Settled · 11 Partly settled · 7 Open · 3 Too vague to settle · 2Underlined: read aloud in Paris · 10
I · The lecture
A Wednesday morning in Paris
In the summer of 1900 Paris was full of the future. The World’s Fair had brought moving pavements, talking films and a palace lit by electricity, and fifty million visitors came to look. In the middle of it, about two hundred and fifty mathematicians met at the Sorbonne for their second international congress. On the morning of Wednesday the eighth of August, in a side session on history and teaching, a thirty-eight-year-old professor from Göttingen stood up to give a talk that was not even on the printed programme, because he had finished writing it too late.
Paris, summer 1900. The World’s Fair seen from under the Eiffel Tower, looking down the Champ de Mars to the Palace of Electricity.Hermann Minkowski, who talked Hilbert into it.
David Hilbert had been unsure what to speak about. His closest friend, Hermann Minkowski, had urged him months earlier to attempt something nobody did at such meetings: forget past results and look ahead. A lecture on the problems of the coming century, Minkowski wrote, might still be talked about decades later. It was one of the better predictions in the history of science.
Who of us would not be glad to lift the veil behind which the future lies hidden; to cast a glance at the next advances of our science and at the secrets of its development during future centuries?
David Hilbert, opening words, Paris 1900
He began by asking what makes a problem worth a mathematician’s life. It must be clear, he said, and he borrowed a line from an old French mathematician: a theory is not complete until you can explain it to the first person you meet in the street. It must be hard enough to tempt us and not so hard that it mocks us. And it should be a signpost, pointing past itself. He gave examples. Johann Bernoulli’s challenge about the curve of fastest descent had created a whole branch of analysis. Fermat’s claim about sums of powers, useless in itself, had forced the invention of new kinds of number.
The Palace of Electricity and the Château d’Eau, the showpiece of the 1900 exposition. The mathematicians met a few kilometres away at the Sorbonne.
Then he read out his problems. There was time for only ten. The published version, which appeared later that year, had twenty-three, running from the nature of the infinite to the stacking of spheres. An American mathematician, Mary Winston Newson, the first woman from the United States to earn a doctorate in mathematics at a European university, translated it into English in 1902, and it is her translation that most of the world has read.
The room’s reaction was polite and brief. The effect arrived slowly and then did not stop. To solve one of Hilbert’s problems became the surest way to make a name; Hermann Weyl said that whoever did so passed into the honours class of the mathematical community. Whole fields grew up around single entries on the list. Several of the mathematicians who changed the world in the twentieth century did it by crossing off a number between one and twenty-three.
What follows is the whole list. For each problem I give the question as plainly as I can, and then what happened to it. The colours are a judgement call, and specialists argue about several of them, because Hilbert’s wording was often loose enough that a problem could be solved under one reading and left open under another. Where that is so, I say so.
II · The list
The twenty-three, one by one
Use the buttons to see only the settled problems, or only the open ones. Problems marked “read aloud in Paris” are the ten Hilbert had time to present in person. The rest appeared only in print.
1
Georg Cantor, who asked it first
Problem 1Set theoryRead aloud in Paris
The continuum hypothesis
Cantor had shown that infinities come in sizes: the counting numbers are a smaller infinity than the points on a line. Is there any size strictly in between? Cantor believed not, and could not prove it.
The answer turned out to be that the usual axioms of mathematics cannot say. Kurt Gödel showed in 1940 that the hypothesis cannot be disproved from them. Paul Cohen showed in 1963 that it cannot be proved either, and invented a new technique, forcing, to do it. Whether that closes the question or only shows we need better axioms is still argued.
Undecidable from the standard axiomsℵ₀ < ? < 2^ℵ₀
2
Problem 2FoundationsRead aloud in Paris
The consistency of arithmetic
Prove that the axioms of arithmetic can never lead to a contradiction. Hilbert wanted mathematics to certify its own safety, using only the simplest reasoning.
In 1931 Gödel proved that no consistent system rich enough for arithmetic can prove its own consistency. That was not the answer Hilbert hoped for. In 1936 Gerhard Gentzen did prove arithmetic consistent, but only by using a principle stronger than arithmetic itself. People still disagree on whether that counts.
Proved, but not on Hilbert’s termsPA ⊬ Con(PA)
3
Problem 3Geometry
Cutting up tetrahedra
Any two polygons of equal area can be cut into finitely many pieces and rearranged into each other. Is the same true in space? Can two pyramids with equal base and height always be cut and reassembled into one another?
No. Max Dehn, a student of Hilbert’s, found a hidden quantity that cutting and gluing never changes, and showed that a cube and a regular tetrahedron of the same volume have different values of it. He did this within months. It was the first of the twenty-three to fall.
No. Dehn, 1900D(cube) ≠ D(tetrahedron)
4
Problem 4Geometry
Geometries where straight lines are shortest
Describe all the geometries in which the shortest path between two points is a straight line. Euclid’s is one. What are the others?
As stated, the problem is too open-ended to have a single answer; it depends on which axioms you keep. Under one natural reading, Aleksei Pogorelov gave a complete solution in 1973. Most historians file it under problems that were a research programme and not a question.
A programme, not a question
5
Problem 5Groups
Are continuous symmetries automatically smooth?
Sophus Lie built his theory of continuous symmetry, the mathematics of rotations and other smooth motions, by assuming everything could be differentiated. Is that assumption needed, or does continuity alone force smoothness?
Continuity is enough. Andrew Gleason, Deane Montgomery and Leo Zippin proved it in 1952, with Hidehiko Yamabe completing the general case the next year. A broader reading of the problem, the Hilbert–Smith conjecture, remains open.
Yes. Gleason, Montgomery, Zippin, 1952
6
Problem 6PhysicsRead aloud in Paris
Axioms for physics
Do for physics what Euclid did for geometry. Write down a small set of axioms from which mechanics and the theory of probability follow by pure deduction.
Probability got its axioms from Andrey Kolmogorov in 1933, and they are still the ones in use. Physics did not hold still: relativity and quantum theory arrived within a generation, and no single axiomatic foundation for all of physics exists. Mathematicians continue to work on pieces, such as deriving the laws of fluids from the motion of atoms.
Probability yes, physics no
7
Problem 7Number theoryRead aloud in Paris
Is 2 to the power √2 transcendental?
Take an algebraic number other than 0 or 1 and raise it to an irrational algebraic power, as in 2 raised to √2. Is the result always transcendental, meaning it is not the root of any polynomial with whole-number coefficients?
Yes. Aleksandr Gelfond proved it in 1934 and Theodor Schneider independently in 1935. Hilbert had once told a lecture audience that this problem was harder than the Riemann hypothesis and that nobody in the room would live to see it solved. He lived to see it solved.
Yes. Gelfond and Schneider, 19342^√2 = 2.6651441…
8
Bernhard Riemann, about 1850
Problem 8Number theoryRead aloud in Paris
The Riemann hypothesis
The prime numbers look scattered, but Bernhard Riemann found in 1859 that their distribution is governed by the zeros of a single function. He guessed that all the interesting zeros lie on one vertical line. Prove it. Hilbert bundled in Goldbach’s conjecture and the twin prime conjecture for good measure.
Still open after more than 125 years. Computers have checked trillions of zeros and every one is on the line. Hilbert is said to have remarked that if he woke after a thousand years of sleep, his first question would be whether it had been proved. Goldbach and the twin primes are open too.
Openζ(s) = 0 ⟹ Re(s) = ½
9
Problem 9Number theory
The most general reciprocity law
Gauss’s law of quadratic reciprocity describes a surprising symmetry between pairs of primes. Find the most general law of this kind, for any power and in any number system.
Emil Artin’s reciprocity law of 1927 answered it for a large and important class of number systems, and it became the centrepiece of class field theory. The general case is one of the goals of the Langlands programme, which is still under construction.
Artin, 1927, for the abelian case
10
Problem 10Number theory
An algorithm for whole-number solutions
Given a polynomial equation with whole-number coefficients, devise a procedure that decides in a finite number of steps whether it has a solution in whole numbers.
There is no such procedure, and there cannot be. Yuri Matiyasevich proved it in 1970 at the age of twenty-two, finishing two decades of work by Martin Davis, Hilary Putnam and Julia Robinson. The proof needed a definition of “procedure” that did not exist in 1900; it came from the theory of computation.
Impossible. Matiyasevich, 1970p(x₁, …, xₙ) = 0
11
Problem 11Number theory
Quadratic forms over any number field
Expressions such as x² + y² have been studied over the whole numbers since antiquity. Extend the theory of such quadratic forms to the far larger number systems of algebraic number theory.
Helmut Hasse settled the case of fractional solutions in the 1920s with his local–global principle: a quadratic equation can be solved if it can be solved in every simpler “local” setting. The case of whole-number solutions is harder and is still an active area.
Rational case done, 1923
12
Problem 12Number theory
Kronecker’s dream of his youth
A theorem of Kronecker and Weber says a certain family of number systems built on the rationals can all be generated from one function, the exponential. Find the functions that play this role when you start from any other number field.
Known for one further family, where elliptic functions do the job. In 2021 Samit Dasgupta and Mahesh Kakde gave an answer of a new kind for a second large family. The general problem is open, and it was the one Hilbert himself seems to have cared about most among the number theory problems.
Open, with recent progress
13
Problem 13AlgebraRead aloud in Paris
Solving the seventh-degree equation
The general equation of degree seven has a solution that depends on three parameters. Can that three-variable function be built up out of functions of only two variables? Hilbert thought not.
For continuous functions Hilbert was wrong. In 1957 Vladimir Arnold, aged nineteen and working with Kolmogorov, proved that every continuous function of several variables can be assembled from functions of two. If you require the building blocks to be algebraic, which may be what Hilbert meant, the problem is unsolved.
When a group of symmetries acts on polynomials, the polynomials it leaves unchanged form a system of their own. Can that system always be generated from finitely many of its members? Hilbert had proved a famous case himself.
No. Masayoshi Nagata constructed a counterexample in 1959. It was one of the few problems where the expected answer was simply false.
No. Nagata, 1959
15
Problem 15Geometry
Making Schubert’s counting rigorous
Hermann Schubert had a method for counting geometric configurations, for instance the number of lines that meet four given lines in space. It produced correct numbers by reasoning nobody could justify. Put it on a sound footing.
Much of it has been, through the intersection theory developed across the twentieth century, and many of Schubert’s numbers have been confirmed. A complete justification of everything in his book has not been achieved.
Largely justifiedlines meeting four lines: 2
16
Problem 16GeometryRead aloud in Paris
The shapes of curves and the cycles of flows
Two questions in one. How can the separate ovals of a curve defined by a polynomial equation be arranged in the plane? And how many isolated closed loops can the flow of a polynomial differential equation have?
Open in both halves. For the second, it is not even known whether there is a finite upper limit for the simplest non-trivial case, equations of degree two.
Open
17
Problem 17Algebra
Positive functions as sums of squares
A square is never negative, so a sum of squares never is. Does it work the other way? If a quotient of polynomials is never negative, can it always be written as a sum of squares of such quotients?
Yes. Emil Artin proved it in 1927, inventing the theory of real fields along the way.
Yes. Artin, 1927f ≥ 0 ⟹ f = g₁² + ⋯ + gₖ²
18
Problem 18Geometry
Tilings, crystals and stacking oranges
Three questions about filling space. Are there only finitely many kinds of crystal symmetry in each dimension? Is there a solid that tiles space but only in an irregular way? And what is the densest way to pack equal spheres?
All three are settled. The first was answered yes in 1910. Karl Reinhardt found a solid of the required kind in 1928. The third is Kepler’s conjecture from 1611, that the greengrocer’s pyramid is best; Thomas Hales proved it in 1998 with heavy computer assistance, and a fully machine-checked proof was finished in 2014.
Yes, yes, and Hales, 1998π / √18 ≈ 0.7405
19
Problem 19AnalysisRead aloud in Paris
Are the solutions always smooth?
Many laws of nature say that a system settles into the state that minimizes some quantity, such as energy. When the problem is well behaved, are the solutions necessarily as smooth as functions can be?
Yes. Ennio De Giorgi in 1957 and John Nash in 1958 proved it independently, by different methods, neither knowing of the other.
Yes. De Giorgi and Nash, 1957
20
Problem 20Analysis
Do boundary value problems always have solutions?
Given a minimization problem of the same kind and conditions on the boundary, is there always a solution, provided one is willing to widen what counts as a solution?
Yes, in a form that took most of the century to work out. There was no single breakthrough. The answer is spread over the work of many analysts and became the modern theory of partial differential equations.
Yes, through decades of work
21
Problem 21AnalysisRead aloud in Paris
Differential equations with prescribed behaviour
A solution of a linear differential equation changes in a definite way when you follow it around one of the equation’s singular points. If you specify in advance how the solutions should change, is there always an equation that does it?
For decades the answer was taken to be yes, on the strength of a 1908 proof by Josip Plemelj. The proof had a gap. In 1989 Andrei Bolibrukh found a counterexample, so the answer to the question in its classical form is no.
No. Bolibrukh, 1989
22
Problem 22AnalysisRead aloud in Paris
Uniformization
A circle can be described by one equation in x and y, or by letting a single variable run and reading off both coordinates. Can every curve given by an analytic relation be described in this second, uniform way?
For curves, yes. Henri Poincaré and Paul Koebe proved the uniformization theorem independently in 1907, one of the summits of nineteenth-century mathematics reached seven years into the twentieth. Hilbert also asked about relations in more variables, and that part is not finished.
Curves: yes, 1907
23
Problem 23Analysis
Develop the calculus of variations further
Not a question at all. Hilbert closed the list by urging mathematicians to push forward the calculus of variations, the study of problems where one seeks the best possible curve or surface.
They did, and the subject is unrecognizable from what it was in 1900. But there is nothing here to mark as proved or refuted.
An exhortation
III · The scorecard
A hundred and twenty-five years later
Counting is harder than it looks, since a problem that asked three questions can be two-thirds solved, and a problem that received an answer nobody wanted may or may not be finished. By my count the board stands like this.
11
Settled
1 · 3 · 5 · 7 · 10 · 14 · 17 · 18 · 19 · 20 · 21
7
Partly settled
2 · 6 · 9 · 11 · 13 · 15 · 22
3
Open
8 · 12 · 16
2
Too vague to settle
4 · 23
Look at the settled column again and notice how many of the answers were not what Hilbert expected. The third problem, the fourteenth and the twenty-first were answered no. The tenth was answered by proving that the thing he asked for cannot exist. The first was answered by showing that the axioms are silent. He predicted the thirteenth would be impossible and a nineteen-year-old showed that, read one way, it could be done. A list of problems is supposed to be a list of things we do not know. This one also turned out to be a list of things we were wrong about.
Each circle is a problem, placed at the year of its decisive result. The third fell within months. The eighteenth took ninety-eight years. Three have no circle at all.
The timeline shows something else. The answers did not come in order of difficulty, or in any order. They came when the tools existed. The tenth problem could not even be stated precisely until the 1930s, when the idea of an algorithm was defined. The first needed forcing, which Cohen had to invent. The last part of the eighteenth needed computers that could check thousands of cases, and then needed software able to verify the computers. Hilbert was setting homework that depended on mathematics nobody had imagined yet. He knew it. That was the point of the exercise.
IV · Königsberg, 1930
“We must know. We will know.”
Underneath the list was a creed. Hilbert believed that every well-posed mathematical question has an answer and that the human mind can find it. In Paris he had put it as a call that every mathematician hears: there is the problem; seek its solution; you can find it by pure reason, for in mathematics there is no ignorabimus. The Latin word means “we shall not know,” and it was the slogan of a fashionable pessimism about the limits of science. Hilbert despised it.
Thirty years later he went home to Königsberg, the city where he was born, to be made an honorary citizen. On the eighth of September 1930 he gave an address that was also recorded for radio; you can still hear his voice, high and emphatic. He ended with six words that were later cut into his gravestone in Göttingen.
Wir müssen wissen. Wir werden wissen.
David Hilbert, Königsberg, 8 September 1930
We must know. We will know. What gives the story its edge is the calendar. The day before, in the same city, at a small conference on the foundations of mathematics, a quiet twenty-four-year-old from Vienna named Kurt Gödel had mentioned a result he had just obtained. In any consistent system of axioms strong enough for arithmetic, there are true statements that the system can never prove. Almost nobody in the room grasped what he had said. It meant that Hilbert’s second problem could not be solved in the way Hilbert intended, and that the dream of one finite set of axioms settling everything was over.
Kurt Gödel as a student in Vienna, 1925, five years before Königsberg.
It is tempting to tell this as a defeat, and it is usually told that way. I think that is too simple. Gödel did not show that there are questions without answers. He showed that truth is larger than any one system of proof, which is a different and in some ways a grander thing, and it is one of the threads in our essay on whether mathematics is invented or discovered. Hilbert’s faith that the mathematical world is orderly and open to reason has held up very well. Eleven of his problems have been settled, and not one has been abandoned. Even the continuum hypothesis, which the axioms cannot decide, is still being worked on by people who think a better axiom will decide it.
V · A footnote
The twenty-fourth problem
For a hundred years the list had twenty-three entries. In 2000 the historian Rüdiger Thiele, reading through Hilbert’s notebooks in the Göttingen library, found a twenty-fourth that Hilbert had planned to include and then left out. It is unlike the others. It does not ask whether something is true. It asks for a way to tell when a proof is the simplest possible: for criteria of simplicity, and for a demonstration that certain proofs cannot be made simpler.
Every mathematician has the intuition he was reaching for. Some proofs are long, clever and unsatisfying, and some are so short and inevitable that they seem to have been there all along. Compare a few of the classics in the original papers and you will feel it. Hilbert wanted that feeling turned into a theorem. A century later logicians and computer scientists are working on exactly this, under names he never heard, and the problem he crossed out may prove to be as fertile as the ones he kept.
VI · After Hilbert
What a good list does
On the twenty-fourth of May 2000, a hundred years after the lecture and in the same city, the Clay Mathematics Institute announced seven Millennium Prize Problems, each with a prize of one million dollars. It was a conscious tribute. Only one problem appears on both lists: the Riemann hypothesis, Hilbert’s eighth, the question of how the primes are distributed. You can watch the Riemann hypothesis explained to see why it has outlasted everyone who has tried it. Of the seven Millennium problems, one has been solved in a quarter of a century. Grigori Perelman proved the Poincaré conjecture and then declined the million dollars.
Others have made lists. None has had the effect of Hilbert’s, and it is worth asking why. Part of it is that he was close to the last person who could survey the whole of mathematics; the problems range over set theory, number theory, algebra, geometry, analysis and physics, and he had done first-rate work in most of them. Part of it is taste. He picked questions that were easy to state and that sat at the joints of the subject, so that answering them meant building something. And part of it is tone. The lecture is not cautious. It tells young mathematicians that the hardest questions are the ones worth their time and that those questions will give way.
That confidence is the real legacy, more than any single theorem. It runs through the whole history of mathematics, from Euclid onward: the belief that the world of number and form is not a fog but a country, with definite features, that rewards whoever sets out to map it. Three of Hilbert’s problems are still open. Somebody reading about them for the first time this year will close one.
Hilbert’s grave, Göttingen. Photo: Julian Herzog, CC BY 4.0
Wir müssen wissen.Wir werden wissen.
We must know. We will know. · Hilbert’s epitaph, Göttingen
✦Read more✦
Keep wandering
A few more pieces in the same spirit — math, design, and slow attention.