2026-10-01 · chapter Yang–Mills and the mass gapPart 6 Constructive field theorythe setters' words

Where four dimensions stops — the ultraviolet is a prospect in finite volume; what is stuck is infinite volume and a uniform gap

§6 of the problem statement (Jaffe–Witten 2000, JW below) divides the state of four dimensions into two sentences. Known

'Some results are known for Yang–Mills theory on a 4-torus T⁴ approximating R⁴, and, while the construction is not complete, there is ample indication that known methods could be extended to construct Yang–Mills theory on T⁴.' / 'Yet, even if this were accomplished, no present ideas point the direction to establish the existence of a mass gap that is uniform in the volume. Nor do present methods suggest how to obtain the existence of the infinite volume limit T⁴ → R⁴.'

The construction on the torus $T^4$ (a box with opposite faces glued together) is not complete, but there are ample signs that it can be made by extending known methods. For a gap uniform in the volume and for $T^4\to\mathbb R^4$, no present idea points the direction. Writing $\mu_L$ for the measure on a box of side $L$ and $\Delta_L$ for the gap of the Hamiltonian, the objects are

($\mu_{a,L}$ is the measure at lattice spacing $a$; formula (6) of 6-01). Example: the setters distinguish the strength of their assessments in words (Table 1). This chapter maps that difference onto the stages.

KnownTable 1 Words of strength in the problem statement (checked in the text of the Clay PDF).

We overlay the six stages of 6-01 with what has been reached in four dimensions and the words of the literature. What has been reached is one of three: Closed (there is a theorem), Provisional (a theorem under a hypothesis, or a sketch without a detailed proof) and Not yet (no theorem). Stages 1–3 are inside the double limit

stage 4 is outside it, and stages 5 and 6 are properties of $\mu$. The first sentence before formula (1) points to the inside, the second to the outside and stage 6.

KnownThe sentences in this figure are the same as in Table 2 (from the texts of the problem statement, the report, the papers and the surveys). The brackets below show that the first sentence of §6 points to stages 1–3 and the second to stages 4 and 6.

KnownTable 2 What has been reached at each stage for four-dimensional Yang–Mills (as a continuum theory).

Within what this chapter has read, the results in four dimensions after 2000 are of two kinds. One is on the side with the lattice spacing fixed (§05). The other is an $SU(2)$ model with a Higgs field (one of the matter fields coupled to gauge fields; 3-07) added, in a special scaling limit taken by linking the lattice spacing, the coupling and the length of the Higgs field (in any dimension $d\ge2$). That limit is Gaussian (a free field), and the author writes 'The question of constructing a non-Gaussian scaling limit remains open' (Chatterjee 2026). Known No literature claiming to close continuum stages 3–6 of pure Yang–Mills was found within what this chapter has read.

Stages 1 and 2 are closed in four dimensions too, on the lattice. JW §6.5 writes 'the lattice approximation reduces the functional integration to a finite-dimensional integral', and says reflection positivity holds as well (Osterwalder–Seiler 1978). Known In the notation of Chatterjee (2018), with $P$ the set of plaquettes (smallest squares) and $U_p$ the product of the group elements once round $p$,

This is because the eigenvalues of $U_p\in SU(N)$ have absolute value 1, and the Haar measure $\mathrm dU_\ell$ has total weight 1. Example: a four-dimensional periodic lattice with 4 points on a side has $6\times4^4=1536$ plaquettes, so for $SU(2)$, $e^{-6144\beta}\le Z\le1$. Computed

In exchange, the lattice breaks rotation symmetry, and a continuum cutoff breaks gauge invariance (6-07 §06). On reflection positivity, JW §6.5 writes that if it is lost in regularisation there are few ways of recovering it (quoted in 0-03 §03). Known A four-dimensional regularisation keeping all four of gauge invariance, reflection positivity, Euclidean invariance and ultraviolet control at once was not found in the literature this chapter read (this chapter's reading).

We read the content in 6-07 (Bałaban 1989 is estimates with a hypothesis, MRS 1993 a sketch). The setters write in §6.5: Known

'One presumably needs to revisit known results at a deep level, simplify the methods, and extend them.'

Review, simplify, extend. What is asked for at stage 3 is not new ideas but this work. The closed form can be written with Wilson loops $W_\gamma$ (the $\frac1N\operatorname{tr}$ of the product of group elements along a closed path $\gamma$; 5-01). For lattice paths $\gamma_a\to\gamma$

(ren is the renormalisation of the next paragraph; the form follows Chatterjee 2018 §5, §6).

Why renormalisation is needed can be seen in Problem 5.2 (the continuum limit) of Chatterjee 2018. On an infinite four-dimensional lattice, writing $\gamma_{R/a,T/a}$ for the rectangular path of physical size $R\times T$ ($R/a$ and $T/a$ steps on the lattice), the problem is to find sequences, as $\beta\to\infty$, such that

($\sigma$ is the string tension of the continuum theory; the original's symbol is $d$). $R+T$ is proportional to the perimeter, and $RT$ is the area. Since $c\to\infty$, as it stands $\log\langle W\rangle\to-\infty$, that is, the expectation collapses to 0. Adding back the term $c(R+T)$ proportional to the perimeter before taking the limit is the renormalisation here. Known But Chatterjee himself writes 'it is not clear to me whether the renormalization term c(R + T) is indeed necessary', so whether this term is really needed is also an open question.

Example: for the abelian $U(1)$ four-dimensional lattice there is a construction of the continuum limit (Driver 1987). Chatterjee 2018 §6 describes Driver as using in four dimensions the method Gross showed for three-dimensional $U(1)$ as convergence in a sense different from the usual one. Known For an abelian group the terms of the lattice action higher than quadratic (self-interactions of the field) do not survive in the continuum limit, and the limit is expected to be the free electromagnetic field: that is this chapter's explanation (Driver's text not read).

For the non-abelian case there are two steps before reaching formula (4). (i) Ultraviolet stability only gives convergence of subsequences (Chatterjee 2018 §6). Convergence of the whole sequence, that is, uniqueness of the limit, is separate work (6-07 §05). (ii) Even if the limit exists, it may be Gaussian (a free field). Not being a free field, that is, non-triviality, is also separate work (6-08 §10). Footnote 2 does not accept weak existence by compactness as a solution, unless the properties of the limit are shown separately (end of §6.5).

The setters' words are a single sentence: 'Nor do present methods suggest how to obtain the existence of the infinite volume limit' (§6). From the inside, MRS 1993 §I says removing the infrared cutoff makes the coupling large, and writes 'there is no easy solvable model of these phenomenons around which to expand'. Known Laying out the sequence of renormalisation group steps (6-05) along the axis of scales,

$a_0$ is the coarsest scale the hypothesis reaches ($g^2(a_0)\le\gamma$; as in 6-07, $\gamma$ is used as an upper bound on $g^2$), and $\ell_*$ is the rough-guide length where extending the running gives $g^2=1$; for $SU(2)$ with $\gamma=0.3$, $\ell_*/a_0\approx10^{10.7}$ (two loops; Table 2 of 6-07 §04). Computed Reading $\ell_*$ as the scale of confinement is a physics estimate. Physics In the organisation of this book, what is missing is the middle: one step that crosses the scales where the coupling grows to $O(1)$ with estimates independent of the volume (6-04 §10).

Example: with the lattice spacing fixed, stage 4 is closed on the strong-coupling side (Osterwalder–Seiler 1978). Shen–Zhu–Zhu 2023 showed, for $SU(N)$ with $|\beta|\lt1/(16(d-1))$ ($1/48\approx0.0208$ when $d=4$), uniqueness of the infinite-volume measure and exponential decay of correlations. This $\beta$ is the value in 't Hooft's convention. In this convention the inverse coupling in front of the action is written $\beta N$, so converting to the $\beta$ of formula (3) multiplies it by $N$, and the condition is $|\beta|\lt N/48$ ($N^2/48$ in this book's Wilson $\beta$; the conversions among the five conventions are in Table 4 of 5-01). On the weak-coupling side even uniqueness is generally unknown for non-abelian groups (Chatterjee 2018 §3), and exponential decay has been shown only for finite groups (Adhikari–Cao 2025, 5-10). $U(1)$ has a non-confining phase at large $\beta$ (Guth 1980, Fröhlich–Spencer 1982), and that there is no phase transition in the non-abelian case is only believed Physics (5-06).

Stage 5 is the Osterwalder–Schrader axioms (4-10). Reflection positivity holds at each lattice spacing, and being an inequality it is inherited by the limit (6-01 §07). Recovering rotation symmetry is separate work, and setting up the Hilbert space is itself part of the problem (JW §6.5). Known

§4 adds conditions. (i) Local fields corresponding to gauge-invariant local polynomials in the curvature such as $\operatorname{Tr}F_{ij}F_{kl}(x)$ (footnote 1: there is no natural one-to-one correspondence, because of renormalisation). (ii) Agreement at short distances with the predictions of asymptotic freedom and perturbative renormalisation (including the stress tensor and the operator product expansion). (iii) Non-triviality (6-08). And

($H\ge0$ is the Hamiltonian, $m$ the mass). Example: since $H\ge0$, $m\lt\infty$ is the same as the spectrum of $H$ containing a positive point, that is, $H\ne0$. If $m=\infty$, there is no spectrum in $(0,\infty)$, and $H=0$. Further, using the form that states uniqueness of the vacuum with respect to translations (exactly one translation-invariant unit vector; the wording of Axiom 3 in nLab's "Wightman axioms"), the spectral condition (energy and momentum lie in the forward cone) makes the momentum 0 as well, every vector becomes translation invariant, and the Hilbert space consists of the vacuum alone. Under axioms of this form, $m\lt\infty$ can be read as a condition excluding the empty theory (this chapter's reading). The summary in §3 of the problem statement takes the form "a Poincaré-invariant vacuum vector unique up to a phase", from which the same conclusion does not follow immediately. The example that fails (iii) is four-dimensional $\phi^4$, whose limit from the lattice is Gaussian (Aizenman–Duminil-Copin 2021).

On a lattice in a finite box the gap exists automatically (6-01 §08), and the question is its independence of the side $L$. JW §6.5 writes 'New ideas are needed to prove the existence of a mass gap that is uniform in the volume of space-time. Such a result presumably would enable the study of the limit as T⁴ → R⁴' (§5 points the same way; Table 1). Known

The setters write that a uniform gap in finite volume (stage 6) could open the road to infinite volume (stage 4) (the 'may play a fundamental role' of §5, the 'presumably would enable' of §6.5). In 6-01 stage 4 was placed first because the gap is defined for the theory on $\mathbb R^4$, not because of the order of proof. The mechanism is a one-line estimate. For the box Hamiltonian $H_L$, vacuum $\Omega_L$ and gap $\Delta_L$, separating the vacuum component from the rest by the spectral decomposition,

(correlations separated by imaginary time $t$), and if $\Delta_L\ge\Delta_*\gt0$ independently of $L$, the rate of decay does not depend on $L$ either. Known Example: on the strip of the two-dimensional Ising model, $E_1-E_0$ closes like $1/L$ at the critical point and closes exponentially at low temperature (2-09 §07, §08). For weakly coupled $\phi^4_2$ and $\phi^4_3$ the expansion in a small parameter gave stages 4 and 6 at once (6-04 §08), but the long distances of four-dimensional Yang–Mills have no such parameter.

JW §5 writes, as a consequence of the gap, 'For any positive constant C < ∆ and for any local quantum field operator O ... such that ⟨Ω, OΩ⟩ = 0, one has (2) |⟨Ω, O(x)O(y)Ω⟩| ≤ exp(−C|x − y|), as long as |x − y| is sufficiently large' ($x,y$ are points of space. Below this inequality is written JW (2), to distinguish it from formula (2) of this chapter). Known The point is $C\lt\Delta$ and a "sufficiently large" distance. The right-hand side needs no constant because $C$ is taken strictly smaller than $\Delta$, so constant factors and power factors alike lose, far enough away, to $e^{-(\Delta-C)r}$. The equal-time correlation of a free scalar field of mass $\Delta$ is

which has the same form as the Euclidean two-point function ($K_1$ is the modified Bessel function of the second kind. The point field is a stand-in for a local operator; smearing changes the form far away only by a constant factor). Normalising to 1 at $r=1/\Delta$, we compare with $e^{-Cr}$, and also set mass 0 (correlation proportional to $1/r^2$) beside it.

ComputedTable 3 The range of $r$ in which the normalised correlation is at most $e^{-Cr}$ ($r$ in units of $1/\Delta$, examined for $r\le5000$).

ComputedThe numbers in this figure were computed on this machine (formula (8). The vertical dotted lines are the distances where JW (2) starts to hold and starts to fail, for mass $\Delta$).

How to read it: (a) for $C\lt\Delta$ it holds beyond $r$ of about 2. (b) For $C=\Delta$ it still holds thanks to the power $r^{-3/2}$, but the problem statement only speaks of $C\lt\Delta$. (c) For $C\gt\Delta$ it always fails far away. (d) For mass 0, every $C\gt0$ fails far away. The slope far away, $-\frac{\mathrm d}{\mathrm dr}\log$, is $1.153\Delta$ at $r=10$ and $1.015\Delta$ at $r=100$, approaching $\Delta$ from above. Computed Lattice measurements sum over space to extract the zero-momentum component, remove the same kind of power factor from correlations in the time direction, and look at sums of exponentials $e^{-mt}$ (5-08). For a free field, the correlation in the time direction obtained by integrating the Euclidean two-point function over space is exactly $e^{-\Delta|t|}/(2\Delta)$. Known To cross from stage 6 to stage 4, JW (2) and formula (7) must hold independently of the size of the box.

What this chapter can say

Sources and reproduction

Next chapter to read: 6-11 Obstacles other than divergence, which lists the reasons stages 4–6 stop as obstacles distinct from divergence. A chronology and a table by dimension are in 9-01.

← 6-09 Stochastic quantisation and regularity structuresContents6-11 Obstacles other than divergence →