Clay Mathematics Institute Millennium Prize Problem (announced 2000, US $1,000,000) · official statement by Arthur Jaffe & Edward Witten · field: constructive quantum field theory / mathematical physics
How to read this page. This problem sits at the seam between mathematics and physics, so the notation is a mix of both. We write G for a compact simple Lie group (think SU(2), SU(3)): the "gauge group." A G-connection A is the mathematical object physicists call a gauge field; its field strength (curvature) is F = dA + A∧A. The coupling constant is g; the energy/momentum scale is μ. ⟨·⟩ denotes a vacuum expectation value (a correlation function), H the Hamiltonian (energy operator), |Ω⟩ the vacuum state. "Euclidean" means we have Wick-rotated time to imaginary values, turning oscillatory path integrals into (hopefully) convergent probability-like measures; "Minkowski" means physical, real-time spacetime. The phrase mass gap means the lightest excitation above the vacuum has strictly positive energy. Everything below is honest about the boundary between what is proven and what is conjectured.
0 · The claim, in one line
Show that a mathematically rigorous quantum Yang–Mills theory with gauge group G exists on four-dimensional spacetime ℝ4, and prove that there is a number Δ > 0 (the mass gap) below which the theory has no non-vacuum states. In the physics slogan: the classical theory has massless fields, yet the quantum theory must produce only massive particles, with a gap you can compute is positive.
1 · The object: what it is and why anyone cares
In 1954 Chen Ning Yang and Robert Mills asked a structural question: electromagnetism is a gauge theory whose symmetry group is the abelian circle U(1): what happens if you demand the same local symmetry principle but with a non-abelian group, where the symmetry operations do not commute? The answer is Yang–Mills theory, defined by the action
S[A] = −¼ ∫ℝ4 tr(Fμν Fμν) d4x, Fμν = ∂μAν − ∂νAμ + g[Aμ, Aν]The Yang–Mills action: the squared curvature of a connection, integrated over spacetime. The commutator term [A,A] is the entire difficulty: it makes the theory nonlinear and self-interacting. Set it to zero and you recover Maxwell's equations.
Why care? Because this is not a toy. The Standard Model of particle physics (the single most precisely tested theory humans possess) is built on the Yang–Mills construction with gauge group SU(3) × SU(2) × U(1). The SU(3) factor is quantum chromodynamics (QCD), the theory of quarks and the strong nuclear force. The predictions of that framework agree with experiment to many significant figures. And yet nobody has proven the theory it is built on exists as a mathematical object.
Chase the "why" to its root and you hit two intertwined physical facts that the mathematics must explain:
Confinement. The force carriers of QCD (gluons) and the quarks they bind are never seen in isolation. You cannot pull a single quark out of a proton; the energy required grows without bound. Only colorless bound states (protons, neutrons, pions) exist as free particles.
Mass generation from nothing. The classical Yang–Mills field, like the photon, is massless: there is no mass term in the action, and adding one by hand would break gauge invariance. Experimentally the strong force is short-ranged, which demands a massive lightest excitation. The mass must therefore be generated by the quantization itself, non-perturbatively, with no mass present in the starting point. That gap Δ is, physically, why the strong force does not reach across the room the way electromagnetism does, and why the lightest glueball/hadron is heavy rather than massless.
So the "why" bottoms out here: the mass gap is the mathematical fingerprint of the two most basic qualitative facts about the strong force. Proving it exists means proving that the theory underlying all of nuclear physics is logically consistent: not merely a recipe that happens to produce good numbers.
2 · The precise statement / the mathematical setup
The official Clay problem, as formulated by Arthur Jaffe and Edward Witten, asks for two things at once. First, existence: construct, for any compact simple gauge group G, a quantum Yang–Mills theory on ℝ4 that satisfies a standard axiom system for relativistic quantum field theory: the Wightman axioms, or equivalently (via reconstruction) a Euclidean field theory satisfying the Osterwalder–Schrader axioms, or a Haag–Kastler net of observable algebras. Second, mass gap: prove the resulting theory has Δ > 0.
What "existence" means precisely
One must produce a Hilbert space ℋ carrying a unitary representation of the Poincaré group, a unique Poincaré-invariant vacuum |Ω⟩, and quantum fields (operator-valued distributions) whose vacuum correlation functions (the Wightman functions) satisfy:
Relativistic covariance under the Poincaré group;
Spectrum condition: the energy–momentum operators have joint spectrum in the forward light cone (energy is non-negative in every frame);
Locality / microcausality: fields at spacelike separation commute;
Existence and uniqueness of the vacuum.
In the Euclidean approach one instead constructs the Schwinger (correlation) functions as moments of a probability measure on the space of (gauge-equivalence classes of) connections, and verifies the Osterwalder–Schrader axioms: in particular reflection positivity, the subtle condition that lets you Wick-rotate back to a genuine Minkowski Hilbert-space theory with a positive-definite inner product. Osterwalder and Schrader proved (in the mid-1970s) that reflection positivity plus regularity and symmetry is exactly what is needed to reconstruct a Wightman theory from Euclidean data.
What "mass gap" means precisely
spec(H) ⊆ {0} ∪ [Δ, ∞), H|Ω⟩ = 0, Δ > 0The Hamiltonian annihilates the vacuum (energy zero), and every other state has energy at least Δ. There is a clean "gap" between the vacuum and the rest of the spectrum.
Equivalently, in Euclidean language, all connected correlation functions decay exponentially with a rate bounded below:
|⟨O(x) O(0)⟩c| ≤ C · e−Δ|x| as |x| → ∞Correlations die off exponentially fast with a definite rate Δ. A massless theory would decay only polynomially (power-law); the exponential rate IS the mass.
The one structural gift: asymptotic freedom
The one thing that makes 4-D Yang–Mills more tractable than a generic interacting field theory is the behavior of its coupling with scale. In 1973 David Gross and Frank Wilczek, and independently H. David Politzer, computed the one-loop beta function and found it negative (Nobel Prize 2004):
μ (dg/dμ) = β(g) = −b0 g3 + O(g5), b0 = 11 C2(G) / (48π2) > 0As you probe higher energies μ, the coupling g shrinks toward zero. For SU(N), C₂(G) = N. The positive sign of b₀ is asymptotic freedom: unique to non-abelian gauge theory in 4-D.
This means that at short distances the theory becomes weakly coupled, so the ultraviolet (high-energy) behavior is, in principle, controllable by perturbative renormalization. The catastrophe is at the other end.
Schematic of the running coupling. At high energy (right) the coupling shrinks: asymptotic freedom makes the ultraviolet well-behaved. At low energy (left), near the dynamically generated scale Λ, the coupling grows large and perturbation theory collapses. The mass gap lives in that strongly-coupled infrared region, precisely where our calculational tools fail.
3 · Every connection: what it is equivalent to and what hangs on it
Confinement. A rigorous mass gap for pure Yang–Mills is widely regarded as the mathematical core of quark confinement. In lattice language, confinement is the statement that large Wilson loops obey an area law rather than a perimeter law; a positive string tension and a positive mass gap are the two faces of the same non-perturbative physics.
Constructive QFT as a whole. No non-trivial, interacting, Poincaré-invariant quantum field theory has ever been constructed in four dimensions satisfying the Wightman axioms. Yang–Mills is the most physically important candidate. Solving it would be the first genuine 4-D construction and would validate the entire Osterwalder–Schrader / Glimm–Jaffe program at the dimension that matters.
The Standard Model's foundations. Everything downstream (QCD, the quark model, nuclear binding) currently rests on a theory assumed to exist. The gap problem is the load-bearing assumption.
Renormalization made rigorous. A clean construction would turn the physicist's renormalization procedure (removing infinities order by order) into a controlled mathematical limit, closing a conceptual gap open since the 1940s.
Lower-dimensional analogues as stepping stones. The problem is solved in 2 and 3 dimensions in meaningful senses (see §4), which is both encouraging and a sharp reminder that dimension four is special: the coupling is dimensionless there, the UV is marginal, and the lower-dimensional methods do not simply port over.
4 · Every serious attack: and exactly how far each got
Lattice gauge theory (Wilson, 1974)
Kenneth Wilson replaced continuous spacetime with a hypercubic lattice of spacing a, putting group elements on the links. On the lattice the theory is mathematically well-defined (a genuine, finite, probability measure) with no ultraviolet infinities. In the strong-coupling regime one can prove, via convergent cluster/strong-coupling expansions, both confinement (area law) and a mass gap. This is a real theorem, but it is a theorem about the lattice at fixed spacing and large coupling. The Millennium problem is the continuum limit: let a → 0 while tuning g (per asymptotic freedom) so that physical quantities stay finite, and prove the resulting continuum theory exists, satisfies OS/Wightman, and keeps its gap. That limit is unproven.
Rigorous renormalization-group control (Bałaban, 1980s)
Tadeusz Bałaban, in a long and formidable series of papers through the 1980s, established ultraviolet stability bounds for lattice Yang–Mills via a block-spin renormalization-group analysis: showing the free energy is controlled uniformly as the cutoff is removed, in a weak-coupling small-field regime. This is the deepest rigorous progress toward the continuum limit and is universally respected. But it is a one-sided bound: it does not construct the full measure, does not control the large-field/strongly-coupled regions that drive the infrared physics, and does not deliver the mass gap. Related constructive efforts by Jacques Magnen, Vincent Rivasseau and Roland Sénéor (early 1990s) pushed the program further on a finite volume with cutoffs but likewise stop short of a complete construction.
Two and three dimensions (largely done)
2-D Yang–Mills. In two dimensions the theory is almost topological and essentially solvable. The continuum 2-D Yang–Mills measure was constructed and studied rigorously by Bruce Driver, Leonard Gross, Ambar Sengupta, Thierry Lévy and others (1990s onward), building on earlier work of Migdal and Witten's exact partition-function formulas. Wilson-loop expectations are computed exactly.
3-D progress. In three dimensions the Yang–Mills measure is super-renormalizable and far better understood; recent work (2010s–2020s) by Sourav Chatterjee, Ajay Chandra, Ilya Chevyrev, Martin Hairer, Hao Shen and collaborators (notably using stochastic quantization and regularity structures to build the Langevin dynamics of the 2-D and 3-D theories and to give gauge-invariant state spaces) represents serious rigorous advances. Chatterjee has also obtained results on lattice gauge theories, confinement at strong coupling, and the large-N limit.
The honest takeaway: the lower-dimensional successes are real mathematics, but they lean on the coupling being dimensionful (super-renormalizability), which makes the UV easy. In 4-D the coupling is dimensionless and the problem is genuinely marginal: the easy structural crutch is gone.
Supersymmetric and large-N windows
In 1994 Nathan Seiberg and Edward Witten exactly solved the low-energy effective theory of N=2 supersymmetric Yang–Mills, exhibiting mass generation (monopole condensation) and confinement in that extended theory. This is a landmark: but it is physics-level rigor about a different, supersymmetric theory, not a construction of ordinary Yang–Mills satisfying the Clay axioms.
5 · Why it breaks down: the honest crux
Wall 1: The infrared is strongly coupled, and perturbation theory is blind to the gap. Asymptotic freedom (§2) is a gift at high energy but a curse at low energy: as μ → Λ the coupling g grows without bound and the perturbative expansion diverges. The mass gap is a non-perturbative effect: it is invisible to any finite order of the expansion in g, because a term like e−c/g2 has every Taylor coefficient equal to zero at g = 0. So the single most powerful tool in quantum field theory (perturbative expansion around the free theory) provably cannot, even in principle, detect the quantity we are asked to prove positive. You need a method that sees the whole measure at once, and no such method is under full rigorous control in 4-D.
Wall 2: The continuum limit of the lattice is uncontrolled where it matters. The lattice gives a bona fide theory with a gap at strong coupling and fixed spacing. To reach the continuum you must send spacing a → 0 and coupling to zero together along the asymptotic-freedom trajectory, and prove: (i) the correlation functions converge, (ii) the limit satisfies reflection positivity so a Hilbert space reconstructs, and (iii) the mass gap, measured in physical units, stays bounded below and does not collapse to zero as the lattice is refined. Bałaban's renormalization-group bounds control one direction (UV stability, small fields) but not the coupled large-field, strong-coupling regime that generates the gap. Nobody has shown the gap survives the limit. The logic fails at the join: the regime where we can prove a gap (strong coupling) and the regime we can renormalize (weak coupling) are opposite ends of the same axis, and the bridge between them is exactly the unproven part.
Wall 3: There is no rigorous non-trivial measure on the space of connections modulo gauge. The formal path integral integrates e−S[A] over all connections A, divided by the (infinite-dimensional, non-compact) group of gauge transformations. Gauge fixing introduces the Gribov ambiguity: Vladimir Gribov (1978) showed that global gauge-fixing conditions like Lorenz/Coulomb gauge do not select a unique representative; the gauge slice intersects each orbit multiple times. So even writing down the object to be constructed (a genuine probability measure on the quotient space with the right support, invariant under the right symmetries, with finite non-trivial moments) is open in 4-D. Every candidate requires an ultraviolet cutoff to be well-defined, and removing that cutoff without the measure degenerating to a Gaussian (free, trivial) or to zero is precisely what remains unproven. This is the constructive-QFT wall: in 4-D, no interacting measure has ever been built.
Wall 4 (speculative / fringe): The holographic shortcut, and why it does not close the problem.Speculative. Via the AdS/CFT correspondence, Edward Witten gave (late 1990s) a heuristic argument that certain confining gauge theories have a mass gap, read off from the spectrum of a dual higher-dimensional gravitational theory. This is genuinely illuminating physics and strongly suggests the answer is "yes, there is a gap." But it is not a proof of the Clay statement for three independent reasons: (a) the gauge/gravity duality is itself a conjecture without a rigorous mathematical formulation; (b) the cleanest holographic examples are supersymmetric or deformed theories, not pure 4-D Yang–Mills with a compact simple group; and (c) deriving a gap from a dual gravity description presupposes both theories exist: it relocates the existence problem rather than solving it. A related fringe current is the steady trickle of claimed "elementary" proofs of the mass gap posted to preprint servers (various authors over the past two decades). None has survived expert scrutiny; the recurring failure mode is a hidden assumption that the continuum theory already exists, or an infrared manipulation that silently assumes the gap it purports to derive. These are honestly classed as questionable and are not accepted by the community.
6 · Status: confirmed
The problem is open. As of this writing there is no accepted solution; the Clay Millennium Prize for Yang–Mills Existence and Mass Gap remains unclaimed. Partial results are substantial and real (a complete rigorous theory in 2-D, strong progress in 3-D, Wilson's lattice formulation, strong-coupling confinement and gap theorems on the lattice, and Bałaban's ultraviolet renormalization-group bounds) but the four-dimensional construction has not been achieved.
A genuine proof would have to do all of the following, with no gaps:
Construct, for an arbitrary compact simple gauge group G, a probability measure (or equivalently a set of Schwinger functions) on ℝ4 as a controlled continuum limit with the ultraviolet cutoff fully removed: not merely bounded, but convergent.
Verify the Osterwalder–Schrader axioms, reflection positivity in particular, so that a Minkowski Wightman theory with a positive-definite Hilbert space reconstructs.
Prove the constructed theory is non-trivial (not secretly a free field).
Prove the Hamiltonian's spectrum has a strictly positive gap Δ > 0 above the unique vacuum (uniformly, surviving the continuum limit) and ideally give a lower bound.
Until an argument does all four and withstands expert refereeing, the honest label is: OPEN.
Credit where it is due. The theory is due to Chen Ning Yang and Robert Mills (1954). Asymptotic freedom (the one structural handle on the problem) was discovered by David Gross, Frank Wilczek, and H. David Politzer (1973; Nobel Prize in Physics 2004). The lattice formulation is Kenneth Wilson's (1974). The axiomatic frameworks are owed to Arthur Wightman and Lars Gårding, to Konrad Osterwalder and Robert Schrader, and to Rudolf Haag and Daniel Kastler; the constructive program to James Glimm and Arthur Jaffe. The deepest rigorous renormalization-group progress is Tadeusz Bałaban's, with constructive work by Jacques Magnen, Vincent Rivasseau, and Roland Sénéor. Two- and three-dimensional constructions are owed to Bruce Driver, Leonard Gross, Ambar Sengupta, Thierry Lévy, and (in recent stochastic-quantization advances) Sourav Chatterjee, Ajay Chandra, Ilya Chevyrev, Martin Hairer, and Hao Shen. The Seiberg–Witten exact solution of N=2 supersymmetric gauge theory (1994) and the holographic mass-gap arguments are due to Nathan Seiberg and Edward Witten. The official Millennium Problem statement was written by Arthur Jaffe and Edward Witten. The unsolved problem is claimed by no one here: if it is ever solved, that proof is offered freely to the world.