2026-10-01 · chapter Yang–Mills and the mass gapPart 0 Entranceproblem statement

Reading the problem statement word by word — each word closes a loophole to an easy answer

On 24 May 2000, at a meeting at the Collège de France in Paris, the Clay Mathematics Institute announced seven prize problems (the Institute's website, and the preface of the book that contains the problem statements). The official problem statement for one of them is Arthur Jaffe and Edward Witten's description "Quantum Yang–Mills Theory" (14 pages) Known. Below, § is a section of the problem statement, and "section 06" is a section of this chapter. The problem is the one sentence in §4.

Yang–Mills Existence and Mass Gap. Prove that for any compact simple gauge group G, a non-trivial quantum Yang–Mills theory exists on $\mathbb R^4$ and has a mass gap $\Delta\gt0$. Existence includes establishing axiomatic properties at least as strong as those cited in [45, 35].

Restated: For any compact simple gauge group $G$, prove that a non-trivial quantum Yang–Mills theory exists on $\mathbb R^4$ and has a mass gap $\Delta\gt0$. Existence includes establishing properties at least as strong as the axiomatic properties cited in reference [45] (the book of Streater–Wightman) and [35] (the two papers of Osterwalder–Schrader).

Writing the skeleton as a logical formula makes the order of the quantifiers visible ($\sigma(H)$ is the spectrum of the Hamiltonian $H$; section 06).

One theory $\mathcal T_G$ is needed for each group, and the gap $\Delta_G$ may differ from group to group. §4 also imposes one more thing: that the supremum of the gap $m_G$ is finite ($m_G\lt\infty$; section 06). The meaning of the words is fixed by the surrounding paragraphs. Quantum fields and the axioms are in §3, local fields and the short-distance condition in the first paragraph of §4, the mass gap in the second paragraph of §4, clustering and the extensions of the problem in §5, and finite-volume estimates and footnote 2 in §6 and §6.5.

Example. For $G=SU(3)$, what is asked for is a theory of only the 8 kinds of gauge field (gluons), with no quarks (8 is the dimension of $SU(3)$; Table 1). QCD with quarks added is not the object (section 09).

The gauge group is the group of symmetries by which the values of the field may be rotated independently at each point (3-01). Compact means bounded and closed as a group of matrices. Simple is meant in the sense of Lie groups: the Lie algebra (1-03) is not commutative and has no non-trivial ideal (a subspace that bracketing with any element does not take outside). $SU(2)$ has the centre $\{\pm1\}$, so as an abstract group it is not simple. In the problem statement, §1 names $G=SU(3)$ as the gauge group of the strong force (it has the centre $\mathbb Z_3$, so it is not simple as an abstract group), and §6.6 also treats $SU(N)$, so we read it in the sense of Lie groups.

A gauge field has as many components as the dimension of the group. Choosing a basis $T^a$ of the Lie algebra, the bracket (commutator) can be expanded in the basis as $[T^a,T^b]=if^{abc}T^c$. These coefficients $f^{abc}$ are the structure constants, the multiplication table of the Lie algebra (1-03). With them,

$C_A$ is 0 for a commutative group and positive for a compact simple group Known. Its value changes with how the size of the basis is taken, so we fix a normalisation. The basis $T^a$ is taken orthonormal for an inner product invariant under the group action, and that inner product is chosen so that the squared length of the longest root (1-04) is 1. For $SU(N)$, the standard normalisation of physics, $\operatorname{tr}T^aT^b=\tfrac12\delta^{ab}$ in the fundamental representation, corresponds to this. In this normalisation $C_A$ equals the dual Coxeter number $h^\vee$, a positive integer determined by each group, and for $SU(N)$ it is $N$ Known. In the normalisation common in mathematics, "squared length 2", it becomes $C_A=2h^\vee$ (checked from the structure constants for $SU(2),SU(3),SU(4),SO(5),SO(7),SO(8)$ Computed). The compact simple Lie algebras are exhausted by the series $A_n,B_n,C_n,D_n$ and the exceptional types $G_2,F_4,E_6,E_7,E_8$ (the Killing–Cartan classification; 1-04) Known, and "any" means all of these.

ComputedTable 1 The dimension and the positive integer $h^\vee$ determined by each group (list of values in 1-04). Every row agrees with the standard values. The procedure of computation is in the source table. $b_0$ is the one-loop coefficient of section 07.

The general form of the one-loop coefficient is $b_0=(11C_A-4n_fT_R)/(12\pi)$ ($n_f$ is the number of kinds of quark, and $T_R$ a positive number determined by the representation of the quarks; PDG 2024) Known. In a theory without quarks, $n_f=0$ gives $b_0=11C_A/12\pi$, and from $C_A\gt0$ it is $b_0\gt0$ for every simple group. At the level of one-loop perturbation theory, the coupling becomes weaker at shorter distances (section 07). The numerical value of $b_0$ depends on the normalisation of the coupling $g$ (the normalisation of $C_A$ above), so comparing its size between groups has no meaning. The point is that it is positive. For $U(1)$ it is $b_0=0$, and that theory is the free field of the massless photon. There is no gap. For a product of two simple groups the action splits into a sum, giving two theories that do not act on each other.

Detour: for non-compact groups

$\mathbb R^4$ is the set of all quadruples of real numbers; it is not a finite box, not a periodic box (the four-dimensional torus $T^4$), and not a lattice. On the approximations, symmetry is lost. The lattice breaks continuous translations and rotations, and $T^4$ keeps translations but breaks rotations.

The sentence in §4 does not state the signature of $\mathbb R^4$. §3 introduces the Wightman axioms as a theory on "$\mathbb R^4$ with Minkowski signature", and §6, in a parenthesis describing the state of constructive field theory, writes "in the Euclidean formulation, replace Euclidean space-time $\mathbb R^4$ by $\mathbb R^2$ or $\mathbb R^3$", from which one can read that the stage of the Euclidean version is $\mathbb R^4$ Known. There are two ways of measuring length.

The two are connected by the theory of Osterwalder and Schrader (§3; formally, the substitution $t=-i\tau$). How this works is treated in 0-04 and 2-04 and in the existing article One sign is different.

Example. §6 says of $T^4$ that "the construction is not complete, but there is good indication that known methods can be extended", and that methods for obtaining the $T^4\to\mathbb R^4$ limit are not suggested by present techniques Known. This gap goes straight into the problem. Why four is special is in 1-12 and in the existing article What happens only in four dimensions.

Symmetry is not the only thing lost in approximation. On reflection positivity §6.5 writes 'few methods exist to recover reflection positivity in case it is lost through regularization—such as with dimensional regularization, …', giving dimensional regularisation and Pauli–Villars as examples Known. The lattice keeps reflection positivity and breaks rotations, and continuum regularisations do the opposite (4-04, 4-10, 6-11).

"Exists" does not mean that some limit exists. §4 says to define a quantum field theory in the sense of §3, with local fields corresponding to gauge-invariant local polynomials in the curvature $F$ and its covariant derivatives Known. The curvature is written $F=dA+A\wedge A$ in §1, and the local polynomial §4 gives as an example is $\operatorname{Tr}F_{ij}F_{kl}(x)$.

The gauge field $A$ itself is not asked for. $A$ changes under gauge transformations and is not an observable. A "quantum field" is defined in §3 as "an operator-valued distribution on space-time". The value at a point $\mathcal O(x)$ has no meaning; $\mathcal O(f)=\int\mathcal O(x)f(x)\,d^4x$, smeared with a test function $f$, is an operator (1-11).

Table 2 The properties included in "existence" (a summary of §3 of the problem statement; the exact form is in references [45] and [35]; in this book, 4-09 and 4-10).

Example. The free Klein–Gordon field was the field that served as a model for the first axiom system (§6.2), but it is excluded by "non-trivial", next. §6.5 states plainly that "establishing the Hilbert space of quantum mechanics is part of the solution of this Millennium problem" Known.

The problem statement does not define "non-trivial". It is usually read as "not a free field (a field without interaction)". §6.2 says that the Gaussian measure with mean 0 and covariance $(-\nabla^2+m_0^2)^{-1}$ ($\nabla^2$ the Laplacian) gives the free field of mass $m_0$, and that non-linear fields need non-Gaussian measures Known. Whether something is Gaussian can be told from higher correlations. For a Gaussian random variable $X$ with mean 0,

and in several variables too, four-point expectations can be written as sums of products of two-point expectations (Isserlis' formula; 1-08) Known.

Example. ComputedFor the density $\propto e^{-x^4}$ the ratio (5) is 2.18844 (agreeing with $\Gamma(\tfrac14)\Gamma(\tfrac54)/\Gamma(\tfrac34)^2$). For the density $\propto e^{-x^2/2-\lambda x^4}$, with $\lambda=1,\ 0.1,\ 0.01$ it is 2.319, 2.537, 2.843, returning to 3 as $\lambda\to0$. Triviality of a field theory is the phenomenon in which each approximation is non-Gaussian, yet in the limit ratios of this kind return to the Gaussian value.

To be exact, the one-variable ratio is a guide. For four-dimensional $\lambda\varphi^4$, it is a theorem that if the ultraviolet is cut off by a lattice and the scaling limit near the critical point is taken with lattice spacing 0 and infinite volume, the limit is Gaussian (Aizenman–Duminil-Copin 2021; 1-08) Known. Yang–Mills differs in that its coupling becomes weak at short distances, and a non-trivial limit is expected, but there is no proof Physics. A free field with mass has a gap (Table 3), so the gap alone cannot shut out free fields (6-08). §6 says that at present not a single non-trivial relativistic field theory satisfying the Wightman (or other reasonable) axioms is known in four dimensions Known.

The definition in §4 is written in terms of the spectrum $\sigma(H)$ of the Hamiltonian $H$ (what corresponds to the set of eigenvalues for a matrix; 1-10). The vacuum is an eigenvector with energy 0, and the energy is never negative. On top of that, having an empty interval just above 0 is called a gap.

$\sup\Delta$ is the supremum of all $\Delta$ satisfying the condition, and we call it the mass $m$ Known. The problem statement does not give the reason for imposing $m\lt\infty$. Reading it from the formula shows what it excludes. If $\sigma(H)=\{0\}$, every $\Delta$ satisfies the condition, and $m=\infty$. Self-adjoint (1-10) is the property of Hermitian matrices (matrices whose eigenvalues are all real) extended to infinite-dimensional operators. For self-adjoint $H$, $\sigma(H)=\{0\}$ is the same as $H=0$, so $m\lt\infty$ means "there is some state of positive energy above the vacuum".

The reason for calling it "mass" lies in the relation between energy and momentum ($\hbar=c=1$; 2-04). A particle of mass $\mu$ has $E=\sqrt{|\vec p|^2+\mu^2}\ge\mu$. For the mass operator $M=\sqrt{H^2-\vec P^2}$ of §6.2, $H\ge M$, and at momentum 0, $H=M$. A state with $M\gt0$ can have its momentum brought to 0 by a Lorentz transformation (using the Lorentz covariance of the axioms), so the gap of $H$ and the gap of $M$ have the same width Known. If there is a state with $M=0$ besides the vacuum, both gaps are 0, and the conclusion is the same. This equivalence is not written in the problem statement itself.

ComputedThe figure is drawn on the spot from the formula $E=\sqrt{p^2+m^2}$ (momentum in one direction only; the shading is continuous spectrum).

Table 3 The spectra of four examples.

Rows 1 and 2 have the same $\sigma(H)$. What the problem statement asks for is only the gap of $\sigma(H)$; an isolated one-particle state is an extension of the problem (section 09). §5 gives clustering as a consequence of the gap. For local fields $\mathcal O$ and $0\lt C\lt\Delta$ with $\langle\Omega,\mathcal O\Omega\rangle=0$, if the two points are far enough apart,

Known. The gap determines how fast fluctuations at distant places approach independence. The body of the definition is in 4-11, and how to read it for finite systems is in 2-05.

Quantum Yang–Mills theory has no definition before it is constructed. What ties the constructed object to "the quantum Yang–Mills theory of $G$" is the condition in the first paragraph of §4. The correlation functions of the quantum fields must agree at short distances with the predictions of asymptotic freedom and perturbative renormalisation theory found in textbooks (original: 'should agree at short distances with the predictions of asymptotic freedom and perturbative renormalization theory, as described in textbooks'). The predictions include the existence of a stress tensor and an operator product expansion, whose local singularities have the form predicted by asymptotic freedom Known.

Asymptotic freedom is the property that the strength of the coupling changes with scale and becomes weak at short distances (large momentum $\mu$) (Gross–Wilczek and Politzer, 1973; references [21] and [38] of the problem statement). In a theory without quarks, $\alpha=g^2/4\pi$ runs in the following way (formula (9.3) of PDG 2024 with $n_f=0$; on the right, the one-loop solution) Known.

$\Lambda$ is a constant of integration, the scale at which the coupling defined in perturbation theory diverges.

Example. ComputedFor $SU(3)$, $b_0=0.8754$, and when $\mu/\Lambda=10,\ 100,\ 1000$, $\alpha=0.248,\ 0.124,\ 0.083$. Each time the momentum grows tenfold, $1/\alpha$ grows by the same amount $2b_0\ln10\approx4.03$. The weakening is logarithmic, and slow.

Solving (8) gives $\Lambda=\mu\,e^{-1/(2b_0\alpha)}$, and $e^{-1/x}$ approaches 0 faster than any $x^k$ as $x\to0^+$, so all coefficients of the power series in $\alpha$ are 0 Known. The classical action has no constant with the dimension of mass, and physics holds that the mass of the quantum theory takes the form $\Delta=c\Lambda$ Physics. This is consistent with "the mass gap is not visible classically" in §6.6. The problem statement does not specify the value of the coupling constant $g$ (4-06, 4-07).

Footnote 1 (§4) says that, because of the subtleties of renormalisation, there is no natural one-to-one correspondence between classical "differential polynomials" and quantum operators, and that the spaces of those of dimension at most $d$ are expected to correspond Known. The dimension is the mass dimension: $A$ has 1, $F$ has 2, and $\operatorname{Tr}F^2$ has 4.

Footnote 2 (end of §6.5) says that weak existence (compactness) is specifically excluded as a solution of the existence part, unless the properties of the limit (the mass gap and the axioms) are established by other methods Known. The text to which the footnote is attached says, in effect, "showing a mass gap uniform in the volume needs new ideas; with that, one could study the $T^4\to\mathbb R^4$ limit". Uniform means that the right-hand side holds for the gap $\Delta_L$ for each box size $L$.

The left-hand side alone holds even for the massless free field. With the field set to 0 on the walls of a box of side $L$, the energy of the lowest vibration is $\sqrt3\,\pi/L$, and the gap closes as $L\to\infty$ (2-05).

A compactness argument is an argument of the type "a bounded sequence has a convergent subsequence" (Bolzano–Weierstrass). For a sequence of probability measures too, given a uniform estimate (tightness), a weakly convergent subsequence can be taken (Prokhorov's theorem) Known. Tight means that every measure in the sequence puts almost all of its probability (at least $1-\varepsilon$) on a common compact set (for $\mathbb R^n$, a bounded closed set), and weak convergence means that the expectations of all bounded continuous functions converge. What is weak is that the limit can change with the choice of subsequence. Whether the limit is unique, and which of the gap or the axioms it satisfies, cannot be known from compactness alone (some properties, like reflection positivity, are preserved under weak convergence; 5-03). Figure 2 uses the sequence $a_n=\cos n$ to show how the limit is not unique.

ComputedThe numbers in this figure were computed on this machine. Rule: $n_1$ is the smallest $n$ with $|\cos n-t|\lt1/2$, and $n_{k+1}$ is the smallest $n$ with $n\gt n_k$ and $|\cos n-t|\lt1/(k+2)$. The figure computes the same rule on the spot, and agrees with the python3 results (Table 4).

ComputedTable 4 Subsequences extracted with $n\le400$.

The remainders of $n$ divided by $2\pi$ are uniformly distributed (Weyl's equidistribution theorem), so a subsequence converging to any $t\in[-1,1]$ can be taken Known. The difference between the empirical distribution of $\cos n$ for $n\le100000$ and the arcsine distribution was below 0.0001 in all 10 intervals Computed. Boundedness alone does not make the limit unique.

To be exact, in lattice methods there is a measure $\mu_a$ for each lattice spacing $a$, and one wants to take $a\to0$. Even if a uniform estimate gives a subsequential limit, whether it is unique and whether it satisfies the axioms and the gap are separate problems, and footnote 2 asks for exactly that to be shown. It has been stated that what can be expected from the known results on the ultraviolet side in four dimensions is also a limit by a compactness argument using subsequences, and that the limit is not necessarily unique (the introduction of Magnen–Rivasseau–Sénéor 1993; 6-07) Known.

§5 lists the following as natural extensions, separate from the body of the problem Known. The existence of isolated one-particle states (an "upper gap" in addition to the mass gap). A proof of confinement. The existence of other four-dimensional gauge theories with added fields that preserve asymptotic freedom, and their dynamics (gap, confinement, chiral symmetry breaking). Extending the existence theorem to arbitrary four-dimensional manifolds. Written as a formula, the upper gap means that the spectrum of $M$ has the following form (this chapter's way of writing; it corresponds to "$m$ is an isolated eigenvalue of $M$" in §6.2).

In a theory with only one kind of particle of mass $m$ and no bound states, the two-particle continuous spectrum begins at $m'=2m$ (row 1 of Table 3).

§1 names three properties that QCD should have at the quantum level: the mass gap, quark confinement and chiral symmetry breaking. Only the first entered the body of the problem, and the object, too, is pure Yang–Mills without quarks. The problem statement does not ask for the value of the gap either. Glueball masses and their ratio to the string tension are objects of physical measurement (5-08).

Example. Problem 5.1 of Chatterjee (2018) asks that on the lattice ($\beta$ is the inverse of the lattice coupling; the larger it is, the weaker the coupling; 5-01) the correlations of plaquette variables decay exponentially (the correlation length $\xi(\beta)$ is finite), and that $\xi(\beta)\to\infty$ as $\beta\to\infty$. It is a formulation on the lattice, and a different thing from Clay's requirements, which include the continuum limit, $\mathbb R^4$, the axioms and non-triviality. Even if it were shown, that alone would not solve the Clay problem (5-09, 9-05).

We take the conditions of (1) word by word, together with the $m_G\lt\infty$ mentioned in section 01, and list what each word directly shuts out.

Table 5 A word-by-word list (this chapter's reading).

The words of Table 5 fall into two kinds according to what happens when they are removed. One kind makes the claim false when removed. "Simple" is one: as in the example below, a theory without a gap gets mixed in. The other kind makes the claim easy when removed. Removing "$\mathbb R^4$" leaves only finite-dimensional integrals on a finite lattice; removing "existence (axioms)" lets through something for which merely a limit exists; removing "footnote 2" makes a subsequential limit enough. Removing "$m\lt\infty$" lets a theory with only the vacuum satisfy the gap condition in form. In this chapter's reading, most of the words of the problem statement are of the latter kind, and close the loopholes to easy answers one by one.

The shutting-out overlaps. The free field with mass is caught by both "non-trivial" and "short distances". Even when one word is removed, another word may shut out the same thing. Table 5 is a guide to the role of each word.

Example. Removing "simple" means showing a gap for $G=U(1)$ as well, but that theory is the free electromagnetic field, which, as row 3 of Table 3 shows, has no gap. The claim with the word removed is false by this one example.

What this chapter can say

Sources and reproduction

Next chapter: 0-04 Four dimensions in mathematics and four dimensions in physics. For digging deeper into each word, go to the chapters named at the end of each section.

← 0-02 One pictureContents0-04 Four dimensions in mathematics and four dimensions in physics →