2026-10-01 · chapter Yang–Mills and the mass gapPart 9 Where things stand, and the tasksmisconceptions

Common misconceptions — taking half-right ways of speaking back to the words of the problem statement

Many of the commonly seen ways of speaking are half right, and to that extent the slip is harder to see. This chapter divides each item into the way of speaking, the correct core, the slip and a precise way of saying it. The standard is the one sentence of the problem statement (0-03); lining up its conditions,

We read each word in one line Known.

• $G$: a compact simple group. All of them, from $SU(2)$ and $SU(3)$ to the exceptional group $G_2$ (0-03).
• $\mathcal T_G$: quantum Yang–Mills theory on $\mathbb R^4$. A theory not on a lattice, nor in a finite box, but on the infinitely extended continuous four dimensions (0-03).
• Axioms: the list of conditions a quantum field theory must satisfy (the Wightman axioms or the Osterwalder–Schrader axioms) (0-03).
• Non-trivial: not collapsed to a Gaussian theory without interaction (a free field) (0-03).
• Agrees with asymptotic freedom at short distances: the behaviour at short distances matches the prediction of perturbation theory (asymptotic freedom; (4) of §04) (0-03).
• $\Delta_G\gt0$: the spectrum of the Hamiltonian $H$ has no points in the interval $(0,\Delta_G)$ above the vacuum energy 0 (0-03).

Most slips happen by replacing one word of (1) with a nearby word. Examples: "$\Delta$" → "the mass of a particle" (§05), "existence" → "the numbers appear to converge" (§07), "$\mathbb R^4$" → "the lattice" (§08).

Words used in this chapter

We borrow words from Parts 4 and 5 in advance. Here we give only one-line definitions, and leave the details to the chapters on the right Known.

The way of speaking: physics proceeds by handling the infinities along the way cleverly, but mathematics does not allow that. That is why it is hard.

The correct core: the continuum theory needs renormalisation (4-04) to control quantities that diverge as the lattice spacing $a\to0$. Bałaban's work is aimed mainly at this ultraviolet control (6-07)Known.

The slip: having granted that a construction on $T^4$ can be expected from extensions of known methods, the setters write 'no present ideas point the direction to establish the existence of a mass gap that is uniform in the volume' (problem statement §6)Known. On a finite periodic lattice with the lattice spacing fixed, the theory is a finite-dimensional integral and there are no infinities (6-11).

$dU_\ell$ is the Haar measure of weight 1, and $S_W\ge0$ the Wilson action (5-01). Example: for $SU(3)$ on a lattice of side 32 there are $4\cdot32^4$ links, each with 8 variables (the dimension of $SU(3)$), and the integral has $4\cdot32^4\times8=33{,}554{,}432$ variables Computed. Many, but finite.

What is hard is the theory with the box extended to $\mathbb Z^4$. Lecture material (Chatterjee 2020) wrote 'In 3D and 4D, mass gap has not been established at large β for any theory.' Known. Since then, for finite groups, Adhikari–Cao have shown exponential decay at large $\beta$ (§04)Known. For continuous non-abelian groups, the gap at large $\beta$ in three and four dimensions has still not been shown Known. What is missing is not the value of an infinity but an estimate uniform in the box and in $\beta$. The gap itself is also thought to have the form $e^{-1/(2b_0g^2)}$, invisible to perturbation theory (0-02)Physics.

Overstatement in the opposite direction: "the ultraviolet is already settled" is also wrong. Bałaban's four-dimensional theorem (1989) comes with a hypothesis, and the theorem removing the hypothesis 'has not been published yet'. The limit of gauge-invariant observables is also at the level of taking subsequences by compactness, and the limit need not be unique (preface of Magnen–Rivasseau–Sénéor 1993; 6-07 §05). Footnote 2 of the problem statement does not accept such weak existence as a solution Known.

The way of speaking: lower the dimension and it is done; only four dimensions remain.

The correct core: $\varphi^4$ has been constructed in two and three dimensions (problem statement §6.2), and the two-dimensional abelian Higgs model has been shown up to the axioms. The problem statement calls it 'the only complete example of an interacting gauge theory satisfying the axioms' Known.

The slip: for non-abelian Yang–Mills the story is different. The two-dimensional construction (6-02) relies on exact solvability, the plaquettes becoming independent once the gauge is fixed, and there are no propagating degrees of freedom. In three dimensions, 'a construction of the 3D YM measure ... even on T3, remains open' (Chevyrev 2022), and Bałaban (1985) goes as far as uniform estimates of the partition function on finite lattices (6-06)Known. The core of the difference is the dimension of the coupling constant.

Example: for $d=3$, $g^2\ell$ ($\ell$ a length) is the dimensionless coupling, and it gets smaller at shorter distances. Interaction weakening at short distances is the foothold for the construction. For $d=4$ this foothold survives only as a slow logarithmic change. Precisely, for $d=4$, $g^2$ becomes dimensionless, and the dependence on scale is only the logarithm from the one-loop running ((4) of §04) Known (a table by dimension is in 9-01).

Even in two dimensions, for the continuum limit and mass gap of asymptotically free sigma models there was no claimed proof up to October 2026 (a manuscript with a claim made public on 2026-10-06 is unverified; 9-01 Table 5).

The way of speaking: the photon ($U(1)$) is massless, but gluons are non-abelian and interact with themselves, so there is a gap.

The correct core: in four-dimensional $U(1)$ lattice theory, the area law breaks down at large $\beta$ (Guth 1980; the fully rigorous proof is Fröhlich–Spencer 1982), and the theory is in a massless phase (Fröhlich–Spencer 1982)Known. Both are said to be results mainly for Villain-type actions (actions of a different form from the Wilson action of (2)). The one-loop coefficient is also determined by the group.

The Casimir number $C_A$ of the adjoint representation is $N$ for $SU(N)$ and 0 for $U(1)$, and a theory without matter fields is asymptotically free only when non-abelian (4-06)Known.

The slip: the lattice theorems divide by group and dimension (Figure 1, Table 1). At strong coupling (small $\beta$), exponential decay holds for every group, and the area law also holds for loops in representations that act non-trivially on the centre (Osterwalder–Seiler 1978)Known. For groups with trivial centre ($G_2$, $SO(3)$, finite groups with trivial centre), large loops are said to follow the perimeter law even at strong coupling, through screening by gluons Physics.

The area law at strong coupling is not evidence of confinement, because it holds at small $\beta$ even for four-dimensional $U(1)$, which does not confine (Chatterjee 2018). Example: three-dimensional $U(1)$ is abelian, yet the area law holds for every $\beta$ (Göpfert–Mack 1982; said to be for a Villain-type action)Known. Four-dimensional finite groups (including non-abelian ones) have exponential decay even at large $\beta$ (Adhikari–Cao)Known, but are believed not to confine Physics.

What matters is the following. For continuous non-abelian groups it is believed that there is no phase transition at finite $\beta$ and that the correlation length diverges as $\beta\to\infty$ Physics. In that limit the gap measured in units of the lattice spacing, $1/\xi$, approaches 0, but physics holds that the gap in physical units (for example $m/\sqrt\sigma$) stays positive Physics. This is what needs to be shown (5-06, 5-10). An argument that does not distinguish groups would wrongly conclude that four-dimensional $U(1)$ also has a gap.

KnownTable 1 The content of Figure 1 (theorems from primary sources and prospects that are "believed". The columns are for the side of large $\beta$). At small $\beta$, exponential decay is a theorem in every row, and the area law is a theorem for loops in representations that act non-trivially on the centre (in the row of finite groups, for groups and representations whose centre is non-trivial). Behaviour at intermediate $\beta$ differs from group to group, and four-dimensional $\mathbb Z_N$ (for large $N$) is said to have a massless phase (note in the table of sources).

The way of speaking: the mass gap is the mass of the lightest glueball (a bound state of gluons), about 1.7 GeV.

The correct core: physics holds that there is an isolated one-particle state (the $0^{++}$ glueball) at the lower end of the gap, and measures it on the lattice. For pure $SU(3)$, $r_0m=4.21(11)(4)$ (Morningstar–Peardon 1999) (5-08)Physics.

There are three slips. First: what the problem statement asks for is a gap in the spectrum, and isolated one-particle states are one of the 'natural extensions' (0-03)Known. Second: what the lattice measures is the decay of correlations. Consider the state $O\Omega$ obtained by applying a gauge-invariant operator $O$ to the vacuum $\Omega$. $\rho$ is the weight (spectral measure) of how much of the energy-$E$ component this state has. The connected correlation of $O$ with itself separated by imaginary time $\tau$ can be written with $\rho$ as follows (0-04 §07, 4-11)Known.

$m_{\mathrm{eff}}$ approaches the lower end of the support of $\rho$ as $\tau\to\infty$, but does not tell whether that lower end is an isolated point or the edge of a continuous spectrum. We compare two models ($\Delta=1$).

$(x)_+^{1/2}$ is $\sqrt x$ for $x\gt0$ and 0 otherwise. From $\int_{E_0}^\infty(E-E_0)^{1/2}e^{-E\tau}dE=\Gamma(\tfrac32)e^{-E_0\tau}\tau^{-3/2}$, B has $m_{\mathrm{eff}}=1+\tfrac{3}{2\tau}$. Example: A comes within 1% at $\tau\approx3.2$ ($c=1$) to $4.2$ ($c=5$), while B only comes within it at $\tau=150$ (Table 2). Fitting a single exponential in an intermediate window, B shows a "particle" heavier than $\Delta$ Computed. Reading the flat part as the mass of one particle is an interpretation at the level of physics Physics.

ComputedThe numbers in this figure were computed on this machine (the closed form of (6) computed on the spot).

ComputedTable 2 $m_{\mathrm{eff}}(\tau)$ ($\Delta=1$, $Z=1$). The last row is the first $\tau$ within 1%. The difference between the closed form and numerical integration is below $4\times10^{-5}$ for $\tau\le16$.

Third: the unit GeV does not come out of pure Yang–Mills itself; it borrows the scale $r_0$ of lattice QCD with quarks. The lattice papers also state that pure gauge theory and QCD are different theories (Athenodorou–Teper)Physics. "Gluons acquire a mass" is also a different matter: the mass term $m^2\,\mathrm{tr}\,A_\mu A_\mu$ is not gauge-invariant Known.

The way of speaking: confinement (that a quark cannot be taken out alone) and the gap are the same thing; show one and the other follows.

The correct core: physics sees the string tension $\sigma$ and the gap $m$ as two faces of one scale (7-01). Precisely, physics holds that the ratio $m/\sqrt\sigma$ approaches a finite positive number independent of $\beta$ and the lattice spacing Physics. On the lattice there are also conditional theorems. Chatterjee (2021), for $G$ a closed connected subgroup of $U(n)$, assumed decay in the following strong form Known.

$B$ is a cubic box, $\delta$ the boundary condition of the box, $f,g$ local functions taking values in $[-1,1]$ (functions determined only by the variables near a single link), and $\mathrm{dist}(f,g)$ the distance between them. The constants $K_1,K_2$ do not depend on $B$, $\delta$, $f$ or $g$, and are determined by $G$, $\beta$ and $d$ alone. This order is the point: if the constants were allowed to differ from box to box and from boundary condition to boundary condition, the assumption would say almost nothing.

The theorem (Theorem 2.4 of Chatterjee 2021) is that if (7) holds, centre symmetry is not broken, and the area law holds for rectangular loops in representations that act non-trivially on the centre Known. The author writes that at small $\beta$ (7) can be shown for almost any lattice gauge theory, while also writing that it is not clear whether the decay believed in four dimensions includes this strong form.

The slip: (a) the problem statement does not ask for confinement; it only lists it among the 'natural extensions' Known. (b) "Any compact simple group" includes $G_2$ and others with trivial centre Known, and such groups are believed possibly not to confine at weak coupling (the remark on $SO(3)$ in the same paper)Physics. The problem statement asks for a gap there too. (c) Example: for three-dimensional $U(1)$ the area law has been shown (Göpfert–Mack; said to be for a Villain-type action). On the other hand, the gap at large $\beta$ is said not to have been shown as of the lecture material (2020). For finite groups, exponential decay at large $\beta$ is a theorem (Adhikari–Cao)Known, and they are believed not to confine Physics (Figure 1).

However, the author cautions that the "decay" spoken of for theories such as finite groups refers to the decay of two-point correlations under a particular boundary condition, which differs from the strong form of (7) (uniform under arbitrary boundary conditions) (the remark after Theorem 2.4 of Chatterjee 2021)Known. So the existence of theories that decay but do not confine does not contradict Theorem 2.4.

The way of speaking: large-scale computations give glueball masses precisely, so the existence of the gap is known.

The correct core: example: for $SU(2)$, $m_{0^{++}}/\sqrt\sigma$ is 3.844(61) (Lucini–Teper 2001) and 3.781(23) (Athenodorou–Teper 2021), in good agreement Physics.

The slip: what is being confirmed is consistency under the assumption that it exists (7-13). In a finite box at finite $\beta$ the gap is always positive (7-13)Known, so a positive number coming out does not test the claim. The value in the continuum is

an extrapolation that assumes this form Physics, and the $\sigma$ in the denominator passes through a correction that treats the flux tube as a Nambu–Goto string (7-13).

The way of speaking (lattice): show on the lattice that the correlation length is finite for every $\beta$, and that solves the prize problem.

The correct core: Problem 5.1 of Chatterjee (2018) is a lattice version of the uniform gap: for any infinite-volume limit with a four-dimensional non-abelian compact Lie group $G\subseteq U(N)$ ($N\ge2$), show the following about the length $\xi(\beta)$ over which plaquette correlations decay Known.

The slip: the continuum limit, the axioms on $\mathbb R^4$, non-triviality and agreement at short distances are not in (9), and (9) alone does not solve the prize problem. The converse implication is not obvious either (5-09)Known.

The way of speaking (finite box): if it can be built on a finite box $T^4$ (the continuous four-dimensional torus), all that remains is to enlarge the box.

The slip: the problem statement writes that a construction on a compact space alone 'would represent a major breakthrough', and places the wall of the uniform gap beyond it Known. Example: on a periodic lattice box (the lattice $T^4$) too, reflection positivity gives a transfer matrix, and the gap of the finite box is positive Known. The gap of a finite box is a quantity one expects to be there. What is needed is $T^4\to\mathbb R^4$ and a gap that does not collapse there (6-10).

What this chapter can say

Sources and reproduction

Next chapter to read: 10-01 The line of construction and Assumption H.

← 9-04 The list of open tasksContents10-00 Entrance — the staircase, and the random walk, the ruler and the clock →