.

    Status reviewed .

    Why it matters

    A proof would realize every finite abstract group as arithmetic symmetries over the rationals. A disproof would identify a finite group excluded from all such Galois extensions.

    Sources

    Research on ProofAtlas

    Also ranked #58 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #181 of 286
    Math-interested nonspecialists
    #61 of 286
    90% bootstrap rank range
    57–73
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #65 of 286
    OpenAI GPT comparisons only
    #74 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  64. Rank 064Polynomial-time unknot recognitionGiven a knot diagram, can an algorithm decide whether it represents an unknotted loop in time bounded by a polynomial in the input size?

    Exact statement

    Determine whether there is a polynomial-time algorithm that, given a knot diagram or an equivalent finite encoding of a knot in three-space, decides whether the knot is the unknot.

    Geometry & topology

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A positive answer would make unknot recognition efficiently decidable in the worst case. A negative answer would show that every algorithm for this exact recognition task exceeds polynomial time on some inputs.

    Sources

    Also ranked #301 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #110 of 286
    Math-interested nonspecialists
    #66 of 286
    90% bootstrap rank range
    58–76
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #66 of 286
    OpenAI GPT comparisons only
    #66 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  65. Rank 065Smallest Sierpiński numberIs 78,557 the smallest positive odd integer k for which k × 2ⁿ + 1 is composite for every positive integer n?Open · subcases solved

    Exact statement

    Prove or disprove that 78557 is the smallest Sierpinski number, equivalently that for every odd k with 1 <= k < 78557 there exists a positive integer n such that k*2^n+1 is prime.

    Number theory & arithmetic geometry

    Also known as

    Selfridge's Conjecture on the Smallest Sierpinski Number

    Status

    Open · subcases solved

    Some smaller candidates remain unresolved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #154 of 286
    Math-interested nonspecialists
    #64 of 286
    90% bootstrap rank range
    58–74
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #72 of 286
    OpenAI GPT comparisons only
    #62 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  66. Rank 066Irrationality of zeta(5)Is the infinite sum 1 + 1/2⁵ + 1/3⁵ + ⋯, called ζ(5), impossible to express exactly as a ratio of two integers?

    Exact statement

    Prove that zeta(5) = sum_{n>=1} 1/n^5 is irrational.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would establish irrationality for this particular odd zeta value. A rational evaluation would instead identify an exact arithmetic relation for ζ(5), without deciding the behavior of other zeta values.

    Sources

    Also ranked #429 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #161 of 286
    Math-interested nonspecialists
    #65 of 286
    90% bootstrap rank range
    59–72
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #63 of 286
    OpenAI GPT comparisons only
    #78 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  67. Rank 067Kolakoski sequence densityIn the sequence 1, 2, 2, 1, 1, 2, … that describes its own run lengths, does the proportion of 1s approach one half?

    Exact statement

    Let K=(K_j)_{j>=1} be the sequence over {1,2} that starts with 1 and equals its own run-length encoding. Then lim_{N->infinity} #{1<=j<=N: K_j=1}/N exists and equals 1/2.

    Combinatorics & discrete geometry

    Also known as

    Kolakoski Sequence Half-Density Conjecture

    Status

    Open

    The ordinary sequence remains unresolved; density results for other self-generating sequences do not settle it.

    Status reviewed .

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #219 of 286
    Math-interested nonspecialists
    #63 of 286
    90% bootstrap rank range
    57–79
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #91 of 286
    OpenAI GPT comparisons only
    #54 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  68. Rank 068P versus BPP (Does BPP = P?)Can every decision problem solved by a randomized polynomial-time algorithm with bounded error on both answers also be solved deterministically in polynomial time?

    Exact statement

    Determine whether every language computable by a probabilistic polynomial-time algorithm with bounded two-sided error (BPP) can be computed by a deterministic polynomial-time algorithm, i.e., whether BPP = P.

    Cryptography, coding, information & optimization

    Status

    Open

    Earlier note (2026-09-14): P versus BPP remains open. Recent pseudorandom-generator and derandomization advances retain additional assumptions, not a deterministic simulation of every BPP language. October 7, 2026: OpenAI’s October 6 release reports partial progress. The theorem derandomizes the logarithmic-workspace subclass of polynomial-time bounded-error computation. Such machines form a restricted subcase of the full BPP=P question. The publisher gives no deterministic simulation for arbitrary polynomial-space work tapes used by polynomial-time BPP machines, so it does not establish the unrestricted equality BPP=P. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    Equality would remove randomness from every computation in this class while preserving polynomial-time solvability. Separation would identify a decision problem for which bounded-error randomization provides power beyond deterministic polynomial time.

    Sources

    Research on ProofAtlas

    Also ranked #25 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #131 of 286
    Math-interested nonspecialists
    #67 of 286
    90% bootstrap rank range
    59–74
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #71 of 286
    OpenAI GPT comparisons only
    #65 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  69. Rank 069Algebraic Independence of e and piIs it impossible for any nonzero polynomial in two variables with rational coefficients to become zero when its variables are replaced by e and π?

    Exact statement

    Prove that no nonzero polynomial P in Q[X,Y] satisfies P(e,pi)=0.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would exclude every polynomial relation between these two constants. It would also imply that their sum and product are transcendental, while making no blanket claim about arbitrary exponential expressions involving them.

    Sources

    Also ranked #168 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #45 of 286
    Math-interested nonspecialists
    #76 of 286
    90% bootstrap rank range
    56–86
    Included in the Top 100
    99.9% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #54 of 286
    OpenAI GPT comparisons only
    #108 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  70. Rank 070The Exponential Time Hypothesis for 3-SATThe 3-SAT problem asks whether Boolean variables can satisfy constraints with three literals per clause. Is there a positive constant c such that no deterministic algorithm solves all n-variable instances in time 2^(cn)?

    Exact statement

    Prove or disprove the Exponential Time Hypothesis: there exists a constant c > 0 such that 3-SAT cannot be solved in deterministic time 2^{c n}.

    Theoretical computer science

    Status

    Open

    ETH remains open. The 2026 source treats positive-exponent hardness as a conjecture; conditional algorithmic lower bounds do not establish unrestricted exponential hardness.

    Status reviewed .

    Why it matters

    A proof would establish an exponential barrier to deterministic 3-SAT algorithms measured by variable count. A negative resolution would remove the asserted positive lower bound on the exponential rate.

    Sources

    Research on ProofAtlas

    Also ranked #43 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #180 of 286
    Math-interested nonspecialists
    #69 of 286
    90% bootstrap rank range
    62–79
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #73 of 286
    OpenAI GPT comparisons only
    #69 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  71. Rank 071Sausage conjecture for ball packingsIn every dimension at least five, does a straight row of touching unit balls minimize the volume of the convex enclosure among packings of the same number of balls?Open · subcases solved

    Exact statement

    Prove or disprove that for every dimension d >= 5 and every number n of unit balls, the sausage arrangement, whose centers are equally spaced on a line segment, minimizes the convex-hull volume among all packings of n unit balls in R^d.

    Combinatorics & discrete geometry

    Also known as

    Fejes Toth Sausage Conjecture for Finite Ball Packings

    Status

    Open · subcases solved

    The full question remains open; some special cases have been solved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #41 of 286
    Math-interested nonspecialists
    #88 of 286
    90% bootstrap rank range
    62–84
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #69 of 286
    OpenAI GPT comparisons only
    #87 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  72. Rank 072Borel normal number conjecture for irrational algebraic numbersMust every real irrational root of an integer polynomial have equally frequent digit blocks of each fixed length in every integer base of at least two?

    Exact statement

    Prove or disprove that every real irrational algebraic number is normal to every integer base b >= 2.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would impose precise long-run frequencies on all digit patterns of algebraic irrationals. It would connect numbers defined by polynomial equations to the same block frequencies as uniformly distributed random digits.

    Sources

    Also ranked #128 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #80 of 286
    Math-interested nonspecialists
    #71 of 286
    90% bootstrap rank range
    66–86
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #76 of 286
    OpenAI GPT comparisons only
    #77 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  73. Rank 073Hilbert's Sixteenth Problem, Second PartFor planar differential equations whose right-hand sides are polynomials of a fixed degree at least two, what is the maximum number of isolated closed trajectories, and how can they be arranged? Is that maximum finite for every degree?

    Exact statement

    For every degree n >= 2, prove finiteness and determine the maximal number and possible configurations of limit cycles of planar polynomial vector fields of degree n.

    Probability, ergodic theory & dynamics

    Status

    Open

    Earlier note (2026-09-14): Hilbert's sixteenth problem, part II, remains open. The checked purported quadratic bound faces a published critique; no universal bound for each degree was established. This is not a claim of a formal retraction absent such evidence. October 7, 2026: OpenAI’s October 6 release reports partial progress. The exact Hilbert XVI target has three requirements for every degree: uniform finiteness, the maximal number, and possible configurations of limit cycles. The first paper establishes a finite degree-dependent bound but states no exact maxima or configuration classification. The second settles a quintic Lienard subclass with maximum two, not all degree-five planar fields. Thus uniform boundedness is resolved while the full compound target remains only partially covered. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A resolution would determine how polynomial degree limits the number and arrangement of isolated periodic motions in planar systems. It would provide a complete degree-by-degree description of the possibilities across all vector fields in scope.

    Sources

    Research on ProofAtlas

    Also ranked #40 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #71 of 286
    Math-interested nonspecialists
    #74 of 286
    90% bootstrap rank range
    67–83
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #58 of 286
    OpenAI GPT comparisons only
    #114 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  74. Rank 074Conway’s thrackle conjectureIf a graph is drawn so every pair of edges meets exactly once, at a shared endpoint or a proper crossing, can it have more edges than vertices?

    Exact statement

    What is the maximum number of edges in a thrackle (a planar graph drawing where every pair of edges intersects at exactly one point, either at a common endpoint with distinct tangents or at a proper crossing)? In particular, is the number of edges at most n?

    Combinatorics & discrete geometry

    Also known as

    Conways Thrackle Conjecture

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #56 of 286
    Math-interested nonspecialists
    #77 of 286
    90% bootstrap rank range
    65–87
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #105 of 286
    OpenAI GPT comparisons only
    #60 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  75. Rank 075Gilbreath’s conjectureStart with the primes and repeatedly take absolute differences between neighbors. Does every new row begin with 1?

    Exact statement

    Prove or disprove that in the triangular array formed from the primes by iterated absolute differences of consecutive entries, every row after the first has leading entry 1.

    Number theory & arithmetic geometry

    Also known as

    Gilbreath's Conjecture

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #162 of 286
    Math-interested nonspecialists
    #70 of 286
    90% bootstrap rank range
    67–85
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #85 of 286
    OpenAI GPT comparisons only
    #70 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  76. Rank 076Infinitely many cousin primesAre there infinitely many pairs of primes that differ by exactly four?

    Exact statement

    Determine whether there are infinitely many primes p such that p+4 is also prime, equivalently infinitely many prime pairs with gap 4.

    Number theory & arithmetic geometry

    Also known as

    Infinitude of Cousin Primes

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #60 of 286
    Math-interested nonspecialists
    #78 of 286
    90% bootstrap rank range
    67–88
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #68 of 286
    OpenAI GPT comparisons only
    #90 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  77. Rank 077Resolvent Degree of the General Degree-Seven Polynomial (Algebraic Form of Hilbert's Thirteenth Problem)How many variables are inherently needed in algebraic functions that solve the general degree-seven polynomial over the complex numbers, even when those functions may be composed in successive steps? Is the minimum exactly three, so two-variable algebraic functions cannot suffice?Open · subcases solved

    Exact statement

    Prove that the resolvent degree rd_C(7) of the general degree-seven polynomial equals 3; equivalently, establish the lower bound rd_C(7) > 2.

    Algebra, representation & category theory

    Status

    Open · subcases solved

    Status review date not recorded in this edition.

    Why it matters

    A proof would establish the proposed lower bound on the number of variables needed in algebraic solution procedures for general septics. A negative answer would permit a reduction to at most two variables.

    Sources

    Research on ProofAtlas

    Also ranked #266 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #175 of 286
    Math-interested nonspecialists
    #72 of 286
    90% bootstrap rank range
    68–86
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #89 of 286
    OpenAI GPT comparisons only
    #67 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  78. Rank 078Repeated numbers in Pascal’s triangleIs there one fixed limit on how often any number other than 1 can appear anywhere in Pascal’s triangle?Open · subcases solved

    Exact statement

    Determine whether there is a value S in N such that any number different from 1 appears in Pascal's triangle at most S times.

    Probability, ergodic theory & dynamics

    Also known as

    Singmaster's conjecture on Pascal's triangle

    Status

    Open · subcases solved

    The full question remains open; some special cases have been solved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #111 of 286
    Math-interested nonspecialists
    #73 of 286
    90% bootstrap rank range
    69–86
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #86 of 286
    OpenAI GPT comparisons only
    #71 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  79. Rank 079Densest packing of regular tetrahedraWhat is the greatest fraction of three-dimensional space that identical regular tetrahedra can fill without overlapping?Open · subcases solved

    Exact statement

    For packings P of R^3 by congruent regular tetrahedra with pairwise disjoint interiors, define their upper density as limsup_{R->infinity} vol((union P) intersect B(0,R))/vol(B(0,R)). Determine the supremum delta_T of this density over all such packings ; attainment is a known general fact for congruent convex bodies.

    Combinatorics & discrete geometry

    Also known as

    Maximum Packing Density of Regular Tetrahedra

    Status

    Open · subcases solved

    An optimal packing is known to exist, but its density has not been determined. Known periodic constructions do not establish the optimum among all packings.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #63 of 286
    Math-interested nonspecialists
    #82 of 286
    90% bootstrap rank range
    69–91
    Included in the Top 100
    99.9% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #80 of 286
    OpenAI GPT comparisons only
    #80 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  80. Rank 080Fermat-Catalan Conjecture on Finiteness of Coprime Power TriplesAre there only finitely many pairwise coprime positive-integer triples (x,y,z) satisfying x^p+y^q=z^r for some integers p,q,r greater than one with 1/p+1/q+1/r less than one?Open · subcases solved

    Exact statement

    Prove that there are only finitely many pairwise coprime positive-integer triples (x,y,z) for which x^p + y^q = z^r for integers p,q,r > 1 satisfying 1/p + 1/q + 1/r < 1.

    Number theory & arithmetic geometry

    Status

    Open · subcases solved

    Status review date not recorded in this edition.

    Why it matters

    A proof would establish finiteness across all permitted exponent choices together. It would show that allowing the exponents to vary cannot produce infinitely many distinct primitive base triples.

    Sources

    Also ranked #351 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #185 of 286
    Math-interested nonspecialists
    #75 of 286
    90% bootstrap rank range
    71–87
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #87 of 286
    OpenAI GPT comparisons only
    #72 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  81. Rank 081Conway’s 99-graph problemCan a 99-vertex graph give each vertex 14 neighbors, with one shared neighbor for each adjacent pair and two for each nonadjacent pair?

    Exact statement

    Determine whether there exists a strongly regular graph with parameters (99,14,1,2), equivalently a 99-vertex 14-regular graph in which each adjacent pair has exactly one common neighbor and each nonadjacent pair has exactly two common neighbors.

    Combinatorics & discrete geometry

    Also known as

    Conway's 99-Graph Problem

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #75 of 286
    Math-interested nonspecialists
    #79 of 286
    90% bootstrap rank range
    71–89
    Included in the Top 100
    99.9% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #104 of 286
    OpenAI GPT comparisons only
    #64 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  82. Rank 082Invariant Subspace Problem for Hilbert Space OperatorsMust every bounded linear operator on a separable infinite-dimensional complex Hilbert space map some closed linear subspace into itself, other than the zero subspace or the whole space?

    Exact statement

    Does every bounded linear operator T on an infinite-dimensional separable complex Hilbert space H have a nontrivial closed invariant subspace, that is, a closed subspace M with {0} != M != H such that T(M) is a subset of M?

    Analysis & PDE

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would guarantee that every such operator has a smaller closed space on which its action can be studied. A counterexample would show that even this basic form of reduction can fail.

    Sources

    Research on ProofAtlas

    Also ranked #106 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #66 of 286
    Math-interested nonspecialists
    #81 of 286
    90% bootstrap rank range
    72–89
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #61 of 286
    OpenAI GPT comparisons only
    #124 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  83. Rank 083Cramer Conjecture on the Limsup of Normalized Prime GapsAre the largest long-run gaps between consecutive primes on the scale of the square of the logarithm of the smaller prime, with the limiting upper ratio exactly one?

    Exact statement

    Prove that lim sup_{n -> infinity} (p_{n+1} - p_n) / (log p_n)^2 = 1, where p_n is the n-th prime.

    Number theory & arithmetic geometry

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    The assertion fixes an exact asymptotic constant for the largest normalized gaps, beyond a bound of the same order. It requires gaps approaching this normalized size arbitrarily far out while excluding any larger limiting upper value.

    Sources

    Research on ProofAtlas

    Also ranked #153 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #68 of 286
    Math-interested nonspecialists
    #84 of 286
    90% bootstrap rank range
    71–93
    Included in the Top 100
    99.7% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #62 of 286
    OpenAI GPT comparisons only
    #132 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  84. Rank 084Artin's primitive root conjectureFor each integer a that is neither a square nor −1, do the primes for which powers of a generate every nonzero residue class form the exact positive proportion predicted by Artin, including its correction factor?

    Exact statement

    Prove or disprove that every integer a other than -1 that is not a perfect square is a primitive root modulo p for a positive natural density of primes p, with the density given by Artin's predicted correction factor.

    Number theory & arithmetic geometry

    Status

    Open

    October 7, 2026: OpenAI’s October 6 release reports partial progress. The source treats every admissible integer base but only proves at least c_a x/(log x)^2 primitive-root primes in each sufficiently large dyadic interval, hence infinitude. The full target requires positive natural density among primes with Artin's predicted correction factor; the stated lower bound permits density zero and gives no asymptotic constant. The simultaneous-prime-base companion is conditional on four inputs and must not upgrade this to the full density conjecture. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A proof would quantify how often a fixed integer generates the multiplicative group modulo a prime. It would supply the predicted density for each permitted integer, beyond merely guaranteeing infinitely many such primes.

    Sources

    Research on ProofAtlas

    Also ranked #188 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #119 of 286
    Math-interested nonspecialists
    #80 of 286
    90% bootstrap rank range
    75–90
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #88 of 286
    OpenAI GPT comparisons only
    #73 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  85. Rank 085Graph reconstruction conjectureCan every finite simple undirected graph with at least three vertices be recovered, up to relabeling, from the complete collection of graphs obtained by deleting each vertex in turn, with repeated results counted?Open · subcases solved

    Exact statement

    Prove or disprove that every finite simple undirected graph on at least three vertices is determined up to isomorphism by the multiset of its vertex-deleted subgraphs.

    Combinatorics & discrete geometry

    Status

    Open · subcases solved

    Status review date not recorded in this edition.

    Why it matters

    A proof would show that this collection of partial views determines the whole graph uniquely. A counterexample would give two different graphs that remain indistinguishable from all their vertex-deleted views.

    Sources

    Research on ProofAtlas

    Also ranked #298 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #195 of 286
    Math-interested nonspecialists
    #83 of 286
    90% bootstrap rank range
    80–93
    Included in the Top 100
    100% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #95 of 286
    OpenAI GPT comparisons only
    #75 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  86. Rank 086Coding Theory and Combinatorics Problem CC-6: Existence of Finite Projective Planes of Non-Prime-Power OrdersIn a finite projective plane, any two points determine a line and any two lines meet. Can one exist with 13 points on each line, meaning order 12, or at another composite non-prime-power order not ruled out by the Bruck–Ryser–Chowla theorem?

    Exact statement

    Does there exist any finite projective plane of order 12, or of any other composite non-prime-power order not ruled out by the Bruck-Ryser-Chowla theorem?

    Cryptography, coding, information & optimization

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    One construction at an admissible order would settle the existence question affirmatively. A negative resolution would need to rule out every order in the stated class, not only order 12.

    Sources

    Research on ProofAtlas

    Also ranked #362 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #67 of 286
    Math-interested nonspecialists
    #89 of 286
    90% bootstrap rank range
    75–97
    Included in the Top 100
    98.1% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #77 of 286
    OpenAI GPT comparisons only
    #94 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  87. Rank 087An odd weird numberCan an odd positive integer have divisors smaller than itself whose sum exceeds it, while no selection of those divisors sums to it exactly?Open · subcases solved

    Exact statement

    Determine whether there exists an odd positive integer n whose proper divisors have sum greater than n but no subset of proper divisors sums to n.

    Number theory & arithmetic geometry

    Also known as

    Existence of an odd weird number

    Status

    Open · subcases solved

    The full question remains open; some special cases have been solved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #147 of 286
    Math-interested nonspecialists
    #85 of 286
    90% bootstrap rank range
    77–94
    Included in the Top 100
    99.7% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #97 of 286
    OpenAI GPT comparisons only
    #76 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  88. Rank 088Kobon triangle problemFor any number of straight lines in the plane, what is the largest number of nonoverlapping triangles they can form?Open · subcases solved

    Exact statement

    Determine the maximum number of nonoverlapping triangles formed by an arrangement of n straight lines in the plane, for every n.

    Combinatorics & discrete geometry

    Status

    Open · subcases solved

    Some line counts are settled, but the general maximum remains unresolved.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #148 of 286
    Math-interested nonspecialists
    #86 of 286
    90% bootstrap rank range
    76–97
    Included in the Top 100
    99% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #103 of 286
    OpenAI GPT comparisons only
    #79 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  89. Rank 089Transcendence of Apery's ConstantIs Apéry's constant, ζ(3) = 1 + 1/2³ + 1/3³ + ⋯, transcendental, meaning it satisfies no nonzero polynomial equation with rational coefficients?Open · disputed claim

    Exact statement

    Prove or disprove that the value zeta(3) is transcendental over the rational numbers.

    Number theory & arithmetic geometry

    Status

    Open · disputed claim

    Status review date not recorded in this edition.

    Why it matters

    A proof would rule out every algebraic equation for this infinite sum. A negative answer would place it among algebraic numbers and determine a different arithmetic nature for this particular odd zeta value.

    Sources

    Also ranked #170 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #168 of 286
    Math-interested nonspecialists
    #87 of 286
    90% bootstrap rank range
    82–96
    Included in the Top 100
    99.2% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #79 of 286
    OpenAI GPT comparisons only
    #95 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  90. Rank 090Four rational distances to a squareIs there a point in the plane whose distances to all four corners of a unit square are rational numbers?

    Exact statement

    Determine whether there exists a point in the plane whose distances to all four vertices of the unit square are rational.

    Combinatorics & discrete geometry

    Also known as

    Four-Distance Problem for the Unit Square

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #85 of 286
    Math-interested nonspecialists
    #90 of 286
    90% bootstrap rank range
    80–98
    Included in the Top 100
    98.2% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #147 of 286
    OpenAI GPT comparisons only
    #58 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  91. Rank 091Exact growth rate of cap setsWhat is the exact exponential growth rate of the largest sets in {0,1,2}ⁿ with no three distinct points summing to zero modulo 3?

    Exact statement

    Let r(n) be the maximum cardinality of a subset A of F_3^n having no three distinct elements x,y,z with x+y+z=0. Determine the exact exponential growth constant c=lim_{n->infinity} r(n)^(1/n)=sup_{n>=1} r(n)^(1/n). Cartesian products make r supermultiplicative and the limit exists.

    Combinatorics & discrete geometry

    Status

    Open

    Status review date not recorded in this edition.

    Sources

    Also ranked #460 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #47 of 286
    Math-interested nonspecialists
    #99 of 286
    90% bootstrap rank range
    77–101
    Included in the Top 100
    94.2% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #55 of 286
    OpenAI GPT comparisons only
    #147 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  92. Rank 092Graceful tree conjectureCan every n-vertex tree be labeled with 1 through n, each used once, so the absolute differences across its edges are all distinct?

    Exact statement

    Prove that every tree admits a graceful labeling, i.e., a bijective labeling of its vertices by {1,...,n} such that the edge-difference labels are pairwise distinct.

    Combinatorics & discrete geometry

    Also known as

    Ringel-Kotzig Graceful Tree Conjecture

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #102 of 286
    Math-interested nonspecialists
    #92 of 286
    90% bootstrap rank range
    81–100
    Included in the Top 100
    96.6% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #82 of 286
    OpenAI GPT comparisons only
    #100 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  93. Rank 093Infinitely many Wieferich primesAre there infinitely many primes p for which p² divides 2^(p−1) − 1?

    Exact statement

    Prove or disprove that there are infinitely many Wieferich primes, i.e., primes p with 2^(p-1) congruent to 1 modulo p^2.

    Number theory & arithmetic geometry

    Also known as

    Infinitude of Wieferich Primes

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #124 of 286
    Math-interested nonspecialists
    #91 of 286
    90% bootstrap rank range
    83–100
    Included in the Top 100
    95% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #67 of 286
    OpenAI GPT comparisons only
    #119 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  94. Rank 094The Hadamard Matrix Existence ConjectureFor every positive integer k, does there exist a 4k×4k matrix of plus and minus ones whose distinct rows have dot product zero?

    Exact statement

    Prove that there exists a Hadamard matrix of order 4k for every positive integer k.

    Cryptography, coding, information & optimization

    Status

    Open

    Status review date not recorded in this edition.

    Why it matters

    A proof would ensure the existence of perfectly orthogonal sign patterns in every order divisible by four. A counterexample would identify an order where that combination of size and orthogonality is impossible.

    Sources

    Research on ProofAtlas

    Also ranked #291 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #123 of 286
    Math-interested nonspecialists
    #93 of 286
    90% bootstrap rank range
    85–103
    Included in the Top 100
    88.2% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #94 of 286
    OpenAI GPT comparisons only
    #93 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  95. Rank 095Global Langlands Functoriality ConjectureLanglands L-groups encode symmetries used to organize automorphic representations. For quasisplit reductive groups over number fields, does every admissible map between their L-groups transfer these representations while matching their local arithmetic data at all but finitely many places?Open · subcases solved

    Exact statement

    For every number field F, quasisplit connected reductive F-groups G-prime and G, and admissible L-homomorphism rho:L(G-prime)->L(G) (over Gamma_F, commuting with its projections, in Arthur Principle1.1's convention), every automorphic representation pi-prime of G-prime(A_F), in the inclusive Langlands sense cited there, has an automorphic representation pi of G(A_F) whose unramified Satake classes equal the images under rho_v of those of pi-prime outside a finite set of places. This is existence of a weak transfer, not uniqueness or a prescribed transfer function.

    Number theory & arithmetic geometry

    Status

    Open · subcases solved

    Global functoriality remains open, with solved subcases. Existing restricted transfers and recent conjectural use do not establish the full specified transfer theorem; evidence of its current general status remains limited.

    Status reviewed .

    Why it matters

    A proof would guarantee the stated transfer across every group map in scope, linking automorphic representations through their local parameters. The existence claim permits more than one transfer and leaves its exceptional places finite.

    Sources

    Also ranked #7 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #139 of 286
    Math-interested nonspecialists
    #94 of 286
    90% bootstrap rank range
    85–106
    Included in the Top 100
    79.2% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #74 of 286
    OpenAI GPT comparisons only
    #129 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  96. Rank 096Tammes problem: points on a sphereFor any number of points on a unit sphere, how far apart along its surface can you make the closest pair?

    Exact statement

    For a general number N of points on the unit 2-sphere, determine the maximal possible minimal spherical distance (or the maximal minimal angular separation) between the points.

    Applied & computational mathematics

    Also known as

    Best Packing of N Equal Circles on the 2-Sphere for General N

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #197 of 286
    Math-interested nonspecialists
    #95 of 286
    90% bootstrap rank range
    90–104
    Included in the Top 100
    78.6% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #107 of 286
    OpenAI GPT comparisons only
    #88 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  97. Rank 097Bunyakovsky conjectureIf a polynomial with integer coefficients cannot split into lower-degree integer polynomials, eventually takes positive values, and has no prime dividing every value, must it produce primes at infinitely many positive integer inputs?Open · disputed claim

    Exact statement

    Prove or disprove that every irreducible polynomial f in Z[x] with positive leading coefficient and no fixed prime divisor takes prime values for infinitely many positive integer inputs.

    Number theory & arithmetic geometry

    Status

    Open · disputed claim

    Status review date not recorded in this edition.

    Why it matters

    A proof would establish infinitely many prime values for every polynomial satisfying these conditions, including x² + 1. A counterexample would reveal an obstruction beyond factorization, sign and fixed prime divisors.

    Sources

    Also ranked #98 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #127 of 286
    Math-interested nonspecialists
    #96 of 286
    90% bootstrap rank range
    89–105
    Included in the Top 100
    77.7% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #117 of 286
    OpenAI GPT comparisons only
    #83 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  98. Rank 098Lebesgue’s universal covering problemWhat is the smallest possible area of a convex shape that can contain every planar set of diameter at most 1 after a rigid motion, including reflection?

    Exact statement

    Determine the infimum of the area of a convex planar set K such that every planar set S with diameter at most 1 has a Euclidean isometric copy contained in K.

    Geometry & topology

    Also known as

    Lebesgue's universal covering problem

    Status

    Open

    This exact question remains open.

    Status review date not recorded in this edition.

    Sources

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #87 of 286
    Math-interested nonspecialists
    #98 of 286
    90% bootstrap rank range
    90–106
    Included in the Top 100
    74.8% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #113 of 286
    OpenAI GPT comparisons only
    #89 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  99. Rank 099Extension of Kroneckers Theorem on Abelian Fields to Any Algebraic Base FieldCan all finite abelian extensions of any number field be generated by values of explicit analytic or arithmetic functions? These are extensions whose field symmetries commute; the goal is an explicit recipe for their generators.Open · subcases solved

    Exact statement

    For each number field K, construct generators for its finite abelian extensions as special values of explicitly defined analytic or arithmetic functions associated with K, extending roots of unity over Q and complex multiplication over imaginary quadratic fields.

    Number theory & arithmetic geometry

    Status

    Open · subcases solved

    Hilbert's twelfth problem over arbitrary number fields remains open, with solved subcases. Dasgupta–Kakde and the independent ICMS account cover a totally real p-adic construction, not every number field or the original complex-analytic formulation. October 6, 2026: The October 4 manuscript Stark units for real quadratic fields and reciprocity laws claims a real-quadratic advance using explicit units and reciprocity. At a fixed modulus, its Proposition 6 identifies a particular fixed subfield; the units need not generate the entire ray class field for every modulus. This claimed specialized progress does not settle Hilbert’s twelfth problem over arbitrary number fields. The manuscript discloses extensive LLM generation followed by human checking and editing; full independent validation was not established in the evidence reviewed.

    Status reviewed .

    Why it matters

    A construction would give explicit generators for every extension in scope through a common special-value description. An obstruction for one number field would show a limit to that proposed universal construction.

    Sources

    Also ranked #35 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #232 of 286
    Math-interested nonspecialists
    #97 of 286
    90% bootstrap rank range
    92–110
    Included in the Top 100
    53.9% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #109 of 286
    OpenAI GPT comparisons only
    #102 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
  100. Rank 100Extended Riemann hypothesis for Dedekind zeta functionsEvery number field—a finite algebraic extension of the rationals—has a Dedekind zeta function. Must all its zeros with real part between zero and one lie on the middle line, with real part one-half?Open · disputed claim

    Exact statement

    For every algebraic number field K (a finite extension of Q, including K=Q), does every zero s of its Dedekind zeta function ζ_K(s) in the critical strip 0<Re(s)<1 satisfy Re(s)=1/2?

    Number theory & arithmetic geometry

    Status

    Open · disputed claim

    Earlier note (2026-09-14): Dedekind ERH remains open, with a disputed proof claim. Zhang's posting was revised to v19 on September 14, 2026; objections or prior assessments of v18 do not review v19. Independent acceptance of the newest claim was not established. October 7, 2026: OpenAI’s October 6 release reports partial progress. The stated half-plane theorem applies to zeta_Q and, by the usual abelian L-function factorization, to Dedekind zeta functions in the abelian settings represented by its Dirichlet/finite-order Hecke factors. It supplies neither the critical line 1/2 nor all number fields, so it is only restricted zero-free progress for the universal Dedekind-zeta target. No independent proof audit or Lean verification was performed for this update.

    Status reviewed .

    Why it matters

    A proof would supply the zeta-function hypothesis used in effective Chebotarev estimates for how primes split in number fields, with their other conditions retained. Including the rational field would also establish the ordinary Riemann Hypothesis.

    Sources

    Also ranked #12 in the Top 500 by importance

    Ranking details and uncertainty

    Audience and model-specific ranks refer to all 286 contenders, including those outside this list.

    English-reading general adults
    #46 of 286
    Math-interested nonspecialists
    #110 of 286
    90% bootstrap rank range
    83–115
    Included in the Top 100
    45.8% of the 1,000 conditional bootstrap refits
    Anthropic Claude comparisons only
    #96 of 286
    OpenAI GPT comparisons only
    #111 of 286

    These ranges describe variation within the ranking procedure, not measured public recognition, proof confidence or likelihood of a solution. How the ranking works.

    Link to this problem
Former Top 100 entries (6)

Former Top 100 entries

These entries retain their fame rank and edition date from the edition in which they were listed. A problem leaving the list is not necessarily solved; each entry records its status and scope.

Source-reported resolutions do not by themselves establish acceptance of a ProofAtlas proof or formalization.

  • Former rank 046Unique Games ConjectureFormer fame rank #46 · Source-reported proved

    Exact statement

    For every fixed pair of real constants epsilon, delta > 0 with epsilon + delta < 1, there exists an integer k >= 2, depending only on epsilon and delta, such that the following promise problem is NP-hard under deterministic polynomial-time many-one reductions. An instance is a finite directed constraint graph G=(V,E) with E nonempty and, for each edge e=(v,w), a permutation pi_e of [k]. For a labeling L:V->[k], an edge is satisfied if pi_e(L(v))=L(w); val(G) is the largest fraction of satisfied edges over all labelings. Distinguish val(G) >= 1-epsilon from val(G) <= delta, with no requirement on instances between these thresholds. The graph is the growing input; epsilon, delta and k are fixed constants. NP-hardness means that every language in NP has such a reduction mapping yes inputs to the first promised set and no inputs to the second.

    Status and scope

    Source-reported proved

    Resolved (proved) on the basis of the OpenAI mathematics release of 6 October 2026. Exact scope reviewed; proofs not independently audited, refereeing not independently confirmed, and Lean artifacts not executed. Family 102: The manuscript asserts deterministic polynomial-time reductions from 3SAT to fixed-alphabet Unique Games with completeness 1-epsilon and soundness delta for every fixed epsilon,delta in (0,1/2). This covers the ranked range epsilon,delta>0, epsilon+delta<1: choose smaller positive parameters below 1/2 and below the requested values, giving stronger completeness and soundness promises. The alphabet remains fixed independently of the graph input. This is the ordinary UGC target, not merely a conditional approximation consequence.

    Status reviewed .

    Sources

    Read the Top 500 history entry

    Link to this former entry
  • Former rank 060Erdős problem 3: divergent harmonic series and long arithmetic progressionsFormer fame rank #60 · Source-reported proved

    Exact statement

    Determine whether every subset A of the natural numbers whose reciprocal sum diverges must contain arbitrarily long arithmetic progressions.

    Status and scope

    Source-reported proved

    Resolved (proved) on the basis of the OpenAI mathematics release of 6 October 2026. Exact scope reviewed; proofs not independently audited, refereeing not independently confirmed, and Lean artifacts not executed. Family 159: The first assertion covers every set of positive integers with divergent reciprocal sum and yields arithmetic progressions of every finite length. This is precisely the frozen reciprocal-sum conjecture, with no positive-density hypothesis or restriction to a single progression length. The quantitative r_k(N) estimate supplies a stronger finite-range statement but is not needed to identify the exact affirmative resolution.

    Status reviewed .

    Sources

    Read the Top 500 history entry

    Link to this former entry
  • Former rank 064Hadwiger ConjectureFormer fame rank #64 · Source-reported disproved

    Exact statement

    Prove that every simple graph with no K_{t+1} minor is t-colourable for every integer t >= 1.

    Status and scope

    Source-reported disproved

    Resolved (disproved) on the basis of the OpenAI mathematics release of 6 October 2026. Exact scope reviewed; proofs not independently audited, refereeing not independently confirmed, and Lean artifacts not executed. Family 157: The first paper gives finite simple graphs with chromatic number greater than their Hadwiger number h(G), even with fractional chromatic number greater than h(G). Taking t=h(G), each example has no K_{t+1} minor but is not t-colourable, directly contradicting the ranked all-t assertion. The separate linear list-coloring bound with an unspecified universal multiplier C does not restore the conjectured coefficient one.

    Status reviewed .

    Sources

    Read the Top 500 history entry

    Link to this former entry
  • Former rank 077Nonexistence of Landau–Siegel zeros for quadratic Dirichlet L-functionsFormer fame rank #77 · Source-reported proved

    Exact statement

    Does there exist an absolute constant δ>0 such that, for every primitive quadratic Dirichlet character χ of conductor q≥3, L(β,χ)≠0 for every real β with 1−δ/log q<β<1? The logarithm is natural and the same δ must work for all conductors.

    Status and scope

    Source-reported proved

    Resolved (proved) on the basis of the OpenAI mathematics release of 6 October 2026. Exact scope reviewed; proofs not independently audited, refereeing not independently confirmed, and Lean artifacts not executed. Family 003: The companion explicitly supplies one absolute positive c with (1-beta) log q >= c for every real zero of each primitive nonprincipal real Dirichlet L-function of conductor q>=3. Primitive quadratic characters are included, and setting delta=c excludes exactly the open real interval in the target, uniformly in conductor.

    Status reviewed .

    Sources

    Read the Top 500 history entry

    Link to this former entry
  • Former rank 083Irrationality of Catalan's ConstantFormer fame rank #83 · Source-reported proved

    Exact statement

    Prove or disprove that Catalan's constant G = beta(2) = sum over n >= 0 of (-1)^n / (2n+1)^2 = 0.9159655941... is an irrational number.

    Status and scope

    Source-reported proved

    Resolved (proved) on the basis of the OpenAI mathematics release of 6 October 2026. Exact scope reviewed; proofs not independently audited, refereeing not independently confirmed, and Lean artifacts not executed. Family 005: The alternating reciprocal odd-square series in the abstract is exactly the frozen definition G=beta(2). The claimed irrationality therefore affirms the full target, with no additional parameter, genericity, or conditional hypothesis; no transcendence assertion is inferred.

    Status reviewed .

    Sources

    Read the Top 500 history entry

    Link to this former entry
  • Former rank 087Convergence of the Flint Hills SeriesFormer fame rank #87 · Source-reported proved

    Exact statement

    Determine whether the Flint Hills series sum_{n=1}^infty 1/(n^3 sin(n)^2) converges.

    Status and scope

    Source-reported proved

    Resolved (proved) on the basis of the OpenAI mathematics release of 6 October 2026. Exact scope reviewed; proofs not independently audited, refereeing not independently confirmed, and Lean artifacts not executed. Family 017: The abstract expressly concludes convergence of exactly sum 1/(n^3 sin^2 n), with angles in radians, from the stronger irrationality-exponent assertion for pi. The frozen target only asks whether this series converges, so the stated consequence fully decides it affirmatively; no claim about normality or an effective irrationality bound is required.

    Status reviewed .

    Sources

    Read the Top 500 history entry

    Link to this former entry

How the ranking works

What is being estimated?

Familiarity with each exact open question among two modelled English-reading audiences: general adults and math-interested nonspecialists, weighted 50/50. Recognition of a broader topic or a solved neighbouring problem does not automatically count.

The ranking uses equal weights for the OpenAI GPT and Anthropic Claude model families. Audience scores, not rank positions, are averaged to produce the combined ordering.

For example, the Collatz conjecture is #3 here and #175 in the Top 500 by importance: the lists assess different things.

How much should the order mean?

The source compared 286 contenders and ran 1,000 conditional bootstrap refits. Each entry shows its 90% rank range and how often it appeared in the Top 100 during those refits.

The ranges depend on the chosen contenders, model families and comparison procedure. They are not population recognition percentages, accuracy estimates, or measures of mathematical truth. Ranges above 100 mean that an entry may fall outside the list under refitting.

The top three are not separated by their 90% ranges; their displayed order is the point estimate.

Different editions may compare different contender pools; their scores are not directly comparable. Status notes and their review dates concern the mathematics separately from this ranking.

Changes since Sep 19, 2026

Changes refer to the source ranking editions of 19 September and 7 October 2026. Scores from different contender pools are not directly comparable; rank changes do not establish changes in public recognition.

Entered this edition

Left this edition