Papex

Open problem · Mathematics · Open

Sums of three cubes

Can every whole number that does not leave remainder 4 or 5 on division by 9 be written as a sum of three integer cubes?

Summary

Numbers leaving remainder 4 or 5 after division by 9 are never a sum of three cubes; whether every other integer is remains unknown. Large searches found cubes summing to 33 and 42 in 2019, while seven numbers below 1,000 are still unresolved.

Why it is open

An integer of the form x³ + y³ + z³ cannot leave remainder 4 or 5 on division by 9, and several authors have conjectured that every other integer has such a form, Jean-Louis Colliot-Thélène and Olivier Wittenberg note, citing among others a 1992 paper of Roger Heath-Brown [3]. Over the rationals the question is settled: every integer, and every rational number, is a sum of three cubes of rational numbers; for cubes of integers the question is open [3].

Colliot-Thélène and Wittenberg do not rule out that polynomial identities could prove the conjecture, but L. N. Vaserstein showed that any identity x(t)³ + y(t)³ + z(t)³ = t with rational polynomials needs one of them to have degree at least 5 [3]. A disproof needs a reason why some integer has no representation, beyond remainders modulo 9. In 2012 Colliot-Thélène and Wittenberg showed that the reciprocity laws in the integral Brauer–Manin obstruction, a general method that can explain why some equations have no integer solutions, exclude no integer outside those two classes [3].

Searching alone cannot show that a number has no representation. A program can list every sum of three cubes, but, as Bjorn Poonen notes, such a listing does not by itself let one test whether a given number such as 33 belongs to it; in 2008 he opened his survey of undecidability in number theory with 33 as an unsolved case [1]. A resolution would be a proof for every integer not of the form 9n ± 4, or one such integer proved to have no representation. No institution or prize sets the bar.

What is known

Early searches (1953–2016). In 1953 L. J. Mordell asked whether 3 has representations other than 1³ + 1³ + 1³ and 4³ + 4³ + (−5)³; the next year J. C. P. Miller and M. F. C. Woollett ran the first computer search for one, on the electronic delay storage automatic calculator at Cambridge, and extended it to every positive k ≤ 100 [6]. Noam Elkies later gave an algorithm that finds rational points near the curve X³ + Y³ = 1 by lattice basis reduction [5]. With it, Andreas-Stephan Elsenhans and Jörg Jahnel searched all positive n < 1,000 other than cubes and twice cubes, with |x|, |y|, |z| < 10¹⁴, and found 3,519 new solutions; in 2009 they reported that 14 numbers below 1,000 remained open: 33, 42, 74, 114, 165, 390, 579, 627, 633, 732, 795, 906, 921 and 975 [2]. In 2016 Sander Huisman raised the bound to 10¹⁵ and found the first solution for 74, which left 33 and 42 as the only open cases below 100 [4].

33, 42 and a new 3 (2019–2021). Andrew Booker's 2019 algorithm finds every solution whose smallest coordinate is below a bound, rather than the largest as in Elkies' method [5], refining an approach proposed by Heath-Brown and co-authors in 1993 [6]. With it he found the first solutions for 33 and 795 [5]: 33 = 8,866,128,975,287,528³ + (−8,778,405,442,862,239)³ + (−2,736,111,468,807,040)³ [5]. His search ruled out solutions with smallest coordinate up to 10¹⁶ for the 11 open numbers below 1,000 [6]. In September 2019 Booker and Andrew Sutherland ran an improved version on Charity Engine, a crowd-sourced grid of about 500,000 personal computers, and found solutions for 42, 165 and 906, then a third representation of 3, which answered Mordell's question [6]: 3 = 569,936,821,221,962,380,720³ + (−569,936,821,113,563,493,509)³ + (−472,715,493,453,327,032)³ [6]. A lower-priority search during 2020 found a solution for 579 in January 2021 [6].

What remains (2021–2026). Booker and Sutherland listed seven numbers below 1,000 as unresolved: 114, 390, 627, 633, 732, 921 and 975 [6]. They are "reluctant to make any unconditional claims" that their searches rule out solutions they did not find, because of possible undetected hardware or software errors on personal computers that typically lack error-correcting memory [6]. In October 2026 FrontierMath: Open Problems, a list of unsolved problems that Epoch AI poses to AI systems, gave 114 as the smallest number whose status is unknown and also asked for a fourth representation of 3 [10].

Heath-Brown's conjecture. For cube-free k, Heath-Brown conjectured that the number of solutions with |x|, |y|, |z| ≤ B grows like c_k log B, with a constant c_k that is positive exactly when k is not congruent to ±4 modulo 9, so that every such k should have infinitely many representations [8]. Elsenhans and Jahnel found that the numbers with the most solutions in their data, 792 and 720, fit this prediction well; for 33 and 42 it suggested "there could be a few solutions, but not many" [2]. Booker and Sutherland computed the predicted densities to high precision and compared them with Huisman's data, which they found gives "strong evidence for Heath-Brown's conjecture, at least on average over k" [6]. In 2025 Tim Browning and Florian Wilsch published a general heuristic for integer points on affine cubic surfaces that specialises to Heath-Brown's conjecture for sums of three cubes [8].

Main approaches

Direct search. Look for representations with ever larger bounds [2, 5, 6]. Strongest evidence: searches have found representations for every number below 100 not of the form 9n ± 4, and for all but seven below 1,000 [6]. Main difficulty: a search can find a solution but not prove that none exists, and its negative results carry the caveat about undetected errors [6].

Heuristic counting. Predict how many solutions lie below each height and compare with data [2, 6, 8]. Strongest evidence: the data of Elsenhans and Jahnel and of Booker and Sutherland agree with Heath-Brown's prediction [2, 6], and Browning and Wilsch's general heuristic reproduces it [8]. Main difficulty: a heuristic is not a proof; for the related surface x³ + y³ + z³ = 1, Browning and Wilsch found the numerical data for their prediction "less compelling" [8].

Obstructions. Look for a reason why some integer has no representation [3]. Strongest evidence: such constraints exist: for k = 3, cubic reciprocity forces x, y and z to be congruent modulo 9, as J. W. S. Cassels observed [3, 5], which Colliot-Thélène and Wittenberg interpret as a Brauer–Manin obstruction to strong approximation [3]. Main difficulty: the same authors proved that the integral Brauer–Manin obstruction rules out no integer outside the classes of 4 and 5 modulo 9 [3].

Analytic methods. Count representations with the circle method, an analytic technique for counting solutions, under standard conjectures about L-functions [7, 9]. Strongest evidence: Victor Wang's preprint shows that standard hypotheses, mainly the Ratios and Square-free Sieve Conjectures, would imply that almost all integers not congruent to ±4 modulo 9 are sums of three cubes [7]. With Browning and Jakob Glas he proved an analogue for polynomials over a finite field of characteristic greater than 3: a positive proportion are sums of three cubes of minimal degree, assuming a form of the Ratios Conjecture [9]. Main difficulty: these results depend on unproved hypotheses, and "almost all" leaves room for exceptions [7, 9].

Open sub-questions

  1. 114 and the other six. Does a search for 114, 390, 627, 633, 732, 921 and 975 find solutions near the heights at which Heath-Brown's predicted density, as tabulated by Booker and Sutherland, makes one solution expected [6]?
  2. A fourth representation of 3. Does x³ + y³ + z³ = 3 have a fourth solution, and how far beyond the range of Booker and Sutherland's verification search, which found no other solution with |z| up to that of their third solution, must one look [6, 10]?
  3. Certified negative results. Can the ranges in which large distributed searches found no solution be confirmed independently, so that Booker and Sutherland's caveat about undetected errors can be lifted for the seven open numbers [6]?
  4. Fewer hypotheses. Can the conditional result that almost all integers not congruent to ±4 modulo 9 are sums of three cubes be proved with weaker hypotheses, first perhaps over polynomials over a finite field [7, 9]?
  5. Data for x³ + y³ + z³ = 1. Do counts of solutions at larger heights bring the data for this surface closer to Browning and Wilsch's prediction [8]?

Work on this in Papex

If you have a testable answer to one of the sub-questions above, you can draft it as a private hypothesis in Papex, or prepare the draft from Claude or ChatGPT with the Papex connector. You review the exact version, sign it and decide whether to publish it; a public record that names this problem is listed on this page. A Papex record shows who signed which version and when. It is not peer review and does not mean the hypothesis is correct.

Key references

  1. Poonen, B. (2008). Undecidability in number theory. Notices of the American Mathematical Society 55(3), 344–350. Source · Checked 16 Oct 2026
  2. Elsenhans, A.-S. & Jahnel, J. (2009). New sums of three cubes. Mathematics of Computation 78(266), 1227–1230 (published online 2008). 10.1090/S0025-5718-08-02168-6 · Checked 16 Oct 2026
  3. Colliot-Thélène, J.-L. & Wittenberg, O. (2012). Groupe de Brauer et points entiers de deux familles de surfaces cubiques affines. American Journal of Mathematics 134(5), 1303–1327. 10.1353/ajm.2012.0036 · Checked 16 Oct 2026
  4. Huisman, S. G. (2016). Newer sums of three cubes. Preprint, arXiv:1604.07746 (version 1, 26 April 2016), not peer reviewed. Source · Checked 16 Oct 2026
  5. Booker, A. R. (2019). Cracking the problem with 33. Research in Number Theory 5(3), 26. 10.1007/s40993-019-0162-1 · Checked 16 Oct 2026
  6. Booker, A. R. & Sutherland, A. V. (2021). On a question of Mordell. Proceedings of the National Academy of Sciences 118(11), e2022377118. 10.1073/pnas.2022377118 · Checked 16 Oct 2026
  7. Wang, V. Y. (2021). Sums of cubes and the Ratios Conjectures. Preprint, arXiv:2108.03398 (version 2, 19 April 2023), not peer reviewed. Source · Checked 16 Oct 2026
  8. Browning, T. & Wilsch, F. (2025). Integral points on cubic surfaces: heuristics and numerics. Selecta Mathematica 31(4), 81. 10.1007/s00029-025-01074-1 · Checked 16 Oct 2026
  9. Browning, T., Glas, J. & Wang, V. Y. (2026). Sums of three cubes over a function field. Forum of Mathematics, Sigma 14, e112. 10.1017/fms.2026.10259 · Checked 16 Oct 2026
  10. Epoch AI. Sum of three cubes. FrontierMath: Open Problems, epoch.ai, accessed 2026-10-16. Source · Checked 16 Oct 2026

Page reviewed 16 Oct 2026

Source

Open, as listed by Wikipedia · Unsolved problems in mathematics. Checked 08 Oct 2026.

The status is the source’s. Papex does not decide whether a problem is solved.

Records on this problem

No public record addresses this problem yet.

More in Mathematics